Asset-level vulnerability and mitigation

Abstract

The aim is to gain insight into the hazard vulnerability of the parcel / property. [Solution] A method, system, and apparatus for receiving a request for a damage propensity score for a parcel, receiving imaging data for the parcel, including street view imaging data for the parcel, extracting characteristics of vulnerability features of the parcel from the imaging data using a machine learning model including multiple classifiers, determining a damage propensity score for the parcel from the characteristics of vulnerability features using the machine learning model, and providing a representation of the damage propensity score for display.

Description

[Technical field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority to U.S. patent application Ser. No. 17 / 158,585, entitled “ASSET-LEVEL VULNERABILITY AND MITIGATION,” filed Jan. 26, 2021, the disclosure of which is incorporated herein by reference. [Background technology]

[0002] Wildfires are becoming more of a problem as land development encroaches on wildland-urban interfaces and environmental changes result in extended droughts. Insurers and risk assessment managers look at the various assets that exist on a parcel and generate wildfire risk assessments using regression techniques and known vulnerabilities. Generating a risk assessment for a parcel can require asset inspections and other one-time static assessments, and as a result, changes to the parcel can create the need for costly updated reassessments to stay aware of current risks. Summary of the Invention

[0003] This specification describes systems, methods, devices, and other techniques related to using machine learning to gain insight into the hazard vulnerability of an area / asset from imaging data captured of the area / asset.

[0004] In general, one innovative aspect of the subject matter described herein can be embodied in a method of receiving a request for a damage propensity score for a parcel and receiving imaging data for the parcel, the imaging data including street view imaging data for the parcel. A machine learning model including a plurality of classifiers extracts characteristics of a plurality of vulnerability features of the parcel from the imaging data and determines a damage propensity score for the parcel from the characteristics of the plurality of vulnerability features. A representation of the damage propensity score is provided for display.

[0005] These and other implementations can each optionally include one or more of the following features: In some embodiments, the method further includes generating a set of partitions of characteristics from the characteristics of the plurality of vulnerability characteristics.

[0006] In some embodiments, the method further includes generating a three-dimensional model of the compartment from the characteristics of the plurality of vulnerability features and the imaging data of the compartment.

[0007] In some embodiments, the imaging data for the compartment includes imaging data captured within a threshold time from the time of the request.

[0008] In some embodiments, receiving the request for the damage propensity score includes receiving hazard event data for the hazard event and determining a damage propensity score for the partition for the hazard event from the characteristics of the plurality of vulnerability features and the hazard event data for the hazard event. The method may further include receiving updated hazard event data for the hazard event and determining an updated damage propensity score for the partition for the hazard event from the characteristics of the plurality of vulnerability features, the hazard event data, and the updated hazard event data.

[0009] In some embodiments, the method further includes determining, with the machine learning model, one or more mitigation steps for the compartment, determining, with the machine learning model, an updated damage propensity score based on the one or more mitigation steps, and providing a representation of the one or more mitigation steps and the updated damage propensity score, wherein the one or more mitigation steps can include adjustments to characteristics of the plurality of vulnerability features extracted from the imaging data.

[0010] In some embodiments, determining the one or more mitigation steps further comprises iterating an updated injury propensity score determination based on the adjusted characteristics of the plurality of vulnerability features. In some embodiments, determining the updated injury propensity score further comprises determining that the updated injury propensity score satisfies a threshold injury propensity score.

[0011] In some embodiments, determining one or more mitigation steps includes determining one or more mitigation steps for a particular type of hazard event, wherein the one or more mitigation steps for a first type of hazard event are different from the one or more mitigation steps for a second type of hazard event.

[0012] In some embodiments, the method further includes generating training data for the machine learning model, the generating including: receiving, for a hazard event, a plurality of parcels located within a vicinity of the hazard event, where each parcel of the plurality of parcels has received at least a threshold exposure to the hazard event; receiving, for each parcel of the plurality of parcels, imaging data for the parcel including street view imaging data; and extracting from the imaging data characteristics of a plurality of vulnerability features of a first subset of the parcels of the plurality of parcels that did not burn during the hazard event and a second subset of the parcels of the plurality of parcels that burned; and providing the training data to the machine learning model.

[0013] In some embodiments, extracting the characteristics of the plurality of vulnerability features includes providing the imaging data to a plurality of classifiers. Extracting the characteristics of the plurality of vulnerability features can include identifying a plurality of objects in the imaging data by the plurality of classifiers.

[0014] In some embodiments, the method further includes receiving additional structural characteristics for each compartment of the plurality of compartments, extracting from the additional structural characteristics a second set of vulnerability features for a first subset of compartments of the plurality of compartments that did not burn during the hazard event and a second subset of compartments of the plurality of compartments that did burn, and providing the second set of vulnerability features to the machine learning model.

[0015] In some embodiments, the additional structural characteristic includes post-hazard event inspection of multiple compartments.

[0016] The present disclosure also provides a non-transitory computer-readable storage medium coupled to one or more processors and storing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations in accordance with implementations of the methods provided herein.

[0017] It will be understood that the methods and systems according to the present disclosure can include any combination of the aspects and features described herein, i.e., the methods and systems according to the present disclosure are not limited to the combinations of the aspects and features specifically described herein, but include any combination of the aspects and features provided.

[0018] Certain embodiments of the subject matter described herein can be implemented to achieve one or more of the following advantages: An advantage of the present technology is that it can develop a novel understanding of hazard vulnerability for a substantially larger number of vulnerability features than conventional methods using trained machine learning models that consider the configuration of characteristics of the vulnerability features depending on a particular set of hazard conditions and degrees of exposure, which may be more complex than the sum of risk factors and may reflect non-obvious features that contribute to the degree of damage suffered or damage / no damage outcome. Hazard vulnerability assessments for specific hazards and degrees of exposure can be determined for parcels using imaging data and may not require additional property inspections. Hazard vulnerability assessments can be utilized in determining property valuations, sales, taxes, etc.

[0019] Utilizing street view imagery of parcels can provide access to unique features of parcels that are not otherwise available using other imaging data, such as features reflecting the current state of the house on the parcel (e.g., vines growing on the side of the house, location of the car parked in the driveway). For example, vulnerability propensity scores can be determined under real-time hazard conditions where mitigation responses can be updated as hazard conditions change to identify vulnerable parcels based on the respective vulnerability of each parcel under current conditions of the hazard event. Optimized mitigation steps, such as risk reduction plans and / or cost-benefit estimates, can be determined in an iterative process by the trained machine learning model based on the extracted characteristics of the parcel's vulnerability features and in response to the hazard event.

[0020] Applications of the technology generally include insurance risk assessment, real-time risk assessment and response, and general natural disaster hazard assessment and mitigation. More specifically, the technology can be utilized by local, state, or national governments to more accurately perform risk assessments and design and enact risk mitigation plans.

[0021] The details of one or more embodiments of the subject matter described 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 description of the drawings]

[0022] [Figure 1] FIG. 1 is a block diagram of an example operating environment for a hazard vulnerability system 102. [Figure 2A] 1 shows satellite / air-based imagery including multiple example plots before and after a hazard event. [Figure 2B] 1 illustrates street view-based imagery of an example section before and after a hazard event. [Figure 2C] 13A-13C show street view-based images of another example section before and after a hazard event. [Diagram 3] FIG. 1 is a flow diagram of an example process of a hazard vulnerability system. [Figure 4] FIG. 13 is a flow diagram of another example process of a hazard vulnerability system. [Diagram 5] FIG. 1 is a block diagram of an exemplary computer system.

[0023] Like reference numbers and designations in the various drawings indicate like elements. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0024] overview The technology of this patent application is directed to using machine learning to gain insight into the hazard vulnerability of an area / asset from imaging data capturing the area / asset.

[0025] More specifically, the technology of the present application utilizes trained machine learning models to identify vulnerability features and identify characteristics of the vulnerability features in imaging data for parcels, such as street view imaging data, LIDAR data, high resolution satellite imagery, aerial imagery, infrared imagery, user-provided imagery, etc. The characteristics of the vulnerability features can be utilized to generate vulnerability propensity scores and / or identify mitigation strategies to reduce hazard vulnerability for particular parcels depending on the particular hazards and degree of exposure.

[0026] Generating training data for training the machine learning model can include selecting a set of parcels located within a vicinity of a hazard event, e.g., within a radius of a burn scar. The hazard event can be, e.g., a wildfire, a flood, a tornado, etc., and each of the set of parcels experiences some degree of exposure to the hazard event. For each parcel in the set of parcels, imaging data is collected that captures the parcel before the hazard event, e.g., a photo of the home / property before experiencing a wildfire.

[0027] Vulnerability features of a compartment can be defined using existing risk assessment data, such as defensible space, architectural structure, or other features known to be associated with increased / decreased hazard vulnerability. Vulnerability features can be further extracted from imaging data indicative of a compartment having a damaged / non-damaged and / or damaged outcome for a particular hazard event, and one or more neural networks can be utilized to process the imaging data and extract additional vulnerability features determined to distinguish between damaged / non-damaged and / or damaged outcomes of the compartment.

[0028] Properties of vulnerability features, such as materials of roof construction, distance between tree and house, manufacturing information of building materials, frontage, type of fence, irrigation, etc., can be extracted from the imaging data of parcels using multiple classifiers and utilizing object recognition techniques. Training data can be generated for multiple sets of parcels and their respective hazard events and can include extracted properties of vulnerability features, location of parcels relative to the hazard events, degree of exposure / damage during the hazard events, and damage / non-damage outcomes. Additionally, public records of parcels, information about hazard events, etc. can be utilized in generating training data for training the machine learning model.

[0029] The trained machine learning model can receive a request for a vulnerability assessment of a particular parcel and a request for a hazard event including the degree of exposure. Imaging data can be collected for the parcel, e.g., using the parcel's known address, geographic location, etc., and can include only imaging data captured within a threshold of time, e.g., collected within the past six months. The imaging data can reflect the current state of the parcel, e.g., the current state of vegetation surrounding the parcel, the location of vehicles, structures (e.g., sheds) built within the parcel, etc. The machine learning model can receive imaging data, public records (e.g., construction year, setbacks, variance permits, etc.), and other relevant geospatial information (e.g., neighborhood residential density, distance to fire / emergency services, distance to major roads, etc.) as inputs and extract characteristics of the parcel's vulnerability characteristics. The determined vulnerability propensity score can be provided as an output.

[0030] In some embodiments, the model can determine mitigation steps to reduce the risk score for a parcel based on the parcel's vulnerability characteristics. Determining mitigation steps by the machine learning model can include identifying characteristics of the vulnerability characteristics that can be adjusted (e.g., trimming vegetation, changing roof material, changing siding material), and iterating the risk score determination based on the adjusted characteristics of the vulnerability characteristics. Permutations of mitigation steps can be evaluated for various hazard event scenarios to provide an optimized subset of mitigation steps for a particular parcel.

[0031] In some embodiments, real-time hazard vulnerability can be determined for a particular parcel based on a real-time hazard event or potential future hazard risk, e.g., an ongoing wildfire, a drought occurrence, severe weather patterns, etc. As the hazard event evolves, e.g., as the degree of exposure changes, the hazard vulnerability of the parcel can be updated and real-time alerts can be generated accordingly, e.g., to inform homeowners or emergency responders of the real-time hazard vulnerability and / or countermeasures.

[0032] Example Operating Environment 1 is a block diagram of an example operating environment 100 for a hazard vulnerability system 102. The hazard vulnerability system 102 can be hosted on one or more local servers, a cloud-based service, or a combination thereof.

[0033] The hazard vulnerability system 102 can be in data communication with a network, which can be configured to enable the exchange of electronic communications between devices connected to the network. The network can include, for example, one or more of the Internet, Wide Area Networks (WAN), Local Area Networks (LAN), analog or digital wired and wireless telephone networks (e.g., public switched telephone network (PSTN), Integrated Services Digital Network (ISDN), cellular networks, and Digital Subscriber Line (DSL)), radio, television, cable, satellite, or any other distribution or tunneling mechanism for carrying data. The network can include multiple networks or sub-networks, each of which can include, for example, wired or wireless data paths. The network can include a circuit-switched network, a packet-switched data network, or any other network capable of carrying electronic communications (e.g., data or voice communications). For example, the network may include a packet-switched network based on Internet protocol (IP), asynchronous transfer mode (ATM), PSTN, IP, X.25, or frame relay, or a network based on other equivalent technologies, and may support voice, for example, using VoIP, or other equivalent protocols used for voice communications. The network may include one or more networks that include wireless data channels and wireless voice channels. The network may be a wireless network, a broadband network, or a combination of networks that include wireless and broadband networks.

[0034] The hazard vulnerability system 102 includes a training data generator 104 and a damage propensity model 106. Optionally, the hazard vulnerability system 102 includes a mitigation engine 108. Although described herein with reference to the training data generator 104, the damage propensity model 106, and the mitigation engine 108, the operations described may be performed by more or fewer subcomponents.

[0035] The training data generator 104 includes a vulnerability feature extractor 110, a compartment hazard event module 112, and a vector generator 114. The training data generator 104 receives imaging data 116 from a repository, such as satellite and / or aerial imagery 118, street view imagery 117, etc., and provides as output training data 120. The output training data 120 can be utilized to train the damage propensity model 106.

[0036] The damage propensity model 106 includes multiple classifiers 107, e.g., one or more neural networks or machine learning models, e.g., random forests. The classifiers may be configured to classify the damage propensity as a binary outcome, e.g., damaged or not damaged, or may be configured to classify the degree of damage propensity, e.g., using a regression task. In some embodiments, the classifiers may be utilized to estimate damage to specific subcomponents of the compartment, e.g., a building roof, a building siding, etc., to further refine the damage propensity model 106.

[0037] The damage propensity model 106 can receive training data 120 including a number of training vectors generated using the imaging data 116 and a large sample of different hazard events documented in the historical hazard event data 124. The damage propensity model 106 can be trained to make inferences about the damage propensity of a particular compartment based in part on the characteristics of the compartment's vulnerability features. The multiple classifiers 109 can include, for example, the same or different sets of classifiers described with reference to the classifier 107, and can process the received imaging data 116 to identify and label characteristics of the vulnerability features extracted from the imaging data 116.

[0038] Satellite / aerial imagery 118 includes any imagery that captures a geographic region and provides information about the geographic region. Information about the geographic region may include, for example, information about one or more areas located in the geographic region, such as structures, vegetation, terrain, etc. Satellite / aerial imagery may be, for example, Landsat imagery, or other forms of aerial imagery. Satellite / aerial imagery 118 may be, for example, RGB imagery or hyperspectral imagery. Satellite / aerial imagery 118 may be captured using satellite technology, such as Landsat, or drone technology. In some implementations, satellite / aerial imagery may be captured using other high altitude technology, such as drones, weather balloons, airplanes, etc. In some embodiments, synthetic aperture radar (SAR) imagery may be utilized in addition to satellite imagery as described herein.

[0039] In some implementations, satellite or other aerial imagery may be captured using radar-based imaging, such as LIDAR imagery, RADAR imagery, or another type of imaging using the electromagnetic spectrum, or a combination thereof. Satellite / aerial imagery 118 may include images of geographical areas including various natural features, including different topography, vegetation, bodies of water, and other features. Satellite / aerial imagery 118 may include images of man-made developments, such as housing constructions, roads, dams, retaining walls, etc.

[0040] Street view imagery 117 includes any imagery that captures aspects of one or more parcels from a frontage perspective, for example, captured from a road or sidewalk facing the parcel. In some embodiments, street view imagery 117 may be captured by one or more cameras mounted on a vehicle and configured to capture street view imagery 117 of the parcel as the vehicle passes by the parcel. Optical and LIDAR street view imagery 117 may be utilized to capture depth information about the parcel. In some embodiments, street view imagery 117 may have a high spatial resolution, for example, street view imagery may have a spatial resolution of less than one centimeter.

[0041] In some embodiments, the street view image 117 may be captured by a user, e.g., a homeowner of a parcel, from a frontage view of the property. The street view image may be captured using a built-in camera on a smart device, e.g., a smartphone or tablet, and / or may be captured using a handheld camera. In some embodiments, the street view image may be captured by a land surveyor, insurance adjuster, parcel appraiser, or other person documenting aspects of a parcel.

[0042] In some embodiments, the street view image 117 may include additional views of the parcel, such as views captured from the side of the parcel, the rear of the parcel, etc. For example, a user may capture a street view image 117 of the back yard of their property, or a side view of their property.

[0043] In some embodiments, the imaging data 116 may include images of a plot before and after a hazard event, e.g., before and after a wildfire. The imaging data 116 associated with a particular hazard event may include a burn scar, e.g., an area damaged by the hazard event. Further description of the imaging data 116 is provided with reference to Figures 2A-2C.

[0044] The training data generator 104 receives parcel data 122 from a repository of historical hazard event data 124. The parcel data 122 may include, for example, insurance assessments, land surveys, appraisals, building / construction records, code inspections / violations, and other public records.

[0045] In some embodiments, parcel data 122 includes public records from post hazard event damage reports, such as from a Damage Inspection (DINS) database maintained by CalFIRE, from post hazard insurance inspections, etc. The post hazard event damage reports can include direct inspections of structures exposed to a hazard event, such as a wildfire, and can include information regarding the vulnerability characteristics of the structures, such as roof type, eave type, construction material, etc., as well as damage levels. Parcel data 122 can include damaged / non-damaged data and / or damage severity data for hazard events for parcels within a radius of the hazard event, such as within a burn radius.

[0046] The training data generator 104 receives as input imaging data 116. The imaging data 116 captures one or more parcels at a particular location, e.g., one or more homes, and at a particular time, e.g., before or after a hazard event. For example, a street view image 117 may capture a home at a particular street address and a first date / time, e.g., before a hazard event.

[0047] The vulnerability feature extractor 110 can include multiple classifiers 107 that can identify vulnerability features, e.g., objects, in the imaging data 116. For each parcel shown in the imaging data 116, the vulnerability feature extractor 110 can extract vulnerability features and provide parcel vulnerability features F1, F2, ... FN as output to the vector generator module 114. Continuing with the above example, vulnerability features F1, F2, ... FN are extracted for a house at a particular street address and a first date / time, e.g., roof structure, vegetation, frontage distance, land slope, etc.

[0048] Vulnerability features may include, but are not limited to, building materials, defensible spaces, slope of the lot, proximity to roads, etc. Vulnerability features may further include objects, e.g., trees, vehicles, etc. In some embodiments, ground truth labeling may be utilized to identify vulnerability features, e.g., by human experts or in an automated / semi-automated manner. Vulnerability features utilized by insurance adjusters / risk assessment managers may be identified in imaging data 116.

[0049] The vulnerability features may further include vulnerability features of the parcel captured in the imaging data 116 that were not traditionally identified as risk hazards by insurance adjusters / risk assessments. In other words, features of the parcel that may not traditionally be labeled as hazard risks may be extracted as possible vulnerability features to generate training data. For example, the structure of the driveway, the distance between the parked car and the house, the type of grass seed used in the lawn, etc. The damage propensity model 106 may define vulnerability features extracted from the imaging data 116 that may not otherwise be predicted as significant by traditional means, for example, as significant, so that the vulnerability features may be processed by a machine learning model to determine which of the possible vulnerability features are significant in the damage propensity of the parcel, for example, have a statistical effect on the damage / non-damage outcome and / or damage severity outcome. In this manner, novel and non-obvious features may be identified as having significance in damage propensity.

[0050] Each vulnerability feature of a parcel describes an aspect of the parcel as shown in the imaging data 116, e.g., the street view imagery 117 and / or the satellite / aerial imagery 118. For each vulnerability feature extracted from the imaging data 116, one or more characteristics C1, C2, ... CN of the vulnerability feature are extracted, e.g., F1{C1, C2, ... CN}. Further details of the feature extraction are described with reference to Figures 2A-2C. The characteristics of the vulnerability feature may include quantifiable and / or qualifiable aspects of the vulnerability feature. For example, a vulnerability feature that is a roof structure may be characterized by construction material, shingle spacing, age, and roof maintenance. In another example, a vulnerability feature that is a parcel slope may be characterized by a slope measurement of 0.5°.

[0051] The partition hazard event module 112 receives as input historical hazard event data 124 including a record of past hazard events, and partition data 122 for each partition shown in the imaging data 116 processed by the vulnerability feature extractor 110. The partition hazard event module 112 provides the partition data 122 to the vector generator 114.

[0052] The historical hazard event data 124 can include times of hazard events, such as a start time and an end time of a hazard event, such as a start time when a wildfire begins and an end time when a wildfire is fully contained or fully extinguished. The historical hazard event data 124 can include the geographic location, such as GPS coordinates, of an impact area affected by a hazard event, such as an area that includes a burn scar.

[0053] Parcel data 122 for a particular parcel can include official records of the parcel before and after a hazard event, e.g., before and after a wildfire. For example, parcel data 122 can include post-hazard insurance / assessment records, e.g., damage assessments from after a hazard event. In other words, parcel data 122 can include parcel damage / non-damage outcomes, e.g., damaged vs. non-damage, for a particular hazard event. Parcel hazard event module 112 can utilize the damage / non-damage outcomes and / or damage severity outcomes of multiple parcels located within a burn radius 208 of a hazard event as ground truth for training data 120.

[0054] In some embodiments, home parcel data 122 may include structural characteristics of the parcel collected prior to the hazard event, such as building records, building materials, roof type, and the like.

[0055] The training data generator 104 can generate training data from images of parcels using imaging data 116 occurring before and after a hazard event, and from parcel data 122 for parcels from historical hazard event data 124 corresponding to before and after the event, e.g., a wildfire. The vulnerability feature extractor 110 can extract vulnerability features and associated characteristics of parcels in the imaging data 116 that each appear within a radius of the hazard event, e.g., within a distance of a burn scar.

[0056] The vector generator 114 receives the extracted vulnerability features and characteristics of the vulnerability features from the feature extraction module 110, and optionally receives as input the partition data 122 from the partition hazard event module 112 for a particular partition. In some embodiments, the partition data 122 may be used as ground truth for the training data 120, e.g., the partition data 122 including damage / non-damage and / or damage severity results for a hazard event may be used to label a partition as either "burned" or "non-burned."

[0057] The vector generator 114 can generate training data 120 from the extracted vulnerability features, characteristics of the vulnerability features, and partition data 122 for each partition, e.g., a respective training vector V. Further details of the generation of the training data are described with reference to FIG.

[0058] The damage tendency model 106 can receive the training data 120 as an input to use the training data 120 to train a machine learning model, e.g., the damage tendency model 106. In some implementations, the damage tendency model 106 can be trained using a number of training vectors generated using a large sample of different locations and compartment data representing a variety of historical hazard events. In one example, the training data 120 provided to the damage tendency model 106 can include thousands of compartments subjected to many different hazard events.

[0059] The hazard vulnerability system 102 receives a request 126 from a user of a user device 128. The user device 128 may include, for example, a mobile phone, a tablet, a computer, or another device including an operating system 129 and an application environment 130 through which the user may interact with the hazard vulnerability system 102. In one embodiment, the user device 128 is a mobile phone including an application environment 130 configured to display a view 132 including at least a portion of a parcel. In one embodiment, as shown in FIG. 1, the application environment 130 displays the view 132 including a street view of a house and surrounding assets, such as, for example, trees, bushes, shrubs, etc.

[0060] The request 126 may include a location of the parcel specified by a user of the user device 128. The location of the parcel may include a geographic location, e.g., GPS coordinates, a street address, etc., and may be entered by the user into the application environment 130.

[0061] The request 126 may further include a request for a damage propensity score, e.g., a request for a relative vulnerability to a hazard event, and the request 126 may specify a particular hazard event, e.g., a particular real-time hazard event, or may specify a general type of hazard event, e.g., wildfire, flood, earthquake, etc. In one example, a user may submit a request 126 specifying a street address of a parcel and request a damage propensity score for that parcel to a real-time hazard event, e.g., an occurring wildfire. In another example, a user may submit a request 126 specifying a location of the parcel including GPS coordinates and request a flood-specific damage propensity score for the parcel.

[0062] In some embodiments, the damage propensity score may be a relative measure of risk for a particular parcel to be damaged by a hazard event. The damage propensity score may be a general measure of risk for a particular parcel to be damaged by a certain type of hazard event, such as a wildfire, or may be a specific measure of risk for a particular parcel to be damaged by a particular hazard event, such as a real-time flood event.

[0063] In some embodiments, the damage propensity score can include a percent loss, e.g., percent damage, for a given parcel under a particular hazard scenario. For example, the damage propensity score can be a 10% loss for a particular parcel under a particular wildfire scenario.

[0064] In some embodiments, an end user, e.g., a property owner, insurance adjuster, government official, etc., can provide a complete probabilistic hazard model, i.e., a distribution of hazard characteristics and associated probabilities, so that the hazard vulnerability system can provide an expected average annual loss (AAL) for a particular parcel.

[0065] In some embodiments, the request 126 may specify a location that includes multiple parcels, e.g., a neighborhood, a street that includes multiple homes, a complex that includes multiple buildings, etc. A user may be interested in determining individual damage propensity scores for each structure in a location that includes multiple parcels, or may be interested in determining a global damage propensity score for multiple parcels.

[0066] In some embodiments, the hazard vulnerability system 102 receives as input a request 126 including a request for mitigation steps to reduce the hazard vulnerability of a parcel. The mitigation engine 108 can receive the request for mitigation steps and identify a set of mitigation steps 136 that a user can take to reduce the hazard vulnerability of the parcel based on the imaging data 116 and the damage propensity score 140. Mitigation steps are quantifiable and / or quantifiable measures that can be taken by a user, e.g., a homeowner, to reduce the damage propensity score 140. Mitigation steps can include, for example, removal / reduction of vegetation adjacent to a structure, building materials used for the structure, etc. For example, a mitigation step can be trimming leaves within a two-foot radius surrounding a house. In another example, a mitigation step can be changing the siding material of a house. In yet another example, a mitigation step can be digging an irrigation ditch to collect runoff from a flooded creek area.

[0067] In some embodiments, the mitigation steps may be provided within the application environment 130 on the user device 128, and the mitigation steps 136 may be visually identified, e.g., an indicator may be overlaid on the view 132 of the lot. For example, a tree with branches overhanging a roof may be visually identified within the application environment, e.g., with a box surrounding the tree and / or the branches.

[0068] In some embodiments, the mitigation engine 108 can receive real-time event data 142, e.g., real-time data about an occurring hazard event, and update the mitigation step 136 in real-time to provide a user with a real-time response to the hazard event. The real-time event data 142 can include, e.g., hazard spread, weather patterns, mitigation events, emergency response, etc. For example, real-time event data 142 for a wildfire can include real-time perimeter of the fire, percent of control by firefighters, evacuation data, wind advisories, etc. In another example, real-time event data 142 for a flood can include real-time river / creek levels, flood levels, rain / weather forecasts, evacuation data, etc.

[0069] Feature Extraction As described above with reference to FIG. 1, the vulnerability feature extractor 110 can receive imaging data 116 including satellite / aerial imagery 118 and street view imagery 117 as inputs and extract vulnerability features with associated characteristics. FIG. 2A is a schematic diagram of an example pair of satellite images including multiple segments before and after a hazard event. The satellite images 200a and 200b are captured at capture times T1 and T2, respectively, where T1 is a time occurring before the hazard event and T2 is a time occurring after the hazard event. Additionally, T1 and T2 can be selected based on the segment data 122 of the segments included in the satellite images 200a, 200b, where the segment data includes records of the segments at a time T1′ before the hazard event and a time T2′ after the hazard event.

[0070] Satellite images 200a and 200b show the same geographic area 202 that includes a set of segments 204. Satellite image 200a is captured at a time T1 that occurs within a first threshold of time before the start of a hazard event, and satellite image 200b is captured at a time T2 that occurs within a second threshold of time after the end of the hazard event.

[0071] The satellite image 200b captured after the hazard event includes a burn scar 206 resulting from the hazard event, e.g., a fire. The burn scar 206 may indicate an area of ​​the geographic region 202 that was damaged / affected by the hazard event. The burn radius 208 defines a perimeter that encompasses the burn scar 206, surrounds the burn scar 206, and includes additional areas that buffer the burn scar. The burn radius 208 may include additional radii extending outward from the burn scar, e.g., an additional 100 feet, an additional 1000 feet, and an additional 5000 feet. The burn radius 208 may include a section A that was damaged / affected by the hazard event and a section B that was not damaged / affected by the hazard event. Section A may be a section located within the burn scar 206 and that was damaged / affected by the hazard event. Section B may be a section located outside the burn scar 206 but within the burn radius 208, or section B may be a section located within the burn scar 206 but not damaged / affected by the hazard event.

[0072] A satellite image 200b captured after a hazard event may include multiple burn scars 206 and burn radii, where the burn scars and / or burn radii may overlap one another, and a section may be located in the overlapping area of ​​the burn scars and / or the overlapping area of ​​the burn radii.

[0073] As described with reference to Fig. 1, the vulnerability feature extractor 110 receives the satellite images 200a, 200b and extracts vulnerability features F1, F2, ... FN from the images 200a, 200b and respective characteristics of the vulnerability features. The vulnerability features extracted from the satellite images 200a, 200b may include, for example, roof structure, location of the parcel relative to natural formations (e.g., forest / tree coverage, waterways, etc.), location of the parcel relative to man-made features (e.g., roads, irrigation ditches, farmland, etc.). The respective characteristics may include, for example, building materials for the roof structure, e.g., ceramic, metal, wood, etc. In another example, the characteristics of the parcel's location relative to the man-made features may include the relative distance between the parcel and the man-made features, e.g., the distance from the house to the street.

[0074] In some embodiments, vulnerability features may be extracted from satellite imagery 118 for multiple parcels appearing in the satellite imagery. Additional vulnerability features may be extracted for each of the parcels of the multiple parcels using higher resolution imagery, for example using street view imagery 117.

[0075] 2B is a schematic diagram of an example pair of street view images of parcel A captured before and after a hazard event. Street view images 220a and 220b show street views captured from street level and facing the parcel, for example, of a house 222 and surrounding vegetation 224 located on parcel A. Street view image 220b, captured after the hazard event, includes hazard event damage 226, for example, to the house, an adjacent storage unit, and nearby trees.

[0076] Vulnerability features are extracted from the street view images 220a, 220b. As described with reference to Figure 1, the vulnerability feature extractor 110 receives imaging data 116 including street view images 117, e.g., images 220a, 220b, and extracts vulnerability features F1, F2, ...FN. Referring back to Figure 2B, the vulnerability features extracted from the street view images 220a, 220b include a group of trees (F1), a group of bushes (F2), an outhouse (F3), and a roof structure (F4) of the house 222. More or fewer vulnerability features may be extracted from the street view images 220a, 220b, and the examples provided are not limiting.

[0077] For each vulnerability feature extracted from the street view images 220a, 220b, the vulnerability feature extractor identifies characteristics of the respective vulnerability feature. For example, a group of trees F1 may have associated quantifiable characteristics, such as the distance of the trees to the house 222, the number of trees, the height of the trees, the closeness of the group of trees, etc., and quantifiable characteristics, such as the health of the trees, being covered with ivy, etc. In another example, a roof structure F4 may have associated characteristics, such as building material, eaves structure, roof slope, age of the roof, roof maintenance, fullness of the gutters, etc.

[0078] In some embodiments, vulnerability feature extractor 110 can identify vulnerability features in section A, e.g., street view image 220b, that reflect damage from a hazard event. In other words, vulnerability feature extractor 110 can focus on vulnerability features whose characteristics reflect damage as a result of a hazard event, e.g., roof structures showing burning / smoke damage, burnt trees, etc.

[0079] Figure 2C is a schematic diagram of an example pair of street view images of parcel B captured before and after a hazard event. Street view images 240a and 240b show street views captured from street level and facing the parcel, for example, of a house 242 and surrounding vegetation 224 located on parcel B, which is shown outside the burn scar 206 of the hazard event and within the burn radius 208. Unlike parcel A described with reference to Figure 2B, parcel B in Figure 2C is shown as having sustained no damage from the hazard event.

[0080] Similarly, as described with reference to Figure 2B, vulnerability features extracted from the street view images 240a, 240b include a group of trees (F5), a group of bushes (F6), the frontage space between the house and the street (F7), and the roof structure (F8) of the house 246. More or fewer vulnerability features may be extracted from the street view images 240a, 240b and the examples provided are not limiting.

[0081] For each vulnerability feature extracted from the street view images 240a, 240b, the vulnerability feature extractor identifies characteristics of the respective vulnerability feature. For example, a group of trees F5 may have associated quantifiable characteristics, such as the distance of the trees to the house 242, the number of trees, the height of the trees, the closeness of the group of trees, etc., and quantifiable characteristics, such as the health of the trees, being covered with ivy, etc. In another example, the frontage space F7 between a house and a street may have characteristics including, for example, distance, slope, type of land cover (e.g., cement vs. grass), etc.

[0082] The vulnerability feature extractor 110 provides vulnerability features and characteristics extracted from the imaging data 216, for example, 200a, 200b, 220a, 220b, 240a, 240b, to the vector generator 114 to generate the training data 120.

[0083] Example Process 3 is a flow diagram of an example process for the hazard vulnerability system 102. The system 102 receives 302 a request for a damage propensity score for a parcel. The request 126 may be provided to the hazard vulnerability system 102 by a user via a graphical user interface of the application environment 130. The request 126 may include a request for a damage propensity score 140 for a particular parcel and may further include a request for one or more mitigation steps 136 to reduce the damage propensity score 140. The request 126 may further specify a particular hazard event, such as an ongoing wildfire, or a general hazard event type, such as a flood, and request a damage propensity score accordingly.

[0084] The system receives 304 imaging data for the parcel, including street view imaging data for the parcel. The hazard vulnerability system 102 can receive 116 imaging data, including street view images 117, from a repository of imaging data. Each image can include a particular designated parcel that is of interest to the user. In some embodiments, the user can capture additional street view images of the parcel with a camera, such as the built-in camera of the user device 128, and upload them to the hazard vulnerability system 102.

[0085] In some embodiments, the system receives imaging data for the compartment that was captured within a threshold time from the time of the request 126, for example, within six months, within two weeks, within one hour, etc.

[0086] The machine learning model including the classifiers extracts 306 characteristics of vulnerability features of the parcel from the imaging data. The damage propensity model 106 can receive the imaging data 116 and extract characteristics of vulnerability features using a number of classifiers 107. The vulnerability feature extractor 110 receives images of the parcel from the imaging data 116 and extracts vulnerability features F1, F2, ... FN of the parcel, and for each vulnerability feature FN, the vulnerability feature extractor 110 extracts one or more characteristics C1, C2, ... CN of the vulnerability feature FN. In one embodiment, the vulnerability feature extractor 110 receives street view images of the parcel capturing the house and surrounding assets and uses a number of classifiers to identify vulnerability features including, for example, the house, vegetation, frontage area, fences, etc. The vulnerability feature extractor 110 identifies characteristics of each of the extracted vulnerability features, for example, building materials used for the house, for example, siding type, roof type, eaves structure, etc.

[0087] The machine learning model determines 308 a propensity score for the parcel from the characteristics of the vulnerability features. The damage propensity model 106 is trained on training data 120 including multiple hazard events, e.g., hundreds of hazard events, and multiple parcels for each hazard event, e.g., thousands of parcels, as described in more detail below with reference to FIG. 4, such that the model 106 can infer between the characteristics of the parcel's vulnerability features and the parcel's propensity for damage. The model 106 generates an damage propensity score 140 as an output.

[0088] The system provides 310 a representation of the propensity score for display. The injury propensity score 140 can be provided within a graphical user interface of the application environment 130. The propensity score 140 can be represented visually, for example, as a numerical value, and / or can be presented with contextual clues including, for example, a relative scale of hazards, color coding (high, medium, or low risk). The propensity score 140 can be presented with contextual information to allow a user to better understand the significance of the propensity score 140.

[0089] In some embodiments, the characteristics of the vulnerability features of the compartment and the imaging data may be utilized by the hazard vulnerability system 102 to generate a three-dimensional model of the compartment. The three-dimensional model of the compartment may be displayed within a graphical user interface to assist a user in understanding the injury propensity score and / or one or more mitigation steps 136 for optimizing the injury propensity score.

[0090] In some embodiments, the hazard vulnerability system 102 can determine one or more mitigation steps 136 to provide to the user as a way to reduce risk, e.g., optimize the damage propensity score 140. The mitigation engine 108 can receive the damage propensity score and the characteristics of the vulnerability features of the parcel and determine one or more mitigation steps 136. The mitigation steps 136 can be identified by the damage propensity model 106 based on an inference of which characteristics of the vulnerability features reduce damage propensity. The mitigation steps 136 can include, for example, pruning tree branches away from the house. In another example, the mitigation steps 136 can include changing roof construction materials, relocating an outhouse, removing brush from the frontage area of ​​the house, etc. The mitigation engine 108 can provide the proposed updated characteristics of the vulnerability features to the damage propensity model 106. The damage propensity model 106 can receive the proposed updated characteristics of the vulnerability features and determine an updated damage propensity score 140.

[0091] In some embodiments, the mitigation engine 108 may determine the mitigation step 136 by calculating the slope of the damage propensity score 140 for each vulnerability feature vector. The vulnerability feature vector with the threshold magnitude slope and / or the largest magnitude slope subset of the set of slopes may be utilized as the basis for the mitigation step 136. In other words, the vulnerability features with the greatest impact (larger magnitude slope) on the damage / non-damage and / or damage degree outcome may be the focus of the mitigation step since they may affect the damage propensity score 140 more than the vulnerability features with less impact (smaller magnitude slope) on the outcome. For example, a roof structural material may be selected as the mitigation step if the slope of the damage propensity score 140 for the vulnerability feature vector for the roof structural material is at least the threshold magnitude or is ranked highest among the slopes of the vulnerability feature vectors of the compartment.

[0092] In some embodiments, neural networks may be utilized to infer vulnerability features within the imaging data that may have the greatest impact on the injury propensity score 140 determined by the injury propensity model 106 .

[0093] In some embodiments, the graphical user interface of the application environment 130 can present a visual representation of the mitigation step 136 on the user device 128. For example, the visual indicator 138 can identify the mitigation step and can be indicated, for example, to remove vegetation from the lot. The visual indicator 138 can be a bounding box surrounding the identified mitigation step 136, a graphical arrow, or other indicator, or the like. The visual indicator 138 can include text-based information regarding the mitigation step 136 that describes, for example, how and why the mitigation step 136 reduces the damage propensity score 140 of the lot.

[0094] In some embodiments, the system 102 can perform an optimization process by iterating the proposed characteristics for the vulnerability characteristics of the compartment until an optimized, e.g., lowest, damage propensity score is found and calculating an updated propensity score 140. The optimization process can continue until a threshold damage propensity score is met.

[0095] In some embodiments, the hazard vulnerability system 102 can further incorporate a cost analysis of the mitigation steps 136. In other words, the hazard vulnerability system 102 can determine a mitigation step 136 that balances optimizing the injury propensity score 140 while keeping the cost of the mitigation step 136 below a threshold cost.

[0096] In some embodiments, the hazard vulnerability system 102 can determine one or more mitigation steps based on a particular type of hazard event, e.g., flood versus wildfire, where the mitigation steps for a first type of hazard event are different from the mitigation steps for a second type of hazard event. For example, mitigation steps corresponding to a flood hazard can include digging runoff trenches and updating gutter systems, while mitigation steps corresponding to a wildfire can include trimming vegetation surrounding a home.

[0097] In some embodiments, the hazard vulnerability system 102 receives a request 126 for a real-time damage propensity score 140 corresponding to a real-time hazard event. The hazard vulnerability system 102 can receive real-time event data 142 for the hazard event and determine a propensity score for the subdivision corresponding to the hazard event from the characteristics of the vulnerability features and the real-time event data. The hazard vulnerability system 102 can re-evaluate the propensity score 140 in real-time based in part on updated hazard event data for the hazard event, e.g., changing weather conditions, containment, etc., and provide the updated propensity score 140 to the user in response to the updated hazard event data.

[0098] In some embodiments, the mitigation engine 108 receives real-time event data 142. The real-time event data can be utilized to provide real-time mitigation steps 136 to a user via a user device 128 to reduce a parcel damage score 140 corresponding to an ongoing hazard event. For example, the real-time event data including wildfire spread and containment, weather patterns, and emergency responder alerts can be utilized to help homeowners take immediate steps to address wildfire spread and reduce the propensity of their property to be damaged by a wildfire.

[0099] 4 is a flow diagram of another exemplary process of a hazard vulnerability system. The hazard vulnerability system 102 generates training data for a machine learning model (402). Training a machine learning model, e.g., a damage propensity model 106, includes generating training data 120 including a large sample set of imaging data 116 and compartment data 122, e.g., historical hazard event data 124 including thousands of images for hundreds of hazard events. The generated training data represents various imaging conditions, e.g., weather conditions, lighting conditions, seasons, etc., for various hazard events, e.g., hazards of different scales, hazard spreads, locations, hazard types, so that the damage propensity model 106 trained on the training data 120 can be generalized to develop heuristics for a wide range of imaging conditions and hazard events.

[0100] The system receives 404 a plurality of compartments located within a vicinity of the hazard event, where each compartment of the plurality of compartments received at least a threshold exposure to the hazard event. The system may receive historical hazard event data 124 for the hazard event, where the historical hazard event data 124 may include compartment data 122 for each compartment of the plurality of compartments that was located within a vicinity of the hazard event, e.g., within a burn radius 208.

[0101] Proximity to a hazard event may be defined as being located within a burn radius 208 surrounding the burn scar 206. As shown in FIG. 2A, the burn radius 208 may define an extended area surrounding the burn scar 206. In some embodiments, proximity to a hazard event may be a threshold distance from the perimeter of the burn scar 206, for example, within 1 mile, within 100 feet, within 5 miles, etc.

[0102] The threshold exposure is a minimum amount of exposure to a hazard event by a parcel, and can be defined, for example, by the parcel's proximity to the hazard event, by the amount of time the parcel is actively exposed to the hazard event, for example, the amount of time the asset is actively exposed to a wildfire, etc. In some embodiments, the threshold exposure can be defined using emergency responder metrics, for example, high risk or evacuation zones considered by emergency responders. In one example, a parcel can meet the threshold exposure by being considered within an evacuation zone of a wildfire. In another example, a parcel can meet the threshold exposure by having flood waters (or a wildfire, or a tornado, or an earthquake, etc.) contact at least a portion of the parcel. In another example, a parcel can meet the threshold exposure by being located within a burn scar or within a threshold vicinity of a burn scar.

[0103] In some embodiments, each of a plurality of compartments located within a burn scar can be considered to have received a threshold amount of exposure.

[0104] In some embodiments, the fire radiative power (FRP) of a fire can be calculated from mapped remote sensing derived measurements of fire intensity. For example, using satellite data of active fires, any structure that is within a particular area and has a given threshold FRP value can be counted as experiencing the same threshold exposure.

[0105] The system receives (406) imaging data for the parcel, including street view imaging data, for each parcel of the plurality of parcels located within the vicinity of the hazard event. The imaging data 116 may include satellite / aerial imagery 118 and street view imagery 117 collected by the hazard vulnerability system 102 from repositories of collected imagery located in various databases and sources, for example. Each image in the imaging data 116 includes a capture time at which the image was captured and includes a geographic area that includes the parcel of the plurality of parcels. The satellite / aerial imagery 118 includes a geographic area captured at a particular resolution and includes location information, e.g., GPS coordinates, that defines the geographic area captured within a frame of the image. The street view imagery 117 includes a street level view of one or more parcels, e.g., one parcel, two parcels, etc., of the plurality of parcels captured at a particular resolution and includes location information, e.g., a street address, that defines the location of the parcel captured in the street view imagery 117.

[0106] The system extracts 408 a plurality of vulnerability feature characteristics from the imaging data for a first subset of compartments that did not burn during the hazard event and a second subset of compartments that burned during the hazard event. As described with reference to Figures 1 and 2A-2C, the vulnerability feature extractor 110 may receive the imaging data 116 and extract the vulnerability features using a plurality of classifiers. In some embodiments, extracting the vulnerability feature characteristics with the plurality of classifiers includes identifying objects in the imaging data 116 with the plurality of classifiers.

[0107] In some embodiments, the system can receive parcel data 122 for a first subset of parcels that did not burn during the hazard event and a second subset of parcels that did burn. The parcel data 122 can include additional structural characteristics for each of the parcels, such as post-hazard event inspections, architectural / construction records, appraisals, insurance assessments, etc. The system can extract vulnerability characteristics and vulnerability characteristic characteristics for the first subset of parcels that did not burn and the second subset of parcels that did burn from the additional structural data.

[0108] The system can generate training vectors from the extracted vulnerability features and characteristics of the vulnerability features. In some embodiments, the vector generator module 114 generates training vectors from the extracted vulnerability features and characteristics of the vulnerability features for the machine learning model.

[0109] The system 102 can use the extracted vulnerability features and characteristics of the vulnerability features for each of the multiple compartments to generate training data 120 for a particular hazard event. Damage / non-damage records and / or damage severity records for a particular compartment of the multiple compartments from the historical hazard event data 124 can be used as ground truth for the burn / non-burn outcome of each compartment.

[0110] The system provides the training data to the machine learning model 410. The training data 120 is provided to a machine learning model, such as the damage propensity model 106, to train the damage propensity model 106 to make inferences about the damage propensity of a particular compartment to a general hazard event type or to a particular hazard event.

[0111] 5 is a block diagram of an exemplary computer system 500 that can be used to perform the operations described above. The system 500 includes a processor 510, a memory 520, a storage device 530, and an input / output device 540. Each of the components 510, 520, 530, and 540 can be interconnected using, for example, a system bus 550. The processor 510 can process instructions for execution within the system 500. In one implementation, the processor 510 is a single-threaded processor. In another implementation, the processor 510 is a multi-threaded processor. The processor 510 can process instructions stored in the memory 520 or the storage device 530.

[0112] The memory 520 stores information within the system 500. In one implementation, the memory 520 is a computer-readable medium. In one implementation, the memory 520 is a volatile memory unit. In another implementation, the memory 520 is a non-volatile memory unit.

[0113] The storage device 530 can provide mass storage for the system 500. In one implementation, the storage device 530 is a computer-readable medium. In various different implementations, the storage device 530 can include, for example, a hard disk device, an optical disk device, a storage device shared over a network by multiple computing devices (e.g., a cloud storage device), or some other large capacity storage device.

[0114] The input / output device(s) 540 provide input / output operations to the system 500. In one implementation, the input / output device(s) 540 may include one or more network interface devices, such as an Ethernet card, a serial communication device, such as an RS-232 port, and / or a wireless interface device, such as a 502.11 card. In another implementation, the input / output device(s) may include a driver device configured to receive input data and send output data to other input / output devices, such as keyboards, printers, and display devices 560. However, other implementations, such as mobile computing devices, mobile communication devices, set-top boxes, television client devices, etc., may also be used.

[0115] Although an exemplary processing system is illustrated in FIG. 5, implementations of the subject matter and functional operations described herein can be implemented in other types of digital electronic circuitry, or computer software, firmware, or hardware, or in combinations of one or more of them, including the structures disclosed herein and structural equivalents thereof.

[0116] The term "configured" is used herein in connection with systems and computer program components. A system of one or more computers configured to perform a particular operation or action means that the system has installed thereon software, firmware, hardware, or a combination thereof that, when in operation, causes the system to perform the operation or action. A system of one or more computer programs configured to perform a particular operation or action means that one or more programs contain instructions that, when executed by a data processing device, cause the device to perform the operation or action.

[0117] Embodiments of the subject matter and functional operations described herein can be implemented in digital electronic circuitry, tangibly embodied computer software or firmware, computer hardware, or a combination of one or more of them, including the structures disclosed herein and their structural equivalents. Embodiments of the subject matter described herein can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible, non-transitory storage medium for execution by or to control the operation of a data processing apparatus. The computer storage medium can be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them. Alternatively or additionally, the program instructions can be encoded on an artificially generated propagated signal, e.g., a machine-generated electrical, optical, or electromagnetic signal generated to encode information to be transmitted to a receiving device suitable for execution by a data processing apparatus.

[0118] The term "data processing apparatus" refers to data processing hardware and encompasses any kind of apparatus, device, and machine for processing data, including, by way of example, a programmable processor, a computer, or multiple processors or computers. An apparatus may also be or further include special purpose logic circuitry, such as, for example, an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit). An apparatus may optionally include, in addition to hardware, code that creates an execution environment for a computer program, such as code that constitutes a processor firmware, a protocol stack, a database management system, an operating system, or one or more combinations thereof.

[0119] A computer program, which may also be referred to or described as a program, software, software application, app, module, software module, script, or code, may be written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages, and may be deployed in any form, including standalone programs or modules, components, subroutines, or other units suitable for use in a computing environment. A program may correspond to a file in a file system, but need not. A program may be stored in one or more scripts stored in a portion of a file that holds other programs or data, for example, in a markup language document, in a single file dedicated to the program in question, or in multiple cooperating files, for example, files that store 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 data communication network.

[0120] As used herein, the term "engine" is used broadly to refer to a software-based system, subsystem, or process that is programmed to perform one or more specific functions. Generally, an engine is implemented as one or more software modules or components and installed on one or more computers in one or more locations. In some cases, one or more computers are dedicated to a particular engine. In other cases, multiple engines may be installed and run on the same computer(s).

[0121] The processes and logic flows described herein may be performed by one or more programmable computers executing one or more computer programs to perform functions by operating on input data to generate output. The processes and logic flows may also be performed by special purpose logic circuitry, such as, for example, an FPGA or an ASIC, or by a combination of special purpose logic circuitry and one or more programmed computers.

[0122] A computer suitable for executing a computer program can be based on a general-purpose or dedicated microprocessor, or both, or any other type of central processing unit. In general, the central processing unit receives instructions and data from a read-only memory, or a random access memory, or both. The essential elements of a computer are a central processing unit for implementing or executing instructions, and one or more memory devices for storing instructions and data. The central processing unit and memory may be supplemented by, or incorporated in, special purpose logic circuitry. In general, a computer also includes one or more mass storage devices, such as magnetic disks, magneto-optical disks, or optical disks, for storing data, or is operatively coupled to receive data from, or transfer data to, or both. However, a computer need not have such devices. Additionally, a computer may be embedded in another device, such as a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a Global Positioning System (GPS) receiver, or a portable storage device, such as a universal serial bus (USB) flash drive, to name just a few.

[0123] Computer-readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media, and memory devices, including, by way of example, semiconductor memory devices, e.g., EPROM, EEPROM, and flash memory devices, magnetic disks, e.g., internal hard disks or removable disks, magneto-optical disks, and CD-ROM and DVD-ROM disks.

[0124] To provide for interaction with a user, embodiments of the subject matter described herein can 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, and a keyboard and pointing device, such as a mouse or trackball, by which the user can provide input to the computer. Other types of devices can be used to provide interaction with the user as well, for example, feedback provided to the user can be any form of sensory feedback, such as visual feedback, auditory feedback, or tactile feedback, and input from the user can be received in any form, including acoustic, speech, or tactile input. In addition, the computer can interact with the user by sending documents to and receiving documents from a device used by the user, for example, sending a web page to a web browser on the user's device in response to a request received from the web browser. The computer can also interact with the user by sending text messages or other types of messages to a personal device, such as a smartphone running a messaging application, and receiving a response message from the user in return.

[0125] A data processing device for implementing machine learning models may also include, for example, dedicated hardware accelerator units for handling the typical, computationally intensive parts of machine learning training or production, i.e., inference, workloads.

[0126] The machine learning model can be implemented and deployed using a machine learning framework such as the TensorFlow framework, the Microsoft Cognitive Toolkit framework, the Apache Singa framework, or the Apache MXNet framework.

[0127] An embodiment of the subject matter described herein can be implemented in a computing system that includes a back-end component, e.g., a data server, or includes a middleware component, e.g., an application server, or includes a front-end component, e.g., a client computer having a graphical user interface, a web browser, or an app through which a user can interact with an implementation of the subject matter described herein, or includes any combination of one or more such back-end, middleware, or front-end components. The components of the system can 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.

[0128] 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. In some embodiments, a server sends data, e.g., HTML pages, to a user device for the purpose of, e.g., displaying data to and receiving user input from a user interacting with the device acting as a client. Data generated at the user device, e.g., a result of user interaction, can be received at the server from the device.

[0129] Although the present specification contains details of many individual implementations, these are not to be construed as limiting the scope of any feature or the scope of what may be claimed, but as descriptions of features specific to particular embodiments. Certain features described herein in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable subcombination. Furthermore, although features may be described above as acting in a certain combination, and may even be initially claimed as such, one or more features from the 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.

[0130] Similarly, although operations are depicted in the figures in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown, or in a sequential order, or that all of the illustrated operations be performed, to achieve desired results. In certain circumstances, multitasking and parallel processing may be advantageous. Furthermore, the separation of various system components in the embodiments described above should not be understood as requiring such separation in all embodiments, and it should be understood that the program components and systems described may generally be integrated together in a single software product or packaged in multiple software products.

[0131] Thus, certain embodiments of the subject matter have been described. Other embodiments are within the scope of the following claims. In some instances, the actions recited in the claims can be performed in a different order and still achieve desirable results. Moreover, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In certain embodiments, multitasking and parallel processing may be advantageous.

Claims

1. A method comprising: receiving a request for an injury propensity score for a partition, the request being in response to an occurrence of a real-time hazard event; receiving hazard event data relating to the real-time hazard event, the hazard event data including a current set of hazard conditions for the real-time hazard event and a degree of exposure of the compartment to the real-time hazard event; receiving imaging data for the section, the imaging data including street view imaging data for the section; extracting characteristics of a plurality of vulnerability features of the subdivision from the imaging data by a machine learning model including a plurality of classifiers and by using object recognition, the plurality of vulnerability features including structures and vegetation; determining, by the machine learning model, from the characteristics of the plurality of vulnerability features, the damage propensity score for the partition in response to the hazard event data relating to the real-time hazard event, the damage propensity score for the partition indicating a measure of risk to the partition of damage from the real-time hazard event, the damage propensity score including the characteristics of the plurality of vulnerability features; providing a representation of the damage propensity scores for the compartments corresponding to the real-time hazard events for display; and A method comprising:

2. The method of claim 1 , further comprising generating a set of partition characteristics from the characteristics of the plurality of vulnerability features.

3. The method of claim 1 , further comprising generating a three-dimensional model of the compartment from the characteristics of the plurality of vulnerability features and imaging data of the compartment.

4. The method of claim 1 , wherein the imaging data for the partition comprises imaging data captured within a threshold time from the time of the request.

5. receiving updated hazard event data for the real-time hazard event; determining an updated damage propensity score for the partition for the real-time hazard event from the characteristics of the plurality of vulnerability features, the hazard event data, and the updated hazard event data; The method of claim 1 further comprising:

6. determining, for the parcel, one or more mitigation steps for mitigating a risk score for the parcel via the machine learning model; determining, by the machine learning model, an updated injury propensity score based on the one or more mitigation steps; and providing a representation of the one or more mitigation steps and the updated injury propensity score; The method of claim 1 further comprising:

7. The method of claim 6 , wherein the one or more mitigation steps include adjustments to the characteristics of the plurality of vulnerability features extracted from the imaging data.

8. The method of claim 7, further comprising: iterating an updated injury propensity score determination based on the adjusted characteristics of the plurality of vulnerability features; The method of claim 6 further comprising:

9. determining that the updated injury propensity score meets a threshold injury propensity score; The method of claim 8 further comprising:

10. determining the one or more mitigation steps determining the one or more mitigation steps for a particular type of hazard event, the one or more mitigation steps for a first type of hazard event being different from the one or more mitigation steps for a second type of hazard event; The method according to claim 6.

11. generating training data for the machine learning model, receiving, for a hazard event, a plurality of compartments located within a vicinity of the hazard event, each of the plurality of compartments having received at least a threshold exposure to the real-time hazard event; receiving, for each partition of the plurality of partitions, imaging data for the partition including street view imaging data; extracting from the imaging data a plurality of vulnerability signature characteristics of a first subset of compartments of the plurality of compartments that did not burn during the real-time hazard event and a second subset of compartments of the plurality of compartments that did burn; generating said training data, providing the training data to a machine learning model; The method of claim 1 further comprising:

12. The method of claim 11 , wherein extracting the characteristics of the plurality of vulnerability features comprises providing the imaging data to the plurality of classifiers.

13. The method of claim 12 , wherein extracting the plurality of vulnerability feature characteristics further comprises identifying a plurality of objects in the imaging data with the plurality of classifiers.

14. receiving additional structural characteristics for each section of the plurality of sections; extracting a second plurality of vulnerability signatures for the first subset of compartments of the plurality of compartments that did not burn during the real-time hazard event and for the second subset of compartments of the plurality of compartments that burned during the real-time hazard event from the additional structural characteristics; providing the second plurality of vulnerability features to the machine learning model; The method of claim 11 further comprising:

15. The method of claim 14 , wherein the additional structural characteristic comprises a post hazard event inspection of the plurality of compartments.

16. A non-transitory computer storage medium containing a computer program, the computer program, when executed by a data processing apparatus, causing the data processing apparatus to: receiving a request for an injury propensity score for a partition, the request being in response to an occurrence of a real-time hazard event; receiving hazard event data relating to the real-time hazard event, the hazard event data including a current set of hazard conditions for the real-time hazard event and a degree of exposure of the compartment to the real-time hazard event; receiving imaging data for the section, the imaging data including street view imaging data for the section; extracting characteristics of a plurality of vulnerability features of the subdivision from the imaging data by a machine learning model including a plurality of classifiers and by using object recognition, the plurality of vulnerability features including structures and vegetation; determining, by the machine learning model, from the characteristics of the plurality of vulnerability features, the damage propensity score for the partition in response to the hazard event data relating to the real-time hazard event, the damage propensity score for the partition indicating a measure of risk to the partition of damage from the real-time hazard event, the damage propensity score including the characteristics of the plurality of vulnerability features; providing a representation of the damage propensity scores for the compartments corresponding to the real-time hazard events for display; and A non-transitory computer storage medium containing instructions to cause a computer to perform operations including

17. generating training data for the machine learning model, receiving, for a hazard event, a plurality of compartments located within a vicinity of the hazard event, each of the plurality of compartments having received at least a threshold exposure to the real-time hazard event; receiving, for each partition of the plurality of partitions, imaging data for the partition including street view imaging data; extracting from the imaging data a plurality of vulnerability signature characteristics of a first subset of compartments of the plurality of compartments that did not burn during the real-time hazard event and a second subset of compartments of the plurality of compartments that did burn; generating said training data, providing the training data to a machine learning model; 20. The non-transitory computer storage medium of claim 16, further comprising:

18. receiving additional structural characteristics for each section of the plurality of sections; extracting a second plurality of vulnerability signatures for the first subset of compartments of the plurality of compartments that did not burn during the real-time hazard event and for the second subset of compartments of the plurality of compartments that burned during the real-time hazard event from the additional structural characteristics; providing the training data to a machine learning model; 20. The non-transitory computer storage medium of claim 17, further comprising:

19. A user device; Interacting with the user device; receiving a request for an injury propensity score for a partition, the request being in response to an occurrence of a real-time hazard event; receiving hazard event data relating to the real-time hazard event, the hazard event data including a current set of hazard conditions for the real-time hazard event and a degree of exposure of the compartment to the real-time hazard event; receiving imaging data for the section, the imaging data including street view imaging data for the section; extracting characteristics of a plurality of vulnerability features of the subdivision from the imaging data by a machine learning model including a plurality of classifiers and by using object recognition, the plurality of vulnerability features including structures and vegetation; determining, by the machine learning model, from the characteristics of the plurality of vulnerability features, the damage propensity score for the partition in response to the hazard event data relating to the real-time hazard event, the damage propensity score for the partition indicating a measure of risk to the partition of damage from the real-time hazard event, the damage propensity score including the characteristics of the plurality of vulnerability features; providing, to the user device, a representation of the damage propensity scores of the partitions corresponding to the real-time hazard events for display; one or more computers operable to perform operations including: A system comprising:

Images