Temporal bounds of wildfires

Abstract

To provide methods, systems and apparatus, including computer programs encoded on computer storage media, for generating a temporal range of a fire.SOLUTION: In some implementations, a server obtains a date when a fire occurred within a region. The server obtains satellite imagery of the region from before the date when the fire occurred. The server generates a first statistical distribution from the satellite imagery. The server determines a start date of the fire using the first statistical distribution. The server obtains second satellite imagery of the region from before and after the start date. The server selects a second set of imagery from the second satellite imagery from before the start date. The server generates a second statistical distribution from the second set of imagery. The server determines an end date of the fire using the second statistical distribution. The server provides the start date and the end date for output.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] This specification relates generally to computer simulations, and one particular implementation relates to generating wildfire time extents using statistical distributions and satellite imagery. [Background technology]

[0002] Wildfires can include forest fires, brush fires, rural fires, etc., and can occur in various geographic regions. Wildfires can originate in one geographic area and spread to another geographic area over a period of time. Wildfires can be caused by factors such as climate, vegetation, and even human activity, to name a few. Summary of the Invention

[0003] The techniques described herein generate a time range of historical wildfires across a geographic region. In particular, a server can generate a time range that includes a start date and an end date of a wildfire given its spatial boundaries in the geographic region, and the dates the wildfire occurred in the geographic region. The start date corresponds to the date the wildfire began. The end date corresponds to the date the wildfire ended.

[0004] The technique for generating a time range of a wildfire is important because this range may be utilized and needed by other applications. For example, a server may provide the time range as training data for a machine learning or another machine learning model pipeline for predicting the duration of a fire. In another example, the server may provide the generated time range to a graphical user interface that can display the spread and duration of a wildfire. In another example, the server may also provide the generated time range to a process that seeks to identify the start date of a currently burning wildfire. In another example, an individual may want to analyze other sensory data before and after a wildfire occurred. Automatically detecting the start and end dates of a wildfire can help complete these investigations. For example, an analyst may need the temporal boundaries of a wildfire while attempting to estimate the annual number of tree deaths in a particular area due to wildfire.

[0005] In some implementations, the server can receive the dates that wildfires occurred within the geographic region. This date can be a start date, an end date, or any date between the start date and the end date. The geographic region can be defined by location data that can include, for example, longitude and latitude coordinates and a region or area surrounding these location coordinates.

[0006] The server can communicate with a satellite image database to obtain satellite imagery of various geographic regions. The satellite image database can store satellite imagery, such as infrared and other satellite data, of various geographic regions of the Earth over various time periods. The satellite imagery can also include other processed satellite imagery products, such as MODIS MCD45A1, which includes pixel-by-pixel combustion and quality assurance information. The satellite image database can index satellite imagery by several factors, such as location data, time, date, and geographic area boundary.

[0007] The server can identify and retrieve satellite imagery using location data associated with the geographic region and an input date that the fire occurred. In particular, the server can provide the location data and input date to a satellite imagery database and receive satellite imagery from a time period before the provided input date. In some implementations, the server can provide the location data and input date to a satellite imagery database and receive satellite imagery from a time period before and after the provided input date.

[0008] In some implementations, the server can generate a statistical distribution based on acquired satellite imagery of a particular geographic area from a period prior to an input date. As described further below, the server can generate a baseline of satellite imagery corresponding to a period prior to an input date, with each image in the baseline showing a geographic area over a defined period. Once the server generates the statistical distribution, the server can identify a start date for the wildfire.

[0009] In some implementations, the server can determine the start date of the wildfire by comparing satellite imagery to a statistical distribution. The server can obtain satellite imagery corresponding to a time period for the geographic region prior to an input date, for example, a time period for the geographic region three months prior to the input date. The three months prior to the input date are selected to ensure that the statistical distribution is free of and does not include fire pixels. Additionally, the server can filter the satellite imagery within a minimum known perimeter of the wildfire, as described further below. The server can compare the satellite imagery to the statistical distribution to identify the start date of the wildfire.

[0010] In some implementations, the server can determine the end date of the wildfire by generating an additional statistical distribution. The server can generate the second statistical distribution based on acquired satellite imagery of a particular geographic area from a period prior to the determined start date. For example, the server filters the acquired satellite imagery within the most recent known perimeter of the wildfire in the geographic area. As described further below, the server can improve the accuracy of identifying the end date by generating the additional statistical distribution using the determined start date rather than the provided input date. In generating the second statistical distribution, the server can generate another baseline of satellite imagery corresponding to a period prior to the determined start date, with each image in the baseline showing a geographic area spanning a defined period. Once the server generates the second statistical distribution, the server can identify the end date of the wildfire.

[0011] In some implementations, the server can determine an end date of the wildfire by comparing the satellite imagery to a second statistical distribution. The server can obtain satellite imagery corresponding to a period of time in the geographic region after the determined start date. Selection of satellite imagery after the determined start date period is necessary to determine the end date, and selection of satellite imagery before the provided input date period is necessary to determine the start date.

[0012] The server can then provide the time range of the wildfire, for example, the determined start and end dates of the wildfire, for output, which may be to a connected client device, a display, or one or more other applications.

[0013] In one general aspect, a method is implemented by a server, the method including: obtaining, by the server, a date on which a fire started in a geographic region; obtaining, by the server, first satellite images of the geographic region before the date on which the fire started in the geographic region; selecting, by the server, a first set of images from the first satellite images before the date on which the fire started in the geographic region; generating, by the server, a first statistical distribution from the first set of images; determining, by the server, a start date of the fire based on a comparison of the first satellite images to the first statistical distribution; and generating, by the server, second satellite images of the geographic region before the determined start date and after the determined start date of the fire in the geographic region. acquiring, by a server, a second set of images from second satellite images taken before a start date of the fire in the geographic region, wherein an amount of the second set of images is less than an amount of the first set of images; generating, by the server, a second statistical distribution from the second set of images taken before the start date of the fire in the geographic region; determining, by the server, an end date of the fire based on a comparison of the second satellite images taken after the start date with the second statistical distribution; and providing, for output by the server, a range including the start date of the fire and the end date of the fire in the geographic region.

[0014] Other embodiments and aspects of the present disclosure include corresponding systems, apparatus, and computer programs encoded on computer storage devices and configured to perform the actions of the methods. One or more computer systems may be so configured by software, firmware, hardware, or a combination thereof installed on the systems that, when operated, causes the systems to perform the actions. One or more computer programs may be so configured by features having instructions that, when executed by a data processing device, cause the systems to perform the actions.

[0015] These and other embodiments can each optionally include one or more of the following features, alone or in combination. For example, one embodiment includes all of the following features in combination:

[0016] In some implementations, the method includes obtaining a date that the fire occurred within the geographic region includes obtaining, by a server, a location that describes the geographic region; and obtaining, by the server, a date that the fire occurred within the geographic region, wherein the date may correspond to a start date of the fire, an end date of the fire, or another date within a range.

[0017] In some implementations, the method includes obtaining a first satellite image of the geographical area prior to a date on which a fire occurred within the geographical area includes providing, by a server, a location describing the geographical area on which the fire occurred and the date on which the fire occurred to a satellite image database, and obtaining, by the server, in response to providing the location and date, a first satellite image showing the geographical area from a period prior to the date.

[0018] In some implementations, the method includes selecting a first set of images from a first satellite imagery prior to a date that a fire occurred in the geographic region, including identifying, by a server, a time period for generating a first statistical distribution, the time period corresponding to a period prior to the date; selecting, by the server, the first set of images from the first satellite imagery based on the time period; determining, by the server, an indication of whether one or more pixels from the first set of selected images indicate a fire; and adjusting, by the server, the first set of selected images by comparing the one or more pixels that indicate the fire to a threshold.

[0019] In some implementations, the method includes generating a first statistical distribution from the first set of images includes, in response to adjusting the selected first set of images, generating, by the server, one or more parameters of the first statistical distribution by: determining, by the server, a number of pixels indicative of fires for each day in the first set of images; determining, by the server, an average of the number of pixels indicative of fires for each day in the first set of images; and generating, by the server, the first statistical distribution by setting the determined average to the one or more parameters.

[0020] In some implementations, the method includes determining a start date of the fire based on a comparison of the first satellite image with the first statistical distribution, the method including: identifying, by a server, a satellite image from the first satellite image corresponding to a day before the date the fire started; determining, by the server, a number of pixels from the identified satellite image that appear to indicate a fire; generating, by the server, a likelihood that the satellite image from the first satellite image includes an indication of a fire based on the first statistical distribution and the determined number of pixels that appear to indicate a fire; comparing, by the server, the likelihood with a threshold; and, in response to determining that the likelihood does not exceed the threshold, obtaining, by the server, another satellite image from the first satellite image that corresponds to another day before the date the fire started to determine the start date.

[0021] In some implementations, the method further includes, in response to determining that the likelihood exceeds a threshold for a particular day, the method including: obtaining, by the server, additional satellite images for a predetermined number of days prior to the particular day; for each day of the predetermined number of days, determining, by the server, a number of active fire pixels from the additional satellite images for that day; generating, by the server, a likelihood that the additional satellite images for that day do not contain an indication based on the first statistical distribution and the number of determined pixels that appear to indicate a fire; and, in response to determining that each day of the predetermined number of days does not indicate a fire, determining, by the server, that the day corresponding to the identified satellite image corresponds to the start date of the fire.

[0022] In some implementations, the method includes obtaining second satellite images of the geographic area before and after a determined start date of a fire in the geographic area includes providing, by a server, a location describing the geographic area where the fire occurred and a start date of the fire to a satellite image database, and obtaining, by the server in response to providing the location and start date, second satellite images showing the geographic area from a period before and after the start date.

[0023] In some implementations, the method includes selecting a second set of images from a second satellite imagery prior to a start date of a fire occurring in the geographic region, wherein an amount of the second set of images is less than an amount of the first set of images; identifying, by a server, a period for generating a second statistical distribution, wherein the period corresponds to a period before the start date; selecting, by the server, the second set of images from the second satellite imagery based on the period; determining, by the server, an indication of whether one or more pixels from the selected second set of images indicate a fire; and adjusting, by the server, the selected second set of images by comparing the one or more pixels indicating the fire to a threshold.

[0024] In some implementations, the method includes generating a second statistical distribution from a second set of images prior to a start date of a fire occurring in the geographic region, in response to adjusting the selected second set of images, by: determining, by the server, a number of pixels indicative of a fire for each day in the second set of images; determining, by the server, an average of the number of pixels indicative of a fire for each day in the second set of images; and generating, by the server, the second statistical distribution by setting the determined average to the one or more parameters.

[0025] In some implementations, the method includes determining an end date of the fire based on a comparison of a second satellite image occurring after the start date to the second statistical distribution, the method including: identifying, by a server, a satellite image from the second satellite image corresponding to a date after the start date; determining, by the server, a number of pixels from the identified satellite image for a date that appear to indicate a fire; generating, by the server, a likelihood that the satellite image from the second satellite image includes an indication of a fire based on the second statistical distribution and the determined number of pixels that appear to indicate a fire; comparing, by the server, the likelihood with a threshold; and, in response to determining that the likelihood exceeds the threshold, obtaining, by the server, another satellite image from the first satellite image corresponding to another date after the date the fire started to determine the end date.

[0026] In some implementations, the method further includes, in response to determining that the likelihood exceeds a threshold for a particular day, the method including: obtaining, by the server, additional satellite images for a predetermined number of days after the particular day; for each day of the predetermined number of days, determining, by the server, a number of pixels from the additional satellite image for that day; generating, by the server, a likelihood that the additional satellite image for that day does not contain an indication based on the second statistical distribution and the number of determined pixels that appear to indicate a fire; and, in response to determining that each day of the predetermined number of days does not indicate a fire, determining, by the server, that the day corresponding to the identified satellite image corresponds to an end date of the fire.

[0027] In some implementations, the method includes the first and second statistical distributions being different Poisson distributions.

[0028] The details of one or more embodiments of the subject matter herein are set forth in the accompanying drawings and the description below. Other features, aspects, and advantages of the subject matter will become apparent from the description, drawings, and claims. [Brief explanation of the drawings]

[0029] [Figure 1] FIG. 1 is a block diagram illustrating an example of a system for generating wildfire time coverage. [Figure 2A] FIG. 1 is a block diagram illustrating an example of a system for receiving satellite imagery and generating a time extent of a wildfire from the satellite imagery. [Figure 2B] FIG. 1 is a block diagram illustrating an example of a system for generating statistical distributions when determining the time extent of a wildfire. [Figure 3] 1 is a flow diagram illustrating an example of a process for generating a time range of a wildfire.

[0030] Like reference numbers and names in the various drawings indicate like elements. The components, their connections and relationships, and their functions shown herein are merely examples and are not intended to limit the implementations described and / or claimed herein. DETAILED DESCRIPTION OF THE INVENTION

[0031] 1 is a block diagram illustrating an example of a system 100 for generating a time range of a wildfire. The system 100 includes an image server system 102, a satellite image database 104, a temporal database 106, and a spatial database 108. The system 100 also includes a monitor 126 connected to the image server system 102 for displaying and allowing a user to interact with the image server system 102. Briefly, the system 100 can generate a time range of a wildfire identified from the date the wildfire occurred and the corresponding location of the wildfire. The system 100 can generate multiple statistical distributions and compare captured satellite images to the generated statistical distributions to identify the time range of the wildfire, e.g., start date and end date.

[0032] System 100 typically attempts to generate a time range for a wildfire because wildfire start and end dates are often missing from databases that store fire information. By generating and storing this information, system 100 can add to and improve upon the overall wildfire modeling and monitoring. This information can also be useful for various wildfire applications, as described below.

[0033] In some implementations, image server system 102 may include one or more servers or computers connected locally or over a network. System 100 may include a network, which may be, for example, a local network, a Wi-Fi network, an intranet, an Internet connection, a Bluetooth connection, or any other connection that allows image server system 102 to communicate, e.g., send and receive, with various databases and various computers. Figure 1 shows various operations in steps (A) through (J), which may be performed in the order shown or in another order.

[0034] The image server system 102 can also communicate with satellites via a satellite network. The satellites may capture media, such as images and video, of geographic regions of the Earth. The image server system 102 can communicate with the satellites to request and receive captured media of various geographic regions of the Earth, as will be described further below.

[0035] In some implementations, the satellite image database 104 can include one or more databases that store satellite images showing one or more geographic regions of the Earth. The satellite images can include high-quality satellite images, medium-quality satellite images, noisy satellite images, and other satellite images. Additionally, the satellite image database 104 can store satellite images showing one or more geographic regions of the Earth over a period of time. For example, the satellite images can include media, such as images and videos, of a particular geographic region of the Earth over the past one year, five years, ten years, fifteen years, or more from the current time. The satellite image database 104 can include satellite images for each day over a past period of time. In some cases, the satellite image database 104 can include multiple images or multiple videos from each day of a geographic region.

[0036] Satellite image database 104 may store satellite media for multiple geographic regions of the Earth over various time periods. For example, satellite image database 104 may store satellite imagery for geographic region 1 through geographic region N, and for each of those geographic regions, store satellite imagery for various time periods. In these examples, satellite image database 104 may store satellite imagery for geographic region 1 for each day over the past 10 years and satellite imagery for geographic region 2 for each day over the past 20 years.

[0037] In some implementations, the dimensions of the satellite image may correspond to the camera utilized on the satellite and the satellite's positional location. The satellite may orbit the Earth at a particular distance from the Earth and capture images of the Earth from that distance. For example, the satellite may orbit the Earth in a low orbit at a distance of 800 kilometers (km) or 497.097 miles from the Earth. In some implementations, the camera associated with the satellite may have zoom capabilities.

[0038] In some implementations, the resolution of the satellite cameras may be set to a standard resolution to ensure uniformity across all satellite images. For example, the camera's resolution may be set to 375 meters horizontally and vertically when capturing a geographic region of the Earth. In other implementations, different satellites may have different resolution capabilities. Thus, the satellite cameras may adjust their zoom capabilities to ensure uniform resolution across satellite images from each of the satellites.

[0039] In some implementations, the time database 106 can include one or more databases that store time periods for satellite images captured by satellites and stored in the satellite image database 104. The time periods can include, for example, defined dates, date ranges, and time ranges. For example, the time database 106 can store a time range from January 1, 2018 to January 1, 2019, or a time range from January 1, 2018, 12:00 PM to January 1, 2019, 12:01 PM.

[0040] Additionally, the time database 106 can store more specific date ranges. More specific date ranges can include specificity down to the hour and minute range. For example, the time database 106 can store a time range from January 1, 2018, 12:00 PM to January 1, 2019, 12:30 PM.

[0041] In some implementations, a satellite can provide satellite images along with metadata to the satellite image database 104. The metadata can include location data for the geographic area from which the satellite image was captured, data identifying the satellite from which the satellite image was captured, the date and time the satellite image was captured, and location data indicating where the satellite was located in space when it captured the satellite image. For example, if a satellite captured an image of the San Diego geographic area, the satellite can associate the following metadata with the image: (1) location data for the geographic area as 32.7515, -117.1364; (2) the date when the satellite image was captured, May 19, 2019, 12:05 PM PT; (3) data indicating the name and model of the satellite from which the image was captured, e.g., satellite 111 and model 1; and (4) location data for the satellite when the image was captured, such as 800 km above the location coordinates of 32.7515, -117.1364. The satellite image database 104 can match the wildfire date and location provided by the image server system 102 with the satellite image metadata to identify the requested satellite image.

[0042] The image server system 102 can then distribute the metadata to each of the databases in the system 100 accordingly. For example, when a satellite provides a satellite image to the image server system 102, the image server system 102 can provide the satellite image and metadata to a satellite image database 104 for storage. In addition, the image server system 102 can extract time data from the metadata and provide the time data to a time database 106. The time data can indicate the date and time the satellite image was captured by the satellite. In addition, the image server system 102 can extract location data of the geographic region from the metadata and provide the location data to a spatial database 108. The image server system 102 can perform this extraction and distribution process for each satellite image obtained from a satellite.

[0043] In some implementations, spatial database 108 may include one or more databases that store location data for locations identified in satellite images captured by satellites and stored in satellite image database 104. Location data may include, for example, latitude and longitude coordinates, addresses, names of landmarks, and dates that identify cities, states, counties, and other data that identify locations on Earth.

[0044] The spatial database 108 may include one or more bounding boxes that indicate the location of an area, e.g., a geographic region or coordinates, indicating where a wildfire started. The bounding box may include, for example, a polygon that indicates the geographic region where the wildfire started or an area around the geographic region where the fire started. The bounding box may be applied to the satellite imagery so that the image server system 102 can determine where to count fire pixels. For example, as described further below, the image server system 102 may count the number of fire pixels within the bounding box applied to the satellite imagery.

[0045] Included within the polygon may be one or more geographic features and one or more characteristics of the wildfire. For example, the one or more characteristics of the wildfire may include one or more unburned areas, one or more actively burning areas, one or more recently burned areas, and one or more scar areas resulting from the fire. The spatial database 108 may be populated by the image server system 102 or another external service.

[0046] Image server system 102 can store data in temporal database 106 and spatial database 108 in response to image server system 102 acquiring a satellite image. In some implementations, image server system 102 can store temporal and spatial data from the metadata of the satellite image to ensure that satellite image database 104 can provide the appropriate satellite image when requested. In this case, if image server system 102 requests a satellite image outside the data contained in temporal database 106 and spatial database 108, satellite image database 104 can return an error message indicating that the requested data is outside the satellite image range.

[0047] In some implementations, processes external to the image server system 102 may perform the population of the satellite image database 104, the temporal database 106, and the spatial database 108. In this case, the image server system 102 may perform processes related to determining the temporal extent of a wildfire by accessing the temporal database 106, the spatial database 108, and the satellite image database 104.

[0048] During phase (A), the image server system 102 can receive an input date 110 that the wildfire occurred. As shown in the system 100, the input date 110 corresponds to July 5, 2019. In some implementations, an external process can access the time database 106 and provide the input date 110 to the image server system 102. In some implementations, a user interacting with the image server system 102 can request a time range for the wildfire and provide the input date 110 to the image server system 102. Additionally, the input data can be provided by, for example, a fire metadata database. The input date 110 can indicate the date that the wildfire occurred and can additionally provide an indication to the image server system 102 that the satellite image database 104 includes satellite imagery of the wildfire at a corresponding geographic location. In other implementations, the image server system 102 can provide a request to the time database 106 to return the date that the wildfire occurred at a particular geographic location.

[0049] During phase (B), the image server system 102 can receive a location 112 where the wildfire occurred based on the wildfire's input date 110. As shown in the system 100, the wildfire's location can correspond to location coordinates, such as, for example, 33.8121N, -117.91899E, etc. The location 112 can also include other descriptors, such as, for example, the name of a landmark, the name of a city, the name of a geographic area, or a specific address.

[0050] In some implementations, the image server system 102 can retrieve a polygon from the spatial database 108 that indicates where the wildfire first started. The image server system 102 can obtain the polygon based on an input date 110 and a location 112. To limit the amount of processing performed by the image server system 102, the image server system 102 can apply the obtained polygon to the retrieved satellite image after retrieving the satellite image from the satellite image database 104. In this case, the polygon can spatially limit the number of pixels searched by the image server system 102 within the satellite image.

[0051] Input date 110 does not necessarily indicate to imager server system 102 the start or end date of the wildfire, even if input date 110 is in fact the start or end date of the wildfire. Rather, input date 110 may correspond to another date, such as a day between the start and end dates, when the wildfire was active or ongoing. Image server system 101 can determine where input date 110 falls within the time range of the wildfire by continuing the process, as described further below.

[0052] During stage (C), the image server system 102 can provide the satellite image database 104 with an input date 110 when the wildfire occurred and a location 112 where the wildfire occurred. The image server system 102 can provide the input date 110 and location 112 to the satellite image database 104 to obtain satellite imagery for the location and time range before the input date 110. In some implementations, the image server system 102 can indicate to the satellite image database 104 a time period before the input date 110 for obtaining satellite imagery. The time period can include, for example, one year before the input date 110, two years before the input date 110, five years before the input date 110, or any other time period before the input date 110 that is sufficient to construct a statistical distribution. For example, the image server system 102 needs satellite imagery for at least one year before the input date 110 to construct a statistical distribution, and as a result, the image server system 102 may request satellite imagery from the satellite image database 104 that is more than two years before the input date 110.

[0053] In some implementations, the satellite image database 104 can index its stored satellite images by location. For example, the image server system 102 can additionally provide the locations 112 to the satellite image database 104, which can access satellite images that include the locations 112 within their field of view, e.g., location coordinates or another geographic or location description. The satellite image database 104 can determine the locations 112 within the field of view of the corresponding satellite image by analyzing the metadata of the satellite image.

[0054] The satellite image database 104 can then filter images that include the location 112 by time range. For example, the satellite image database 104 can exclude or filter satellite imagery that has a timestamp and date after the input date 110 and that falls outside of a time period before the input date 110. For example, the image server system 102 can instruct the satellite image database 104 to provide all satellite imagery that shows the location 112 from two years prior to the date July 5, 2019, such as from July 5, 2017 to July 5, 2019. The image server system 102 can exclude all satellite imagery that shows the location 112 before the date July 5, 2017 and after the date July 5, 2019.

[0055] The satellite image database 104 can identify and provide satellite imagery 114 that meets the criteria of the image server system 102 and return it to the image server system 102. As shown in the system 100, the satellite imagery 114 can include one or more images or videos of satellite imagery that includes the location 112 and can have timestamps that fall within a time range from two years prior to the input date 110 to the input date 110. In some examples, the satellite imagery 114 can include a satellite image for each day from July 5, 2017 to July 5, 2019, or can include multiple satellite images or videos for each day from July 5, 2017 to July 5, 2019. In some cases, one or more of the days between July 5, 2017 and July 5, 2019 may not include satellite imagery if satellites did not capture imagery around the location 112 on those days.

[0056] In some implementations, each medium of the satellite imagery 114 can include a specified resolution. For example, as described above, each medium can include a resolution of 325 meters horizontally and 325 meters vertically. This resolution allows the image server system 102 to visually inspect not only the location 112 but also the area surrounding the location 112. Because a wildfire may originate from a particular location and move to another location, by viewing a larger area proximate to the location 112, such as within a bounding box around the location 112, the image server system 102 can refine its determination of the start and end dates of the wildfire. For example, the bounding box or polygon can correspond to an area in the satellite image that indicates the area encompassing the wildfire. The bounding box can include the minimum area covering the wildfire in the satellite image or an area encompassing both the wildfire and outside the wildfire.

[0057] Additionally, if the location 112 is on an edge of the satellite image 114, the satellite image database 104 can also provide satellite images adjacent to the location 112 in the edged satellite image. In this case, the image server system 102 can ensure that the location 112 shown in the satellite image is always surrounded by the area of ​​the image, e.g., the image resolution, even if the location 112 is on, e.g., a vertical or horizontal edge of the satellite image. In some implementations, the image server system 102 can indicate to the satellite image database 104 a boundary region or area around the location 112. Then, once the satellite image database 104 identifies a satellite image that meets the criteria of the image server system 102, the satellite image database 104 can identify additional satellite images to provide if the initially identified satellite image does not fall within the boundary region or area criteria.

[0058] In some implementations, the satellite image database 104 may provide the identified satellite images 114 to the image server system 102 over a network. In other implementations, the satellite image database 104 may provide an index of the identified satellite images 114 to the image server system 102 for retrieval of the satellite images 114. In some cases, the satellite image database 104 may also provide a link to the identified satellite images 114 for rapid retrieval by the image server system 102. The link may include, for example, a zip file, access to cloud storage, or some other form for downloading the satellite images 114.

[0059] During stage (D), the image server system 102 can generate a statistical distribution based on the identified satellite images 114. In some implementations, the image server system 102 can generate a statistical distribution for comparison with the satellite images to determine the likelihood of wildfires in the satellite images. In particular, the statistical distribution corresponds to a frequency model of satellite noise in a particular geographic region over a period of time. In some examples, the statistical distribution can be modeled using a Poisson distribution, a Gaussian distribution, or a normal distribution.

[0060] In some implementations, the satellite imagery 114 may show various landmarks, buildings, roads, and other geographic features. Other geographic features may include, for example, rivers, oceans, lakes, hills, and plains. The satellite imagery 114 may also show people, vehicles, animals, and other features commonly found in the geographic region. The satellite imagery 114 may also include noise characteristics such as glare, flashing lights, and distorted pixels that may make it difficult for the image server system 102 to distinguish wildfires from noise.

[0061] In some implementations, the satellite imagery can show various fire areas. These areas can include, for example, one or more unfired areas, one or more actively burning areas, one or more recently burned burned areas, and one or more scar areas resulting from the fire. The unfired areas can indicate another geographic area where fire is not occurring. The actively burning areas can include, for example, areas showing active fire, smoke, or embers. The recently burned burned areas can indicate geographic areas where fire once flared but is no longer active, but the geographic areas are still hot and dangerous. The scar areas can indicate geographic areas where fire once flared but is no longer active, and the geographic areas are no longer hot or dangerous.

[0062] The image server system 102 can generate and utilize statistical distributions to improve the accuracy of detecting active wildfires in satellite imagery. In particular, the image server system 102 can begin the process of comparing the satellite imagery to the generated statistical distribution by first identifying the number of pixels in the satellite imagery 114 that appear to be indicative of a wildfire. Some of the identified pixels may not be indicative of a wildfire, but rather may represent noisy pixels. To mitigate this problem, the image server system 102 can compare the number of identified pixels in the satellite imagery 114 that appear to be indicative of a wildfire to the generated statistical distribution to filter out images with noisy characteristics.

[0063] To be able to make such a comparison, the image server system 102 can construct a statistical distribution using historical satellite images retrieved from the satellite image database 104 that show the same location 112 as the location 112, but that do not show an active wildfire. For example, the image server system 102 must first identify a period prior to the input date 110 within the retrieved satellite image 114 that does not show a wildfire. In addition to identifying a date three months prior to the input date 110 that ensures no wildfires are found in the corresponding satellite image, the image server system 102 can consult historical weather forecasts and other meteorological databases to confirm whether fires have existed in a particular geographic area, such as the geographic area identified in the satellite image. By generating a satellite image baseline, the image server system 102 can identify any pixels in the satellite image that fall outside the baseline's norm and indicate that the corresponding satellite image appears to show a wildfire. In this case, the image server system 102 first determines a date three months prior to the input date 110. In the example of the system 100, the date three months before the input date 110 is April 5, 2019.

[0064] The image server system 102 may determine a date three months prior to the input date 110 because wildfires generally do not last longer than three months. By identifying a date three months prior to the input date 110, the image server system 102 may safely assume that the same wildfire identified by the input date 110 and location 112 is not active or lit in the corresponding satellite imagery, e.g., a satellite imagery dated three months prior to the input date 110. In some implementations, if the image server system 102 identifies a wildfire that is still burning in the satellite imagery 114 on a date three months prior to the input date 110, the image server system 102 may identify an even earlier date. In some cases, the image server system 102 may identify whether one or more other wildfires were active during the period between the input date 110 and the date three months prior to the input date 110. The one or more other wildfires may correspond to wildfires that had a different origin and that did not cause the wildfire identified by the input date 110. The image server system 102 may access one or more public wildfire databases to determine whether other wildfires were active during this time window. However, the occurrence of other wildfires active within the same geographic area as the wildfire identified by the input date 110 is rare.

[0065] For example, the image server system 102 may move the date back one month from April 5, 2019, to March 5, 2019. If the image server system 102 determines that there are no pixels in the satellite imagery 114 on March 5, 2019 that appear to indicate a wildfire, the image server system 102 can set March 5, 2019 as the end date of the time range for constructing the statistical distribution. The image server system 102 can iteratively perform the process of identifying the end date of the time range, for example, going back one day, one month, or one week, until a wildfire-free start date is indicated in the corresponding satellite imagery. In this case, the image server system 102 cannot set this determined end date as the actual start date of the wildfire because the wildfire may not actually become active for, for example, one day, one week, or one month after the end date.

[0066] In some implementations, the image server system 102 can determine a date that is one year before the determined end date for the time range. The date that is one year before the determined end date of the time range is known as the start date of the time range. The image server system 102 can identify start and end dates for the time range for identifying satellite images and building a statistical distribution. Continuing with the example of the system 100, the image server system 102 can determine the start date of the time range to be April 5, 2018, one year before the end date of, for example, April 5, 2019.

[0067] However, in order for image server system 102 to set the start date to April 5, 2018, image server system 102 can use data from an external database of historical fires to ensure that satellite imagery showing location 112 does not show an active wildfire between the start date (e.g., April 5, 2018) and the end date (e.g., April 5, 2019). If image server system 102 identifies data from the external database of historical fires that indicates that a wildfire existed or was active in the geographic region from April 5, 2018 to April 5, 2019, for example, image server system 102 can adjust the start date or end date accordingly. The manner in which image server system 102 detects or identifies active wildfires in satellite imagery is described further below.

[0068] To determine whether satellite images within a time range indicate an active wildfire, the image server system 102 can identify the number of fire pixels in the satellite image and compare the number of fire pixels to a threshold. For example, the image server system 102 can run a fire detection algorithm on the satellite image to determine how many pixels contain a fire. The fire detection algorithm can include, for example, an active fire detection algorithm. If the number of fire pixels in the satellite image is less than the threshold, the image server system 102 can indicate that the particular satellite image does not contain a fire.

[0069] For example, the image server system 102 may select a subset of satellite images from the satellite imagery 114 between April 5, 2018 and April 5, 2019, inclusive. The image server system 102 may process each image or video from each day within that time range and search for pixels that appear to indicate a wildfire. If the image server system 102 detects pixels that appear to indicate a wildfire on April 8, 2018, e.g., if the number of pixels is greater than a threshold, but not on other days, the image server system 102 may adjust the time range to April 9, 2018 to April 5, 2019. Alternatively, if the image server system 102 detects pixels that appear to indicate a wildfire on April 1, 2019, e.g., if the number of pixels is greater than a threshold, the image server system 102 may adjust the time range to April 5, 2018 to March 31, 2019.

[0070] In some cases, if the image server system 102 detects pixels that appear to indicate a wildfire in the middle of a time range, such as December 1, 2018 or November 1, 2018, for example, if the number of pixels is greater than a threshold, the image server system 102 can exclude those days from the time range and adjust the time range into multiple consecutive sub-time ranges. For example, the consecutive sub-time ranges can include April 5, 2018 - October 31, 2018 - November 2, 2018 - November 31, 2018 - December 2, 2018 - April 5, 2019. Thus, the image server system 102 can create a single consecutive time range or multiple consecutive sub-time ranges to construct a satellite image distribution.

[0071] In some implementations, the image server system 102 can determine a date nine months prior to a date three months prior to the input date 110. For example, if the image server system 102 determines that satellite imagery for the date April 5, 2019 does not appear to indicate a wildfire, the image server system 102 can identify a date nine months prior to the date April 5, 2019 for the start of the time range. The end date of the time range would be April 5, 2019. The date nine months prior to April 5, 2019 corresponds to July 5, 2018. Thus, the image server system 102 can construct a statistical distribution based on the time range from July 5, 2018 to April 5, 2019.

[0072] If the image server system 102 bases the statistical distribution on a shorter nine-month time range, e.g., July 5, 2018 to April 5, 2019, the image server system 102 can generate the statistical distribution more quickly and save processing speed / power. However, if the image server system 102 bases the statistical distribution on a longer one-year time range, e.g., April 5, 2018 to April 5, 2019, the image server system 102 can generate a more accurate and representative statistical distribution of the locations 112. In some examples, wildfires may typically occur in a particular region, e.g., the California region, during a standard three-month fire season beginning around August and September. For fires occurring during this fire season, a nine-month window can be selected instead of a one-year window to reduce the likelihood that a satellite image that accidentally includes a wildfire from the previous year's fire season will be selected in the baseline distribution. In some implementations, the image server system 102 can remove the fire season corresponding to the time range from the satellite imagery to reduce the likelihood of accidentally including a fire within the time range or baseline distribution.

[0073] In response to identifying satellite images within the identified time range from the satellite imagery 114, the image server system 102 can determine the number of pixels that appear to indicate wildfires for each day in the identified satellite images. For example, the image server system 102 can analyze the satellite images for each day during the identified time range from April 5, 2018 to April 5, 2019 to determine the number of fire pixels for each day. In some implementations, the image server system 102 can apply a polygon retrieved from the spatial database 108 to the identified satellite images. By cropping the identified satellite images to the polygon, the image server system 102 can reduce the amount of pixels searched across the identified satellite images. In this case, the image server system 102 can search across the pixels in the identified satellite images within the retrieved polygon.

[0074] For example, the image server system 102 may count 10 fire pixels on April 5, 2018, 11 fire pixels on April 6, and continue counting satellite imagery for each day until April 5, 2019. The same process is similar for analyzing satellite imagery for a time range of one year, or for a time range having multiple consecutive sub-time ranges.

[0075] In some implementations, the image server system 102 can normalize the number of fire pixels detected on each day. For example, the image server system 102 can analyze one satellite image on April 5, 2018, 50 satellite images on April 6, 2018, and 10 satellite images on April 7, 2018. The image server system 102 can divide the average number of fire pixels detected on each day to normalize the detections. The reason for normalizing is to ensure that the generated statistical distribution generates a smooth curve. If the values ​​are not normalized, the generated distribution curve may contain jagged edges, distorting the probability prediction.

[0076] In some implementations, after determining the number of fire pixels for each day within the satellite imagery for the identified time range or within a polygon of the satellite imagery, the image server system 102 can generate a statistical distribution. For example, the image server system 102 can generate a frequency histogram of the number of identified pixels within the satellite imagery that appear to indicate an ongoing wildfire. The frequency histogram can be shown as a graphical representation of the number of days on the X-axis and the number of fire pixels on the Y-axis.

[0077] The image server system 102 can then determine the mean of the frequency histogram and fit other data descriptors, such as the mean or standard deviation, to a statistical distribution. For example, the image server system 102 can fit the mean of the frequency histogram to a Poisson distribution or another statistical distribution. In a Poisson distribution, the mean is equivalent to the variance. For the system 100, the average number of pixels identified as indicating wildfires in a given time interval, e.g., one year, is equivalent to the variance of the Poisson distribution.

[0078] In some implementations, if a significant number of wildfire events are unintentionally included in a baseline distribution, the baseline distribution may no longer have an equivalent mean and variance. In other words, the baseline distribution may no longer follow a Poisson distribution. In this case, the image server system 102 can perform a test to determine whether the mean and variance of the baseline distribution, e.g., a statistical distribution, are equivalent. The test is performed, for example, by analyzing the mean of the samples in the statistical distribution and analyzing the variance of the samples. If the image server system 102 detects that the mean and variance of the statistical distribution are not equivalent, the image server system 102 can adjust the time range and corresponding satellite images and construct a new statistical distribution based on the satellite images from the adjusted time range. The image server system 102 can repeat this process until a statistical distribution with an equivalent mean and variance is constructed. Once the statistical distribution is constructed, only a single parameter, such as the average number of pixels that appear to indicate a wildfire, is required to determine the probability of an event. The following equation shows the probability distribution function of a Poisson random variable:

[0079]

number

[0080] In Equation 1 above, which describes the Poisson distribution, the value λ corresponds to the average number of pixels that appear to indicate a wildfire from the baseline. The value k corresponds to the number of times a random variable or event occurs, e.g., the number of pixels from the input image that appear to indicate a wildfire. The resulting value, f(k;λ), corresponds to a probability that represents the Poisson probability distribution function. In some implementations, the image server system 102 can return a Poisson distribution 116 once the value λ is determined.

[0081] The Poisson distribution 116 can provide a resultant value during comparison with the input image, which indicates the likelihood of an event occurring within a given time period. In particular, the probability density function (PDF) of the Poisson distribution 116 can return the probability that a particular day with k counted pixels occurred naturally from a distribution of noise. Thus, the lower the value output by the PDF, the more likely it is that an external force, e.g., a wildfire, contributed to the k counts. Alternatively, the higher the value output by the PDF, the less likely it is that an external force contributed to the k counts, indicating that the image more closely resembles an image from a baseline distribution. For example, if the probability corresponds to a low value, such as 0.01, the image server system 102 can determine that the image is likely to show a wildfire. Alternatively, if the probability corresponds to a high value, such as 0.90, the image server system 102 can determine that the image is likely not to show an ongoing wildfire.

[0082] During stage (E), the image server system 102 can determine the start date of the wildfire based on the generated statistical distribution and the satellite images 114. For example, as shown in system 100, the image server system 102 can generate the statistical distribution 116 during stage (D). The image server system 102 then compares each of the satellite images 114 beginning with an input date 110 of July 5, 2019 to the generated statistical distribution 116.

[0083] First, the image server system 102 determines from the satellite image 114 the number of pixels that appear to indicate a wildfire in the satellite image on July 5, 2019. In some implementations, the image server system 102 can use a fire detection algorithm to determine whether the pixels indicate a wildfire. In other implementations, the image server system 102 can analyze the brightness of the pixels to determine whether the pixels indicate a wildfire. For example, the image server system 102 can determine the number of pixels that appear to indicate a wildfire in the satellite image on July 5, 2019 by analyzing the brightness of each pixel and comparing the brightness of the pixel to a threshold. For example, the image server system 102 can determine the brightness of a pixel by averaging the red, blue, and green (RGB) values ​​of the pixel. In some implementations, the image server system 102 can analyze wavelengths outside of the RGB values ​​to determine which pixels are on fire. For example, wavelengths outside of RGB can include infrared wavelengths. The lower the average, the less bright the pixel, and vice versa. In another example, the image server system 102 can determine the brightness of a pixel by calculating a relative brightness value using the following formula: Y=0.2126*R+0.7152*G+0.0722*B (2)

[0084] Equation 2 above shows that the relative brightness value of a pixel is calculated based on the RGB values ​​of the pixel. For example, if a pixel's red value is 100, its green value is 100, and its blue value is 100, then the relative brightness value is 100. The greater the relative brightness value, the greater the brightness of the pixel. Although the brightness of a pixel does not indicate whether the pixel indicates a wildfire or noise, the brightness of a pixel may be an indication that the pixel appears to be on fire. In some implementations, the image server system 102 can also analyze the intensity of a pixel to determine whether a wildfire is present within the pixel. In other implementations, the image server system 102 can rely on external sources to determine whether a corresponding pixel indicates a wildfire. For example, the image server system 102 can utilize the Visible Infrared Imaging Radiometer Suite (VIIRS) Active Fire Product developed by NASA to determine whether a pixel was on fire by combining infrared bands from satellite images with the RGB bands of the pixel. Typically, each satellite includes its own detection algorithm, which is optimized for the specifications of the corresponding instrument.

[0085] In some implementations, the image server system 102 can apply polygons retrieved from the spatial database 108 to the satellite imagery for a particular day. The image server system 102 can determine the number of pixels within the polygon of the satellite imagery that appear to indicate a wildfire on July 5, 2019, by analyzing the brightness or intensity of each pixel and comparing the brightness of the pixels to a threshold. By incorporating the polygons onto the satellite imagery, the image server system 102 can reduce processing complexity and speed up processing time because fewer pixels are searched within the satellite imagery.

[0086] The image server system 102 may calculate the brightness of each pixel in the satellite imagery for July 5, 2019, or within a polygon applied to the satellite imagery for July 5, 2019. Then, for each pixel, the image server system 102 may compare the brightness of each pixel to a threshold value. For example, the image server system 102 may specify a threshold brightness value of 50 or some other threshold value. If a particular pixel of the satellite imagery is greater than the threshold value, the image server system 102 may consider the pixel to be included in the count. The image server system 102 repeats this process for each pixel of the satellite imagery.

[0087] In some implementations, the image server system 102 can identify pixels in the satellite imagery that are indicative of a wildfire. Depending on the situation, pixels that may be indicative of a wildfire may correspond to areas within a polygon, such as an actively burning area and a potentially burned area where a recent fire has occurred. If a recently burned area includes one or more embers from the fire, the image server system 102 can detect and identify pixels associated with those embers in the satellite imagery as being indicative of a fire. However, pixels associated with non-burned areas and scar areas resulting from the fire are not detected by the image server system 102 as being indicative of a fire.

[0088] The image server system 102 can then input the number of counted pixels in the satellite image into the value of k in Equation 1 of the Poisson distribution 116. For example, the image server system 102 may fit the frequency histogram to a Poisson distribution to determine the following equation:

[0089]

number

[0090] As shown in Equation 3, the number 5 in place of the value λ corresponds to the average number of pixels that appear to indicate a wildfire from the baseline distribution for each day. Furthermore, the value λ can describe the broadness of the Poisson distribution curve. Equation 3 represents the probability density function or continuous probability density function of the Poisson distribution. The value k corresponds to the number of pixels found in the satellite image on July 5, 2019 that appear to indicate a wildfire. For example, if the pixel count for a particular image corresponds to a value of 6, Equation 3 should yield the following value:

[0091]

number

[0092] As shown in Equation 4, the image server system 102 calculated a 14.62% probability that a six fire pixel event is likely to occur in the baseline distribution. Furthermore, the calculated probability indicates the number of events (e.g., fire pixels) that occurred in the satellite imagery in a single day (e.g., July 5, 2019) relative to the known average percentage of fire pixels occurring in a single year (e.g., April 5, 2018 - April 5, 2019), as determined from the baseline distribution.

[0093] In some implementations, the calculated probability can correspond to a metric that is a proxy for the probability that the satellite image is not on fire. Given this proxy, the image server system 102 can determine whether the satellite image shows a wildfire depending on which side of a threshold the likelihood falls. If the likelihood is greater, the likelihood can indicate that the satellite image shows a wildfire. Alternatively, if the likelihood is smaller, the likelihood can indicate that the satellite image does not show a wildfire.

[0094] The image server system 102 may then compare the output probability to a threshold to determine whether the satellite image for July 5, 2019 has a similar number of fire pixels as the baseline distribution. For example, the image server system 102 may set the threshold at 5% or 0.05. If the image server system 102 determines that the output probability is less than the threshold, the image server system 102 may determine that the satellite image likely shows a fire. Alternatively, if the image server system 102 determines that the output probability is greater than the threshold, the image server system 102 may determine that the satellite image does not contain a fire.

[0095] If the image server system 102 determines that the output probability is less than the threshold, the image server system 102 repeats the process of comparing the average number of pixels that appear to indicate a wildfire with the previous day. For example, the image server system 102 can obtain a satellite image for July 4, 2019 from the satellite imagery 114. The image server system 102 can then determine, for example, using a fire detection algorithm, the number of pixels in the satellite image for July 4, 2019 that appear to indicate a wildfire. The image server system 102 can determine a probability from a statistical distribution based on the count of fire pixels for that day. The image server system 102 can then compare the probability from the statistical distribution to a threshold to determine whether the satellite image contains a likelihood of a fire.

[0096] The image server system 102 repeats this process until the probability from the statistical distribution is greater than a threshold. In some implementations, if the image server system 102 determines that the probability that a satellite image for a particular day shows a wildfire is greater than a threshold, the image server system 102 can consider the satellite image to be fire-free. In this case, the image server system 102 can identify the first occurrence of the satellite image that generates the probability based on the statistical distribution and the number of pixels greater than the threshold. In some implementations, the image server system 102 can consider that date to be the date the wildfire started. As shown in the system 100, this start date 118, or wildfire start date, corresponds to June 1, 2019, because satellite imagery does not show an ongoing wildfire, but satellite imagery for June 2, 2019, July 4, 2019, and July 5, 2019 shows a fire.

[0097] In other implementations, the image server system 102 can perform additional processes if the statistical distribution outputs a probability greater than a threshold for the identified date. In particular, the image server system 102 can perform verification that the identified date that the wildfire started is actually accurate and not determined from an error. For example, the image server system 102 can determine that the identified date that the wildfire started was not determined by noise, such as clouds or smoke, found in satellite imagery. To verify, the image server system 102 can calculate a probability using the statistical distribution 116 for each day one week prior to the identified start date.

[0098] For example, if the image server system 102 determines June 1, 2019 as the identified date, the image server system 102 can perform an additional validation process to calculate a probability using the statistical distribution 116 for each day prior to June 1, 2019, for example, for each day going back to May 25, 2019. The image server system 102 can, for example, obtain satellite imagery for each day from May 25, 2019 through May 31, 2019, and count the number of fire pixels for each day from May 25, 2019 through May 31, 2019. Then, for the pixel counts from each day, the image server system 102 can calculate a corresponding probability using the statistical distribution 116 from each day. If each corresponding probability is greater than a threshold, e.g., indicating a high enough probability to be classified as a fire-free day, the image server system 102 can consider June 1, 2019, to be a true date on which a wildfire occurred. Instead, if one of the days in the previous week period indicates a probability lower than the threshold, for example, if one of these days indicates a wildfire, the image server system 102 can continue to iterate backward to find the actual start date of the wildfire and then the "fire-free" days for the entire previous week. The same process applies to finding the end date of the wildfire.

[0099] In some implementations, the image server system 102 may need to make adjustments to the identified start or end date. For example, the image server system 102 may determine that satellites are unable to identify a wildfire until two days after the start date. To resolve this issue, the image server system 102 may subtract two days from the identified start date to ensure accurate time range detection. For example, if the image server system 102 identified June 1, 2019 as the start date, the image server system 102 may move the date back two days, such as May 30, 2019, to account for inconsistent detection with the satellite. Other days, such as three, four, or five days, may also be possible.

[0100] In some implementations, the image server system 102 can adjust the identified start date based on the rate of change of the fire's intensity. For example, if the image server system 102 notices that the satellite imagery changed by 60 in intensity from one day to the next, the image server system 102 can move the date back by three days, e.g., to May 29, 2019. If the image server system 102 notices that the satellite imagery changed by 40 in intensity from one day to the next, the image server system 102 can move the date back by four days, e.g., to May 28, 2019. Typically, a rapidly spreading fire requires fewer adjustments than a slowly spreading fire. The process is similar for adjusting the identified end date.

[0101] During stage (F), the image server system 102 can generate a second statistical distribution using the wildfire start date 118. As shown from the system 100, the wildfire start date 118 occurred one month and four days before the provided input date 110. The image server system 102 can determine the wildfire end date 124 based on the second generated statistical distribution.

[0102] Initially, the image server system 102 can provide the start date 118 and the location 112 to the satellite image database 104 to obtain satellite imagery for generating the second statistical distribution. Step (F) is similar to step (C) in that the date and location are provided to the satellite image database 104, and the satellite image database 104 returns satellite imagery showing the location. However, in step (F), the image server system 102 can also instruct the satellite image database 104 to provide satellite imagery from both before and after the start date 118. The image server system 102 can generate the second statistical distribution based on the satellite imagery from before the start date 118, and can determine the end date 124 of the wildfire based on a comparison of the satellite imagery from after the start date 118 to the second statistical distribution.

[0103] For example, the image server system 102 can indicate to the satellite image database 104 a period around the start date 118 for acquiring satellite imagery. The period around the start date 118 can be determined based on an amount of time needed to construct the second statistical distribution, e.g., a period before the start date 118, and an amount of time needed to identify an end date, e.g., a period after the start date 118. For example, the image server system 102 can identify a one-year and three-month period that is one year before the start date 118 for constructing the second statistical distribution and three months after the start date 118 for determining the end date 124. In another example, the image server system 102 can request a large number of satellite imagery, such as a two-year, five-year, or more-year period centered around the start date 118.

[0104] During step (G), the satellite image database 104 may provide the identified satellite images 120a and 120b to the image server system 102 over the network. The satellite image database 104 may provide the satellite images 120a and 120b over the network as an email, a zip file, or another format. In some examples, the satellite image database 104 may send an index to the image server system 102 for retrieving the satellite images 120a and 120b.

[0105] For example, satellite image 120a may correspond to a satellite image acquired before start date 118. Satellite image 120b may correspond to a satellite image acquired after start date 118. Image server system 102 may generate a second statistical distribution based on satellite image 120a. Additionally, image server system 102 may determine end date 124 based on the generated second statistical distribution and satellite image 120b.

[0106] During stage (H), the image server system 102 can generate a second statistical distribution based on the identified satellite image 120a and the determined start date 118. Stage (H) is similar to stage (D). However, during stage (H), the image server system 102 does not need to determine the date three months prior to the input date because the start date 118 is known. In this case, the image server system 102 can identify the time region prior to the start date 118.

[0107] In some implementations, the image server system 102 can return an error, the identified time range, or both. The image server system 102 can return an error if an end date is requested for a wildfire that is still active, if satellite imagery is unavailable or of low quality for the time and geographic location of the historical fire, if satellite data is unavailable or of low quality for the time range used to construct the noise distribution, or otherwise, to name a few. The error can indicate, for example, "insufficient satellite imagery," "wildfire still active," or "low quality satellite imagery." In other examples, the image server system 102 can indicate that if the baseline distribution is too close to the current day, e.g., within one day, several days, or one week, the image server system 102 can move the baseline distribution to an earlier time period accordingly.

[0108] For example, the image server system 102 may generate a time range that extends one year prior to the start date 118. In this example, the image server system 102 may generate a time range from June 1, 2018 to June 1, 2019 for acquiring satellite imagery to construct the second statistical distribution. As was done in step (C), the image server system 102 may ensure that no wildfires are active within the satellite imagery during this time range. If the image server system 102 detects that there is an ongoing wildfire within this satellite imagery, for example, or within a polygon of the satellite imagery, the image server system 102 may adjust the time range accordingly to avoid the satellite imagery showing an active wildfire.

[0109] In some implementations, once the time range is identified, the image server system 102 can generate the second statistical distribution 122. In particular, the image server system 102 can generate the second statistical distribution 122 based on a subset of the satellite images 120a that meet the criteria of the identified time range, e.g., June 1, 2018 - June 1, 2019. The second statistical distribution 122 can also be in the form of a Poisson distribution, a normal distribution, or a Gaussian distribution, to name a few. In some cases, the value λ for the second statistical distribution 122 can be different from the value λ for the first statistical distribution 116.

[0110] During phase (I), image server system 102 can determine an end date 124 of the wildfire based on generated second statistical distribution 122 and satellite imagery 120b. The functionality of phase (I) is similar to the functionality of phase (E). However, image server system 102 compares each of the satellite images from satellite imagery 120b beginning with a start date of June 1, 2019, to generated second statistical distribution 122.

[0111] For example, the image server system 102 can obtain a satellite image for June 1, 2019 from the satellite imagery 120a. The image server system 102 can then determine the number of pixels in the satellite image for June 1, 2019 that appear to indicate a wildfire by counting each of the pixels that appear to indicate a wildfire. For example, the image server system 102 can identify the number of pixels within a polygonal region of the satellite image for June 1, 2019 that appear to indicate a wildfire. The image server system 102 can determine a probability from the second statistical distribution 122 based on the count of fire pixels in the image. The image server system 102 can then compare the probability from the statistical distribution to a threshold to determine whether the satellite image contains a likelihood of a fire.

[0112] If image server system 102 determines that the probability that a satellite image for a particular day, e.g., June 2, 2019, shows a wildfire is below a threshold, image server system 102 can assume that the satellite image appears to show a fire. In this case, image server system 102 identifies a satellite image for the next day in the future from satellite image 120b, e.g., June 3, 2019. Image server system 102 can traverse forward in time to find the wildfire's end date. In particular, image server system 102 repeats this process until the probability from the statistical distribution is greater than a threshold and the probability for some consecutive number of days immediately following the potential end date is also greater than the threshold. This number of days may be 6, 7, 8, or some other number, to name a few examples. This process for verifying that a wildfire's potential end date is indeed correct is similar to verifying that the dates of a wildfire's previous potential states are correct. However, instead of calculating probabilities using statistical distribution 116 for each day one week prior to the potential start date, image server system 102 calculates probabilities in this process using statistical distribution 122 for each day one week after the potential end date to identify "fire-free" days. Image server system 102 can continue to iterate forward to identify the actual end date of the wildfire if the next one-week period results in one or more days indicating a wildfire.

[0113] In some implementations, if the image server system 102 determines that the probability that a satellite image for a particular day shows a wildfire is greater than a threshold, the image server system 102 can consider the satellite image to be fire-free. The image server system 102 can then repeat the process of calculating probabilities for successive days, such as one week, after the potential day using a statistical distribution to identify whether the potential day corresponds to an actual end date. If the image server system 102 determines that the generated probability for each day after the potential date is greater than a threshold, the image server system 102 can consider that date to be the date the wildfire ended.

[0114] In some implementations, the image server system 102 can determine the wildfire's end date 124 by analyzing pixels within one or more burned areas and one or more scarred areas resulting from the recent fire. The image server system 102 can determine that these corresponding pixels within a polygon in the satellite image correspond to the burned or scarred areas. Additionally, the image server system 102 can expand the polygonal area overlaid on the satellite image when identifying the end date 124. By expanding, the image server system 102 can identify pixels within the maximum perimeter of the polygon to ensure that the wildfire has not spread to other areas in the satellite image.

[0115] In some implementations, the image server system 102 can include multiple surrounding polygons in satellite imagery spanning the lifespan of a wildfire. In this case, the image server system 102 can crop the polygon closest to the satellite data after the input date 110 when constructing the statistical distribution 116. Additionally, the image server system 102 can crop the oldest polygon when constructing the statistical distribution 122. The image server system 102 can select the closest polygon for the statistical distribution 116 and the oldest polygon for the statistical distribution 122 to reduce noise when determining the start date. If the image server system 102 views satellite imagery in an area known not to have burned as a result of a wildfire during a particular day, the image server system 102 will only count noisy pixels that do not contain the wildfire. Therefore, if the image server system 102 only counts fire pixels, the accuracy of the start date prediction will be reduced.

[0116] As shown in system 100, a satellite image on June 1, 2019 appears to show a wildfire, a satellite image on June 2, 2019 appears to show a wildfire, a satellite image on August 4, 2019 appears to show a smaller wildfire, and a satellite image on August 5, 2019 appears to show no wildfire. Assuming image server system 102 does not detect a wildfire on any days after August 5, 2019, for example, from August 6, 2019 to August 13, 2019, image server system 102 can define an end date 124, or the date the wildfire ended, as August 5, 2019.

[0117] During stage (J), image server system 102 may provide an output time range for the wildfire to one or more pipelines. As shown in system 100, image server system 102 has determined that the time range for the wildfire is from June 1, 2019 to August 5, 2019 based on first statistical distribution 116, second statistical distribution 122, and corresponding satellite imagery. Furthermore, image server system 102 may determine that this time range is accurate because the length of the time range is less than three months, e.g., less than the maximum amount of time that a fire may last, and input date 110 falls within the time range.

[0118] In some implementations, the image server system 102 can provide the time range to one or more internal or external pipelines. The internal pipelines can be, for example, different machine learning models as training data. The different machine learning models can implement applications such as monitoring wildfires and their spread. Other internal pipelines can include, for example, different graphical representations of wildfires or other user interfaces for showing the spread of wildfires. External pipelines can be, for example, other external systems performing wildfire detection, dispatching responders to extinguish wildfires, and remotely monitoring crises such as wildfires or other climate disasters. The image server system 102 can also provide the time range of the wildfire to other systems. For example, as shown in system 100, the image server system 102 can output the time range to a monitor 126 for review.

[0119] To verify the authenticity of the process of system 100, image server system 102 can retrieve satellite images with known start and end dates from satellite image database 104. For example, image server system 102 can retrieve multiple satellite images corresponding to dates in the middle of an identified time range. Image server system 102 can then provide each satellite image from the multiple satellite images to the process illustrated in system 100 to determine whether image server system 102 can identify the same time range, e.g., the same start and end dates. If image server system 102 can identify the same time range with each satellite image, the process illustrated in system 100 appears to be functioning properly.

[0120] 2A is a block diagram illustrating an example of a system 200 for receiving satellite imagery and generating a time extent of a wildfire from the satellite imagery. System 200 includes similar components and performs similar functions as system 100. For example, system 200 includes a satellite 202, an image server system 204, a temporal database 214, a spatial database 216, and a satellite image database 218. System 200 also includes a network 205, which may include either a satellite network, a local connection, or another connection over the Internet.

[0121] During phase (A), image server system 204 may send request 203 to satellite 202 over network 205. Request 203 may instruct satellite 202 to capture one or more satellite images of a geographic region on Earth. Request 203 may include location coordinates, e.g., latitude and longitude coordinates, of the area for which satellite 202 is to capture one or more images as it travels around the Earth.

[0122] During stage (B), satellite 202 may capture one or more satellite images of the geographic area provided by request 203. Satellite 202 may include one or more satellite cameras for capturing the satellite images. For example, as shown in system 200, the geographic area may include one or more mountains, hills, fires, and smoke. Other geographic areas may include more, fewer, or different geographic features than those shown in system 200. Satellite 202 may capture satellite image 206 of the requested geographic area as satellite 202 travels around the Earth.

[0123] For example, satellite 202 may navigate over a requested geographic area on a daily basis and thus provide satellite imagery of the requested geographic area on a daily basis. In some implementations, satellite 202 may capture multiple images, multiple videos, or both of a geographic area per day. In this case, image server system 102 may be able to view multiple satellite images and videos one day, a different set of multiple satellite images and videos the next day, and so on. By providing sets of media, image server system 204 can improve the accuracy of its time-scope wildfire detection.

[0124] In some implementations, the one or more geographic regions captured by the satellite cameras may include one or more locations associated with a wildfire. For example, the one or more locations associated with a wildfire may include one or more unburned regions, one or more actively burning regions, one or more recently burned burned regions, and one or more scarred regions resulting from the fire. As shown in system 201, the one or more locations associated with a wildfire within a geographic region may include one or more unburned regions 211, one or more actively burning regions 210, one or more recently burned burned regions 208, and one or more scarred regions resulting from the fire.

[0125] Each of regions 208, 210, 211, and 212 can be adjacent to one another within a geographic region. When satellite 202 captures one or more images of a geographic region based on the location in request 203, the satellite image can include one or more of these regions. For example, satellite 202 can capture an image including the location from request 203, and the image can include an actively burning region 210, a burned region 208 where a fire recently occurred, and one or more scar regions 212 resulting from the fire. Other satellite images can include other regions, such as each of the regions shown in system 100.

[0126] The satellite 202 can then transmit the captured media 206 to the image server system 204 via the network 205. In some implementations, the satellite 202 can transmit the captured media 206 directly to the satellite image database 218, directly bypassing the image server system 204. In this case, at a later point in time, the image server system 204 can obtain data identifying the captured media 206 to determine the time range of the wildfire represented by the captured media. For example, the data identifying the captured media 206 can include, for example, location data and the particular time that the wildfire represented by the captured media 206 was active.

[0127] During stage (C), the image server system 204 obtains an input date 220 of the wildfire and a corresponding location 222 of the wildfire. Stage (C) of the system 200 is similar to stages (A), (B), and (C) of the system 100. In particular, as shown in the system 200, the image server system 204 obtains an input date 220 of July 5, 2019 and a location 222 of 33.8121N, -117.91899E for the corresponding wildfire. In some implementations, the image server system 204 can obtain a polygon to apply to the satellite imagery to reduce the amount of pixels searched. The polygon can be obtained from the spatial database 216. The polygon can spatially limit the number of pixels searched for fire pixels by the image server system 102. For example, the image server system 204 can apply a polygon 226 to the obtained satellite imagery 224. The corresponding wildfire may correspond to a captured wildfire in the satellite image 206 , but may alternatively refer to a satellite image previously captured and stored in the satellite image database 218 .

[0128] The image server system 204 can provide the satellite image database 218 with the date 220 that the wildfire occurred and the location 222 where the wildfire occurred. Additionally, the image server system 204 can indicate a time period for satellite image retrieval. For example, the image server system 204 can instruct the satellite image database 218 to provide all satellite imagery starting from two years prior and with a date of July 5, 2019, showing the location 222. The image server system 204 can also request satellite imagery from the satellite image database 218 having different time periods, such as one year, three years, five years, or more, to name a few. In response, the satellite image database 218 can return satellite imagery 224 to the image server system 204 based on criteria specified by the image server system 204. The image server system 204 can then apply a polygon 226 to the retrieved satellite imagery 224 to spatially limit the area that the image server system 204 searches for fire pixels.

[0129] During step (D), the image server system 204 may generate a first statistical distribution based on the acquired satellite images 224. Step (D) of the system 200 is similar to step (D) of the system 100.

[0130] In particular, the image server system 204 can determine a time range for identifying a subset of satellite images from the satellite imagery 224. The image server system 204 identifies a date three months prior to the input date 220, such as April 5, 2019. The image server system 204 then determines either a one-year time range (e.g., April 5, 2018 - April 5, 2019) or a nine-month time range (e.g., July 5, 2019 - April 5, 2019) to build a satellite imagery baseline.

[0131] In response to identifying the time range, the image server system 204 can identify a subset of satellite images from the satellite images 224 using the identified time range. For example, the image server system 204 can identify satellite images from April 5, 2018 to April 5, 2019 from the satellite images 224 and ensure that the identified satellite images do not exhibit wildfire pixels, e.g., do not exhibit satellite images within a threshold. The image server system 204 can then generate a first statistical distribution by determining the average number of fire pixels found in the identified satellite images within the time range and fitting the average number of fire pixels to a statistical distribution. The statistical distribution can be, for example, a Poisson distribution, where the mean is equal to the variance.

[0132] During step (E), image server system 204 can determine the number of pixels in the satellite image that appear to indicate an active wildfire from satellite image 224. Step (E) of system 200 is similar to step (E) of system 100. In particular, image server system 204 can determine the number of pixels within a polygon in the satellite image that appear to indicate a wildfire on a particular day, e.g., July 5, 2019, by counting detected fire pixels using a fire detection algorithm or using brightness threshold detection, to name a few.

[0133] During step (F), the image server system 204 can compare the number of detected fire pixels to the generated statistical distribution. Step (F) of the system 200 is similar to step (E) of the system 100. For example, if the generated statistical distribution is a Poisson distribution, the image server system 204 can determine the probability by providing the number of detected fire pixels for a particular image as a value in the Poisson distribution. The Poisson distribution can return a probability indicating the likelihood that an event will occur, for example, the likelihood that a particular image from July 5, 2019, shows a wildfire. The image server system 204 can compare the probability to a threshold to determine whether satellite images from a particular day have a similar number of fire pixels.

[0134] If image server system 204 determines that the probability output by the generated statistical distribution is less than the threshold, then during step (G), image server system 204 retrieves satellite imagery for the previous day. Step (G) of system 200 is similar to step (E) of system 100. For example, image server system 204 retrieves satellite imagery for July 4, 2019, and repeats the process of steps (E) and (F) of system 200. Image server system 204 repeats the process of steps (E), (F), and (G) until the start date of the wildfire is identified. Thus, image server system 204 continues to go back one date at a time until the corresponding satellite imagery showing location 222 no longer indicates an active or ongoing wildfire, e.g., until the probability output by the generated statistical distribution is greater than the threshold.

[0135] Alternatively, if image server system 204 determines that the probability output by the generated statistical distribution is greater than a threshold, then during stage (H), image server system 204 indicates that wildfire start date 228 has been identified, assuming image server system 204 does not detect a wildfire for a predetermined number of days prior to potential start date 228. As shown in system 200, wildfire start date 228 corresponds to June 1, 2019. Stage (H) of system 200 is similar to stages (F) and (G) of system 100.

[0136] For example, during stage (H), the image server system 204 can provide the identified start date 220 and location 222 to the satellite image database 218 to acquire satellite imagery for generating the second statistical distribution. Here, the image server system 204 can also instruct the satellite image database 218 to provide satellite imagery from both before the start date 228 and after the start date 228. For example, the image server system 204 can specify a particular time frame around the start date 228 (e.g., one year and three months) or request a larger time frame around the start date 228 (e.g., two years, five years, or more).

[0137] In some implementations, the satellite image database 218 can return captured satellite images 232 based on criteria identified by the image server system 204. In particular, the satellite images 232 can include satellite images before the identified start date 228, for example, to construct a second statistical distribution, and satellite images after the identified start date 228, for example, to identify an end date of the wildfire.

[0138] During step (I), the image server system 204 may generate a second statistical distribution based on the satellite images 232 before the identified start date 228 and the identified start date 228. Step (I) of the system 200 is similar to step (H) of the system 100. In this step, the image server system 204 may identify a time range before the start date 228 to construct the second statistical distribution. The image server system 204 may extract a subset of satellite images from the satellite images 232 based on the identified time range. The image server system 204 may then fit the subset of satellite images to a statistical distribution, such as a Poisson distribution, a normal distribution, or a Gaussian distribution.

[0139] During stage (J), image server system 204 can determine the number of pixels in the satellite image that appear to indicate a wildfire from satellite imagery 232 after start date 228. Stage (J) of system 200 is similar to stage (I) of system 100. In this case, image server system 204 can count the number of pixels in the satellite image that appear to indicate a wildfire for a particular day, e.g., June 1, 2019, by determining the brightness of each pixel or by using a fire detection algorithm.

[0140] During step (K), the image server system 204 can compare the count of fire pixels for a particular image on a particular day to the generated second statistical distribution. Step (K) of the system 200 is similar to step (F) of the system 200. The generated second statistical distribution can generate a probability indicating the likelihood of an event occurring, for example, the likelihood that a particular image on June 1, 2019, shows a wildfire. The image server system 204 can compare the probability to a threshold to determine whether satellite images on a particular day have a similar number of fire pixels.

[0141] If the image server system 204 determines that the probability output by the generated second statistical distribution is less than the threshold, then during stage (L), the image server system 204 acquires satellite imagery for the next day. Stage (L) of system 200 is similar to stage (I) of system 100. For example, the image server system 204 acquires satellite imagery for June 2, 2019, and repeats the process of stages (J) and (K) of system 200. The image server system 204 repeats the process of stages (J), (K), and (L), advancing one date at a time until an end date for the wildfire is identified and the image server system 204 assumes that it will not detect a wildfire for a predetermined number of days after the potential end date.

[0142] Alternatively, if image server system 204 determines that the probability output by the generated statistical distribution is greater than a threshold, then during stage (K), image server system 204 indicates that the wildfire end date 234 has been identified, assuming that image server system 204 does not detect a wildfire for a predetermined number of days after the potential end date 234. As shown in system 200, wildfire end date 234 corresponds to August 5, 2019. Stage (K) of system 200 is similar to stage (I) of system 100.

[0143] During stage (M), image server system 204 provides wildfire time range 236 for output. Time range 236 indicates that the wildfire range is from June 1, 2019 to August 5, 2019. Stage (M) of system 200 is similar to stage (J) of system 100. For example, image server system 204 can output time range 236 to one or more external or internal pipelines.

[0144] 2B is a block diagram illustrating an example of a system 201 for generating statistical distributions when determining the time extent of a wildfire. System 201 illustrates the processes performed during steps (D) and (H) from system 100 and steps (D) and (I) from system 200. Additionally, system 201 and its processes may also be performed by image server system 102.

[0145] In some implementations, the system 201 exhibits a process for generating a statistical distribution. The image server system 204 can generate a statistical distribution that is a Poisson distribution, a normal distribution, or a Gaussian distribution, to name a few. The process for generating the statistical distribution generally includes, among other things, (i) identifying a first date that is three months prior to the input date, (ii) identifying a second date that is one year or nine months prior to the first date, (iii) selecting a subset of satellite images from the acquired satellite images based on a time range between the first date and the second date, (iv) determining the mean, standard deviation, and other statistical characteristics of the fire pixels from the subset of satellite images, and (v) generating the statistical distribution based on the statistical characteristics of the fire pixels from the subset of satellite images.

[0146] In some implementations, the image server system 204 can generate the statistical distribution without identifying a first date that is three months prior to the input date. In this case, if the image server system 204 determines that the input date corresponds to the actual start date of the wildfire, such as start date 118 generated in the system 100, the image server system 204 can proceed to identify a second date that is one year prior to the input date without identifying the first date. For example, if the input date is the determined start date of the wildfire, e.g., June 1, 2019, instead of identifying a date three months prior to June 1, 2019 to ensure that the wildfire is not present in the satellite imagery, the image server system 204 can identify a second date that is one year prior to the start date, which corresponds to the second date of June 1, 2018.

[0147] As a result, for example, if image server system 204 moves the start date back by three months, image server system 204 may identify a time range of June 1, 2018 to June 1, 2019 instead of March 1, 2018 to March 1, 2019. In some cases, image server system 204 may identify a nine-month time range, e.g., September 1, 2019 to June 1, 2019, instead of a one-year time range. Image server system 204 may identify a nine-month range when a shorter time range is needed due to memory constraints to avoid detecting fire pixels in historical satellite imagery, to name a few examples, or may identify a nine-month range when an external user requests that specific time range.

[0148] During stage (A), image server system 204 may determine a first date 242 that is three months prior to the input date. For example, image server system 204 may determine that first date 242 is April 5, 2019 if the input date corresponds to wildfire input date 110, e.g., July 5, 2019. In some cases, image server system 204 skips this stage if the input date corresponds to the wildfire's determined start date, e.g., start date 118.

[0149] During stage (B), image server system 204 may determine a second date 244 that is one year earlier than first date 242. For example, image server system 204 may determine that second date 244 is April 5, 2019 if first date 242 corresponds to April 5, 2018 and the input date corresponds to wildfire input date 110. However, in this example, image server system 204 may determine that first date 242 is June 1, 2018 if the input date is start date 118.

[0150] During step (C), image server system 204 can select a subset of satellite images from the acquired satellite images. As shown in system 201, image server system 204 can obtain a location 240 of the geographic area where the wildfire occurred and a date 238 that the wildfire occurred at location 240. The date can be, for example, a string indicating July 5, 2019, and the location can be a string or number indicating latitude and longitude coordinates of 33.8121N, -117.91899E. This step is similar to steps (A) and (B) from system 100.

[0151] The image server system 204 can then request satellite imagery from the satellite image database 218 by providing a date 238, a location 240, and a time period 237. The time period 237 can indicate, for example, a period prior to the date 238 for which to obtain satellite imagery. For example, the time period 237 can indicate one year, three years, five years, or another length of time sufficient to construct a statistical distribution.

[0152] In some implementations, image server system 204 may indicate that time period 237 corresponds to a time period before date 228 and a time period after date 238. For example, if image server system 204 is attempting to determine the end date of a wildfire, image server system 204 may instruct satellite image database 218 to identify two years before date 238 and five months after date 238. Other time periods may also be indicated.

[0153] The satellite image database 218 may include satellite images showing various geographic regions of the Earth. For example, the satellite images may include high-resolution images 213b, noisy images 213a, and other types of images such as infrared video and other media.

[0154] In some implementations, image server system 204 can acquire satellite images and select a subset of satellite images from the acquired satellite images. For example, as shown in system 201, image server system 204 can select satellite images 213c from the acquired satellite images. Satellite images 213c can include a set of satellite images representing high-resolution images, noisy-resolution images, low-resolution images, satellite video, and other satellite media. Image server system 204 can select satellite images 213c based on the time period identified in steps (A) and (B) of system 201, one or more time ranges that avoid fire pixel detection in the set of satellite images, and other criteria.

[0155] During step (D), the image server system 204 can determine fire pixel detections for each day over the identified time range. This step is similar to step (D) from system 100. For example, the image server system 204 can analyze each satellite image from satellite images 213c to determine the number of pixels that appear to indicate a wildfire for each day within the set of satellite images. For example, the image server system 102 can analyze the satellite images 113c for each day during the identified time range from April 5, 2018 to April 5, 2019 to determine the number of fire pixels for each day. In another example, the image server system 102 can analyze within a polygon in the satellite images 113c for each day from April 5, 2018 to April 5, 2019 to determine the number of fire pixels for each day within the polygon. For example, the image server system 102 may count two fire pixels on April 5, 2018, count three fire pixels on April 6, and continue counting satellite images for each day until April 5, 2019.

[0156] During step (E), the image server system 204 can generate a statistical distribution in response to determining the number of fire pixels for each day in the satellite images 113c. For example, the image server system 204 can generate one or more parameters for the statistical distribution, which may be a Poisson distribution, such as the mean and variance of the distribution. If the statistical distribution is a Poisson distribution, the image server system 204 can determine the average number of pixels from the set of satellite images 113c that appear to indicate a fire. In this case, the average number of identified pixels that appear to indicate a wildfire over the identified time range may be, for example, 5, which is equivalent to the variance of the Poisson distribution. In response to identifying and fitting the average of fire pixel detections in the satellite images, the image server system 204 can provide the statistical distribution 116 for various applications.

[0157] 3 is a flow diagram illustrating an example of a process 300 for generating a time range of a wildfire. The image server system 102 of the system 100 and the image server system 204 of the system 200 can implement the process 300.

[0158] The image server system obtains 302 dates on which fires have occurred within a geographic region. For example, an external database, such as a time database, can provide the image server system with dates indicating when wildfires have occurred in a geographic region. The dates can be provided in month, day, year format, or another format. The dates can correspond to the start date of the wildfire, the end date of the wildfire, or another date between the start and end dates of the wildfire.

[0159] In some implementations, the image server system can also obtain a location that describes the location of the wildfire based on the date, for example, the location can include location GPS coordinates, the name of a landmark, the name of a city, the name of another geographic area, a specific address, or other location.

[0160] The image server system can also obtain a polygon that indicates an area within the satellite image that contains a wildfire. The polygon can spatially limit the number of pixels within the satellite image that the image server system can search to identify fire pixels. Essentially, the image server system can search for fire pixels within the satellite image within the area spatially defined by the polygon to reduce processing and the amount of pixels the image server system needs to search to identify fire pixels.

[0161] The image server system acquires 304 a first satellite image of the geographic region prior to a date that a fire occurred within the geographic region. In some implementations, the image server system can provide the acquired location describing the geographic region where the wildfire occurred and the acquired date that the wildfire occurred to a satellite image database. In addition, the image server system can also provide a time range indicating a period prior to the input date for acquiring satellite imagery. The period can include, for example, one year prior to the input date, two years prior to the input date, five years prior to the input date, or any other period prior to the input date that is sufficient to construct a statistical distribution.

[0162] The satellite image database can use the acquisition date, acquisition location, and time range to identify satellite images that meet the provided criteria. The satellite images can include one or more satellite images, for example, high quality, medium quality, low quality, noisy, and other satellite images. The satellite images can depict one or more geographic regions of the Earth. In some implementations, the satellite image database can provide the identified satellite images that meet the criteria to an image server system to build a statistical distribution.

[0163] The image server system selects 306 a first set of images from the first satellite imagery that precedes a date when a fire occurred in the geographic region. In some implementations, the image server system can identify the first set of images from the first satellite imagery to construct a statistical distribution. First, the image server system can identify a period prior to the input date in the acquired satellite imagery that does not show a wildfire. The image server system can determine a date three months prior to the input date because wildfires generally do not last longer than three months. By identifying a date three months prior to the input date, the image server system can safely assume that the same wildfire identified by the input date and location is not active or lit in the corresponding satellite imagery.

[0164] In some implementations, the image server system can determine the date one year before a date that is three months before the input date. For example, if the input date corresponds to July 5, 2019, the image server system can determine that the date three months before the input date is April 5, 2019. The image server system then identifies the date one year before April 5, 2019, which corresponds to April 5, 2018. In some implementations, the image server system can use the date nine months before instead of the date one year before.

[0165] In some implementations, the image server system can use the time range to identify a first set of satellite images from the first satellite image to construct a statistical distribution. For example, the image server system can identify satellite images from satellite images retrieved from a satellite image database between dates April 5, 2018 and April 5, 2019, inclusive.

[0166] However, because the image server system uses a first set of satellite images from April 5, 2018, to April 5, 2019, the image server system can ensure that no historical fires are present in the identified satellite images during that time period. For example, the image server system can use an external database of historical fires to make this determination. If the external database of historical fires indicates that no fires were present during this time period, the image server system can run a fire detection algorithm on each satellite image within the time range from April 5, 2018, to April 5, 2019, to determine the number of pixels from each image that contain fire. If the number of fire pixels in the satellite image is below a threshold, the image server system can indicate that the particular satellite image does not contain a fire. Alternatively, the image server system can adjust the time range if it finds that the image contains a fire, e.g., if the number of fire pixels is greater than a threshold.

[0167] The image server system generates a first statistical distribution from the first set of images (308). To generate the first statistical distribution, the image server system can generate one or more parameters for the statistical distribution. For example, the image server system can determine the number of pixels that appear to indicate a fire for each day in the satellite images identified from (306). The image server system can count the number of fire pixels for each day using a fire detection algorithm. In some implementations, the image server system can crop a polygon for each image in the identified satellite images before counting the fire pixels to reduce the number of pixels that need to be analyzed. The image server system can then determine the mean value or average number of fire pixels from each image. The image server system may also determine other data descriptors, such as standard deviation and variance, to fit a particular statistical distribution.

[0168] For example, the image server system can fit the average of the average number of fire pixels to a Poisson distribution. In a Poisson distribution, the mean is equivalent to the variance. Other distributions, such as normal and Gaussian distributions, require the image server system to perform different processes to determine data descriptors that help fit the identified satellite imagery to other distributions. For example, once the data is fitted to a Poisson distribution, the image server system can process the acquired satellite imagery to identify the start date of the wildfire.

[0169] The image server system determines the start date of the fire based on a comparison of the first satellite image to the first statistical distribution (310). In some implementations, the generated statistical distribution can be used to indicate the probability that some event occurred from the baseline of the generated statistical distribution. For example, a Poisson probability density function can return the probability that a particular day with k counted pixels occurred naturally from a distribution of noise. The lower the value output by the PDF, the more likely it is that an external force, e.g., a wildfire, contributed to the k counts. Alternatively, the higher the value output by the PDF, the less likely it is that an external force contributed to the k counts, indicating that the image more closely resembles an image from the baseline distribution.

[0170] From the acquired satellite imagery, the image server system can identify satellite imagery from a day prior to the date on which the wildfire occurred. The image server system can count or determine the number of pixels in the satellite imagery from the previous day that appear to indicate a fire. In some cases, the number of pixels counted can be within a polygonal area cropped by the image server system. The image server system can then provide the number of fire pixels determined for that image on that particular day as an input to the generated statistical distribution.

[0171] The image server system can compare the output probability from the generated statistical distribution, e.g., a Poisson distribution, to a threshold. If the output probability is less than the threshold, the image server system can determine that the corresponding satellite image contains a likelihood of a fire. Alternatively, if the output probability is greater than the threshold, the image server system can determine that the corresponding satellite image does not contain a fire. If a fire is determined, the image server system repeats this process, going back in time, e.g., from July 4, 2019 to July 3, 2019, until it finds satellite images that do not appear to indicate a wildfire.

[0172] In response to the image server system determining that the satellite imagery does not indicate a wildfire, the image server system retrieves additional satellite imagery from the retrieved satellite imagery for a predetermined number of days prior to the particular date. For example, if the image server system determines a potential start date of June 1, 2019, the image server system may identify eight days prior to June 1, 2019, such as May 24, 2019, and retrieve satellite imagery for each of those days. The image server system then attempts to identify and verify whether June 1, 2019 is the actual start date and analyzes a predetermined number of days prior to the potential start date to determine that no fire is present in the satellite imagery corresponding to those days. If no fire is present on those days, the image server system can confidently assert that June 1, 2019 is the actual start date of the fire.

[0173] First, the image server system can count the number of pixels that appear to indicate a fire for each day in a predetermined number of days prior to June 1, 2019. The image server system can then generate a likelihood for each day using the statistical distribution and the number of pixels that appear to indicate a wildfire on each corresponding day. The statistical distribution can return a probability or likelihood for each day. For example, for May 31, 2019, the probability may be 0.56, for May 30, 2019, the probability may be 0.6, and for May 29, 2019, the probability may be 0.7. This process is repeated until May 24, 2019, or until each day in the predetermined number of days has an associated probability. If the probability for each day is greater than a threshold, the image server system can confidently indicate that June 1, 2019 was the actual start date of the fire. If one or more of the day probabilities are below the threshold, the image server system may continue to iterate backwards to find the actual start date of the wildfire, and then the entire week prior to the "fire-free" day.

[0174] The image server system acquires (312) second satellite images of the geographic region before and after the determined start date of a fire occurring within the geographic region. The image server system can provide the acquired locations and the identified start date to a satellite image database. In addition, the image server system can also provide a time period around the identified start date to construct a second statistical distribution, for example, a one-year and three-month time period that is one year before the start date to construct the second statistical distribution and three months after the start date of the satellite images to determine the end date. The satellite image database can use criteria to identify and return second satellite images that meet the criteria. (312) includes functionality similar to (304).

[0175] The image server system selects a second set of images from the second satellite images prior to the start date of the fire in the geographic region, the second set of images being fewer than the first set of images (314). In some implementations, the image server system can identify a second set of images from the second satellite images to construct a second statistical distribution. The image server system can identify a period prior to the start date in the acquired satellite images that does not show the wildfire. The image server system does not need to determine a date three months prior to the start date because the date the fire occurred is already known. Thus, the period prior to the start date for the second satellite images can be shorter or less than the period identified for the first set of satellite images, which included a period three months prior. Instead, the image server system can identify a date one year prior to the start date to construct the second statistical distribution.

[0176] In some implementations, the image server system can identify a second set of satellite images from the second satellite images using this time range to construct a second statistical distribution. For example, the image server system can identify satellite images from satellite images retrieved from the satellite image database between June 1, 2018 and June 1, 2019, inclusive. Additionally, the image server system can ensure that no wildfires are active in the second set of satellite images during the time range between June 1, 2018 and June 1, 2019. If a wildfire is active, the image server system can adjust the time range. (314) includes functionality similar to (306).

[0177] The image server system generates a second statistical distribution (316) from a second set of images prior to the start date of the fire within the geographic region. (316) includes functionality similar to (308). In particular, the image server system can generate one or more parameters for the second statistical distribution. For example, the image server system can determine the number of pixels that appear to indicate a fire for each day of the satellite images identified from (314). The image server uses a fire detection algorithm to count the number of fire pixels for each day and determine the average number of fire pixels from each image. The image server system may also determine other data descriptors, such as standard deviation and variance, to fit a particular statistical distribution, such as a Poisson distribution. Once the data has been fitted to a second statistical distribution, such as a Poisson distribution, the image server system can process the acquired satellite images to identify the end date of the wildfire.

[0178] The image server system determines the end date of the fire based on a comparison of a second satellite image taken after the start date to the second statistical distribution (318). (318) includes functionality similar to (310). In some implementations, the image server system can identify satellite images from satellite images acquired on days after the start date. The image server can count the number of fire pixels in the identified satellite images and provide the determined number of fire pixels as input to the second statistical distribution. If the output probability from the second statistical distribution is less than a threshold, the image server system can determine that the corresponding satellite image contains a likelihood of a fire. Alternatively, if the output probability is greater than a threshold, the image server system can determine that the corresponding satellite image does not contain a fire. The image server system repeats this process, iterating forward one day at a time, for example, from June 1, 2019 to June 2, 2019 to June 3, 2019, until a satellite that does not appear to indicate a wildfire is found.

[0179] In response to detecting a date as a potential end date, e.g., August 5, 2019, the image server system can retrieve additional satellite imagery from the retrieved satellite imagery for a predetermined number of days after the potential end date. Here, the image server system attempts to identify and verify whether August 5, 2019 is an actual end date, and analyzes the predetermined number of days after the potential end date to determine that no fire is present in the satellite imagery corresponding to those days. If no fire is present on those days, the image server system can confidently assert that August 5, 2019 is the actual end date of the wildfire.

[0180] Similar to (310), the image server system can count the number of pixels that appear to indicate a wildfire for each day in a predetermined number of days after August 5, 2019. The image server system can generate a likelihood for each day using a second statistical distribution and the number of pixels that appear to indicate a wildfire on each corresponding day. The statistical distribution can return a probability or likelihood for each day. If the probability for each day is greater than a threshold, the image server system can confidently indicate that August 5, 2019 is the actual end date of the fire. If one or more of the day probabilities are less than the threshold, the image server system can continue to iterate forward in time to find the actual end date of the wildfire, and then find the next entire week of "fire-free" days.

[0181] The image server system provides for output a range that includes a fire start date and a fire end date within a geographic region (320). In some examples, the wildfire time range, e.g., the determined start and end dates, can be provided to one or more other pipelines and can also be provided to a display on the image server system. Additionally, the image server system can provide the time range over a network to a client device of a user that requested the time range.

[0182] All of the embodiments and functional operations of the invention described herein may be implemented in digital electronic circuitry, or in computer software, firmware, or hardware, including the structures disclosed herein and their structural equivalents, or in combinations of one or more of these. Embodiments of the invention may be implemented as one or more computer program products, i.e., one or more modules of computer program instructions encoded on a computer-readable medium for execution by or controlling the operation of a data processing apparatus. The computer-readable medium may be a non-transitory computer-readable storage medium, a machine-readable storage device, a machine-readable storage substrate, a memory device, a composition of matter giving rise to a machine-readable propagated signal, or a combination of one or more of these. The term "data processing apparatus" encompasses all apparatuses, devices, and machines for processing data, including, by way of example, a programmable processor, a computer, or multiple processors or computers. In addition to hardware, an apparatus may include code that creates an execution environment for the computer program in question, such as code constituting processor firmware, a protocol stack, a database management system, an operating system, or a combination of one or more of these. A propagated signal is an artificially generated signal, for example, a machine-generated electrical, optical, or electromagnetic signal generated to encode information for transmission to a suitable receiver device.

[0183] A computer program (also known as a program, software, software application, script, or code) may be written in any form of programming language, including compiled or interpreted languages, and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment. A computer program does not necessarily correspond to a file in a file system. A program may be stored within a portion of a file that holds other programs or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files storing one or more modules, subprograms, or portions of code). A computer program may be deployed to be executed on one computer or on multiple computers located at one site or distributed across multiple sites and interconnected by a communications network.

[0184] The processes and logic flows described herein may be implemented by one or more programmable processors executing one or more computer programs to perform functions by operating on input data and generating output. The processes and logic flows may also be implemented by, and an apparatus may be implemented as, special purpose logic circuitry, such as an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit).

[0185] Processors suitable for executing a computer program include, by way of example, both general-purpose and special-purpose microprocessors, as well as any one or more processors of any kind of digital computer. Generally, a processor receives instructions and data from a read-only memory or a random-access memory, or both. The essential elements of a computer are a processor for executing instructions and one or more memory devices for storing instructions and data. Generally, a computer also includes one or more mass storage devices, e.g., magnetic disks, magneto-optical disks, or optical disks, for storing data, or is operatively coupled to receive data from or transfer data to one or more mass storage devices, or both. However, a computer need not have such devices. Furthermore, a computer may be incorporated into another device, such as a tablet computer, a mobile phone, a personal digital assistant (PDA), a mobile audio player, or a Global Positioning System (GPS) receiver, to name but a few. Suitable computer-readable media for storing computer program instructions and data include, by way of example, semiconductor memory devices such as EPROM, EEPROM, and flash memory devices, magnetic disks such as internal hard disks or removable disks, magneto-optical disks, and CD-ROM and DVD-ROM disks. The processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.

[0186] To provide for user interaction, embodiments of the present invention may be implemented on a computer having a display device, such as a CRT (cathode ray tube) or LCD (liquid crystal display) monitor, for displaying information to the user, as well as a keyboard and pointing device (e.g., a mouse, trackball) by which the user can provide input to the computer. Other types of devices may be used to provide for user interaction; for example, feedback provided to the user may be any form of sensory feedback, such as visual feedback, auditory feedback, or tactile feedback, and input from the user may be received in any form, including acoustic, speech, or tactile input.

[0187] Embodiments of the invention may be implemented in a computing system, or any combination of one or more such back-end, middleware, or front-end components, including back-end components (e.g., as a data server), middleware components (e.g., an application server), or front-end components (e.g., a client computer having a graphical user interface or web browser through which a user may interact with an implementation of the invention). The components of the system may be interconnected by any form or medium of digital data communication, e.g., a communication network. Examples of communication networks include a local area network ("LAN") and a wide area network ("WAN"), e.g., the Internet.

[0188] A computing system may include clients and servers. Clients and servers are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.

[0189] While several implementations have been described in detail above, other modifications are possible. For example, while a client application is described as accessing a delegate, in other implementations, the delegate may be employed by other applications implemented by one or more processors, such as applications running on one or more servers. Additionally, the logic flows depicted in the figures do not require the particular order shown, or sequential order, to achieve desirable results. Additionally, other actions may be provided or removed from the described flows, and other components may be added to or removed from the described systems. Accordingly, other implementations are within the scope of the following claims.

[0190] While this specification contains details of many specific implementations, these should not be construed as limiting the scope of any invention or what may be claimed, but rather as descriptions of features that may be specific to particular embodiments of a particular invention. Certain features described herein in the context of separate embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented in multiple embodiments separately or in any suitable subcombination. Furthermore, while features may be described above as working in certain combinations and as originally claimed as such, one or more features from a claimed combination may, in some cases, be deleted from the combination, and the claimed combination may be directed to a subcombination or a variation of the subcombination.

[0191] Similarly, although operations are depicted in the figures in a particular order, this should not be understood as requiring such operations to be performed in the particular order shown, or in any sequential order, or that all illustrated operations be performed, to achieve desired results. In certain situations, multitasking and parallel processing may be advantageous. Furthermore, the separation of various system modules and components in the above-described embodiments should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems may generally be integrated together in a single software product or packaged in multiple software products.

[0192] Specific embodiments of the present subject matter have been described. Other embodiments are within the scope of the following claims. For example, the actions recited in the claims can be performed in a different order and still achieve desirable results. As an example, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In certain implementations, multitasking and parallel processing may be advantageous.

Claims

1. A computer-implemented method performed comprising: obtaining, by the server, dates on which fires occurred within a geographic region; obtaining, by the server, a first satellite image of the geographic area prior to the date the fire occurred within the geographic area; selecting, by the server, a first set of images from the first satellite imagery prior to the date the fire occurred within the geographic region; generating, by the server, a first statistical distribution from the first set of images; determining, by the server, a start date of the fire based on a comparison of the first satellite image and the first statistical distribution; acquiring, by the server, second satellite images of the geographic region before the determined start date and after the determined start date of the fire within the geographic region; selecting, by the server, a second set of images from the second satellite imagery prior to the start date of the fire in the geographic region, wherein an amount of the second set of images is less than an amount of the first set of images; generating, by the server, a second statistical distribution from a second set of images prior to the start date of the fire in the geographic region; determining, by the server, an end date of the fire based on a comparison of the second satellite image generated after the start date with the second statistical distribution; providing, for output by the server, a range that includes the start date of the fire and the end date of the fire within the geographic region.

2. obtaining the date of occurrence of the fire within the geographic region; obtaining, by the server, a location describing the geographic region; 2. The computer-implemented method of claim 1, comprising: obtaining, by the server, the date the fire started within the geographic region, wherein the date may correspond to the start date of the fire, the end date of the fire, or another date within the range.

3. obtaining the first satellite image of the geographic area prior to the date the fire occurred within the geographic area; providing, by the server, the location describing the geographic area in which the fire occurred and the date the fire occurred to a satellite image database; and obtaining, by the server, in response to providing the location and the date, the first satellite imagery showing the geographic region from a period prior to the date.

4. selecting the first set of images from the first satellite imagery prior to the date the fire occurred within the geographic region; identifying, by the server, a time period for generating the first statistical distribution, the time period corresponding to a time period before the date; selecting, by the server, the first set of images from the first satellite imagery based on the time period; determining, by the server, an indication of whether one or more pixels from the first set of selected images indicate a fire; and adjusting, by the server, the first set of selected images by comparing the one or more pixels indicative of a fire to a threshold.

5. generating the first statistical distribution from the first set of images; adjusting, by the server, one or more parameters of the first statistical distribution in response to adjusting the first set of selected images; determining, by the server, a number of pixels representing fires for each day in the first set of images; determining, by the server, an average number of pixels representing fires for each day in the first set of images; and generating, by the server, the first statistical distribution by setting the determined mean to the one or more parameters.

6. determining the start date of the fire based on the comparison of the first satellite image to the first statistical distribution; identifying, by the server, satellite images from the first satellite images corresponding to days before the date on which the fire occurred; determining, by the server, a number of pixels from the identified satellite imagery for the day that appear to indicate a fire; generating, by the server, a likelihood that the satellite image from the first satellite image contains an indication of a fire based on the first statistical distribution and the number of pixels determined to appear to be indicative of a fire; comparing, by the server, the likelihood to a threshold; and in response to determining that the likelihood does not exceed the threshold, obtaining, by the server, another satellite image from the first satellite image corresponding to another day before the date on which the fire started, to determine the start date.

7. In response to determining that the likelihood exceeds the threshold for a particular day, the method: acquiring, by the server, additional satellite images for a predetermined number of days prior to the particular date; For each day of the predetermined number of days, determining, by the server, a number of active fire pixels from the additional satellite imagery for that day; generating, by the server, a likelihood that the additional satellite imagery for that day will not contain an indication based on the first statistical distribution and the number of pixels determined to appear to be indicative of a fire; 7. The computer-implemented method of claim 6, further comprising: determining, by the server, in response to determining that each day of the predetermined number of days does not show the fire, that the day corresponding to the identified satellite image corresponds to the start date of the fire.

8. acquiring the second satellite imagery of the geographic area before the determined start date and after the determined start date of the fire within the geographic area; providing, by the server, a location describing the geographic area in which the fire occurred and the start date of the fire to a satellite image database; and in response to providing the location and the start date, obtaining, by the server, the second satellite imagery showing the geographic region from time periods before and after the start date.

9. Selecting the second set of images from the second satellite imagery prior to the start date of the fire in the geographic region, wherein the amount in the second set of images is less than the amount in the first set of images, includes: identifying, by the server, a time period for generating the second statistical distribution, the time period corresponding to a time period before the start date; selecting, by the server, a second set of images from the second satellite images based on the time period; determining, by the server, an indication of whether one or more pixels from the selected second set of images indicate a fire; and adjusting, by the server, the second set of selected images by comparing the one or more pixels indicative of a fire to a threshold.

10. generating the second statistical distribution from a second set of images prior to the start date of the fire in the geographic region; adjusting, by the server, one or more parameters of the second statistical distribution in response to adjusting the second set of selected images; determining, by the server, a number of pixels representing fires for each day in the second set of images; determining, by the server, an average number of pixels representing fires for each day in the second set of images; and generating, by the server, the second statistical distribution by setting the determined mean to the one or more parameters.

11. determining the end date of the fire based on a comparison of the second satellite image generated after the start date to the second statistical distribution; identifying, by the server, satellite images from the second satellite images that correspond to days after the start date; determining, by the server, a number of pixels from the identified satellite imagery for the day that appear to indicate a fire; generating, by the server, a likelihood that the satellite image from the second satellite image contains an indication of a fire based on the second statistical distribution and the number of pixels determined to appear to be indicative of a fire; comparing, by the server, the likelihood to a threshold; 11. The computer-implemented method of claim 1, further comprising: in response to determining that the likelihood exceeds the threshold, obtaining, by the server, another satellite image from the first satellite image corresponding to another day after the date on which the fire started, to determine the end date.

12. In response to determining that the likelihood exceeds the threshold for a particular day, the method: acquiring, by the server, additional satellite images for a predetermined number of days after the particular date; For each day of the predetermined number of days, determining, by the server, a number of pixels from the additional satellite imagery for that day; generating, by the server, a likelihood that the additional satellite imagery for that day will not contain an indication based on the second statistical distribution and the number of pixels determined to appear to be indicative of a fire; 12. The computer-implemented method of claim 11, further comprising: determining, by the server, in response to determining that each day of the predetermined number of days does not show the fire, that the day corresponding to the identified satellite image corresponds to the end date of the fire.

13. The computer-implemented method of any one of claims 1 to 12, wherein the first statistical distribution and the second statistical distribution are different Poisson distributions.

14. 1. A system comprising: one or more computers and one or more storage devices that store instructions, the instructions, when executed by the one or more computers, causing the one or more computers to: obtaining, by the server, dates on which fires occurred within a geographic region; obtaining, by the server, a first satellite image of the geographic area prior to the date the fire occurred within the geographic area; selecting, by the server, a first set of images from the first satellite imagery prior to the date the fire occurred within the geographic region; generating, by the server, a first statistical distribution from the first set of images; determining, by the server, a start date of the fire based on a comparison of the first satellite image and the first statistical distribution; acquiring, by the server, second satellite images of the geographic region before the determined start date and after the determined start date of the fire within the geographic region; selecting, by the server, a second set of images from the second satellite imagery prior to the start date of the fire in the geographic region, wherein an amount of the second set of images is less than an amount of the first set of images; generating, by the server, a second statistical distribution from a second set of images prior to the start date of the fire in the geographic region; determining, by the server, an end date of the fire based on a comparison of the second satellite image generated after the start date with the second statistical distribution; providing, for output by the server, a range including the start date of the fire and the end date of the fire within the geographic region.

15. obtaining the date of occurrence of the fire within the geographic region; obtaining, by the server, a location describing the geographic region; 15. The system of claim 14, further comprising: obtaining, by the server, the date that the fire occurred within the geographic region, wherein the date that the fire occurred within the geographic region may correspond to the start date of the fire, the end date of the fire, or another date within the range.

16. obtaining the first satellite image of the geographic area prior to the date the fire occurred within the geographic area; providing, by the server, the location describing the geographic area in which the fire occurred and the date the fire occurred to a satellite image database; and in response to providing the location and the date, obtaining, by the server, the first satellite imagery showing the geographic area from a period prior to the date.

17. selecting the first set of images from the first satellite imagery prior to the date the fire occurred within the geographic region; identifying, by the server, a time period for generating the first statistical distribution, the time period corresponding to a time period before the date; selecting, by the server, the first set of images from the first satellite imagery based on the time period; determining, by the server, an indication of whether one or more pixels from the first set of selected images indicate a fire; and adjusting, by the server, the first set of selected images by comparing the one or more pixels indicative of a fire with a threshold.

18. generating the first statistical distribution from the first set of images; adjusting, by the server, one or more parameters of the first statistical distribution in response to adjusting the first set of selected images; determining, by the server, a number of pixels representing fires for each day in the first set of images; determining, by the server, an average number of pixels representing fires for each day in the first set of images; and generating, by the server, the first statistical distribution by setting the determined mean to the one or more parameters.

19. determining the start date of the fire based on the comparison of the first satellite image to the first statistical distribution; identifying, by the server, satellite images from the first satellite images corresponding to days before the date on which the fire occurred; determining, by the server, a number of pixels from the identified satellite imagery for the day that appear to indicate a fire; generating, by the server, a likelihood that the satellite image from the first satellite image contains an indication of a fire based on the first statistical distribution and the number of pixels determined to appear to be indicative of a fire; comparing, by the server, the likelihood to a threshold; and in response to determining that the likelihood does not exceed the threshold, obtaining, by the server, another satellite image from the first satellite image corresponding to another day before the date on which the fire occurred, to determine the start date.

20. A non-transitory computer-readable medium storing software including instructions executable by one or more computers, the instructions, when executed, causing the one or more computers to: obtaining, by the server, dates on which fires occurred within a geographic region; obtaining, by the server, a first satellite image of the geographic area prior to the date the fire occurred within the geographic area; selecting, by the server, a first set of images from the first satellite imagery prior to the date the fire occurred within the geographic region; generating, by the server, a first statistical distribution from the first set of images; determining, by the server, a start date of the fire based on a comparison of the first satellite image and the first statistical distribution; acquiring, by the server, second satellite images of the geographic region before the determined start date and after the determined start date of the fire within the geographic region; selecting, by the server, a second set of images from the second satellite imagery prior to the start date of the fire in the geographic region, wherein an amount of the second set of images is less than an amount of the first set of images; generating, by the server, a second statistical distribution from a second set of images prior to the start date of the fire in the geographic region; determining, by the server, an end date of the fire based on a comparison of the second satellite image generated after the start date with the second statistical distribution; providing, for output by the server, a range including the start date of the fire and the end date of the fire within the geographic region.

Images