Construction method of underlying surface dust emission level prediction model based on ground-atmosphere coupled remote sensing

Abstract

The present application discloses a method for constructing a model for forecasting the dust emission level of an underlying surface based on ground-atmosphere coupled remote sensing, comprising real-time collection of multi-source observation data, preprocessing the multi-source observation data to obtain original observation data; obtaining key meteorological forecasting factors through the original observation data, and obtaining a sample data set by taking 16 days as an interval statistic; using the mean of aerosol product data as the true value of the dust emission amount, training and testing the sample data set using a random forest algorithm, and constructing a machine learning model for identifying dust emission conditions; using the machine learning model for identifying dust emission conditions to predict and calculate a dust emission index, revising the model results according to the newly added snow accumulation, precipitation phase and precipitation products, and outputting a forecast of the dust emission level for the next ten days; the present invention dynamically adjusts the dust emission level by distinguishing different precipitation types and precipitation amounts, so as to improve the accuracy and pertinence of the forecast.

Description

Technical Field

[0001] The present application relates to the technical field of sandstorm weather level forecasting, and in particular to a method for constructing a surface dust level forecasting model based on ground-atmosphere coupled remote sensing. Background Art

[0002] The existing grid-based real-time information of sandstorm weather mostly adopts the method of direct interpolation of ground-based station observations. Due to the serious uneven distribution of sandstorm source areas in East Asia, especially ground-based observations, coupled with complex underlying surface conditions, it is difficult to obtain comprehensive sandstorm disaster weather distribution information relying solely on ground-based observations.

[0003] Satellite remote sensing can provide atmospheric pollution monitoring under broad background conditions, and retain the original characteristics of particulate matter through remote sensing methods without being destroyed by physical means. It has broad application prospects in the spatial distribution monitoring and long-distance transmission of atmospheric particulate matter. In the field of environmental monitoring and meteorological forecasting, the prediction of the sand emission level of the underlying surface is of great significance for the assessment of sandstorms, soil wind erosion and air quality. At present, the technical solutions widely used in the market mostly rely on ground observation station data combined with meteorological models for prediction, such as using meteorological satellite data combined with parameters such as wind speed and humidity of ground meteorological stations to estimate the sand emission level through empirical formulas or statistical models. However, these existing technical solutions have the following obvious disadvantages: ground observation stations are unevenly distributed, especially in remote or complex terrain areas, data acquisition is difficult, resulting in limited prediction accuracy, traditional models are often based on complex physical equations, with large computational complexity, and it is difficult to achieve fast and efficient real-time predictions. Most of the existing models ignore the dynamic interaction between surface characteristics (such as soil moisture, vegetation cover) and atmospheric conditions, which affects the accuracy of the prediction.

[0004] At present, the application of single observation data in the actual dust situation still has certain limitations, and its temporal and spatial resolution or data integrity still cannot meet the needs of actual quantitative analysis. Compared with reanalysis data, it has the disadvantage of a longer time lag. Therefore, a method for constructing a surface dust level forecasting model based on ground-atmosphere coupled remote sensing is needed. Summary of the invention

[0005] The present application provides a method for constructing a surface sand emission level forecasting model based on ground-atmosphere coupled remote sensing, which can improve the accuracy of actual observation of disastrous weather.

[0006] In the first aspect, the present application provides a method for constructing a surface sand emission level prediction model based on ground-atmosphere coupled remote sensing, which adopts the following technical solution:

[0007] Collect multi-source observation data in real time, and pre-process the multi-source observation data to obtain original observation data;

[0008] The key meteorological forecast factors are obtained through the original observation data. The key meteorological forecast factors include normalized difference vegetation index, soil moisture, soil temperature, precipitation and snow accumulation. Except for the snow accumulation which is the average snow depth of the day, the rest are obtained by taking 16 days as the interval statistics to obtain multi-year sample data sets.

[0009] The mean value of aerosol product data is used as the true value of dust emission, and the random forest algorithm is used to train and test the sample data set to build a machine learning model for identifying dust emission conditions;

[0010] After using the machine learning model to predict and calculate the dust emission index, the initial results of the dust emission level forecast for the next ten days are output through the discrimination conditions.

[0011] The discrimination conditions specifically include:

[0012] When the dust emission index is ≥0.61, the grid point is judged to be dust emission level 1 and marked as extremely high;

[0013] When the dust emission index satisfies 0.61>≥-0.39, the grid point is judged as dust emission level 2 and marked as prone to dust emission;

[0014] When -0.39>Dust Raising Index≥-0.82, the grid point is judged to be dust raising level 3 and marked as dust raising possible;

[0015] When -0.82>Dust Raising Index≥-1.32, the grid point is judged to be dust raising level 4 and marked as not prone to dust raising;

[0016] When the value <-1.32 is satisfied, the grid point is judged to be at level 5 of sand generation and marked as difficult to generate sand.

[0017] The initial results are revised based on the newly added snow data and precipitation phase and precipitation products as supplementary conditions, and the daily sand level forecast for the next ten days is output;

[0018] The specific modification methods include: distinguishing precipitation phases according to the grid points, judging the precipitation type and precipitation amount. If the precipitation type shows that there is a precipitation type, but the precipitation data is zero, the grid point is directly considered as a no-precipitation point.

[0019] When the grid precipitation type is rain or sleet:

[0020] When the 24-hour cumulative precipitation is ≥10mm, the dust emission level forecast for the grid point within 48 hours is 5, and the forecast for 72 to 96 hours is reduced by one level from the original level. The forecast for 120 hours and thereafter is calculated based on the threshold of the dust emission index.

[0021] When 5mm≤24h cumulative precipitation<10mm, the dust emission level of the grid point within 24 hours = Level 5, the 48-hour forecast is reduced by one level from the original level, and the forecast for 72 hours and thereafter is calculated based on the threshold of the sand and dust emission index;

[0022] When 1mm≤24h cumulative precipitation is less than 5mm, the dust emission level of the grid point will be reduced by one level within 24 hours, and the forecast for 48 hours and thereafter will be obtained based on the threshold of the dust emission index;

[0023] When R < 1 mm, the dust emission level of the grid point is calculated according to the threshold of the dust emission index;

[0024] When the grid precipitation type is snowfall:

[0025] When the 24-hour cumulative precipitation is ≥5mm or the new snow accumulation is ≥5mm, the dust emission level of the grid point will be level 5 within 48 hours, and the forecast for 72 to 96 hours will be reduced by one level from the original level. The forecast for 120 hours and thereafter will be calculated based on the threshold of the sand and dust emission index;

[0026] When 2.5mm≤24h cumulative precipitation<5mm or 2.5mm≤new snow accumulation<5mm, the dust raising level within 24 hours is level 5, the 48-hour forecast is downgraded by one level, and the forecast for 72 hours and thereafter is calculated based on the threshold of the dust raising index;

[0027] When the accumulated precipitation of 1mm≤24h is less than 2.5mm or the accumulated precipitation of 1mm≤24h is less than 2.5mm, the dust emission level of the grid point will be reduced by one level within 24 hours. The forecast for 48 hours and thereafter will be obtained based on the threshold of the dust emission index.

[0028] When the 24-hour cumulative precipitation is less than 1 mm, the dust emission level of the grid point is calculated based on the threshold of the dust emission index.

[0029] Furthermore, the multi-source observation data include Sentinel satellite aerosol index product data, atmospheric environment monitoring data and satellite remote sensing observation data.

[0030] Furthermore, the data set includes the average value of soil moisture with the maximum NDVI value calculated at intervals of 16 days, the variance of soil temperature, the accumulated precipitation value, the snow coverage rate, and the average value of aerosol.

[0031] Furthermore, the random forest algorithm divides the sample data set into a training set and a test set, performs data enhancement on the training set data, and extracts influencing factors.

[0032] MSE is used as the loss function to evaluate the scheduling performance. The calculation of MSE is as follows:

[0033]

[0034] where y i is the ith actual data, is the i-th prediction data, and M represents the amount of training data.

[0035] In a second aspect, the present application provides a computer device, which adopts the following technical solution:

[0036] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect when executing the program.

[0037] In a third aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution:

[0038] A computer-readable storage medium stores a computer program that can be loaded by a processor and execute any one of the methods in the first aspect.

[0039] In summary, the present application includes at least one of the following beneficial technical effects:

[0040] This application provides a method for constructing a surface dust emission level forecasting model based on ground-atmosphere coupled remote sensing. By collecting and processing multi-source observation data in real time, special attention is paid to the influence of precipitation phase and precipitation amount in the process of revising the model results. By distinguishing different precipitation types and precipitation amounts, the dust emission level is dynamically adjusted to improve the accuracy and pertinence of the forecast. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of a method for constructing a model for forecasting the level of underlying surface sand emission based on ground-atmosphere coupled remote sensing in an embodiment of the present application. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0043] The present application embodiment discloses a method for constructing a model for forecasting the level of underlying surface sand emission based on ground-atmosphere coupled remote sensing, which adopts the following technical solution:

[0044] Reference Figure 1 A method for constructing a surface sand emission level prediction model based on ground-atmosphere coupled remote sensing includes the following steps:

[0045] S101: collecting multi-source observation data in real time, and preprocessing the multi-source observation data to obtain original observation data;

[0046] In this embodiment, multi-source observation data may include ground automatic station meteorological observation data, atmospheric environment monitoring data and satellite remote sensing observation data. The original observation data include ground automatic station meteorological original data, atmospheric environment original data and satellite remote sensing original data. For example, based on the H-9 geostationary meteorological satellite, after declouding processing, satellite remote sensing observation data can be obtained. Based on the ground observation data of the China National Environmental Monitoring Center and the China Meteorological Administration, ground automatic station meteorological observation data and atmospheric environment monitoring data can be obtained. These data can be synchronized to the local database in real time. As long as a certain collection frequency is preset, real-time acquisition of multi-source observation data can be achieved.

[0047] In one embodiment of the present application, the preprocessing of the multi-source observation data includes:

[0048] According to the underlying surface factors most relevant to dust formation as the key meteorological forecast factors for dust formation, the normalized vegetation index (NDVI), soil moisture, soil temperature, precipitation, and snow cover were selected as the influencing factors of dust emission conditions. The mean value of the aerosol product data of Sentinel 5P was used as the true value of dust emission (-2-2). The maximum value of NDVI, the average value of soil moisture, the variance of soil temperature, the cumulative value of precipitation, the coverage rate of snow and the average value of aerosol were statistically calculated at intervals of 16 days to establish a data set of each factor for 15 days from 2019 to 2024. Random forest regression was used to establish a dust emission condition model. 2000 points were randomly sampled for each image of the processed data. Four types of models were established for target grassland, forest land, cultivated land, and desert by training all points, and used for model prediction. The dust emission index (SR) was calculated by inputting the maximum value of NDVI, the average value of soil moisture, the variance of soil temperature, the cumulative value of precipitation, and the coverage rate of snow at intervals of 16 days. The specific steps of model construction are as follows: according to the underlying surface factors most relevant to dust formation as the key meteorological forecast factors for dust formation, a total of five factors, namely, normalized difference vegetation index (NDVI), soil moisture, soil temperature, precipitation, and snow cover, are selected as the influencing factors of dust emission conditions. The mean value of aerosol product data of Sentinel 5P is used as the true value of dust emission (-2-2). The maximum value of NDVI, the mean value of soil moisture, the variance of soil temperature, the cumulative value of precipitation, the coverage rate of snow cover, and the mean value of aerosol are statistically analyzed at intervals of 16 days to establish a data set of each factor for 15 days from 2019 to 2024. The established data set is divided into training and test sets according to the time interval. The sample data are trained and tested using a variety of machine learning algorithms such as random forest algorithm, decision tree model, and support vector machine. A machine learning model for identifying dust emission conditions is constructed. The performance of the model on the test set is tested using the hit rate (POD), false alarm rate (FAR), false negative rate (FNR), and TS score of the model to evaluate the model's ability to identify dust emission conditions. Tree models such as random forests and decision trees can determine the relative contribution of different predictors to the model by outputting the importance scores of the predictors. Predictors with higher contributions are selected and cross-validation and model tuning are performed using the grid search method (GridSearchCV) in Scikit-learn to select the machine learning model that most accurately identifies sand-raising conditions.

[0049] After obtaining the SR through model prediction calculation, the model results are further revised based on the additional snow accumulation, precipitation phase and precipitation products, and the daily sand level (SL) forecast for the next ten days is output.

[0050] When SR≥0.61 is met, the dust intensity at the grid point is judged to be level 1, and the dust intensity is marked as extremely high;

[0051] When 0.61>SR≥-0.39 is satisfied, the dust intensity at the grid point is judged to be level 2, and the dust intensity is marked as easy to cause dust;

[0052] When -0.39>SR≥-0.82 is satisfied, the dust intensity at the grid point is judged to be dust raising grade 3, and the dust intensity is marked as dust raising potential;

[0053] Dust Level 4 (not easy to raise dust): When -0.82>SR≥-1.32 is satisfied, the dust intensity at the grid point is determined to be dust level 4, and the dust intensity is marked as not easy to raise dust;

[0054] Dust level 5 (difficult to raise dust): When <-1.32 is satisfied, the dust intensity at the grid point is determined to be dust level 5, and the dust intensity is marked as difficult to raise dust;

[0055] Reporting time: 08:00, daily, 240h

[0056] Specific methods for revising model results based on the added snow accumulation, precipitation phase and precipitation products:

[0057] According to the grid points, the precipitation phase is distinguished and judged. The precipitation phase includes the 24-hour cumulative precipitation and the newly added snow accumulation. After judging the precipitation type, the precipitation amount is judged. When the precipitation type shows that there is a precipitation type, but the precipitation data is 0, the data is not involved in the judgment. The grid point is directly considered as a no precipitation point, and the newly added snow accumulation data is used as a supplementary condition for judgment. Specifically:

[0058] 3.1 When the grid precipitation type is rain or sleet:

[0059] 3.1.1 When R ≥ 10 mm, the grid point is predicted to be SL = 5 (difficult to raise sand) within 48 hours, and the forecast from 72 to 96 hours is reduced by one level from the original level. The forecast for 120 hours and thereafter is directly obtained by using the model value according to the SR threshold;

[0060] 3.1.2 When 5mm≤R<10mm, the SL of the grid point is 5 (difficult to raise sand) within 24 hours, the 48-hour forecast is downgraded by one level, and the forecast for 72 hours and thereafter is directly obtained by using the model value according to the SR threshold;

[0061] 3.1.3 When 1mm≤R<5mm, the SL of the grid point will be reduced by one level within 24 hours. The forecast for 48 hours and thereafter will be directly obtained by using the model value according to the SR threshold.

[0062] 3.1.4 When R < 1 mm, the SL of the grid point is directly calculated using the model value according to the SR threshold.

[0063] 3.2 When the grid precipitation type is snowfall:

[0064] 3.2.1 When R ≥ 5 mm or the amount of new snow accumulation ≥ 5 mm, the sand level of the grid point within 48 hours is level 5 (difficult to sand), and the forecast for 72 to 96 hours is reduced by one level from the original level. The forecast for 120 hours and thereafter is directly obtained by using the model value according to the SR threshold;

[0065] 3.2.2 When 2.5mm≤R<5mm or 2.5mm≤new snow accumulation<5mm, the dust level within 24 hours is 5 (difficult to dust), the 48-hour forecast is downgraded by one level, and the forecast for 72 hours and thereafter is directly obtained by using the model value according to the SR threshold;

[0066] 3.2.3 When 1mm≤R<2.5mm or 1mm≤R<2.5mm, the SL of the grid point will be reduced by one level within 24 hours. The forecast for 48 hours and thereafter will be directly obtained by using the model value according to the SR threshold.

[0067] 3.2.4 When R < 1 mm, the SL of the grid point is directly calculated using the model value according to the SR threshold.

[0068] In one embodiment of the present application, a large-scale sandstorm weather process occurred on June 21, 2024 as a verification of this embodiment. A sandstorm weather process occurred in the central and southern part of region A, the central, western and southeastern part of region B, and the northern and southeastern part of other regions.

[0069] The observation data of each factor involved in machine learning and the forecast results of the Dust Emission Index (SR) when relatively rare sandstorm weather occurs in the summer of 2024 (June 24). From top to bottom, they are the cumulative precipitation in the past 15 days (mm), the maximum NDVI in the past 10 days, the average snow depth on June 24 (cm), the variance of soil temperature at a depth of 0-5 cm on June 24 and the past 15 days, the average soil moisture content at a depth of 0-5 cm in the past 15 days (m3 / m-3), and the forecast results of the Dust Emission Index obtained by the machine learning algorithm.

[0070] As can be seen from the above, the present application provides a method for constructing a surface sand-raising grade forecasting model based on ground-atmosphere coupled remote sensing. By real-time collection and processing of multi-source observation data, the present application overcomes the problems of insufficient "spatial quantification", "real-time" and "comprehensiveness of single data" in the monitoring and early warning services of disastrous sandstorm weather, and adapts to the needs of refined, spatiotemporal quantitative monitoring and early warning services of disastrous sandstorm weather. Through the preset model library, multi-source data can be accurately processed to obtain accurate identification results, and then the identification results are fused by combining the distance weight method. Through the comprehensive application of ground meteorological observations, atmospheric environment monitoring, satellite remote sensing observations and other data, a satellite inversion sand and dust model integrating multi-source data is constructed, and a more comprehensive, high-precision, minute-level sand and dust grid monitoring product is constructed. The real-time and accuracy of identification are greatly improved, reaching more than 85%, and it is capable of providing support for the ecological improvement of the East Asian region and the spatiotemporal quantitative and refined development of sandstorm weather monitoring and early warning services.

[0071] In the several embodiments provided in this application, it should be understood that the provided methods and devices can be implemented in other ways. For example, the system embodiments described above are only illustrative; for example, the division of a component is only a logical function division, and there may be other division methods in actual implementation, such as multiple components can be combined or integrated into another system, or some features can be ignored or not executed.

[0072] The embodiment of the present application also discloses a computer device.

[0073] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned non-performing asset valuation method is implemented.

[0074] The embodiment of the present application also discloses a computer-readable storage medium.

[0075] A computer-readable storage medium stores a computer program that can be loaded by a processor and execute any of the above-mentioned non-performing asset valuation methods.

[0076] Among them, computer-readable storage media can be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, apparatus or device; the program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0077] It should be noted that in the above embodiments, the description of each embodiment has different emphases, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0078] The above are all preferred embodiments of the present application, and are not intended to limit the protection scope of the present application. Any feature disclosed in this specification (including the abstract and drawings), unless otherwise stated, can be replaced by other equivalent or alternative features with similar purposes. That is, unless otherwise stated, each feature is only an example of a series of equivalent or similar features.

Claims

1. A method for constructing a surface sand level prediction model based on ground-atmosphere coupled remote sensing, characterized in that: The following steps are involved: Collect multi-source observation data in real time, and pre-process the multi-source observation data to obtain original observation data; The key meteorological forecast factors are obtained through the original observation data. The key meteorological forecast factors include five elements: normalized difference vegetation index, soil moisture, soil temperature, precipitation and snow accumulation. Except for snow accumulation, which is the average snow depth on the day, the other elements are statistically obtained by taking 16 days as the interval to obtain a multi-year sample data set; The mean value of aerosol product data is used as the true value of dust emission, and the random forest algorithm is used to train and test the sample data set to build a machine learning model for identifying dust emission conditions; After using the machine learning model to predict and calculate the dust emission index, the initial results of the daily dust emission level forecast for the next ten days are output through the discrimination conditions. The discrimination conditions specifically include: When the dust emission index is ≥0.61, the grid point is judged to be dust emission level 1 and marked as extremely high; When the dust emission index satisfies 0.61>≥-0.39, the grid point is judged as dust emission level 2 and marked as prone to dust emission; When -0.39>Dust Raising Index≥-0.82, the grid point is judged to be dust raising level 3 and marked as dust raising possible; When -0.82>Dust Raising Index≥-1.32, the grid point is judged as dust raising grade 4 and marked as not prone to dust raising; When the value <-1.32 is satisfied, the grid point is judged as level 5 of sand generation and marked as difficult to generate sand; The initial results are revised based on the newly added snow data and precipitation phase and precipitation products as supplementary conditions, and the daily sand level forecast for the next ten days is output; The specific revision method includes: distinguishing precipitation phases according to the grid points, judging the precipitation type and precipitation amount. If the precipitation type shows that there is a precipitation type, but the precipitation data is zero, the grid point is directly considered as a no-precipitation point. When the grid precipitation type is rain or sleet: When the 24-hour cumulative precipitation is ≥10mm, the dust emission level forecast for the grid point within 48 hours is 5, and the forecast for 72 to 96 hours is reduced by one level from the original level. The forecast for 120 hours and thereafter is calculated based on the threshold of the dust emission index. When 5mm≤24h cumulative precipitation<10mm, the dust emission level of the grid point within 24 hours = Level 5, the 48-hour forecast is reduced by one level from the original level, and the forecast for 72 hours and thereafter is calculated based on the threshold of the sand and dust emission index; When 1mm≤24h cumulative precipitation is less than 5mm, the dust emission level of the grid point will be reduced by one level within 24 hours, and the forecast for 48 hours and thereafter will be obtained based on the threshold of the dust emission index; When the 24-hour cumulative precipitation is less than 1 mm, the dust emission level of the grid point is calculated based on the threshold of the dust emission index; When the grid precipitation type is snowfall: When the 24-hour cumulative precipitation is ≥5mm or the new snow accumulation is ≥5mm, the dust emission level of the grid point will be level 5 within 48 hours, and the forecast for 72 to 96 hours will be reduced by one level from the original level. The forecast for 120 hours and thereafter will be calculated based on the threshold of the sand and dust emission index; When 2.5mm≤24h cumulative precipitation<5mm or 2.5mm≤new snow accumulation<5mm, the dust raising level within 24 hours is level 5, the 48-hour forecast is downgraded by one level, and the forecast for 72 hours and thereafter is calculated based on the threshold of the dust raising index; When the accumulated precipitation of 1mm≤24h is less than 2.5mm or the accumulated precipitation of 1mm≤24h is less than 2.5mm, the dust emission level of the grid point will be reduced by one level within 24 hours. The forecast for 48 hours and thereafter will be obtained based on the threshold of the dust emission index. When the 24-hour cumulative precipitation is less than 1 mm, the dust emission level of the grid point is calculated based on the threshold of the dust emission index.

2. The method for constructing a surface sand emission level prediction model based on ground-atmosphere coupled remote sensing according to claim 1 is characterized in that: The multi-source observation data include Sentinel satellite aerosol index product data, atmospheric environment monitoring data and satellite remote sensing observation data.

3. The method for constructing a surface sand emission level prediction model based on ground-atmosphere coupled remote sensing according to claim 1 is characterized in that: The data set includes the average value of soil moisture with the maximum value of NDVI at intervals of 16 days, the variance of soil temperature, precipitation accumulation, snow cover and the average value of aerosol.

4. The method for constructing a surface sand emission level prediction model based on ground-atmosphere coupled remote sensing according to claim 1 is characterized in that: The random forest algorithm divides the sample data set into a training set and a test set, performs data enhancement on the training set data, and extracts influencing factors; MSE is used as the loss function to evaluate the scheduling performance. The calculation of MSE is as follows: where y i is the ith actual data, is the i-th prediction data, and M represents the amount of training data.

5. A computer-readable storage medium, characterized in that: The computer-readable storage medium comprises a computer program for implementing the method according to any one of claims 1 to 4.

6. A computing device, characterized in that The computing device comprises a memory and a processor, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, the method according to any one of claims 1 to 4 is implemented.

Images