Using mobile data to improve weather information

Abstract

A system (100) for using mobile data to improve weather information is provided. The system (100) includes a weather forecast station (120) configured to receive fixed observation data provided by a plurality of stationary weather stations (110) along with data from a plurality of input weather models (115) and generate integrated weather model estimates based on the fixed observation data and the input weather model data. The system (100) further includes a processor (130) configured to aggregate mobile observation data provided by a plurality of non-stationary sensors (140) and adjust the weather model estimates using the aggregated mobile observation data.

Description

[Technical field]

[0001] Technical Field The embodiments described below relate to weather information, and more particularly, to the use of mobile data to improve weather information. [Background technology]

[0002] background There is growing interest in using mobile observations to improve current-state estimates and (future) predictions produced by weather models. Mobile observations are attractive because they are available in many locations, such as when provided by mobile phones or vehicles. However, these observations have some significant disadvantages compared to observations from stationary sensors (e.g., weather stations), and novel methods are needed to make them useful for improving weather model information.

[0003] For example, individual observations, often obtained from consumer-quality devices or sensors primarily intended for other purposes, tend to be of low quality. In fact, the quality is often significantly worse than that obtained from the weather model itself. This makes it difficult to improve weather models using only individual mobile observations. Also, by definition, mobile observations are made at locations that change periodically. In contrast, non-trivial weather modeling processes require a time series of data from a location to enable statistical processes. Furthermore, in some cases, observations may appear related to information required for or generated by a weather model, but may not be an exact match. For example, vehicle traction control system data is obviously not related to whether conditions such as ice are present on the road, and low traction values, while they may correspond to ice, may also reflect other factors of low traction.

[0004] As a relevant example of this need for mobile observations, the use of mobile observations is of particular interest within the transportation sector, such as to provide accurate roadside weather-related information. This use includes both atmospheric weather conditions such as temperature and wind, as well as "road weather" such as road surface temperature and weather-related conditions (e.g., dry, damp, wet, snow, ice, including amounts). Summary of the Invention [Problem to be solved by the invention]

[0005] Historically, observations of these parameters have been made at Road Weather Information System (RWIS) sites located along roadways and typically operated by government transportation agencies. Because they are expensive to acquire and maintain, the number of existing RWIS is relatively small, and they sample the road network only sparsely.

[0006] Thus, there is a need to use mobile data to improve weather information, particularly mobile data provided by plentiful and readily available mobile sensors, and there is a need to improve the quality and suitability of the mobile data provided by mobile sensors for use in improving weather information. [Means for solving the problem]

[0007] overview A system is provided that uses mobile data to improve weather information. According to one embodiment, the system includes a weather forecast station configured to receive fixed observation data provided by a plurality of stationary weather stations along with data from a plurality of input weather models, and generate an integrated weather model estimate based on the fixed observation data and the input weather model data. The system also includes a processor configured to aggregate mobile observation data provided by a plurality of non-stationary sensors, and adjust the integrated weather model estimate using the aggregated mobile observation data.

[0008] A method of using mobile data to improve weather information is provided, comprising the steps of receiving fixed observation data provided by a plurality of fixed weather stations, receiving data from a plurality of input weather models, generating an integrated weather model estimate based on the fixed observation data and the input weather model data, aggregating mobile observation data provided by a plurality of non-stationary sensors, and adjusting the integrated weather model estimate using the aggregated mobile observation data.

[0009] aspect According to one aspect, a system (100) for using mobile data to improve weather information comprises a weather forecast station (120) configured to receive fixed observation data provided by a plurality of stationary weather stations (110) and to receive data from a plurality of input weather models (115) and to generate an integrated weather model estimate based on the fixed observation data, the input weather model data, the system (100) further comprising a processor (130) configured to aggregate mobile observation data provided by a plurality of non-stationary sensors (140) and to adjust the integrated weather model estimate using the aggregated mobile observation data.

[0010] Preferably, the processor (130) is further configured to determine hypothetical observation data based on the adjustment of the integrated weather model estimate, the hypothetical observation data being an estimate of observation data that would have been provided by a plurality of non-existent fixed weather stations.

[0011] Preferably, the processor (130) being configured to determine virtual observation data based on the adjustment of the integrated weather model estimate comprises the processor (130) being configured to determine climatological values ​​based on the aggregated mobile observation data and to determine the virtual observation data using the climatological values.

[0012] Preferably, the processor (130) is further configured to perform post-processing on a weather forecast model using the virtual observed data.

[0013] Preferably, the processor (130) being configured to use the aggregated mobile observation data to adjust the integrated weather model estimate comprises the processor (130) being configured to spatially correlate the mobile observation data with a virtual observation location (230) to adjust the weather model estimate at the virtual observation location (230).

[0014] Preferably, the processor (130) is configured to spatially correlate the aggregated mobile observation data with the virtual observation locations (130), such that the processor (130) calculates a spatial correlation distance (d bin CORR ) to obtain the spatial correlation distance (d bin CORR ) defines the distance over which observations can be usefully extrapolated to adjust the weather model estimates.

[0015] Preferably, the mobile observation data provided by the plurality of non-stationary sensors (130) is organized into space-time bins (210) defined by location and time parameters.

[0016] Preferably, the system (100) further comprises a plurality of fixed weather stations (110), the plurality of fixed weather stations (110) configured to provide the fixed observation data of meteorological conditions at the locations of the fixed weather stations (110).

[0017] According to one aspect, a method of using mobile data to improve weather information includes receiving fixed observation data provided by a plurality of fixed weather stations; receiving data from a plurality of input weather models; generating an integrated weather model estimate based on the fixed observation data and the input weather model data; aggregating mobile observation data provided by a plurality of non-stationary sensors; and adjusting the integrated weather model estimate using the aggregated mobile observation data.

[0018] Preferably, the method further comprises determining hypothetical observation data based on the adjustment of the integrated weather model estimate, the hypothetical observation data being an estimate of observation data that would have been provided by a plurality of non-existent fixed weather stations.

[0019] Preferably, the step of determining virtual observation data based on said adjustment of said integrated weather model estimate comprises the steps of determining climatic values ​​based on said aggregated mobile observation data and using said climatic values ​​to determine said virtual observation data.

[0020] Preferably, the method further comprises the step of performing post-processing on a weather forecast model using the virtual observed data.

[0021] Preferably, the step of adjusting the integrated weather model estimate using the aggregated mobile observation data comprises the step of spatially correlating the mobile observation data with a virtual observation location to adjust the integrated weather model estimate at the virtual observation location.

[0022] Preferably, the step of spatially correlating the mobile observation data with the virtual observation locations comprises determining a spatial correlation distance, the spatial correlation distance defining a distance over which observations may be usefully extrapolated to adjust the integrated weather model estimates.

[0023] Preferably, said mobile observation data provided by said plurality of non-stationary sensors is organized into space-time bins defined by location and time parameters.

[0024] Like reference numbers represent like elements in all figures. It should be understood that the drawings are not necessarily drawn to scale. [Brief description of the drawings]

[0025] [Figure 1] FIG. 1 illustrates a system 100 that uses mobile data to improve weather information. [Diagram 2] 2 is another diagram of the system 100 shown in FIG. 1. [Diagram 3] FIG. 3 illustrates a method 300 for using mobile data to improve weather information. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0026] Detailed Description 1-3 and the following description provide specific examples to teach those skilled in the art how to construct and use the best mode of the embodiment of using mobile data to improve weather information. For the purpose of teaching the principles of the invention, some conventional aspects have been simplified or omitted. Those skilled in the art will appreciate variations from these examples that are within the scope of the description. Those skilled in the art will appreciate that the features described below can be combined in various ways to form multiple variations of using mobile data to improve weather information. As a result, the embodiments described below are not limited to the specific examples described below, but only by the claims and their equivalents.

[0027] FIG 1 illustrates a system 100 that uses mobile data to improve weather information. As shown in FIG 1, the system includes a fixed weather station 110, an input weather model 115, and a weather forecast station 120. As shown in FIG 1, the fixed weather station 110 and the weather forecast station 120 are communicatively coupled to each other. The weather forecast station 120 may be configured to receive fixed observation data provided by the fixed weather station 110. Additionally or alternatively, the weather forecast station 120 may be configured to receive fixed observation data provided by one or more other fixed weather stations and model data from one or more other input weather models.

[0028] As shown in FIG. 1, the system 100 also includes a processor 130 communicatively coupled to the weather station 120. As shown in FIG. 1, the processor 130 is separate from the weather station 120, although the processor 130 may be part of the weather station 120. Although the system 100 is shown to include a fixed weather station 110, the system 100 may be comprised of the weather station 120 and the processor 130. The processor 130 may be configured to perform post-processing on the weather forecast model. More specifically, the processor 130 may be configured to use hypothetical observation data for locations that do not have a fixed weather station. This hypothetical observation data may be an estimate of observation data that would have been provided by a non-existent fixed weather station. As described in more detail below, the hypothetical observation data may be based on mobile observation data.

[0029] System 100 may also include a non-stationary sensor 140 communicatively coupled to processor 130. Although not shown, non-stationary sensor 140 may alternatively or additionally be communicatively coupled to weather station 120 to receive or transmit data. For example, if processor 130 is part of weather station 120, non-stationary sensor 140 may be communicatively coupled to weather station 120. The data provided by non-stationary sensor 140 may be mobile observation data.

[0030] The fixed weather station 110 may be configured to determine weather conditions at or near the fixed weather station 110. Thus, the fixed weather station 110 may be configured to provide fixed observation data. This fixed observation data may include a number of fixed observations that may be used by the weather forecast station.

[0031] The input weather model 115 is shown communicatively coupled to a weather station 120. The input weather model 115 may include a weather model for an area that includes the system 100, such as a portion of the system. Multiple input weather models 115 may provide data to the weather station 120.

[0032] Weather forecast station 120 may use the fixed point observation data to determine an integrated weather model estimate by generating the data using one or more weather forecast model methods. That is, the integrated weather model estimate may be based on the fixed point observation data. The integrated weather model estimate may include current state estimates (e.g., for locations other than the fixed weather stations) and / or (future) forecasts. Weather forecast station 120 may generate the integrated weather model estimate based on one or more weather forecast models using any suitable method, as described below.

[0033] Weather station 120 may assimilate the stationary observation data into a numerical weather prediction (NWP) model. A feature of this approach is that NWP models typically require an hour or more to compute once the observations are received before a forecast is available. This "latency" may be undesirable if the desired information concerns using the observations to improve current estimates or near-future (first hour or so) forecasts. To among other things compensate for the latency and improve accuracy, the NWP model may be adjusted using further "post-processing" steps using more current stationary observation data. Processor 130 may perform such post-processing. The collection of these steps, including multiple input models and observation data combined in a way that improves each input, is referred to as an "integrated" weather model.

[0034] For example, an NWP estimate of air temperature calculated using observations from an hour or more ago as determined by a weather forecast model may differ by 4° C. from the currently observed temperature for any of a number of reasons. Post-processing performed by processor 130 can adjust this model estimate to match the observations using a computational process that can take several minutes. Furthermore, a fixed weather station's observation-enhanced current state estimate (and associated near-future forecast) can be used to improve similar estimates / forecasts at locations in the area surrounding the observation location. Additionally, the availability of observations at locations that do not have fixed sensors can improve the overall weather model information.

[0035] The input to the post-processing performed by the processor 130 may consist of multiple NWP models along with observations. Post-processing may improve the accuracy of the input NWP models at the observed locations in any of three or more steps, including but not limited to a) statistical correction of the individual NWP models by bias adjustment, b) guidance to a consensus combination of those models to generate an improved estimate / forecast, and c) ensuring that the final estimate / forecast is consistent with the current observations at the observed locations. A fourth step, added to extend the post-processing to non-observed locations, includes adjusting the model information using guidance from nearby observations from stationary observation locations. Stationary observation data may be used in each of the above four steps.

[0036] Fixed point observation data generates a time series of observations at a single location. This observation time series can then be statistically analyzed as part of post-processing at that location. In contrast, mobile observation data from an individual mobile sensor 140 does not generate such a time series for a fixed location and does not include such statistical analysis for the mobile observation data. To enable post-processing at a fixed location using mobile observation data, the processor 130 may aggregate the mobile observation data. Such aggregation may include grouping observations from multiple mobile sensors 140, which are organized into space-time bins associated with limited spatial distances and time periods. Additionally, the aggregated mobile observation data may be interpolated into virtual observation data. The virtual observation data may be for locations other than the location of the mobile observation data or the fixed weather station, as described in more detail below with reference to FIG. 2. The virtual observation data may be used as an input to all NWP generation and post-processing steps.

[0037] FIG. 2 is another diagram of the system 100 shown in FIG. 1. As shown in FIG. 2, the system 100 includes a plurality of the fixed weather stations 110 and the input weather models 115 described above. The fixed weather stations 110 are shown as small circles, and the input weather models 115 are shown as small squares. The system 100 also includes the weather forecast station 120 and the processor 130 described above. The weather forecast station 120 and the processor 130 are communicatively coupled to each other. Although only two of the plurality of fixed weather stations 110 and only one of the input weather models 115 are shown as communicatively coupled to the weather forecast station 120, all of the fixed weather stations 110 and the input weather models 115 may be communicatively coupled to the weather forecast station 120. A plurality of non-stationary sensors 140 are also shown. The plurality of non-stationary sensors 140 are shown as being on or as part of a vehicle, but any suitable non-stationary sensors may be utilized. A number of non-stationary sensors 140 are shown on a two lane road with arrows indicating the direction of movement of the non-stationary sensors 140 .

[0038] As shown in Figure 2, the system 100 also includes space-time bins 210, which are shown to include a portion of a two-lane road shown in Figure 4. As can be appreciated, if multiple non-stationary sensors 140 cross one of the space-time bins 210 during a given period of time, the mobile observation data provided by each of those non-stationary sensors 140 will be within the same space-time bin 210 and will therefore represent conditions, such as weather conditions, at the location and time of the given space-time bin 210. Spatially and temporally coincident mobile observation data from different non-stationary sensors 140 may be aggregated into corresponding space-time bins 210.

[0039] Also shown is a spatial correlation region 220 corresponding to one of the space-time bins 210. The spatial correlation region 220 is an area where observations of the aggregated mobile observation data in the corresponding space-time bin 210 may be usefully extrapolated to adjust a weather forecast model at a given location. Such extrapolations may be the hypothetical observations described above and located at the corresponding hypothetical observation locations 230 shown in FIG. 2 , depicted as boxes with “Xs”. The spatial correlation region 220 is a function of the spatial correlation distance d for the corresponding space-time bin 210, as described in more detail below. bin CORR is defined as follows:

[0040] Note that continuous and discrete variables would require slightly different derivations of this method. Continuous variables are variables whose values ​​tend to vary smoothly in space and time over a continuous range. Examples include temperature, dew point, wind speed, liquid / ice depth on the road, and similar variables. Discrete variables have discrete levels. Discrete variables vary discontinuously in space and / or time. Examples include road conditions (e.g., dry, damp, wet, snow, ice), precipitation occurrence, precipitation type, wiper status, and similar variables.

[0041] A technique for generating virtual observations from mobile observations may include four steps: quality control and aggregation, climatology calculation, spatial correlation calculation, and virtual observation calculation. Quality control and aggregation may be used because mobile observations are generally of lower quality than industry standard fixed point observations (such as observations from weather stations at airports). Therefore, new quality control processes may need to be applied to the mobile observation data. One means for reducing errors from individual mobile devices is to aggregate multiple observations into space-time bins 210. For example, all vehicles located within a 1 km road segment during a 1 hour period may be aggregated into a space-time bin with a spatial size of 1 km and a temporal size of 1 hour. This "aggregated observation" may include multiple statistical parameters such as mean and variance. The number of samples in a space-time bin is related to the accuracy of this aggregated observation, so it is likely that the quality will be higher where more mobile observations are available. Techniques for intercomparison of mobile devices with sensors (i.e., non-stationary sensors), such as when they match in space-time, may also be important.

[0042] For climatology, if "aggregated observations" of space-time bins are stored and accumulated over time, it is possible to calculate climatological values ​​of the aggregated observations. For example, aggregated observations of hourly variables over a month can be averaged over the month to average the variable for that month at each bin location, perhaps with respect to time of day or other parameters, to adequately reflect expected climate change. Climatology may be calculated over a static period, such as monthly, or over a dynamic period, such as the last 30 days.

[0043] For the calculation of spatial correlation, the spatial correlation distance d for each variable and each space-time bin 210 is calculated using binned aggregated observations from the data used to calculate the climatology. bin CORR As mentioned above, it is also possible to calculate the spatial correlation distance d bin CORRtells us the distance to which observations can be usefully extrapolated to adjust forecast models at nearby locations. In some cases, it may be useful to calculate this on a conditional basis, such as the correlation distance for road conditions when snow is the predicted or observed road condition. Also, the spatial correlation distance d bin CORR can also be calculated in terms of spatial directions, such as a different value for each of the two directions along the road. Thus, the spatial correlation distance d bin CORR can be considered to be a vector originating from the location corresponding to a given space-time bin 210. Thus, although the spatial correlation region 220 is shown as a circle, other shapes may be utilized. As with climatology, spatial correlation may be calculated over a static period, such as monthly, or over a dynamic period, such as the last 30 days, and may be a function of various parameters, such as time of day.

[0044] Regarding the calculation of the virtual observation results, the spatial correlation distance d bin CORR allows the computation of a typically non-sparse set of observation equivalents, called "virtual observations", from possibly sparse real-time aggregated mobile observations. The virtual observations can then be input to NWP generation and all traditional post-processing stages (fine-scale model bias adjustment, location-specific consensus combination, real-time error correction, and real-time observation extrapolation). To understand the process for generating virtual observations, for a given day / time, we consider aggregated observations Y of variable Y at location A, A (AggObs) , but not the observations at location B, such as the hypothetical observation location 230 shown in FIG. 2. A (AggObs)can be used to improve weather information at location A, but we also want to use it to improve weather information at location B (which may or may not have one or more nearby locations A, each with valid aggregated observations). To come up with a solution, the joint model estimate (see below) is A (UniModel) and Y B (UniModel) and the separation between points A and B is d A,B and the calculated spatial correlation distance at location A is d A CORR Let us assume that.

[0045] For continuous variables, hypothetical observations Y at locations A and B (Virtual) can be calculated from the joint model estimates for locations A and B and the aggregated mobile observations at location A using the following formula:

[0046] Y n (Virtual) =Y n (UniModel) -Y n (ObsAdj) (n=A,B) In the formula, Y n (ObsAdj) is an observation-based adjustment.

[0047] Observation-based adjustment Y n (ObsAdj) can be defined, for example, as follows:

[0048] Y n (ObsAdj) =Q A K.D. A,B ·(Y A (UniModel) -Y A (AggObs) ) where K is a local or global user-defined sensitivity coefficient between 0 and 1, set to manage the desired overall sensitivity of the system to the mobile observation adjustment process; Q n is a user-defined location-based quality factor between 0 and 1, D A,B is a correlation distance based factor for applying smaller adjustments as the location separation increases.

[0049] Location-based quality factor Q n can be defined, for example, as follows: Q n =H n (s0 / s n )(s n >s0) =H n (s n <=s0) where n is the location index (e.g., "A" or "B"), s n is the standard deviation reflecting the uncertainty in the observations / climatology of variable Y at location n, s0 is a user-defined threshold uncertainty reflecting the maximum tolerable uncorrected error in the observations and climatology of variable Y used to calculate the hypothetical observations; H n is the observation ambiguity coefficient (discussed later in this paper) that is nominally equal to 1 for many variables.

[0050] Correction factor D A,B can be defined, for example, as follows: D A,B =1-(d A,B / (m·d A CORR ))(d A,B <=m·d A CORR ) =0 (d A,B >m·d A CORR ) where m is a user-set sensitivity factor that defines how much the correlation distance should be adjusted.

[0051] Correlation coefficient D A,B There are three important cases. First, at location A, B=A, so the distance is d A,B ->d A,A = 0, the resulting correlation coefficient D A,B = 1. Second, the distance d A,B M·D A CORR If it is equal to or less than the correlation coefficient D A,B is 0 <D A,B <1. In this case, the distance-weighted observation-based adjustment of A is made at location B. Third, the distance d A,B If is larger, the spatial correlation coefficient D A,B = 0. In this case, no adjustments based on the observations of A are made at location B. In other words, the hypothetical observations at A and B are as expected.

[0052] That is, the quality of the mobile observations is sufficiently high (i.e., Q A and K is equal to 1), then the hypothetical observation Y at location A A (virtual) simply takes the value of the aggregated observation at A. As quality decreases, the virtual observation Y at location A A (virtual) Similarly, the hypothetical observation result Y at location B B (virtual) is equal to the aggregate model value adjusted by a fraction of the aggregated observations at A, the fraction being determined by the quality of the aggregated observations at A and the distance of B from A compared to the calculated correlation distance. If no adjustments are made, the virtual observation takes the value of the aggregate model at B.

[0053] For each virtual observation derived from the underlying aggregated observations, an updated quality estimate, such as the standard deviation, can be calculated. For locations affected by multiple observations, we consider n adjacent observation locations A, B, each separated from B by less than a multiple of the correlation distance m. n Adjusted observations from Y n (ObsAdj) The values ​​can be accumulated and added in a normalized manner, such as by distance weighting.

[0054] For discrete values, hypothetical observations Y at locations A and B (Virtual) can be calculated from the joint model estimates for locations A and B and the aggregated mobile observations at location A using the following formula:

[0055] Y A (Virtual) =Y A (AggObs) (Q A >=Q thresh ) =Y A (UniModel) (Q A thresh ) Y B (Virtual) =>Y A (AggObs) (Q A >=Q thresh and d A,B <=Q A K(m d A CORR )) =>Y B (UniModel) (Q A thresh ), (Q A >=Q thresh and d A,B >Q A K(m d A CORR )) In the formula, K,Q A and m is as defined above for continuous variables; Q thresh ​​is the user-defined quality threshold for correcting Y.

[0056] In other words, the discrete integrated model Y A (UniModel) and Y B (UniModel) The values ​​are: a) the observational / climatology quality factor Q A is a user-defined threshold Q thresh and b) the distance d between A and B is greater than A,B is less than the quality-weighted correlation distance for the discrete aggregated observations Y A (AggObs) In locations affected by multiple observations, the discrete aggregate model value Y B (UniModel) The decision to replace Y with one of the n eligible nearby observations can be made by taking a discrete weighted average of the n contributions. The revised uncertainty estimate is A (AggObs) either the original estimate of or a modified version reflecting further processing to obtain a hypothetical observation at location B.

[0057] Further consideration may be required when an observation is a proxy for a model variable but has substantially different characteristics. For example, a vehicle traction system utilizes sensors that provide data that can be considered to be an observation of traction or friction. In some cases, it is desirable to correlate traction or friction with road weather conditions. For example, very low traction values ​​may be used to infer that the road is icy. Similarly, a wiper in an "on" state may be a proxy for precipitation, even though an "on" state may also mean that the wiper is being used during road splash or wash fluid conditions where no precipitation is occurring. Although there is considerable ambiguity in correlating such proxy observations with model variables, it is important to include this process.

[0058] These proxy observation cases are easily accommodated using the described methodology. In such cases, the quality factor Q A can be adjusted to reflect any further uncertainty caused by ambiguous mapping of observations to relevant model variables. For example, we consider the static friction measurement T, which has the range 0 to 1. (AggObs) But, T (AggObs) = 0 corresponds to a frozen road with probability 1, and T (AggObs) = 0.3 corresponds to an icy road with probability 0. Then, Q A The coefficient H in the formula for A can be shown, for example, by:

[0059] H A =1-T A (AggObs) / 0.3(T A (AggObs) <=0.3) =0 (T A (AggObs) >0.3) Based on the nature of the sensor and how its data is transformed to determine the model equivalent, A The particular equation for is believed to be empirical and situational. In this example, actual observations of static friction are used to generate virtual observations of road conditions. Similarly, actual observations of wiper motor speed can be used to generate virtual observations of precipitation rate. The inverse process can be used to relate model variables to their equivalents measured by mobile sensors. The particular equation that works best for any pair of model variables and surrogate mobile observations, and for any parameters included in the equation, can be refined and improved by analysis (both real-time and offline) that statistically compares predicted and observed values ​​of the variables.

[0060] Virtual observations can be used as inputs to NWP models and post-processing in a similar manner to stationary observations. For example, in NWP model bias adjustment, historical virtual observations are incrementally input into the system (as they become available) in the same way as data from stationary observations. If the system can accommodate the uncertainty values ​​associated with each virtual observation, then that information is available. This step may traditionally utilize a Model Output Statistics (MOS) or Dynamic Model Output Statistics (DMOS) process to bias correct each component NWP model. To do so, the model error (Y n (OrigModel) -Y n (virtual) Time series (typically 30-90 days for DMOS) reflecting the model-induced precipitation (defined as Δt) are accumulated and statistically evaluated to determine a regression fit that relates model values ​​to the corresponding observations. The regression fit is used to calibrate the model, Y, which is most commonly a simple bias, but can be more elaborate model corrections such as temperature-dependent bias. n (ModelAdj) Following the standard approach, but using regression based on hypothetical observations, for each NWP model, it can be expressed as:

[0061] Y n (UnbiasedModel) =Y n (OrigModel) -Y n (ModelAdj) The quality of the regression fit can be determined from the regression data and quality controls can be applied in the same way as for fixed point observations.

[0062] In the consensus NWP combination, historical hypothetical observations are input incrementally into the system (as they become available) in the same way as data from stationary observations. This historical observation is then used to calculate model weights for each location. If the system can accommodate the uncertainty values ​​associated with each aggregated or hypothetical observation, then that information is available.

[0063] Error corrections for the current situation and near-future forecasts (typically the first few hours) are performed using each hypothetical observation in a manner similar to fixed point observations. Extrapolation techniques for generating the hypothetical observations are used to make corrections where there are no current observations. If the system can accommodate the uncertainty values ​​associated with each hypothetical observation, then that information is available. This includes using the hypothetical observations to revise confidence estimates calculated as part of the forecast, and using the hypothetical observations to flag "outlier" events where a "bad-case" forecast differs substantially from the hypothetical observations.

[0064] For extrapolation to locations without sensors, if the density of mobile observations is sufficient, the above technique supersedes other methods for extrapolating sensor observations to locations without sensors.

[0065] It should be clear that a "spin-up" period is required to accumulate the necessary climatology and time history before reliable virtual observations can be produced for use in post-processing. Experience with conventional post-processing has shown that this period can be anywhere from 30 to 90 days, depending on the desired fidelity.

[0066] Because virtual observations generalize the concept of observations, the technique works for both fixed and mobile observations. For example, it was previously not possible to effectively extrapolate fixed or stationary observations (such as those from an airport weather station or RWIS) to surrounding locations that themselves do not have observations at any fine-scale of fidelity. No quantitative guidelines were available to identify how far from the observation location the observational knowledge could be applied to improve weather information. For example, it is conceptually expected that observations at an airport in flat terrain will reflect the weather within an area surrounding the airport that is significantly larger than an airport in mountainous terrain. Using climatology and associated correlation distance information created using mobile observations, combined with the geographic extent around the stationary observation, one can accurately extrapolate the correlation distance of the stationary observation, i.e., the stationary observation within the range of any location, not just along a road.

[0067] The hypothetical observations are robust to the lack of observational data. In the absence of observations at a particular location, two situations can arise. At locations where insufficient mobile (or fixed) observations are available to compute a reliable climatology or to determine the correlation distance, the hypothetical observations will have the form of the original integrated model value for that location adjusted by observations within the correlation distance. At locations with the additional feature of not being within the correlation length of the observations over the entire climatology period or in real-time data, the hypothetical observations will be equal to the integrated model value. With a sufficiently dense source of mobile observations, such situations are expected to be rare.

[0068] Due to the nature of the extrapolation method used, the information will be consistent both in locations with abundant and sparse observations: in other words, locations adjacent to locations with no observations will themselves only have small adjustments, ensuring spatial continuity.

[0069] FIG. 3 illustrates a method 300 of using mobile data to improve weather information. As shown in FIG. 3, the method 300 receives stationary observation data provided by a stationary weather station and receives data from a plurality of input weather models (320) at step 310. At step 330, the method 300 generates an integrated weather model estimate based on the stationary observation data and the input weather model data. At step 340, the method 300 aggregates mobile observation data provided by a plurality of non-stationary sensors. At step 350, the method 300 adjusts the integrated weather model estimate using the aggregated mobile observation data. The method 300 may be performed by the system 100 described above, particularly the weather forecast station 120 and / or the processor 130, although any suitable system or system components may be utilized.

[0070] The method 300 may further determine hypothetical observation data based on the adjustment of the weather model estimates, where the hypothetical observation data is an estimate of observation data that would have been provided by a non-existent fixed weather station. For example, hypothetical observation result Y n (Virtual) is the pooled model estimate Y at location n n (UniModel) Adjustment of observation results for Y n (ObsAdj) It can be calculated using:

[0071] Method 300 may also determine hypothetical observation data based on adjustments to weather model estimates by determining climatic values ​​based on the aggregated mobile observation data and using the climatic values ​​to determine hypothetical observation data. For example, as described above, aggregated mobile observations of a variable in hourly aggregated mobile observation data over a month may be averaged over the month to determine an average value of the variable for that month at each location.

[0072] The method may also perform post-processing on the weather forecast model using the hypothetical observations as inputs to the NWP model. For example, as described above, hypothetical observation Y at location n may be n (Virtual) can be used in post-processing for the NWP model estimates at location n. In one particular example, the model error (Y n (OrigModel) -Y n (Virtual) A time series reflecting the model (defined as n (ModelAdj) Determine.

[0073] Adjusting 350 the weather model estimates using the aggregated mobile observation data may comprise spatially correlating the mobile observation data with the virtual observation locations to adjust the weather model estimates at the virtual observation locations. For example, as described above, the aggregated observation data is used to adjust the weather model estimates at location n from the mobile observation data at location A, n (ObsAdj) The observation adjustment Y n (ObsAdj) Using the pooled model estimate Y at location n, n (UniModel) This adjustment can be done by adjusting the virtual observation Y n (Virtual) can be determined.

[0074] The step of spatially correlating the mobile observation data with the virtual observation location includes the spatial correlation distance d bin CORR Such a method may include determining a spatial correlation distance, such as, that defines a distance over which observations may be usefully extrapolated to adjust the integrated weather model estimates. The spatial correlation distance may have a directional and a magnitude component. Thus, multiple spatial correlation distances may define spatial correlation regions of various shapes and sizes, such as the spatial correlation region 220 shown in FIG. 2.

[0075] The mobile observation data provided by the multiple non-stationary sensors may be data for space-time bins previously defined by location and time parameters occupied by the multiple non-stationary sensors. For example, with reference to FIG. 2, the mobile observation data provided by the multiple non-stationary sensors 140 may be data for space-time bins 210 that each non-stationary sensor 140 passes through during a time parameter (e.g., 1 hour) of the space-time bin 210.

[0076] Method 300 may be performed by software similar to that used in NWP modeling and standard post-processing. Thus, system 100, such as weather forecast station 120 and / or processor 130, may be configured to perform steps of method 300 or similar methods. As a result, system 100 and method 300 are designed to accommodate observations of various qualities with unknown spatial and temporal densities. Thus, system 100 and method 300 are applicable to a wide range of mobile observation scenarios and can be realized at any spatial-temporal resolution depending on the characteristics of the available observations.

[0077] System 100 and method 300 provide at least three improvements over the use of model data without access to mobile observations. First, model bias adjustments enabled at every location by mobile observation climatology allow fine spatial scale refinements to model conditions / predictions. Second, model weights for consensus combinations can be calculated for each location, rather than assumed to be constant over large areas. Third, when observations are available in real time, these observations can be used to reduce model errors in real time for current conditions and near-future predictions. System 100 and method 300, by using correlation distances, allow those refinements to be extended to locations where observations are not available.

[0078] The detailed description of the above embodiments is not an exhaustive description of all embodiments contemplated by the inventors within the scope of the present description. Indeed, those skilled in the art will recognize that certain elements of the above embodiments may be combined or removed in various ways to create further embodiments, and such further embodiments are within the scope and teachings of the present description. It will also be apparent to those skilled in the art that all or part of the above embodiments may be combined to create further embodiments within the scope and teachings of the present description.

[0079] Thus, while specific embodiments are described herein for illustrative purposes, various equivalent modifications are possible within the scope of this description, as those skilled in the art will recognize. The teachings provided herein may be applied to other systems and methods that use weather data to improve weather information, as well as the embodiments described above and illustrated in the accompanying drawings. Thus, the scope of the above embodiments should be determined from the claims that follow.

Claims

1. 1. A method comprising: receiving an integrated weather model output; receiving a statistical correlation based on distance relating the weather observation data for space-time bins associated with the limited spatial distance and time period to the virtual observation location; receiving aggregated mobile observation data provided by a plurality of non-stationary sensors for said space-time bins; and adjusting the integrated weather model output based on the statistical correlation and the aggregated mobile observation data.

2. The method of claim 1 , wherein the weather observation data for the space-time bins comprises the aggregate mobile observation data.

3. The method of claim 1 or 2, wherein the statistical correlation relating the weather observation data for the space-time bins to the virtual observation locations is further based on time of day.

4. The method of any one of claims 1 to 3, wherein the statistical correlation relating the weather observation data for the space-time bins to the virtual observation locations is further based on a one-month period.

5. The method of any one of claims 1 to 4, wherein the integrated weather model output is generated using an input weather model that utilises fixed point observation data as input.

6. The method of any one of claims 1 to 5, wherein the statistical correlation is calculated by dividing the distance between the space-time bin and the virtual observation location by a correlation distance for the space-time bin.

7. The method of claim 6 , wherein the statistical correlation is further calculated using a sensitivity to the aggregated movement observation data.

8. 8. The integrated weather model output is at least one hour old. The method according to

9. 1. A system comprising: Memory, a processor coupled to the memory, the processor comprising: Receive integrated weather model output; receiving a statistical correlation based on the distance relating the weather observation data for space-time bins associated with a limited spatial distance and time period to the virtual observation location; receiving aggregated mobile observation data provided by a plurality of non-stationary sensors for said space-time bins; and adjusting the integrated weather model output based on the statistical correlation and the aggregated mobile observation data.

10. The system of claim 9 , wherein the weather observation data for the space-time bin comprises the aggregated mobile observation data.

11. The system of claim 9 or 10, wherein the statistical correlation relating the weather observation data for the space-time bins to the virtual observation locations is further based on time of day.

12. The system of any one of claims 9 to 11, wherein the statistical correlation relating the weather observation data for the space-time bins to the virtual observation locations is further based on a one-month period.

13. The system of any one of claims 9 to 12, wherein the integrated weather model output is generated using an input weather model that utilises fixed point observation data as input.

14. The system of any one of claims 9 to 13, wherein the statistical correlation is calculated by dividing the distance between the space-time bin and the virtual observation location by a correlation distance for the space-time bin.

15. The system of any one of claims 9 to 14, wherein the statistical correlation is further calculated using sensitivity to the aggregated mobile observation data.

Images