TWSC prediction method and system based on water balance and machine learning error optimization

Abstract

The invention discloses a land water storage variation prediction method and system based on water balance and machine learning error optimization, and the method comprises the following steps: S1, obtaining a remote sensing data set in a historical period, and carrying out the optimization processing of the remote sensing data set; s2, according to the optimized remote sensing data set, calculating an estimation error by using a multiple matching method, and estimating land water storage variation in a historical period in combination with a water balance formula; s3, introducing an error regulation factor based on a water balance residual error; s4, constructing a machine learning model according to a relation among an error regulation factor in a historical period, the optimized remote sensing data set and an estimation error, and training the machine learning model; and S5, selecting a prediction time period, calculating an optimized remote sensing data set and an estimation error in the prediction time period, and inputting the remote sensing data set and the estimation error into the machine learning model to obtain an error regulation factor of the prediction time period so as to predict the land water storage variation. According to the invention, the precision of land water storage variation estimation can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of water resources management and prediction, and specifically proposes a method and system for predicting terrestrial water storage changes based on water balance and machine learning error optimization. Background Art

[0002] Terrestrial water storage refers to all water stored on land, including soil water, snow water, groundwater, rivers, lakes and canopy water storage. It is an important state variable of the global water cycle and freshwater resources. Obtaining high-precision terrestrial water storage change (TWSC) data is crucial to accurately understand the changes in the Earth's climate, ecosystems and water cycle. Accurately simulating changes in terrestrial water storage has always been a key challenge in the study of global water cycle changes. One of the important reasons is that the value of TWSC is small and the simulation accuracy is easily affected by the error of the driving variable.

[0003] The GRACE satellite provides an important opportunity to observe TWSC, but is limited by the length of the observation period (from 2002 to the present). In order to obtain historical and future TWSC data under the influence of climate change and human activities, various methods are needed to simulate TWSC. Widely used methods include hydrological models, machine learning models, and water balance methods. Hydrological models obtain the total terrestrial water storage changes by simulating water storage components, but their accuracy is often affected by factors such as the accuracy of climate-driven data, model structure, parameterization, and human activities. Machine learning models simulate TWSC by establishing the relationship between TWSC and different driving factors, but are susceptible to the uncertainty and overfitting of driving factors, and lack consideration of water cycle mechanisms. Compared with hydrological models and machine learning models, TWSC simulations calculated using water balance have the advantages of using fixed variables (precipitation P, evapotranspiration ET, runoff Q) as driving factors, improving the consistency between P, ET, and Q datasets, and can directly obtain TWSC and consider water cycle mechanisms (mass conservation). However, TWSC simulations calculated using water balance are greatly affected by the errors of P, ET, and Q datasets. Therefore, reducing the errors transferred from P, ET, and Q datasets to estimate TWSC is the key to improving the accuracy of TWSC estimation. Summary of the invention

[0004] The main purpose of the present invention is to provide a method and system for predicting terrestrial water storage changes based on water balance and machine learning error optimization, which can optimize the water balance method and improve the TWSC estimation accuracy.

[0005] The technical solution adopted by the present invention is:

[0006] A method for predicting terrestrial water storage changes based on water balance and machine learning error optimization is provided, comprising the following steps:

[0007] S1. Obtain remote sensing data sets of precipitation, evapotranspiration, and runoff in historical periods, fuse observation data, and optimize remote sensing data sets;

[0008] S2. Based on the optimized remote sensing data set, the estimation errors of precipitation, evapotranspiration and runoff are calculated using the multiple matching method, and the terrestrial water storage in the historical period is estimated in combination with the water balance formula;

[0009] S3, introduce an error control factor, which is the difference between the actual terrestrial water storage and the estimated terrestrial water storage minus the estimation error and the water balance residual;

[0010] S4. Construct and train a machine learning model based on the relationship between the error control factor in the historical period, the optimized remote sensing data set, and the estimated error;

[0011] S5. Select a prediction period, calculate the optimized remote sensing data set and estimated error in the prediction period, and input them into the trained machine learning model to obtain the predicted error control factor for the prediction period, and calculate the terrestrial water storage for the prediction period based on the predicted error control factor.

[0012] Following the above technical solution, in step S1, a multivariate linear regression method is specifically used to establish the relationship between the observation data of precipitation, evapotranspiration, and runoff and the remote sensing data set, and the relationship is transferred to the grid without observation sites to obtain an optimized remote sensing data set.

[0013] Following the above technical solution, the multiple matching method is a triple matching method.

[0014] Following the above technical solution, the specific steps for calculating the estimated error in step S2 are: calculating the distance between the optimized remote sensing data set and the true value according to the multiple matching method, using the distance as a weight to estimate the true value, and then making a difference between the optimized remote sensing data set and the estimated true value to obtain the estimated error of precipitation, evapotranspiration, and runoff.

[0015] Following the above technical solution, the machine learning model in step S4 is a random forest model.

[0016] The present invention also provides a land water storage change prediction system based on water balance and machine learning error optimization, comprising:

[0017] The remote sensing data set processing module is used to obtain the remote sensing data sets of precipitation, evapotranspiration, and runoff in the historical period, fuse the observation data, and optimize the remote sensing data sets;

[0018] The estimation module is used to calculate the estimation errors of precipitation, evapotranspiration and runoff based on the optimized remote sensing data set using a multiple collocation method, and estimate the terrestrial water storage in the historical period in combination with the water balance formula;

[0019] An error control factor calculation module is used to calculate the error control factor, which is the difference between the actual terrestrial water storage change and the estimated terrestrial water storage change minus the estimation error and the water balance residual;

[0020] The error control factor prediction module is used to build and train a machine learning model based on the relationship between the error control factor of the historical period and the optimized remote sensing data set and the estimated error; then, according to the selected prediction period, the optimized remote sensing data set and the estimated error in the prediction period are calculated and input into the trained machine learning model to obtain the predicted error control factor of the prediction period;

[0021] The terrestrial water storage prediction module is used to calculate the terrestrial water storage in the prediction period according to the predicted error control factor.

[0022] Following the above technical solution, the data set processing module specifically adopts the multivariate linear regression method to establish the relationship between the observation data of precipitation, evapotranspiration, and runoff and the remote sensing data set, and migrates the relationship to the grid without observation sites to obtain the optimized remote sensing data set.

[0023] Following the above technical solution, the specific process of calculating the estimation error in the estimation module is as follows: the distance between the optimized remote sensing data set and the true value is calculated according to the multiple matching method, the true value is estimated with the distance as the weight, and then the estimated error of precipitation, evapotranspiration and runoff is obtained by subtracting the optimized remote sensing data set from the estimated true value.

[0024] Following the above technical solution, the machine learning model is a random forest model.

[0025] The present invention also provides a computer storage medium, which stores a computer program that can be executed by a processor, and the computer program executes the method for predicting land water storage changes based on water balance and machine learning error optimization described in the above technical solution.

[0026] The beneficial effects of the present invention are as follows: the present invention estimates the errors of various hydrological variables using a multiple matching method, introduces an error control factor in combination with a water balance formula, and constructs a terrestrial water storage change prediction model based on water balance and error optimization, thereby greatly reducing the error in transferring remote sensing data sets of precipitation, evapotranspiration, and runoff to estimate terrestrial water storage, and effectively improving the accuracy of terrestrial water storage estimation. The present invention provides an important reference for solving the technical problem of low accuracy of surface water storage simulation caused by problems in the prior art such as climate-driven data, model structure, and parameterization.

[0027] Furthermore, by adopting the method of multiple linear regression, the relationship between the observation data and remote sensing datasets of precipitation, evapotranspiration, and runoff is established respectively, and the relationship is transferred to the grid without observation sites to obtain the optimized remote sensing dataset, thereby improving the accuracy of the remote sensing dataset in the historical period.

[0028] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0030] Figure 1 This is a flow chart of a method for predicting terrestrial water storage changes based on water balance and machine learning error optimization according to an embodiment of the present invention;

[0031] Figure 2 is a flow chart of a method for predicting land water storage change based on water balance and machine learning error optimization according to another embodiment of the present invention;

[0032] Figure 3 It is a schematic diagram of the structure of a system for predicting land water storage changes based on water balance and machine learning error optimization according to an embodiment of the present invention. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention 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 invention and are not used to limit the present invention.

[0034] It should be noted that the illustrations provided in the embodiments of the present invention are only used to illustrate the basic concept of the present invention in a schematic manner. Therefore, the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout type may also be more complicated.

[0035] In the present invention, it is also necessary to explain that, if the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. appear, the orientation or position relationship indicated is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. In addition, if the terms "first" and "second" appear, they are only used for description and distinction purposes, and cannot be understood as indicating or implying relative importance.

[0036] like Figure 1 As shown, the method for predicting the change in terrestrial water storage based on water balance and machine learning error optimization in an embodiment of the present invention includes the following steps:

[0037] S1. Obtain remote sensing data sets of precipitation, evapotranspiration, and runoff in historical periods, fuse observation data, and optimize remote sensing data sets;

[0038] S2. Based on the optimized remote sensing data set, the estimation errors of precipitation, evapotranspiration and runoff are calculated using the multiple matching method, and the terrestrial water storage in the historical period is estimated in combination with the water balance formula;

[0039] S3, introducing an error control factor, which is the difference between the actual land water storage change prediction and the estimated land water storage change prediction minus the estimation error and the water balance residual;

[0040] S4. Construct and train a machine learning model based on the relationship between the error control factor in the historical period, the optimized remote sensing data set, and the estimated error;

[0041] S5. Select a prediction period, calculate the optimized remote sensing data set and estimated error in the prediction period, and input them into the trained machine learning model to obtain the predicted error control factor for the prediction period, and calculate the predicted land water storage change in the prediction period based on the predicted error control factor.

[0042] Specifically, in step S1, a multivariate linear regression method is specifically used to establish the relationship between the observation data of precipitation, evapotranspiration, and runoff and the remote sensing dataset, and the relationship is migrated to a grid without an observation site to obtain an optimized remote sensing dataset, thereby improving the accuracy of the remote sensing dataset in the historical period.

[0043] Furthermore, in the embodiment of the present invention, the multiple matching method preferably uses a triple matching method, namely the TC method. Take three precipitation remote sensing data sets as an example:

[0044] The first step is to make a difference between different data sets and find the distance d between them. 12 ,d13 and d 23 ;

[0045] d 12 =P1-P2

[0046] d 13 =P1-P3

[0047] d 23 =P2-P3;

[0048] Step 2: Use the Pythagorean equation to calculate the value of d 12 ,d 13 and d 23 Solve to get d 1t ,d 2t and d 3t , that is, the distance between the remote sensing data set and the true value, where the Pythagorean equation is:

[0049]

[0050] Step 3: According to d 1t ,d 2t and d 3t Calculate the weights of the three remote sensing data sets relative to the true value. The calculation formula is as follows:

[0051]

[0052] Step 4: Estimate the true value based on W1, W2 and W3. The calculation formula is as follows:

[0053] P0=P1*W1+P2*W2+P3*W3.

[0054] Specifically, the specific steps for calculating the estimated error in step S2 are: calculating the distance between the optimized remote sensing data set and the true value according to the multiple matching method, estimating the true value with the distance as the weight, and then subtracting the optimized remote sensing data set from the estimated true value to obtain the estimated error of precipitation, evapotranspiration, and runoff.

[0055] In step S4, the random forest model can be selected as the machine learning model. Other machine learning models can also be used, such as GBDT, LightGBM and RF. It is found that the changes in terrestrial water storage predicted by these three machine learning methods are not much different, with correlation coefficients of 0.78, 0.77 and 0.78 respectively. Since RF has higher computational efficiency and lower hyperparameter sensitivity, the random forest model is recommended.

[0056] It can be seen that the above embodiment uses a multiple matching method to estimate the errors of various hydrological variables, and introduces an error control factor in combination with the water balance formula to construct a land water storage change prediction model based on water balance and error optimization. It greatly reduces the error in transferring from remote sensing data sets of precipitation, evapotranspiration, and runoff to estimating land water storage change prediction, and effectively improves the accuracy of land water storage change prediction estimation. The present invention provides an important reference for solving the technical problem of low accuracy of land surface water storage simulation due to problems in the prior art such as climate-driven data, model structure, parameterization, etc.

[0057] like Figure 2 FIG. 1 is another embodiment of the present invention, which is based on the above embodiment and further implemented on the basis of the above embodiment. The method of this embodiment mainly includes the following steps:

[0058] Step S1, obtaining precipitation, evapotranspiration, and runoff data sets for the historical period, and fusing the observation data to improve the accuracy of the historical period data sets;

[0059] Step S2, using the multiple collocation method (TC method) to calculate precipitation, evapotranspiration, and runoff errors, and combining the water balance formula to estimate the TWSC in the historical period;

[0060] Step S3, introducing the error control factor ε to reduce the uncertainty of TWSC estimation;

[0061] Step S4: Use the machine learning model to establish the quantitative relationship between the error control factor ε in the historical period and the driving variables of TWSC

[0062] Step S5: transfer the quantitative relationship to the forecast period of the control factor ε, and use the improved water balance equation to calculate the change of water storage in the basin during the forecast period.

[0063] Step S1 specifically includes:

[0064] The multivariate linear regression method was used to establish the relationship between the observed data and remote sensing data sets of precipitation (P), evapotranspiration (ET), and runoff (Q), and the relationship was transferred to the grid without observation stations to improve the accuracy of the remote sensing data set in the historical period, and the precipitation (P) of the entire basin fused with the observed data was obtained. Raw ), evaporation (T Raw ), runoff (Q Raw ) dataset.

[0065] The expression of the multivariate linear regression method for fusion observation data is:

[0066] y=β0+β1x1+β2x2+β3x3+β4x4

[0067] Where x1, x2, x3 and x4 represent the remote sensing data value, longitude and latitude, and elevation of each grid, respectively; y represents the observed data; β0 represents the intercept term; β1, β2, β3, β4 represent the slope of the linear relationship between y and x.

[0068] As a preferred embodiment of the present invention, the specific process of calculating precipitation, evapotranspiration, and runoff errors using the multiple collocation method (TC) in step S2 includes:

[0069] Calculation of precipitation, evapotranspiration and runoff error r′ using the multiple collocation method (TC method) P , r′ ET , r′ Q First, we estimate P using the multiple collocation method. Raw , E.T. Raw , Q Raw The distance d relative to the true value P ,d ET ,d Q ; Then use the estimated distance d P ,d ET ,d Q As weights, we estimate the true values ​​of P, ET, and Q, expressed as PT′, ETT′, and QT′; finally, we use P Raw , E.T. Raw , Q Raw The difference between the estimated true values ​​PT′, ETT′, QT′ is taken as the error r′ of P, ET, Q P , r′ ET , r′ Q , that is, r′ P =P Raw -PT′, r′ ET =ET Raw -ETT′ and r′ Q =Q Raw -QT′.

[0070] The water balance residual △Res is further calculated in combination with the water balance formula. The expression of △Res is:

[0071] △Res=P Raw -ET Raw -Q Raw -TWSC GRACE

[0072] Where P Raw , E.T. Raw , Q Raw It represents the optimized remote sensing dataset after fusion of observation data, and TWSC is the change in terrestrial water storage obtained through GRACE (GRACE is a gravity satellite that monitors global water storage changes and can invert the remote sensing dataset of terrestrial water storage changes).

[0073] The main innovation of the present invention is to introduce an error control factor ε to reduce the uncertainty of TWSC estimation.

[0074] Since P which has fused the observed data Raw , E.T. Raw , Q Raw The data set has errors, so the theoretical estimate of the true value of TWSC is T The expression is:

[0075] TWSC T =(P Raw +r P )-(ET Raw +r ET )-(Q Raw +r Q )

[0076] In the formula, r P 、r ET 、r Q Indicates P Raw , E.T. Raw , Q Raw Theoretical error.

[0077] However, due to the errors r′ in the precipitation, evapotranspiration and runoff calculated by the TC method P , r′ ET , r′ Q are not equal to their true errors. Therefore, the error control factor ε is introduced to make the estimated TWSC closer to the true value. In fact, the estimated true value of TWSC is TWSC T The expression is:

[0078] TWSC T =P Raw -ET Raw -Q Raw +r′ P -r′ ET -r′ Q +ε

[0079] In the formula, r′ P , r′ ET , r′ Q represents P obtained by TC method Raw , E.T. Raw , Q Raw The true error.

[0080] On this basis, in order to meet the water closure condition, the water balance residual △Res formula is combined with the actual estimated TWSC true value TWSC TThe expression of error control factor ε in the historical period can be obtained:

[0081] ε=r′ ET +r′ Q -r′ P -△Res

[0082] In the formula, r′ P , r′ ET , r′ Q represents P obtained by TC method Raw , E.T. Raw , Q Raw , △Res represents the water balance residual.

[0083] As a preferred solution of the present invention, in step S5, a quantitative relationship between the error control factor ε in the historical period and the driving variable of TWSC is established by using a machine learning model, which specifically includes:

[0084] The machine learning method is used to establish the quantitative relationship between the error control factor ε in the historical period and the driving variables of TWSC. The random forest model is used to establish the relationship between the historical period ε and its influencing factors (i.e., driving variables). The influencing factors include r′ P , r′ ET , r′ Q , P Raw , E.T. Raw and Q Raw , the expression is:

[0085] F(ε,X)=F(ε,(r′ P ,r′ ET ,r′ Q ,P Raw ,ET Raw ,Q Raw ))

[0086] Where P Raw , E.T. Raw , Q Raw The precipitation, evapotranspiration and runoff remote sensing datasets optimized for fusion observation data, r′ P , r′ ET , r′ Q are P obtained by TC method respectively. Raw , E.T. Raw , Q Raw of error.

[0087] As a preferred solution of the present invention, in step S5, the quantitative relationship between the error control factor ε in the historical period and the driving variable of TWSC is transferred to other prediction periods, so as to predict the change of terrestrial water storage, specifically:

[0088] First, obtain various P, ET, and Q data sets for the forecast period, and use the TC method to obtain their error r′ P , r′ ET , r′ Q ;

[0089] Then, the error control factor ε of the forecast period is obtained by using the quantitative relationship between the error control factor ε established in the historical period and the driving variables of TWSC (i.e., using the constructed machine learning model);

[0090] Finally, the change in land water storage during the prediction period is calculated based on the error control factor ε predicted by the model, and its expression is:

[0091] TWSC=P-ET-Qr′ P +r′ ET +r′ Q +ε

[0092] Where TWSC represents the change in land water storage during the forecast period, P, ET, and Q are the precipitation, evapotranspiration, and runoff data sets during the forecast period, and r′ P , r′ ET , r′ Q represents P obtained by TC method Raw , E.T. Raw , Q Raw , and ε is the error control factor of the forecast period.

[0093] like Figure 3 As shown, the present invention also provides a land water storage change prediction system based on water balance and machine learning error optimization, which is mainly used to implement the above method embodiment, and the system includes:

[0094] The remote sensing data set processing module is used to obtain the remote sensing data sets of precipitation, evapotranspiration, and runoff in the historical period, fuse the observation data, and optimize the remote sensing data sets;

[0095] The estimation module is used to calculate the estimation errors of precipitation, evapotranspiration and runoff based on the optimized remote sensing data set using a multiple matching method, and estimate the changes in terrestrial water storage in the historical period in combination with the water balance formula;

[0096] An error control factor calculation module is used to calculate the error control factor, which is the difference between the actual terrestrial water storage change and the estimated terrestrial water storage change minus the estimation error and the water balance residual;

[0097] The error control factor prediction module is used to build and train a machine learning model based on the relationship between the error control factor of the historical period and the optimized remote sensing data set and the estimated error; then, according to the selected prediction period, the optimized remote sensing data set and the estimated error in the prediction period are calculated and input into the trained machine learning model to obtain the predicted error control factor of the prediction period;

[0098] The land water storage change prediction module is used to calculate the land water storage change in the prediction period according to the predicted error control factor.

[0099] Furthermore, the dataset processing module specifically uses the multivariate linear regression method to establish the relationship between the observation data of precipitation, evapotranspiration, and runoff and the remote sensing dataset, and migrates the relationship to the grid without observation sites to obtain the optimized remote sensing dataset.

[0100] Furthermore, the specific process of calculating the estimation error in the estimation module is as follows: the distance between the optimized remote sensing data set and the true value is calculated according to the multiple matching method, the true value is estimated with the distance as the weight, and then the estimated error of precipitation, evapotranspiration and runoff is obtained by subtracting the optimized remote sensing data set from the estimated true value.

[0101] Each module is mainly used to implement each step of the above method embodiment, which will not be described one by one here.

[0102] In summary, the present invention combines water balance and machine learning models to introduce error control factors, which greatly reduces the error transferred from P, ET and Q data sets to estimate TWSC when the water balance formula predicts TWSC, and effectively improves the accuracy of TWSC estimation.

[0103] The present application also provides a computer-readable storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., an SD or DX memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a disk, an optical disk, a server, an App application store, etc., on which a computer program is stored, and when the program is executed by a processor, a corresponding function is realized. When the computer-readable storage medium of this embodiment is executed by a processor, the TWSC prediction method based on water balance and machine learning error optimization of the method embodiment is realized.

[0104] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0105] The order of execution of each step in the above embodiment does not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0106] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the appended claims of the present invention.

Claims

1. A method for predicting terrestrial water storage changes based on water balance and machine learning error optimization, characterized in that: The following steps are involved: S1. Obtain remote sensing data sets of precipitation, evapotranspiration, and runoff in historical periods, fuse observation data, and optimize remote sensing data sets; S2. Based on the optimized remote sensing data set, the estimation errors of precipitation, evapotranspiration and runoff are calculated using the multiple matching method, and the changes in land water storage in the historical period are estimated in combination with the water balance formula; S3, introduce an error control factor, which is the difference between the actual terrestrial water storage change and the estimated terrestrial water storage change minus the estimation error and the water balance residual; S4. Construct and train a machine learning model based on the relationship between the error control factor in the historical period, the optimized remote sensing data set, and the estimated error; S5. Select a prediction period, calculate the optimized remote sensing data set and estimated error in the prediction period, and input them into the trained machine learning model to obtain the predicted error control factor for the prediction period, and calculate the change in land water storage in the prediction period based on the predicted error control factor.

2. The method for predicting land water storage changes based on water balance and machine learning error optimization according to claim 1 is characterized in that: In step S1, a multivariate linear regression method is specifically used to establish the relationship between the observation data of precipitation, evapotranspiration, and runoff and the remote sensing data set, and the relationship is transferred to the grid without observation sites to obtain an optimized remote sensing data set.

3. The method for predicting land water storage changes based on water balance and machine learning error optimization according to claim 1 is characterized in that: The multiple collocation method is a triple collocation method.

4. The method for predicting land water storage changes based on water balance and machine learning error optimization according to claim 1 is characterized in that: The specific steps for calculating the estimated error in step S2 are: calculating the distance between the optimized remote sensing data set and the true value according to the multiple matching method, estimating the true value with the distance as the weight, and then subtracting the optimized remote sensing data set from the estimated true value to obtain the estimated error of precipitation, evapotranspiration, and runoff.

5. The method for predicting terrestrial water storage changes based on water balance and machine learning error optimization according to claim 1, characterized in that: The machine learning model in step S4 is a random forest model.

6. A land water storage change prediction system based on water balance and machine learning error optimization, characterized in that: include: The remote sensing data set processing module is used to obtain the remote sensing data sets of precipitation, evapotranspiration, and runoff in the historical period, fuse the observation data, and optimize the remote sensing data sets; The estimation module is used to calculate the estimation errors of precipitation, evapotranspiration and runoff based on the optimized remote sensing data set using a multiple matching method, and estimate the changes in terrestrial water storage in the historical period in combination with the water balance formula; An error control factor calculation module is used to calculate the error control factor, which is the difference between the actual terrestrial water storage change and the estimated terrestrial water storage change minus the estimation error and the water balance residual; The error control factor prediction module is used to build and train a machine learning model based on the relationship between the error control factors in the historical period and the optimized remote sensing data set and the estimated error; Then, according to the selected prediction period, the optimized remote sensing data set and estimated error in the prediction period are calculated and input into the trained machine learning model to obtain the predicted error control factor for the prediction period; The land water storage change prediction module is used to calculate the land water storage change in the prediction period according to the predicted error control factor.

7. The system for predicting land water storage changes based on water balance and machine learning error optimization according to claim 6 is characterized in that: The dataset processing module specifically uses the multivariate linear regression method to establish the relationship between the observation data of precipitation, evapotranspiration, and runoff and the remote sensing dataset, and migrates the relationship to the grid without observation sites to obtain the optimized remote sensing dataset.

8. The system for predicting land water storage changes based on water balance and machine learning error optimization according to claim 6, characterized in that: The specific process of calculating the estimation error in the estimation module is as follows: the distance between the optimized remote sensing dataset and the true value is calculated according to the multiple matching method, the true value is estimated with the distance as the weight, and then the estimated error of precipitation, evapotranspiration and runoff is obtained by subtracting the optimized remote sensing dataset from the estimated true value.

9. The system for predicting land water storage changes based on water balance and machine learning error optimization according to claim 6, characterized in that: The machine learning model is a random forest model.

10. A computer storage medium, characterized in that: A computer program executable by a processor is stored therein, and the computer program executes the method for predicting changes in terrestrial water reserves based on water balance and machine learning error optimization as described in any one of claims 1 to 5.

Images