A method and system for calculating evapotranspiration based on multi-source remote sensing data

Through multi-source remote sensing data and site measurement data, combined with random forest and artificial neural network model PI-ANN, the PT correction coefficient prediction in evaporation calculation is improved, and the problem of insufficient evaporation calculation accuracy in the prior art is solved. High accuracy and interpretability results are achieved.

CN118535837BActive Publication Date: 2024-11-08BEIJING NORMAL UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410747999.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-06-11
Publication Date
2024-11-08
Estimated Expiration
2044-06-11

AI Technical Summary

Technical Problem

The reliability of the existing evaporation amount calculation method in many areas to be calculated is due to insufficient consideration of aerodynamic components, which makes the default PT correction coefficient used in the Priestly-Taylor formula unable to guarantee the calculation accuracy.

Method used

Using a method based on multi-source remote sensing data, the PT correction coefficient is calculated through the actual measurement of the evaporative dispersion amount on the site, the key influencing factors of the evaporative dispersion amount is determined, and the prediction is performed using the artificial neural network model PI-ANN that fuses physical information to improve the calculation of the PT correction coefficient. .

Benefits of technology

Improve the accuracy of evaporation evaporation calculation, especially in global and regional forecasts, achieve 70% and 88% accuracy, increase the interpretability of the model, and realize the time-series scale expansion from annual to daily scale, suitable for No data predictions for areas with data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118535837B_ABST
    Figure CN118535837B_ABST
Patent Text Reader

Abstract

The present invention relates to the technical field of evapotranspiration calculation, and in particular to an evapotranspiration calculation method and system based on multi-source remote sensing data, including: according to the measured evapotranspiration of multiple stations, using the evapotranspiration Preistly-Taylor formula to calculate the PT correction coefficient #imgabs0# of multiple stations; according to the multi-source remote sensing data of the multiple stations, calculating a number of evapotranspiration influencing factor data, and determining three evapotranspiration key influencing factor data from the several evapotranspiration influencing factor data; inputting the determined three evapotranspiration key influencing factor data and the corresponding PT correction coefficient #imgabs1# as a label and training an artificial neural network model PI-ANN that integrates physical information; using the trained PI-ANN to predict the PT correction coefficient #imgabs2# of the evapotranspiration area to be calculated; according to the predicted PT correction coefficient #imgabs3#, using the Preistly-Taylor formula to calculate the evapotranspiration of the evapotranspiration area to be calculated. By adopting the present invention, the evapotranspiration of the evapotranspiration area to be calculated can be accurately calculated.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of evapotranspiration calculation, and in particular to an evapotranspiration calculation method and system based on multi-source remote sensing data. Background Art

[0002] Evapotranspiration ) is an important part of the hydrological cycle and can affect local climate, water resources and other factors. Accurate acquisition of land evapotranspiration can improve the understanding of changes in the hydrological cycle under the background of climate change, and because of its important impact on surface runoff, it helps to improve water resource management capabilities. At present, the commonly used methods for obtaining evapotranspiration at the basin scale are site measurements (direct and indirect observations), such as the lysimeter method or the eddy covariance method; the other is to use the empirical formula for evapotranspiration for calculation. Among them, the Priestly-Taylor formula has been widely used due to its simple formula form and relatively less required input data, but its accuracy is mainly limited by the PT correction coefficient. Due to insufficient consideration of aerodynamic components, the existing PT correction factor (alpha) Using the default value (1.26) does not guarantee reliability for many regions where evapotranspiration is being calculated. Summary of the invention

[0003] The present invention provides a method and system for calculating evapotranspiration based on multi-source remote sensing data to solve the problems existing in the above-mentioned prior art. The technical solution is as follows:

[0004] On the one hand, a method for calculating evapotranspiration based on multi-source remote sensing data is provided, including:

[0005] S1. Based on the measured evapotranspiration of multiple stations, the PT correction coefficients of multiple stations are calculated using the evapotranspiration Preistly-Taylor formula. ;

[0006] S2. Calculate a number of evapotranspiration influencing factor data based on the multi-source remote sensing data of the multiple stations, and determine three evapotranspiration key influencing factor data from the number of evapotranspiration influencing factor data;

[0007] S3. The three key factors affecting evapotranspiration and the corresponding PT correction coefficients as labels are determined. , input and train the artificial neural network model PI-ANN that integrates physical information;

[0008] S4. Using the trained PI-ANN, the PT correction factor for the area where evapotranspiration is to be calculated Make predictions;

[0009] S5. Correction factor based on predicted PT , use the Preistly-Taylor formula to calculate the evapotranspiration of the area where the evapotranspiration is to be calculated.

[0010] Optionally, the evapotranspiration Preistly-Taylor formula is as follows:

[0011]

[0012] in, is the evaporation rate, is the slope of the saturated water vapor pressure-temperature curve, is a constant, is the net radiation, For soil heat, is the PT correction factor, is the unit conversion factor;

[0013] Apart from , the other parameters in the formula are known quantities, then according to the formula, we can calculate .

[0014] Optionally, the S2 specifically includes:

[0015] According to the multi-source remote sensing data of the plurality of stations, several evapotranspiration influencing factor data are calculated, including but not limited to the following twelve factors: annual average temperature T, annual maximum temperature TM, annual minimum temperature TN, annual average rainfall P, annual average saturated vapor pressure difference VPD, annual average upward longwave radiation flux ulwrf, annual average upward shortwave radiation flux uswrf, annual average downward shortwave radiation flux dswrf, annual average wind speed Wind, annual average relative humidity RH, annual average leaf area index LAI, annual average forest canopy height data TH;

[0016] The random forest method was used to calculate the feature importance ranking of the factors affecting evapotranspiration, and the results were as follows from large to small: TM > P > TN > LAI > T > VPD > dswrf > TH > RH > Wind > uswrf > ulwrf;

[0017] The three-factor combination of annual average rainfall, annual average temperature and annual average downward shortwave radiation flux was determined as the key influencing factor data of evapotranspiration.

[0018] Optionally, the artificial neural network model PI-ANN integrating physical information is composed of an input layer, two fully connected layers and an output layer, and the three key influencing factors of evapotranspiration determined and the corresponding PT correction coefficients as labels are used. The model is input through the input layer. The first fully connected layer fc1 includes 50 neurons, which convert the three key influencing factors of evapotranspiration into fifty intermediate influencing factors, and perform feature conversion and dimension mapping, including: by changing the weight of each connection line between each layer of the network, inputting the weighted sum of all outputs from the previous layer, and adding a fixed bias term to improve the flexibility of the neural network. This conversion enables the network to extract deeper information from the original input data and pass this information to the next layer of the network, and then convert it through the nonlinear activation function ReLU to generate the output of the first fully connected layer. The ReLU is as follows:

[0019]

[0020] The ReLU allows the model to maintain good gradient flow during training because of its simple calculation and effective solution to the gradient vanishing problem;

[0021] The second fully connected layer fc2 includes twenty neurons, which further process the information transmitted by the first fully connected layer. In this way, the feature space of the fully connected layer is expanded, and the model's ability to capture the intrinsic relationship of the data is increased;

[0022] The output layer uses the Sigmoid activation function to limit the output range to the reasonable range of the global PT correction coefficient between 0 and 1.5, which is consistent with the actual physical meaning and makes the prediction results more reasonable and usable.

[0023] Optionally, the loss function of the artificial neural network model PI-ANN integrating physical information is a composite loss function PI-ANN_Loss of machine learning loss and physical information loss. The PI-ANN_Loss aims to optimize the prediction performance of the model while ensuring that its output follows basic physical laws, including mean square error loss MSE_Loss and physical information loss Physics_Loss. Physics_Loss combines physical knowledge so that the composite loss function not only optimizes the statistical performance of the model, but also ensures that the model output is physically interpretable and reasonable. By directly encoding physical laws into the loss function, the decision-making process of the model is more transparent and the interpretability of the output is also enhanced. This part of the loss works by comparing the model output with the expected value calculated based on the physical formula. The formula is as follows:

[0024]

[0025]

[0026]

[0027]

[0028] in Refers to the physical constraint influencing factor, which guides the model to learn predictions that conform to physical laws. It is about the saturated vapor pressure difference Function formula, due to the PT correction factor In the Priestly-Taylor formula, it mainly expresses the regulation effect on air unsaturation, which is similar to Related, so use Improve the loss function and add physical meaning to the constraints. Indicates PT correction factor The predicted value of Indicates PT correction factor The true label value of .

[0029] On the other hand, a system for calculating evapotranspiration based on multi-source remote sensing data is provided, the system comprising:

[0030] The first calculation module is used to calculate the PT correction coefficients of multiple sites based on the measured evapotranspiration of multiple sites using the Preistly-Taylor formula for evapotranspiration ;

[0031] A second calculation module is used to calculate a plurality of evapotranspiration influencing factor data according to the multi-source remote sensing data of the plurality of stations, and determine three evapotranspiration key influencing factor data from the plurality of evapotranspiration influencing factor data;

[0032] Input training module, used to determine the three key factors affecting evapotranspiration and the corresponding PT correction coefficients as labels , input and train the artificial neural network model PI-ANN that integrates physical information;

[0033] Prediction module, used to use the trained PI-ANN to calculate the PT correction factor for the area where evapotranspiration is to be calculated Make predictions;

[0034] The third calculation module is used to correct the PT coefficient according to the prediction , use the Preistly-Taylor formula to calculate the evapotranspiration of the area where the evapotranspiration is to be calculated.

[0035] Optionally, the evapotranspiration Preistly-Taylor formula is as follows:

[0036]

[0037] in, is the evaporation rate, is the slope of the saturated water vapor pressure-temperature curve, is a constant, is the net radiation, For soil heat, is the PT correction factor, is the unit conversion factor;

[0038] Apart from , the other parameters in the formula are known quantities, then according to the formula, we can calculate .

[0039] Optionally, the second calculation module is specifically configured to:

[0040] According to the multi-source remote sensing data of the plurality of stations, several evapotranspiration influencing factor data are calculated, including but not limited to the following twelve factors: annual average temperature T, annual maximum temperature TM, annual minimum temperature TN, annual average rainfall P, annual average saturated vapor pressure difference VPD, annual average upward longwave radiation flux ulwrf, annual average upward shortwave radiation flux uswrf, annual average downward shortwave radiation flux dswrf, annual average wind speed Wind, annual average relative humidity RH, annual average leaf area index LAI, annual average forest canopy height data TH;

[0041] The random forest method was used to calculate the feature importance ranking of the factors affecting evapotranspiration, and the results were as follows from large to small: TM > P > TN > LAI > T > VPD > dswrf > TH > RH > Wind > uswrf > ulwrf;

[0042] The three-factor combination of annual average rainfall, annual average temperature and annual average downward shortwave radiation flux was determined as the key influencing factor data of evapotranspiration.

[0043] Optionally, the artificial neural network model PI-ANN integrating physical information is composed of an input layer, two fully connected layers and an output layer, and the three key influencing factors of evapotranspiration determined and the corresponding PT correction coefficients as labels are used. The model is input through the input layer. The first fully connected layer fc1 includes 50 neurons, which convert the three key influencing factors of evapotranspiration into fifty intermediate influencing factors, and perform feature conversion and dimension mapping, including: by changing the weight of each connection line between each layer of the network, inputting the weighted sum of all outputs from the previous layer, and adding a fixed bias term to improve the flexibility of the neural network. This conversion enables the network to extract deeper information from the original input data and pass this information to the next layer of the network, and then convert it through the nonlinear activation function ReLU to generate the output of the first fully connected layer. The ReLU is as follows:

[0044]

[0045] The ReLU allows the model to maintain good gradient flow during training because of its simple calculation and effective solution to the gradient vanishing problem;

[0046] The second fully connected layer fc2 includes twenty neurons, which further process the information transmitted by the first fully connected layer. In this way, the feature space of the fully connected layer is expanded, and the model's ability to capture the intrinsic relationship of the data is increased;

[0047] The output layer uses the Sigmoid activation function to limit the output range to the reasonable range of the global PT correction coefficient between 0 and 1.5, which is consistent with the actual physical meaning and makes the prediction results more reasonable and usable.

[0048] Optionally, the loss function of the artificial neural network model PI-ANN integrating physical information is a composite loss function PI-ANN_Loss of machine learning loss and physical information loss. The PI-ANN_Loss aims to optimize the prediction performance of the model while ensuring that its output follows basic physical laws, including mean square error loss MSE_Loss and physical information loss Physics_Loss. Physics_Loss combines physical knowledge so that the composite loss function not only optimizes the statistical performance of the model, but also ensures that the model output is physically interpretable and reasonable. By directly encoding physical laws into the loss function, the decision-making process of the model is more transparent and the interpretability of the output is also enhanced. This part of the loss works by comparing the model output with the expected value calculated based on the physical formula. The formula is as follows:

[0049]

[0050]

[0051]

[0052]

[0053] in Refers to the physical constraint influencing factor, which guides the model to learn predictions that conform to physical laws. It is about the saturated vapor pressure difference Function formula, due to the PT correction factor In the Priestly-Taylor formula, it mainly expresses the regulation effect on air unsaturation, which is similar to Related, so use Improve the loss function and add physical meaning to the constraints. Indicates PT correction factor The predicted value of Indicates PT correction factor The true label value of .

[0054] On the other hand, an electronic device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the above-mentioned evapotranspiration calculation method based on multi-source remote sensing data.

[0055] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement the above-mentioned evapotranspiration calculation method based on multi-source remote sensing data.

[0056] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0057] The present invention uses the three key influencing factors of evapotranspiration determined and the corresponding PT correction coefficients as labels based on multi-source remote sensing data and site measured evapotranspiration data. , input and train the artificial neural network model PI-ANN that integrates physical information. After multiple training and parameter adjustments, the trained PI-ANN is obtained. The trained PI-ANN is used to predict the PT correction coefficient Compared with the existing default value of 1.26, the global prediction accuracy is over 70%, and the regional prediction accuracy is over 88%, indicating that the trained PI-ANN prediction results are reliable, not only for the calculation of evapotranspiration has a good improvement effect, and increases the interpretability of the machine learning model, and the training data of the present invention is the average data of each year, and the application in the daily data prediction has also achieved good results, realizing the expansion of the time series scale from the annual scale to the daily scale, which not only improves the evapotranspiration accuracy, but also can be applied to areas with no data in practical applications (no measured data from the site, but multi-source remote sensing data can be obtained). BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0059] Figure 1 This is a flow chart of a method for calculating evapotranspiration based on multi-source remote sensing data provided by an embodiment of the present invention;

[0060] Figure 21 is a schematic diagram of the order of importance of the characteristics of the factors affecting evapotranspiration provided by an embodiment of the present invention;

[0061] Figure 3 It is a schematic diagram of the structure of an artificial neural network model PI-ANN integrating physical information provided by an embodiment of the present invention;

[0062] Figure 4 is the global PT correction factor provided by the embodiment of the present invention Distribution map;

[0063] Figure 5 is a comparison diagram of the improved evapotranspiration calculation result, the original calculation result and the measured value provided by the embodiment of the present invention;

[0064] Figure 6 It is an overall flow chart provided by an embodiment of the present invention;

[0065] Figure 7 This is a block diagram of an evapotranspiration calculation system based on multi-source remote sensing data provided by an embodiment of the present invention;

[0066] Figure 8 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0067] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0068] The embodiment of the present invention provides a method for calculating evapotranspiration based on multi-source remote sensing data, which can be implemented by an electronic device, which can be a terminal or a server. Figure 1 The flowchart of a method for calculating evapotranspiration based on multi-source remote sensing data is shown in FIG. The processing flow of the method may include the following steps:

[0069] S1. Based on the measured evapotranspiration of multiple stations, the PT correction coefficients of multiple stations are calculated using the evapotranspiration Preistly-Taylor formula. ;

[0070] Optionally, the evapotranspiration Preistly-Taylor formula is as follows:

[0071]

[0072] in, is the evaporation rate, is the slope of the saturated water vapor pressure-temperature curve, is a constant, is the net radiation, For soil heat, is the PT correction factor, is the unit conversion factor;

[0073] Apart from , the other parameters in the formula are all known quantities (the embodiment of the present invention uses the Fluxnet2015 data set from the global flux observation network Fluxnet (https: / / fluxnet.org / ), selects the measured latent heat flux data and temperature data of 163 stations, converts the measured latent heat flux data into measured evapotranspiration data, and calculates other parameter data based on the temperature data), then according to the formula, .

[0074] S2. Calculate a number of evapotranspiration influencing factor data based on the multi-source remote sensing data of the multiple stations, and determine three evapotranspiration key influencing factor data from the number of evapotranspiration influencing factor data;

[0075] Optionally, the S2 specifically includes:

[0076] According to the multi-source remote sensing data of the plurality of stations, several evapotranspiration influencing factor data are calculated, including but not limited to the following twelve factors: annual average temperature T, annual maximum temperature TM, annual minimum temperature TN, annual average rainfall P, annual average saturated vapor pressure difference VPD, annual average upward longwave radiation flux ulwrf, annual average upward shortwave radiation flux uswrf, annual average downward shortwave radiation flux dswrf, annual average wind speed Wind, annual average relative humidity RH, annual average leaf area index LAI, annual average forest canopy height data TH;

[0077] The embodiment of the present invention collects the annual average temperature (T), annual average maximum temperature (TM), annual average minimum temperature (TN), annual average saturated vapor pressure difference (VPD) and annual average rainfall (P) of the CRU TS 4.05 data set, and at the same time, collects the annual average upward longwave radiation flux (ulwrf), annual average upward shortwave radiation flux (uswrf), annual average downward shortwave radiation flux (dswrf), annual average wind speed (Wind), annual average relative humidity (RH) and MODIS Version 6.1 products (MCD15A2H and MOD15A2H) extract the annual average leaf area index (LAI) data for each station, where the LAI value is calculated based on the data time range of each station. The multi-year annual average LAI is extracted at each station. In addition, the multi-year annual average forest canopy height data (TH) for each station is extracted based on the global data provided by the Earth Science Laser Altimeter System (GLAS) of the ICESat satellite. All data are preliminarily processed by the station input using python and matlab languages, and the original data are extracted, outliers are removed, and other operations are performed. Finally, the data of 12 evapotranspiration influencing factors are sorted out.

[0078] The random forest method is used to calculate the feature importance ranking of the factors affecting evapotranspiration (python's own calculation package). The results are as follows from large to small: TM > P > TN > LAI > T > VPD > dswrf > TH > RH> Wind > uswrf > ulwrf, such as Figure 2 As shown;

[0079] The three-factor combination of annual average rainfall, annual average temperature and annual average downward shortwave radiation flux was determined as the key influencing factor data of evapotranspiration.

[0080] In order to improve the generalization of the model, the influencing factor combination with the least number of influencing factors and the easiest data to obtain is preferred. By comparing different permutations and combinations of two-factor combinations, three-factor combinations and other combinations, the embodiment of the present invention finally determines that the three-factor combination of annual average rainfall, annual average temperature, and annual average downward shortwave radiation flux can achieve the best training effect. Although the three are not the three influencing factors with the highest feature importance, their combination can maximize the capture of the characteristics of site evapotranspiration changes, and the prediction accuracy reaches 75.1%. At the same time, rainfall, temperature, and downward shortwave radiation flux are data types that are easy to obtain, which is convenient for promotion in more areas without data.

[0081] S3. The three key factors affecting evapotranspiration and the corresponding PT correction coefficients as labels are determined. , input and train the artificial neural network model PI-ANN that integrates physical information;

[0082] like Figure 3 As shown, the artificial neural network model PI-ANN integrating physical information is composed of an input layer, two fully connected layers and an output layer (the embodiment of the present invention has found through research that this network structure is very effective in dealing with various regression and classification problems). The three key influencing factors of evapotranspiration (annual average rainfall, annual average temperature and annual average downward shortwave radiation flux) and the corresponding PT correction coefficients as labels are determined. The model is input through the input layer. The first fully connected layer fc1 includes 50 neurons, which convert the three key influencing factors of evapotranspiration into fifty intermediate influencing factors, and perform feature conversion and dimension mapping, including: by changing the weight of each connection line between each layer of the network, inputting the weighted sum of all outputs from the previous layer, and adding a fixed bias term to improve the flexibility of the neural network. This conversion enables the network to extract deeper information from the original input data and pass this information to the next layer of the network, and then convert it through the nonlinear activation function ReLU to generate the output of the first fully connected layer. The ReLU is as follows:

[0083]

[0084] The ReLU allows the model to maintain good gradient flow during training because of its simple calculation and effective solution to the gradient vanishing problem;

[0085] The second fully connected layer fc2 includes twenty neurons, which further process the information transmitted by the first fully connected layer. In this way, the feature space of the fully connected layer is expanded, and the model's ability to capture the intrinsic relationship of the data is increased;

[0086] The output layer uses the Sigmoid activation function to limit the output range to the reasonable range of the global PT correction coefficient between 0 and 1.5, which is consistent with the actual physical meaning and makes the prediction results more reasonable and usable.

[0087] Optionally, the loss function of the artificial neural network model PI-ANN integrating physical information is a composite loss function PI-ANN_Loss of machine learning loss and physical information loss, wherein PI-ANN_Loss aims to optimize the prediction performance of the model while ensuring that its output follows basic physical laws, including mean square error loss MSE_Loss and physical information loss Physics_Loss. By combining physical knowledge, the composite loss function not only optimizes the statistical performance of the model, but also ensures that the model output is physically interpretable and reasonable. By directly encoding physical laws into the loss function, the decision-making process of the model is more transparent and the interpretability of the output is also enhanced (physical information loss helps the model make reasonable predictions without sufficient data support. Even in environments with sparse data or high noise, the model can improve its generalization ability by following known physical laws. Introducing physical constraints during training can also help avoid overfitting problems caused by excessive model complexity, and is crucial for applications that need to explain model predictions to gain the trust of users or regulators). This part of the loss works by comparing the model output with the expected value calculated based on the physical formula. The formula is as follows:

[0088]

[0089]

[0090]

[0091]

[0092] in Refers to the physical constraint influencing factor, which guides the model to learn predictions that conform to physical laws. It is about the saturated vapor pressure difference Function formula, due to the PT correction factor In the Priestly-Taylor formula, it mainly expresses the regulation effect on air unsaturation, which is similar to Related, so use Improve the loss function and add physical meaning to the constraints. Indicates PT correction factor The predicted value of Indicates PT correction factor The true label value of .

[0093] The obtained data of three key factors affecting global evapotranspiration are input into the trained PI-ANN to predict the global PT correction coefficient. , whose distribution is as follows Figure 4 shown.

[0094] By applying the evapotranspiration calculation method of the embodiment of the present invention to the vorticity station of the Stipa leaf forest in Inner Mongolia, the calculated evapotranspiration value is similar to the measured value. Figure 5 As shown, the original calculation results represent the existing original PT formula method (PT correction factor The calculation results are calculated using the default value of 1.26). The improved calculation results represent the improved PT correction coefficient coefficient of the method of the embodiment of the present invention. The calculation results are shown below. The correlation coefficient R between the calculated evapotranspiration value and the measured value 2 Also known as the coefficient of determination, it is an important indicator to measure the degree of fit of a statistical model. 2 The RMSE increased from 0.86 to 0.88, and the root mean square error (RMSE) decreased from 0.05 to 0.03, indicating that the method of the embodiment of the present invention has a good improvement effect on the calculation of evapotranspiration and increases the interpretability of the machine learning model. In addition, the training data of the embodiment of the present invention is the average data of each year, and the application in the daily data prediction has also achieved good results, realizing the time series scale expansion from the annual scale to the daily scale, which not only improves the evapotranspiration accuracy, but can also be applied to areas without data in practical applications (no measured data of the site, but multi-source remote sensing data can be obtained). The overall flow chart is as follows: Figure 6 shown.

[0095] S4. Using the trained PI-ANN, the PT correction factor for the area where evapotranspiration is to be calculated Make predictions;

[0096] In practical applications, the three key factors affecting evapotranspiration calculated based on multi-source remote sensing data can be input into the trained PI-ANN to predict the PT correction coefficient of the evapotranspiration area to be calculated. .

[0097] S5. Correction factor based on predicted PT , use the Preistly-Taylor formula to calculate the evapotranspiration of the area where the evapotranspiration is to be calculated.

[0098] Other parameter data in the Preistly-Taylor formula obtained from multi-source remote sensing data and the predicted PT correction coefficient , use the Preistly-Taylor formula to calculate the evapotranspiration of the area where the evapotranspiration is to be calculated.

[0099] like Figure 7 As shown, an embodiment of the present invention further provides an evapotranspiration calculation system based on multi-source remote sensing data, the system comprising:

[0100] The first calculation module 710 is used to calculate the PT correction coefficients of multiple sites based on the measured evapotranspiration of multiple sites using the Preistly-Taylor formula of evapotranspiration ;

[0101] The second calculation module 720 is used to calculate a plurality of evapotranspiration influencing factor data according to the multi-source remote sensing data of the plurality of stations, and determine three evapotranspiration key influencing factor data from the plurality of evapotranspiration influencing factor data;

[0102] Input training module 730 is used to convert the determined three key influencing factors of evapotranspiration and the corresponding PT correction coefficients as labels into , input and train the artificial neural network model PI-ANN that integrates physical information;

[0103] Prediction module 740, used to use the trained PI-ANN to calculate the PT correction factor of the area where evapotranspiration is to be calculated Make predictions;

[0104] The third calculation module 750 is used to correct the PT coefficient according to the prediction , use the Preistly-Taylor formula to calculate the evapotranspiration of the area where the evapotranspiration is to be calculated.

[0105] Optionally, the evapotranspiration Preistly-Taylor formula is as follows:

[0106]

[0107] in, is the evaporation rate, is the slope of the saturated water vapor pressure-temperature curve, is a constant, is the net radiation, For soil heat, is the PT correction factor, is the unit conversion factor;

[0108] Apart from , the other parameters in the formula are known quantities, then according to the formula, we can calculate .

[0109] Optionally, the second calculation module is specifically configured to:

[0110] According to the multi-source remote sensing data of the plurality of stations, several evapotranspiration influencing factor data are calculated, including but not limited to the following twelve factors: annual average temperature T, annual maximum temperature TM, annual minimum temperature TN, annual average rainfall P, annual average saturated vapor pressure difference VPD, annual average upward longwave radiation flux ulwrf, annual average upward shortwave radiation flux uswrf, annual average downward shortwave radiation flux dswrf, annual average wind speed Wind, annual average relative humidity RH, annual average leaf area index LAI, annual average forest canopy height data TH;

[0111] The random forest method was used to calculate the feature importance ranking of the factors affecting evapotranspiration, and the results were as follows from large to small: TM > P > TN > LAI > T > VPD > dswrf > TH > RH > Wind > uswrf > ulwrf;

[0112] The three-factor combination of annual average rainfall, annual average temperature and annual average downward shortwave radiation flux was determined as the key influencing factor data of evapotranspiration.

[0113] Optionally, the artificial neural network model PI-ANN integrating physical information is composed of an input layer, two fully connected layers and an output layer, and the three key influencing factors of evapotranspiration determined and the corresponding PT correction coefficients as labels are used. The model is input through the input layer. The first fully connected layer fc1 includes 50 neurons, which convert the three key influencing factors of evapotranspiration into fifty intermediate influencing factors, and perform feature conversion and dimension mapping, including: by changing the weight of each connection line between each layer of the network, inputting the weighted sum of all outputs from the previous layer, and adding a fixed bias term to improve the flexibility of the neural network. This conversion enables the network to extract deeper information from the original input data and pass this information to the next layer of the network, and then convert it through the nonlinear activation function ReLU to generate the output of the first fully connected layer. The ReLU is as follows:

[0114]

[0115] The ReLU allows the model to maintain good gradient flow during training because of its simple calculation and effective solution to the gradient vanishing problem;

[0116] The second fully connected layer fc2 includes twenty neurons, which further process the information transmitted by the first fully connected layer. In this way, the feature space of the fully connected layer is expanded, and the model's ability to capture the intrinsic relationship of the data is increased;

[0117] The output layer uses the Sigmoid activation function to limit the output range to the reasonable range of the global PT correction coefficient between 0 and 1.5, which is consistent with the actual physical meaning and makes the prediction results more reasonable and usable.

[0118] Optionally, the loss function of the artificial neural network model PI-ANN integrating physical information is a composite loss function PI-ANN_Loss of machine learning loss and physical information loss. The PI-ANN_Loss aims to optimize the prediction performance of the model while ensuring that its output follows basic physical laws, including mean square error loss MSE_Loss and physical information loss Physics_Loss. Physics_Loss combines physical knowledge so that the composite loss function not only optimizes the statistical performance of the model, but also ensures that the model output is physically interpretable and reasonable. By directly encoding physical laws into the loss function, the decision-making process of the model is more transparent and the interpretability of the output is also enhanced. This part of the loss works by comparing the model output with the expected value calculated based on the physical formula. The formula is as follows:

[0119]

[0120]

[0121]

[0122]

[0123] in Refers to the physical constraint influencing factor, which guides the model to learn predictions that conform to physical laws. It is about the saturated vapor pressure difference Function formula, due to the PT correction factor In the Priestly-Taylor formula, it mainly expresses the regulation effect on air unsaturation, which is similar to Related, so use Improve the loss function and add physical meaning to the constraints. Indicates PT correction factor The predicted value of Indicates PT correction factor The true label value of .

[0124] The functional structure of the evapotranspiration calculation system based on multi-source remote sensing data provided in the embodiment of the present invention corresponds to the evapotranspiration calculation method based on multi-source remote sensing data provided in the embodiment of the present invention, which will not be described in detail here.

[0125] Figure 81 is a schematic diagram of the structure of an electronic device 800 provided in an embodiment of the present invention. The electronic device 800 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 801 and one or more memories 802, wherein the memory 802 stores at least one instruction, and the at least one instruction is loaded and executed by the processor 801 to implement the steps of the above-mentioned evapotranspiration calculation method based on multi-source remote sensing data.

[0126] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including instructions, which can be executed by a processor in a terminal to complete the above-mentioned evapotranspiration calculation method based on multi-source remote sensing data. For example, the computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0127] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0128] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for calculating evapotranspiration based on multi-source remote sensing data, characterized in that: The method comprises: S1. According to the measured evapotranspiration of multiple stations, the PT correction coefficient α of multiple stations is calculated using the evapotranspiration Preistly-Taylor formula; S2. Calculate a number of evapotranspiration influencing factor data based on the multi-source remote sensing data of the multiple stations, and determine three evapotranspiration key influencing factor data from the number of evapotranspiration influencing factor data; S3, inputting the determined three key influencing factors of evapotranspiration and the corresponding PT correction coefficient α as labels into and training the artificial neural network model PI-ANN integrating physical information; S4. Use the trained PI-ANN to predict the PT correction factor α for the area where evapotranspiration is to be calculated; S5. Calculate the evapotranspiration of the area to be calculated using the Preistly-Taylor formula according to the predicted PT correction coefficient α; The loss function of the artificial neural network model PI-ANN that integrates physical information is a composite loss function PI-ANN_Loss of machine learning loss and physical information loss. The PI-ANN_Loss aims to optimize the prediction performance of the model while ensuring that its output follows basic physical laws, including mean square error loss MSE_Loss and physical information loss Physics_Loss. Physics_Loss combines physical knowledge so that the composite loss function not only optimizes the statistical performance of the model, but also ensures that the model output is physically interpretable and reasonable. By directly encoding physical laws into the loss function, the decision-making process of the model is more transparent and the interpretability of the output is also enhanced. This part of the loss works by comparing the model output with the expected value calculated based on the physical formula. The formula is as follows: PI_feature t =1.5*VPD t -0.5 PI-ANN_Loss=MSE_Loss+Physics_Loss Among them, PI_feature t Refers to the physical constraint influencing factor, which guides the model to learn predictions that conform to physical laws. It is about the saturated vapor pressure difference VPD t Function formula, because the PT correction coefficient α in the Priestly-Taylor formula mainly represents the regulation effect on air unsaturation, and VPD t Related, so use VPD t Improve the loss function and add physical meaning to the constraints. t represents the predicted value of the PT correction factor α, O t Represents the true label value of the PT correction coefficient α.

2. The method according to claim 1, characterized in that The Preistly-Taylor formula for evapotranspiration is as follows: Where ET is evapotranspiration, Δ is the slope of the saturated water vapor pressure-temperature curve, γ is a constant, and R n is the net radiation, G is the soil heat, α is the PT correction factor, and λ is the unit conversion factor; Except for α, all other parameters in the formula are known quantities, so α can be calculated according to the formula.

3. The method according to claim 1, characterized in that The S2 specifically includes: According to the multi-source remote sensing data of the plurality of stations, several evapotranspiration influencing factor data are calculated, including but not limited to the following twelve factors: annual average temperature T, annual maximum temperature TM, annual minimum temperature TN, annual average rainfall P, annual average saturated vapor pressure difference VPD, annual average upward longwave radiation flux ulwrf, annual average upward shortwave radiation flux uswrf, annual average downward shortwave radiation flux dswrf, annual average wind speed Wind, annual average relative humidity RH, annual average leaf area index LAI, annual average forest canopy height data TH; The random forest method was used to calculate the feature importance ranking of the factors affecting evapotranspiration, and the results were as follows: TM>P>TN>LAI>T>VPD>dswrf>TH>RH>Wind>uswrf>ulwrf; The three-factor combination of annual average rainfall, annual average temperature and annual average downward shortwave radiation flux was determined as the key influencing factor data of evapotranspiration.

4. The method according to claim 1, characterized in that The artificial neural network model PI-ANN integrating physical information is composed of an input layer, two fully connected layers and an output layer. The determined three key influencing factors of evapotranspiration and the corresponding PT correction coefficient α as a label are input into the model through the input layer. The first fully connected layer fc1 includes 50 neurons, which convert the three key influencing factors of evapotranspiration into fifty intermediate influencing factors, and perform feature conversion and dimension mapping, including: by changing the weight of each connection line between each layer of the network, inputting the weighted sum of all outputs from the previous layer, and adding a fixed bias term to improve the flexibility of the neural network. This conversion enables the network to extract deeper information from the original input data and pass this information to the next layer of the network, and then convert it through the nonlinear activation function ReLU to generate the output of the first fully connected layer. The ReLU is as follows: ReLU(x)=max(0,x) The ReLU allows the model to maintain good gradient flow during training because of its simple calculation and effective solution to the gradient vanishing problem; The second fully connected layer fc2 includes twenty neurons, which further process the information transmitted by the first fully connected layer. In this way, the feature space of the fully connected layer is expanded, and the model's ability to capture the intrinsic relationship of the data is increased; The output layer uses the Sigmoid activation function to limit the output range to the reasonable range of the global PT correction coefficient between 0 and 1.5, which is consistent with the actual physical meaning and makes the prediction results more reasonable and usable.

5. An evapotranspiration calculation system based on multi-source remote sensing data, characterized in that: The system comprises: The first calculation module is used to calculate the PT correction coefficient α of the multiple stations according to the measured evapotranspiration of the multiple stations using the Preistly-Taylor formula of evapotranspiration; A second calculation module is used to calculate a plurality of evapotranspiration influencing factor data according to the multi-source remote sensing data of the plurality of stations, and determine three evapotranspiration key influencing factor data from the plurality of evapotranspiration influencing factor data; An input training module is used to input the determined three key influencing factors of evapotranspiration and the corresponding PT correction coefficient α as a label into and train the artificial neural network model PI-ANN integrating physical information; A prediction module is used to use the trained PI-ANN to predict the PT correction factor α for the area where evapotranspiration is to be calculated; The third calculation module is used to calculate the evapotranspiration of the area to be calculated evapotranspiration using the Preistly-Taylor formula according to the predicted PT correction coefficient α; The loss function of the artificial neural network model PI-ANN that integrates physical information is a composite loss function PI-ANN_Loss of machine learning loss and physical information loss. The PI-ANN_Loss aims to optimize the prediction performance of the model while ensuring that its output follows basic physical laws, including mean square error loss MSE_Loss and physical information loss Physics_Loss. Physics_Loss combines physical knowledge so that the composite loss function not only optimizes the statistical performance of the model, but also ensures that the model output is physically interpretable and reasonable. By directly encoding physical laws into the loss function, the decision-making process of the model is more transparent and the interpretability of the output is also enhanced. This part of the loss works by comparing the model output with the expected value calculated based on the physical formula. The formula is as follows: PI_feature t =1.5*VPD t -0.5 PI-ANN_Loss=MSE_Loss+Physics_Loss Among them, PI_feature t Refers to the physical constraint influencing factor, which guides the model to learn predictions that conform to physical laws. It is about the saturated vapor pressure difference VPD t Function formula, because the PT correction coefficient α in the Priestly-Taylor formula mainly represents the regulation effect on air unsaturation, and VPD t Related, so use VPD t Improve the loss function and add physical meaning to the constraints. t represents the predicted value of the PT correction factor α, O t Represents the true label value of the PT correction coefficient α.

6. The system according to claim 5, characterized in that The Preistly-Taylor formula for evapotranspiration is as follows: Where ET is evapotranspiration, Δ is the slope of the saturated water vapor pressure-temperature curve, γ is a constant, and R n is the net radiation, G is the soil heat, α is the PT correction factor, and λ is the unit conversion factor; Except for α, all other parameters in the formula are known quantities, so α can be calculated according to the formula.

7. The system according to claim 5, characterized in that The second calculation module is specifically used for: According to the multi-source remote sensing data of the plurality of stations, several evapotranspiration influencing factor data are calculated, including but not limited to the following twelve factors: annual average temperature T, annual maximum temperature TM, annual minimum temperature TN, annual average rainfall P, annual average saturated vapor pressure difference VPD, annual average upward longwave radiation flux ulwrf, annual average upward shortwave radiation flux uswrf, annual average downward shortwave radiation flux dswrf, annual average wind speed Wind, annual average relative humidity RH, annual average leaf area index LAI, annual average forest canopy height data TH; The random forest method was used to calculate the feature importance ranking of the factors affecting evapotranspiration, and the results were as follows: TM>P>TN>LAI>T>VPD>dswrf>TH>RH>Wind>uswrf>ulwrf; The three-factor combination of annual average rainfall, annual average temperature and annual average downward shortwave radiation flux was determined as the key influencing factor data of evapotranspiration.

8. The system according to claim 5, characterized in that The artificial neural network model PI-ANN integrating physical information is composed of an input layer, two fully connected layers and an output layer. The determined three key influencing factors of evapotranspiration and the corresponding PT correction coefficient α as a label are input into the model through the input layer. The first fully connected layer fc1 includes 50 neurons, which convert the three key influencing factors of evapotranspiration into fifty intermediate influencing factors, and perform feature conversion and dimension mapping, including: by changing the weight of each connection line between each layer of the network, inputting the weighted sum of all outputs from the previous layer, and adding a fixed bias term to improve the flexibility of the neural network. This conversion enables the network to extract deeper information from the original input data and pass this information to the next layer of the network, and then convert it through the nonlinear activation function ReLU to generate the output of the first fully connected layer. The ReLU is as follows: ReLU(x)=max(0,x) The ReLU allows the model to maintain good gradient flow during training because of its simple calculation and effective solution to the gradient vanishing problem; The second fully connected layer fc2 includes twenty neurons, which further process the information transmitted by the first fully connected layer. In this way, the feature space of the fully connected layer is expanded, and the model's ability to capture the intrinsic relationship of the data is increased; The output layer uses the Sigmoid activation function to limit the output range to the reasonable range of the global PT correction coefficient between 0 and 1.5, which is consistent with the actual physical meaning and makes the prediction results more reasonable and usable.