A large space scale water resource quantity estimation method
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA INST OF WATER RESOURCES & HYDROPOWER RES
- Filing Date
- 2022-09-20
- Publication Date
- 2026-07-07
AI Technical Summary
Existing technologies are insufficient to effectively predict water resources on a large spatial scale, especially in the arid inland northwest region where there is a lack of hydrological monitoring stations and measured runoff data, making it impossible to accurately account for the impact of future climate change.
By combining climate model outputs with support vector regression (SVR) models, and selecting model indicators such as precipitation, temperature, wind speed, solar radiation, saturated vapor pressure, and potential evapotranspiration, a water resource prediction model is constructed. MATLAB software is used for programming calculations and model training to optimize kernel functions and parameters, thereby enabling the prediction of future water resources.
It achieves efficient prediction of water resources on a large spatial scale. The model has a simple structure, high solution efficiency, reduces dependence on measured data, and has high prediction accuracy and generalization ability.
Smart Images
Figure CN115438870B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method for predicting water resources on a large spatial scale. Background Technology
[0002] Currently, water resource forecasting models are mainly divided into two categories: data-driven statistical methods and hydrological modeling methods based on hydrological and hydraulic physical mechanisms. Statistical methods mainly include mathematical statistics, similarity year methods, and extended runoff methods, while hydrological modeling methods mainly include distributed modeling and lumped hydrological modeling. Among them, data-driven statistical methods have high accuracy but also high data requirements. They mostly rely on historical water resource data for prediction and are difficult to consider the impact of future climate change on water resources. Hydrological modeling methods generally predict surface and groundwater runoff and require validation with measured river runoff data. They are suitable for single or multiple watersheds, but their applicability to large-scale regional water resource forecasting is limited. In the arid inland areas of Northwest China, many rivers lack hydrological monitoring stations, and the measured runoff data from some existing monitoring stations are too short to provide sufficient measured data to support the establishment of hydrological models.
[0003] Currently, conventional hydrological models cannot calculate water resources on a large spatial scale. Water resource prediction on a large spatial scale mainly relies on historical data. In other words, current hydrological models based on measured runoff data face practical difficulties in predicting water resource trends on a large spatial scale. Considering the urgency of formulating strategic measures for water security, this invention proposes a novel water resource prediction method that combines climate model output with SVR. Summary of the Invention
[0004] To overcome the shortcomings of the prior art and solve the technical problem of "how to predict water resource trends on a large spatial scale," this invention provides a method for predicting water resource volume on a large spatial scale, the specific technical solution of which is as follows:
[0005] A method for large-scale spatial water resource estimation includes the following steps:
[0006] Step 1: Select model metrics;
[0007] Step 2: Model construction, mainly including the following steps:
[0008] Step 2.1: Input the historical watershed data of the model indicators and the corresponding historical water resource data, and generate training and test sets based on the model indicators;
[0009] Step 2.2: Create an SVR regression model using kernel functions and determine the model parameters;
[0010] Step 2.3: Input the relevant parameters. The first parameter returned is the predicted value mapped from the multidimensional space, and the second parameter is the mean square error (MSE) and coefficient of determination (R²) between the predicted and actual water resources in the test set. 2 ;
[0011] Step 2.4: Utilize the mean squared error (MSE) and coefficient of determination (R²) returned by the model. 2 The performance of the established SVR regression model was evaluated.
[0012] Step 2.5: Adjust the model parameters or reselect the kernel function type, and repeat steps 2.2 to 2.4 above until the model returns the mean squared error (MSE) and coefficient of determination (R²). 2 The requirements are met;
[0013] Step 3: Input the changes in the model indicators and the future simulation prediction results of the environment, fit the model with the trained SVR model, and output the estimated future water resources.
[0014] Preferably, the model indicators selected in step 1 mainly include precipitation (pre) and maximum temperature (T). max Minimum temperature T min The seven factors are wind speed Win, solar radiation Rs, saturated vapor pressure ea, and potential evapotranspiration Et0.
[0015] Preferably, a radial basis function is used when creating the SVR regression model, and the optimal parameters are determined by cross-validation. The optimal parameters are then used to train the SVR regression model.
[0016] Preferably, the mean squared error returned by the model
[0017] Coefficient of determination
[0018] Preferably, the future simulation prediction results of the model index change environment are obtained in the following way: under different typical concentration paths of RCP2.6, RCP4.5, RCP6.0 and RCP8.5, the simulation results of different experiments of the same mode by five GCMs models are comprehensively compared and the results are arithmetically averaged with equal weights. Then, the same method is applied to integrate the results of multiple models. The GCMs output is reduced to a grid with a resolution of 0.5° and corrected using the trend preservation bias correction method.
[0019] Preferably, the model indicators selected in step 1 mainly include precipitation (pre) and maximum temperature (T). max Minimum temperature T min The seven parameters are: wind speed (Win), sunshine duration, relative humidity, and evaporation rate.
[0020] The beneficial effects of this invention are as follows: By establishing a water resource prediction model that combines climate output patterns with SVR (Self-Range Dynamics) and using climate simulation results under different concentration pathways as input conditions, it is possible to predict future water resources in a large spatial scale study area. This model has a simple structure, high solution efficiency, low requirement for verification with measured data, and can be applied to water resource prediction on a large spatial scale. Attached Figure Description
[0021] Figure 1 This is a flowchart of the model construction and water resource estimation process of this invention;
[0022] Figure 2 This is a comparison chart of the prediction results of the training set and the test set of this invention;
[0023] Figure 3 This is a map showing the estimated water resources produced by the Hotan River Basin according to the present invention.
[0024] Figure 4 This is a map showing the estimated water resources produced in the Yarkand River Basin according to the present invention.
[0025] Figure 5 This is a map showing the estimated water resources produced by the Kashgar River Basin according to the present invention.
[0026] Figure 6 This is a map showing the estimated water resources produced in the Aksu River Basin according to the present invention.
[0027] Figure 7 This is a map showing the estimated water resources produced in the Weigan River Basin according to the present invention.
[0028] Figure 8 This is a diagram showing the estimated water resources produced by the river basin through the hole, as presented in this invention.
[0029] Figure 9 This is a map showing the estimated water resources produced by the small rivers of the Krya River in this invention.
[0030] Figure 10 This is a map showing the estimated water resources produced by the Cherchen River basin and other small river basins according to the present invention.
[0031] Figure 11 This is a diagram showing the estimated water resources produced by the Tarim River main stream according to the present invention.
[0032] Figure 12 This is a map showing the estimated total self-produced water resources in the Tarim River Basin according to the present invention. Detailed Implementation
[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0034] Water resource forecasting based on climate model outputs typically couples with small-basin hydrological models for small-scale runoff prediction. However, this approach is challenging, complex, and impractical for large-scale water resource forecasting. This invention proposes a water resource estimation model that couples climate model outputs with an SVR model, based on intelligent algorithms. This model is simple in structure, highly efficient in solution, requires minimal validation with measured data, and can be applied to large-scale water resource forecasting. The model primarily studies the response relationship between climate elements and water resource quantity under changing environments and predicts future water resource quantities in the study area. The model construction mainly includes three parts: first, the identification and analysis of water resource influencing factors under climate change, i.e., the selection of model indicators; second, model construction and validation; and third, the prediction of future water resource quantities in the study area under climate change.
[0035] SVR was developed by Vapnik et al. by introducing an insensitive loss function based on SVM (Support Vector Machine) classification. This algorithm has no limitation on data dimensionality, has good performance and effectiveness, and can be applied to simulation. It has been successfully applied in many fields.
[0036] The basic idea of SVM is to find an optimal classification surface that separates two samples. Support Vector Machine Regression (SVR) aims to find an optimal classification surface in multidimensional space that minimizes the total error of all training samples from that surface. For a detailed solution process, please refer to the detailed explanation of SVR case studies by Shi Feng et al.
[0037] Based on the aforementioned existing technologies, such as Figure 1 As shown, this application provides a method for estimating water resources at a large spatial scale, including the following steps:
[0038] Step 1: Select model metrics;
[0039] Step 2: Model construction, mainly including the following steps:
[0040] Step 2.1: Input the historical watershed data of the model indicators and the corresponding historical water resource data, and generate training and test sets based on the model indicators;
[0041] Step 2.2: Create an SVR regression model using kernel functions and determine the model parameters;
[0042] Step 2.3: Input the relevant parameters. The first parameter returned is the predicted value mapped from the multidimensional space, and the second parameter is the mean square error (MSE) and coefficient of determination (R²) between the predicted and actual water resources in the test set. 2 ;
[0043] Step 2.4: Utilize the mean squared error (MSE) and coefficient of determination (R²) returned by the model. 2 The performance of the established SVR regression model was evaluated.
[0044] Step 2.5: Adjust the model parameters or reselect the kernel function type, and repeat steps 2.2 to 2.4 above until the model returns the mean squared error (MSE) and coefficient of determination (R²). 2 The requirements are met;
[0045] Step 3: Input the changes in the model indicators and the future simulation prediction results of the environment, fit the model with the trained SVR model, and output the estimated future water resources.
[0046] In step 1, when selecting model indicators, it is necessary to consider the specific circumstances of the watershed's topographic and geomorphological factors as well as hydrological and meteorological elements. For example, when selecting the Tarim River Basin as the research object, the following should be considered:
[0047] The water resources of the Tarim River Basin are mainly influenced by geomorphological features and runoff sources: rainfall and snowmelt are the most direct driving forces for water resources in the nine tertiary zones of the Tarim River Basin; meteorological elements such as temperature, humidity, and radiation are the main influencing factors of runoff generation and snowmelt; geological conditions affect runoff generation by influencing soil infiltration capacity; and human activities indirectly affect the formation process of water resources by altering underlying surface conditions. Given that the runoff generation area of the Tarim River Basin is mountainous, the interference of human activities on the underlying surface of the runoff generation area can be ignored, and the geological conditions of the basin have not changed significantly over many years. Therefore, this invention takes the nine tertiary water resource zones of the Tarim River Basin as examples, and through the analysis of influencing factors on water resources in hilly areas, selects indicators and constructs an indicator system.
[0048] Based on the factors influencing water resources in mountainous areas, the following hydrological and meteorological characteristic parameters of the watershed are considered:
[0049] Precipitation (pre-). The amount of precipitation reflects a region's ability to replenish water resources. Historical precipitation data for the nine water resource tertiary zones in the Tarim River Basin were obtained from national meteorological stations. Future precipitation forecasts reference the prediction results of the GCM model under four typical concentration paths (RCP2.6, RCP4.5, RCP6.0, and RCP8.5).
[0050] Temperature T, including the highest temperature T max and lowest temperature Tmin For hydrological phenomena, temperature affects evaporation during runoff generation, while long-term temperature fluctuations influence precipitation and thus water resources. The data acquisition and correction process for temperature is the same as that for precipitation data.
[0051] Wind speed. Wind is a natural phenomenon caused by air movement. On Earth, wind is related to water sources, carrying water vapor and driving the water cycle. In the hydrological cycle, wind speed also affects the evaporation and transpiration of water surfaces and vegetation.
[0052] Solar radiation Rs. Solar radiation is the primary source of Earth's heat. It raises the Earth's surface temperature, which in turn affects the melting of glaciers and snow, and is also a major driving force for wind.
[0053] Saturated vapor pressure ea. Saturated vapor pressure is directly related to temperature and directly affects precipitation and evaporation, indirectly impacting water resources.
[0054] Potential evapotranspiration (Et0). Potential evapotranspiration, also known as reference crop evapotranspiration, is an important component of the water cycle. Together with precipitation, it determines the wet and dry conditions of a region and is a key factor in the water cycle.
[0055] In summary, preferably, the water resource prediction model of this invention, which combines climate model output with SVR model, mainly selects indicators including precipitation (pre) and maximum temperature (T). max Minimum temperature T min The seven factors are wind speed Win, solar radiation Rs, saturated vapor pressure ea, and potential evapotranspiration Et0.
[0056] In step 2, MATLAB software can be used for programming calculations when building the model, specifically as follows:
[0057] (1) Generate training / test sets
[0058] The generated training and test sets should be interchangeable. This can be achieved by randomly sorting the samples and selecting the first half as the training set and the second half as the test set. Specifically, the original data can be sorted from 1 to x. Then, a random sequence is defined, and each run generates a new sequence, denoted as sequence n, by randomly sorting the data from 1 to x.
[0059] (2) Create / train the SVR regression model
[0060] When creating an SVR regression model, the libsvm package needs to be installed. The function svmpredict in the package can be used to create and train the model. During the design process, the impact of normalization, the type of kernel function, and the parameter values on the regression model must be comprehensively considered. The selection of the kernel function is a crucial step in building an SVR model. While many new kernel functions are emerging, the four widely accepted basic kernel functions are the linear kernel function, the polynomial kernel function, the radial basis function (RBF) kernel function, and the sigmoid kernel function. Because the radial basis kernel function (RBF) involves fewer parameters and is computationally simple, it is preferred in this invention. The model parameters are determined by referring to existing literature and using cross-validation to find the optimal parameters, which are then used to train the model.
[0061] (3) Simulation test
[0062] After building the model, input the relevant parameters. The first parameter returned is the predicted value mapped from the multidimensional space, and the second parameter is the mean squared error (MSE) and coefficient of determination R of the test set. 2 The smaller the MSE value, the better the accuracy of the constructed predictive model in simulating numerical values. R 2 The closer the value is to 1, the higher the reference value of the related equation; conversely, the closer the value is to 0, the lower the reference value. The calculation formulas for both are as follows:
[0063]
[0064]
[0065] Where: y i —The true value of the i-th sample;
[0066] —The predicted value of the i-th sample.
[0067] (4) Performance Evaluation
[0068] Using the mean squared error (MSE) and coefficient of determination (R²) returned by the model 2 This allows for the evaluation of the performance of the established SVR regression model. If the mean squared error (MSE) is too large or the coefficient of determination (R²) is too low... 2 If the value is too small, the model parameters can be adjusted or the kernel function type can be reselected. Steps 2.2 to 2.4 can be repeated until the requirements are met.
[0069] This invention establishes a water resource prediction model that combines climate output patterns with SVR (Self-Regulating Dynamics) and uses climate simulation results under different concentration pathways as input conditions for the model, enabling prediction of future water resources in a large spatial scale study area. The model has a simple structure, high solution efficiency, low requirement for verification with measured data, and can be applied to water resource prediction on a large spatial scale.
[0070] This invention also specifically discloses a modeling method for the Tarim River Basin and a method for large-scale spatial estimation of water resources.
[0071] Based on the modeling requirements, hydrological, meteorological, and water resource data from the Tarim River Basin from 1956 to 2018 were selected to model nine watersheds. To eliminate randomness, 53 samples were randomly selected from each watershed as the training set, and the remaining 10 samples were used as the test set to verify the model performance. To avoid the impact of overfitting on model accuracy, simulation results with high fitting on the training set but low fitting on the test set were discarded. The hydrological and meteorological data used in this invention comes from the China Meteorological Data Network (http: / / data.cma.cn / ), mainly consisting of daily hydrological and meteorological data from national meteorological base stations, covering the period from 1990 to 2018. The main meteorological factors include: 1. Evaporation (mm); 2. Average wind speed (m / s); 3. Precipitation (mm); 4. Temperature (°C); 5. Sunshine duration (h); 6. Relative humidity (%).
[0072] Among them, the potential evaporation Et0 is calculated based on meteorological factors using the Penman formula, and the maximum temperature T is derived from the air temperature. max and lowest temperature T min The model uses sunshine duration instead of solar radiation (Rs) and relative humidity instead of saturated vapor pressure (ea). Therefore, the selected model indicators mainly include precipitation (pre) and maximum temperature (T). max Minimum temperature T min Seven factors are considered: wind speed (Win), sunshine duration, relative humidity, and potential evaporation.
[0073] Water resources data for the study area were obtained from the Third Xinjiang Water Resources Survey and Assessment, covering the period from 1956 to 2016. Water resources data for 2017 and 2018 were sourced from the "Xinjiang Uygur Autonomous Region Water Resources Bulletin." For future water resources forecasting, hydrological and meteorological data were selected under different typical concentration pathways (RCP2.6, RCP4.5, RCP6.0, and RCP8.5). The results of different simulation experiments on the same model using five GCMs models (GFDL-ESM2M, HaDGem2, IPSL_CM5A_LR, MIROC-ESM-CHEM, and NorESM1-M) were comprehensively compared and averaged using equal weights. Then, the same method was applied to integrate the multi-model results. The GCMs output was reduced to a grid with a resolution of 0.5° and corrected using a trend-preserving bias correction method.
[0074] Table 1. Mean Squared Error and Coefficient of Determination for Training and Test Sets
[0075]
[0076]
[0077] like Figure 2 As shown in Table 1, model simulation analysis indicates that the predicted values for each watershed are quite close to the actual values. Specifically, Figure 2 Table 1 shows the results of a single operation across nine watersheds. Figure 2 The actual values are derived from the data of the Third Water Resources Survey and Evaluation and the Water Resources Bulletin.
[0078] Depend on Figure 2 As shown in Table 1, the fitting accuracy of the training and test sets for the nine watersheds is relatively high. Except for the training sets of the Cherchen River basins, the Kashgar River, and the Weigan River basin, and the test set of the Yarkand River basin, as well as the training and test sets of the Tarim River main stream, where the coefficients of determination are between 0.6 and 0.8, indicating strong goodness of fit, the coefficients of determination for the other watersheds are between 0.8 and 1.0, indicating very strong goodness of fit, and the mean squared error is no greater than 0.10. The simulation results for the nine watersheds in the study area demonstrate that the established SVR regression model has excellent generalization ability, good model performance, and high simulation accuracy, making it applicable to watershed or regional determination of future water resources.
[0079] By training and testing the model, and provided that the model's reliability and accuracy meet the requirements, future hydrological and meteorological indicators under different discharge paths in each river basin can be substituted into the model to predict the future water resources of each river basin.
[0080] In climate change research, scenarios are defined as possible pathways of factors influencing future anthropogenic climate change. They can be understood as the potential consequences of varying degrees of proactive policies adopted by humanity in response to greenhouse gas emissions and global climate change. Both the IPCC report and the Fifth Coupled Interaction and Reconciliation Programme of Climate Models (CMIP5) use the Representative Concentration Pathway (RCP) as a possible scenario. The RCP is a predictive scenario for future greenhouse gas emissions. The RCP2.6 scenario indicates that with proactive human efforts to reduce greenhouse gas emissions and proactive measures to address climate change, radiative forcing will peak before 2100 (≈2.6 W / m³). 2 And it has already begun to decline. RCP 8.5 represents a relatively passive response to global climate change, with continued large-scale emissions of greenhouse gases, and radiative forcing reaching 8.5 W / m² by 2100. 2 The levels are around 4.5 and 6.0. RCP4.5 and RCP6.0 are two scenarios between RCP2.6 and RCP8.5, representing radiative forcing stabilizing at 4.5 W / m² by 2100. 2 and 6.0W / m 2 The estimated water resources in the Tarim River Basin for 2030 and 2050 under four emission pathways (RCP2.6, RCP4.5, RCP6.0, and RCP8.5) are as follows: Figures 3 to 12 As shown.
[0081] Depend on Figures 3 to 12 It can be seen that the overall self-produced water resources in the Tarim River Basin will not fluctuate significantly in the future, remaining at 36-38 billion cubic meters per second. 3 The overall water resources volume was 40.75 billion cubic meters lower than the average water resources volume from 2001 to 2016. 3 The average water resources volume from 1956 to 2016 was 36.89 billion cubic meters. 3 The levels remain unchanged. Contrary to previous understanding, the water resources in the Tarim River Basin did not show a significant increasing or decreasing trend under different discharge pathways. The estimated water resources produced by each basin are as follows:
[0082] The self-produced water resources in the Hotan River Basin are expected to remain at 5.6-5.8 billion m³ under four discharge pathways in 2030 and 2050. 3 The overall water resources volume was between 6.34 billion cubic meters, lower than the average water resources volume from 2001 to 2016. 3 The average water resources volume from 1956 to 2016 was basically the same as 5.76 billion cubic meters. 3 The result was the same.
[0083] The self-produced water resources in the Yarkand River Basin are expected to remain at 7.6-8 billion m³ under four discharge pathways in 2030 and 2050. 3 The overall water resources volume was lower than the average of 8.35 billion cubic meters from 2001 to 2016. 3The average water resources volume from 1956 to 2016 was 7.67 billion cubic meters. 3 The result was the same.
[0084] The self-produced water resources of the Kashgar River Basin are expected to remain between 4.9 billion and 5.2 billion cubic meters under four discharge pathways in 2030 and 2050. 3 The overall water resources volume was lower than the average of 5.73 billion cubic meters from 2001 to 2016. 3 The average water resources volume from 1956 to 2016 was 5.18 billion cubic meters. 3 The result was the same.
[0085] The self-produced water resources in the Aksu River Basin are expected to remain between 4.1 and 4.5 billion m³ under four discharge pathways in 2030 and 2050. 3 Between these scenarios, except for the RCP4.5 and RCP6.0 scenarios, the estimated self-produced water resources in 2030 are 4.38 billion m³. 3 and 4.11 billion m 3 This is lower than the average water resources volume of 4.39 billion cubic meters from 2001 to 2016. 3 The rest are all within the average range of 2001–2016 and 1956–2016 (4.58 billion m³). 3 )between.
[0086] The self-produced water resources in the Weigan River Basin are expected to remain between 3.9 and 4.3 billion m³ under four discharge pathways in 2030 and 2050. 3 In between, except for the estimated self-produced water resources in 2030 under the RCP2.6 scenario, which is 4.26 billion m³. 3 The water resource volume was higher than the multi-year average, while the rest were lower than the average water resource volume of 4.22 billion cubic meters from 2001 to 2016. 3 The average water resources volume from 1956 to 2016 was 4.05 billion cubic meters. 3 Fluctuations in the vicinity.
[0087] The self-produced water resources in the Kaikong River Basin are expected to remain at 4.9-5 billion m³ under four discharge pathways in 2030 and 2050. 3 The overall water resources volume was lower than the average of 5.52 billion cubic meters from 2001 to 2016. 3 The average water resources volume from 1956 to 2016 was 5.07 billion cubic meters. 3 The result was the same.
[0088] The self-produced water resources of the Krya River basin are projected to remain between 2.3 and 2.5 billion cubic meters under four discharge pathways in 2030 and 2050. 3 The overall water resources volume was 2.86 billion cubic meters lower than the average water resources volume from 2001 to 2016. 3 The water resources volume is basically the same as the average water resources volume of 2.36 billion cubic meters from 1956 to 2016. 3 The result was the same.
[0089] The self-produced water resources of the Cherchen River basin and its smaller tributaries are projected to remain between 2.1 and 3.4 billion m³ under four discharge pathways in 2030 and 2050. 3 The overall value fluctuated significantly, with the highest value exceeding the average water resources volume of 3.13 billion m³ from 2001 to 2016. 3 The lowest value was lower than the average water resources volume of 2.4 billion cubic meters from 1956 to 2016. 3 .
[0090] The self-produced water resources of the Tarim River main stream are expected to remain between 0.09 and 0.28 billion m³ under four discharge pathways in 2030 and 2050. 3 The overall value fluctuated significantly, with the highest value exceeding the average water resources volume from 2001 to 2016 by 0.25 billion m³. 3 The lowest value was 0.16 billion m³ lower than the average water resources volume from 1956 to 2016. 3 .
[0091] Fluctuations in water resources are primarily related to precipitation. In the Tarim River Basin, precipitation in the nine tertiary zones under four discharge pathways fluctuates within a certain range, indicating that different discharge pathways have little impact on precipitation, thus resulting in minimal changes in water resources. The Cherchen River and the main stream of the Tarim River experience relatively larger fluctuations in water resources. This is mainly due to significant temperature fluctuations in these two basins. The Tarim River Basin has the highest average daily maximum temperature, while the Cherchen River Basin has the lowest average daily minimum temperature. These two basins are significantly affected by temperature, hence their more pronounced fluctuations compared to other basins. Furthermore, future water resource forecasts indicate that the abundant water resources from 2001 to 2018 are temporary, with water resources expected to decline in 2030 and 2050, posing potential risks to future water supply security.
[0092] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.
[0093] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0094] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A method for predicting water resources on a large spatial scale, characterized in that, Includes the following steps: Step 1: Select model metrics; Step 2: Model construction, mainly including the following steps: Step 2.1: Input the historical watershed data of the model indicators and the corresponding historical water resource data, and generate training and test sets based on the model indicators; Step 2.2: Create an SVR regression model using kernel functions and determine the model parameters; Step 2.3: Input the relevant parameters. The first parameter returned is the predicted value mapped from the multidimensional space, and the second parameter is the mean square error between the predicted and actual values of water resources in the test set. MSE and coefficient of determination ; Step 2.4: Utilize the mean squared error returned by the model MSE and coefficient of determination The performance of the established SVR regression model was evaluated. Step 2.5: Adjust the model parameters or reselect the kernel function type, and repeat steps 2.2 to 2.4 above until the model returns the mean squared error. MSE and coefficient of determination The requirements are met; Step 3: Input the future simulation and prediction results of the changing environment of the model indicators, fit the trained SVR model, and output the estimated future water resources. The model indicators selected in step 1 mainly include precipitation (pre) and maximum temperature (T). max Minimum temperature T min Seven factors: wind speed Win, solar radiation Rs, saturated vapor pressure ea, and potential evapotranspiration Et0; When creating the SVR regression model, a radial basis function kernel function is used, and the optimal parameters are determined by cross-validation. The optimal parameters are then used to train the SVR regression model. Mean squared error returned by the model ; Coefficient of determination ; The simulation prediction results of the changing environment of the model indicators are obtained in the following way: under different typical concentration paths of RCP2.6, RCP4.5, RCP6.0 and RCP8.5, the simulation results of the same mode of the five GCMs models are compared and the results are averaged with equal weight. Then, the same method is applied to integrate the results of the multiple models. The output of GCMs is reduced to a grid with a resolution of 0.5° and corrected using the trend preservation bias correction method.