Small watershed hydrological runoff production and pollution load short-term prediction refinement method

Abstract

The invention discloses a small watershed hydrological runoff and pollution load short-term prediction refinement method, which comprises the following steps of: constructing a modeling data set, constructing a small watershed foundation base model, segmenting hydrological response units according to land utilization types, crop planting structures and reservoir pond / wetland distribution data in a refinement manner respectively, and calculating the hydrological runoff and pollution load short-term prediction of the small watershed foundation base model. A small watershed foundation base model corresponding to the optimal segmentation scheme is screened out, and a small watershed depth base model is generated; constructing a reservoir pond / wetland simulation module; coupling the small-watershed depth base model and the reservoir pond / wetland simulation module to form a small-watershed hydrology and water quality early warning model, and optimizing the small-watershed hydrology and water quality early warning model; rolling to generate a hot start file for short-term prediction; constructing a future short-term weather enhancement forecasting model; the small watershed hydrology and water quality early warning model is driven to operate, the future water volume and load output are calculated in a rolling mode, and hydrology and water quality results are output; and calibrating and updating the small watershed hydrology and water quality early warning model. The precision and timeliness of hydrology and water quality prediction of the small watershed are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of simulation and analysis of watershed hydrological and water quality processes, and in particular to a refined method for short-term prediction of hydrological runoff and pollution load in a small watershed. Background Art

[0002] Accurately predicting pollutants and water volume in a hydrologically regulated basin will play a key role in controlling the pollution load in the basin. For example, the deployment of pollution load control work in the basin can be made in advance based on the prediction results. However, existing technologies are mainly aimed at predicting the water volume and pollution load in non-hydrologically regulated basins, such as calculating the load of basins without artificial interception measures or with a small amount of water intercepted by artificial interception measures. When predicting at the scale of hydrologically regulated basins with a large amount of water intercepted by artificial interception measures, the current existing technologies have poor prediction effects. The current existing technologies mainly have the following defects: (1) Inadequate expression of micro-characteristics: Existing watershed models are usually too rough in expressing watershed micro-characteristics, such as planting structure and reservoir / wetland distribution, and are therefore unable to effectively capture the rapid hydrological response and pollution process of small watersheds. (2) Ignoring or simplifying reservoir / wetland processes: In existing mainstream watershed models, the interception and degradation processes of reservoirs / wetlands are usually simplified or ignored, resulting in their contribution to the hydrological cycle and pollution load in small watersheds not being accurately reflected. However, in the simulation of hydrologically regulated watersheds, ponds and artificial interception measures within the watershed have a great impact on runoff generation and load migration and transformation, but the existing methods for hydrologically regulated watersheds do not take these issues into account.

[0003] Therefore, it is urgent to solve the above problems. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a refined method for short-term prediction of hydrological runoff and pollution load in a small watershed. The present invention significantly improves the accuracy, timeliness and response capability of hydrological and water quality predictions.

[0005] Technical solution: To achieve the above objectives, the present invention discloses a method for refining short-term prediction of hydrological runoff and pollution load in a small watershed, comprising the following steps: (1) Collect static and dynamic data of the study area, perform preprocessing, and construct a modeling data set; static data include soil data, elevation DEM data, crop data, land use type data, and reservoir / wetland distribution data; dynamic data include historical meteorological data, historical hydrological and water quality monitoring data, and soil moisture data; (2) Construct a small watershed foundation model based on the modeling data set; (3) The hydrological response units are segmented according to land use type, crop planting structure and reservoir / wetland distribution data, and different hydrological response unit segmentation schemes are generated. The small watershed foundation base model is run for each segmentation scheme, and the simulation results before and after the segmentation are recorded and compared. The model significance, model stability, model physical parameters and model calculation efficiency are analyzed in turn to select the small watershed foundation base model corresponding to the most preferred segmentation scheme as the small watershed deep base model. (4) Construct a reservoir / wetland simulation module to simulate the interception and degradation processes of reservoirs / wetlands, and dynamically characterize the impact of reservoirs / wetlands on micro-hydrological and water quality processes in small-scale watersheds; (5) Couple the small watershed deep base model with the reservoir / wetland simulation module to form a small watershed hydrological and water quality early warning model; use the small watershed deep base model to calculate the water yield and load of the upstream area of ​​the reservoir / wetland as the water inflow and load of the reservoir / wetland simulation module; the reservoir / wetland simulation module calculates the water outflow and load of the reservoir / wetland as the water inflow and load of the downstream area of ​​the reservoir / wetland of the small watershed deep base model; then use the small watershed deep base model to calculate the final water outflow and load of the watershed; (6) The acquired historical hydrological and water quality monitoring data are divided into a training data set and a verification data set, and the model parameters of the small watershed hydrological and water quality early warning model are encoded and processed. Based on the training data set of the historical hydrological and water quality monitoring data, the genetic algorithm is used to gradually screen out the optimal model parameter combination; finally, the verification data set of the historical hydrological and water quality monitoring data is used for verification to obtain the optimized small watershed hydrological and water quality early warning model; (7) Based on the simulation results and real-time automatic monitoring data of the small watershed hydrological and water quality early warning model, a data assimilation algorithm is used to generate a hot start file for short-term prediction on a daily basis, providing dynamic adjustment initial conditions for the small watershed hydrological and water quality early warning model; (8) Construct a future short-term meteorological enhanced forecast model, map the difference between historical rainfall forecast data and measured rainfall data to the correction amount of future meteorological forecast, and output the future meteorological enhanced forecast results; (9) Based on the latest hot start file and future meteorological enhanced forecast results, the meteorological boundary conditions are constructed, and the operation of the small watershed hydrological and water quality early warning model is driven. The future water volume and load generation are calculated in a rolling manner, and the hydrological and water quality results are output and visualized; (10) Automatically collect actual monitoring data every day, compare the output results of the small watershed hydrological and water quality early warning model with the actual monitoring data, and calculate the error rate. If the error rate exceeds the preset threshold, the automatic calibration module is triggered. The automatic calibration module uses the sensitive parameters of the small watershed hydrological and water quality early warning model as the key decision variables, sets the error index as the optimization objective function, and optimizes the parameter combination with the minimum error through the genetic algorithm. The parameters of the small watershed hydrological and water quality early warning model are updated to obtain the updated small watershed hydrological and water quality early warning model, which is then fed back to step (9).

[0006] Optionally, the soil data in step (1) includes soil type and soil physical and chemical properties; the elevation DEM data is used to extract the watershed range and calculate the terrain characteristics; the crop data includes the crop planting structure and growing season; the land use type data includes agricultural land, construction land, and unused land; the reservoir / wetland distribution data marks the range and location of the reservoir / wetland; the historical meteorological data includes rainfall, temperature, humidity, wind direction, wind speed, air pressure, evaporation, cloud cover, sunshine and solar radiation; the historical hydrological and water quality monitoring data includes river water level flow, reservoir / wetland water level and pollutant concentration; the soil moisture data includes soil moisture sensor data; the static data and the dynamic data are spatially superimposed, unified in coordinate system, and format converted to construct a modeling data set.

[0007] Optionally, in the step (2), a hydrological and water quality simulation computing platform suitable for a small watershed is selected as the basic base model of the small watershed, the elevation DEM data is converted into slope space vector data, and the acquired land use type data, soil data, and slope space vector data are integrated to construct a hydrological response unit; the macro-hydrological process and the macro-water quality process are simulated, wherein the simulation of the macro-hydrological process refers to the simulation of the hydrological cycle of the small watershed, and the simulation of the macro-hydrological process includes the simulation of rainfall interception, infiltration, evaporation and transpiration, surface runoff, and the flow generation and confluence process within the region; the simulation of the macro-water quality process includes the simulation of pollutant accumulation, scouring, degradation and transport processes; and the preliminary operation results of the small watershed basic base model are verified based on the regional runoff coefficient and the annual average pollutant concentration survey data.

[0008] Optionally, in step (3), the acquired land use type data is used as the primary classification, and the subdivided land type of the primary classification after investigation is combined as the secondary classification to segment the hydrological response unit; the segmentation method adopts a cyclic iteration method, that is, the primary classification and the secondary classification of the land use type are gradually combined to generate different hydrological response unit segmentation schemes; the small watershed base model is run for each segmentation scheme, and the simulation results before and after the segmentation are recorded, the simulation results include the runoff and pollution load of the most downstream outlet of the watershed, and the differences in the simulation results before and after the segmentation are compared; The hydrological response units are segmented and refined according to the crop planting structure. First, the hydrological response units are segmented and combined using the planting structure in the acquired crop data, that is, different crop planting combinations are divided into separate hydrological response units. The planting combinations include intercropping or rotation. A variety of hydrological response unit segmentation schemes are generated for different planting combinations. The small watershed foundation base model is run for each segmentation scheme, and the simulation results of the small watershed foundation base model are recorded. The simulation results include the runoff and pollution load at the downstream outlet of the watershed, and the differences in simulation results before and after segmentation are compared. The hydrological response units are refined and segmented according to the reservoir / wetland distribution data, and the reservoir / wetland areas in the basin are separated from the original hydrological response units and divided into independent hydrological response units. The small watershed foundation base model is run for each segmentation scheme, and the simulation results of the small watershed foundation base model are recorded. The simulation results include the runoff and pollution load at the downstream outlet of the basin, and the differences in simulation results before and after the segmentation are compared.

[0009] Optionally, the model significance analysis in step (3) refers to comparing the differences in the mean and variance of the output variables of the simulation results before and after the segmentation, the output variables include the runoff and pollution load at the most downstream outlet of the basin, and using statistical test methods to analyze whether the differences are significant. When the statistical difference is significant p<0.05, a model stability analysis is performed. The standard of the model stability analysis is that the standard deviation of the output variables of the simulation results of the segmented model shall not exceed twice that of the original model; the model physical parameter analysis refers to verifying whether the physical parameters of the model after segmentation are consistent with the range of actual observations or literature reports, and the physical parameters include evaporation and runoff coefficients; the model calculation efficiency analysis refers to recording the running time of different hydrological response unit segmentation schemes, and finding the small basin base model corresponding to the segmentation scheme with the shortest simulation time of 1 week.

[0010] Optionally, the interception process simulation calculation in step (4) includes water inflow process calculation, rainfall, evaporation and infiltration process calculation and water outflow process calculation; The calculation of the water inflow process includes the calculation of the water production upstream of the reservoir / wetland. The water production upstream of the reservoir / wetland is output by the deep base model. The water production upstream of the reservoir / wetland is the water inflow of the reservoir / wetland. ; The calculation of rainfall, evaporation and infiltration process refers to the calculation of rainfall, evaporation and infiltration in reservoirs / wetlands. The calculation formula of rainfall in reservoirs / wetlands is: , in, is the length of the reservoir / wetland, is the width of the reservoir / wetland, is the rainfall depth in the reservoir / wetland over a period of time, To adjust the coefficient, the reservoir / wetland calculation time step is converted to the deep base model output time step; The calculation formula for evaporation from reservoirs / wetlands is: , in, is the potential evapotranspiration depth over a period of time, is the adjustment factor used to adjust potential evapotranspiration to actual evaporation; The calculation formula for the infiltration rate of reservoirs / wetlands is: , in, is the infiltration volume per unit area of ​​reservoir / wetland; The calculation of the water discharge process refers to the calculation of the water discharge of the reservoir / wetland. The calculation formula for the water discharge of the reservoir / wetland is: , in is an empirical parameter; The width of the weir at the outlet of the reservoir / wetland; represents the acceleration due to gravity; The depth of the reservoir / wetland exceeds the height of the weir. When the depth of the reservoir / wetland does not exceed the height of the weir The value of is 0; The calculation formula is: , in The amount of water stored in the reservoir / wetland. is the area of ​​reservoir / wetland, Height of reservoir / wetland; Water storage capacity of reservoir / wetland The calculation formula is calculated by the water balance relationship between inflow and outflow and stored water: , for The amount of water stored in reservoirs / wetlands, is the water storage capacity of the reservoir / wetland at the initial moment, The amount of water added to the reservoir / wetland from the first moment to the last moment. , is the reduction in water volume in the reservoir / wetland from the initial moment to the final moment, .

[0011] Optionally, the degradation process calculation in step (4) refers to calculating the regional flow and load of the upstream catchment area of ​​the reservoir / wetland using a deep base model, and calculating the load of the reservoir / wetland degradation over a period of time through a first-order degradation process based on the regional flow and load of the upstream catchment area over a period of time and the load retained in the reservoir / wetland over a period of time; wherein the initial water volume and concentration are set at the beginning of the iteration to calculate the initial value of the load retained in the reservoir / wetland, and the load retained in the reservoir / wetland during the iteration process is the load retained in the reservoir / wetland after the previous iteration.

[0012] Optionally, step (7) specifically includes the following steps: (7.1) Obtain regional soil moisture data, reservoir / wetland water level and water quality data, and river water level, flow and water quality data, and perform data preprocessing; Among them, regional soil moisture data refers to the soil moisture distribution data in the study area obtained based on soil moisture sensor monitoring; reservoir / wetland water level and water quality data refers to the water level monitoring data and pollutant concentration data of reservoirs or wetlands, and the pollutant concentration data includes multiple concentration data of chemical oxygen demand, nitrogen, phosphorus, and biochemical oxygen demand; river water level flow and water quality data refers to the hydrological monitoring data of rivers and pollutant concentration monitoring data of river sections; the specific steps of data preprocessing are: unify the data time step; use linear interpolation to fill in missing data; perform spatial interpolation, that is, convert point data into complete data corresponding to the space of model sub-basins and sub-rivers; finally, perform unit and coordinate system one to ensure that all data formats are consistent and match the input requirements of the small watershed hydrological and water quality early warning model; (7.2) The data assimilation algorithm selects the ensemble Kalman filter to optimize multiple variables at the same time. The multiple variables include soil moisture, water level and pollutant concentration. The data assimilation algorithm is used to assimilate the data and update the model status. (7.3) Extract the updated model status, generate the corresponding hot start file format according to the model requirements, and dynamically generate the hot start file based on the rolling updated observation data and model operation results, so that each model operation uses the latest initial conditions; extract the updated model status including soil moisture, reservoir / wetland water level and water quality, and river water level, flow and water quality, use the assimilated soil moisture distribution as the initial condition of the small watershed hydrological and water quality early warning model, use the updated water level and pollutant concentration to adjust the initial conditions of the reservoir / wetland simulation module in the small watershed hydrological and water quality early warning model, and update the initial state of the river in the small watershed hydrological and water quality early warning model. The initial state of the river includes flow, cross-section water level and water quality.

[0013] Optionally, step (8) specifically includes the following steps: (8.1) Obtain historical rainfall forecast data, measured rainfall data and field monitoring station information, perform data alignment and spatiotemporal synchronization, and perform data cleaning and preprocessing; the historical rainfall forecast data includes the rainfall forecast information at the location of the field monitoring station at each time step, the measured rainfall data comes from the historical rainfall measured data of the field monitoring station, and the measured rainfall data is time series data; the field monitoring station information includes location and terrain; Match historical rainfall forecast data with measured rainfall data according to timestamps, so that historical rainfall forecast data and measured rainfall data are aligned in time and space; use interpolation to fill missing values, check and process abnormal data points, and finally standardize or normalize the data to ensure that the input data are at the same magnitude; (8.2) Calculate the difference between historical rainfall forecast data and measured rainfall data. Construct input variables based on the data in step (8.1), including forecast time series characteristics, forecast meteorological characteristics, and terrain characteristics of each field monitoring station. The output variable is the difference characteristics of forecast and measured rainfall. The calculation formula of the difference characteristics is:

[0014] in is the difference between the historical rainfall forecast data and the measured rainfall data at time t, is the historical rainfall forecast data at time t, is the measured rainfall data at time t; The forecast time series features include the month, day and hour of the timestamp, the forecast meteorological features include the predicted rainfall value, the terrain features include the height and terrain slope, and the difference feature is the difference between the historical rainfall forecast data and the measured rainfall data; (8.3) Construct a future short-term meteorological enhanced forecast model based on the random forest algorithm, use input variables and output variables to build, train and verify the future short-term meteorological enhanced forecast model, and realize the mapping of predicted rainfall to corrected rainfall; The input variables and output variables are divided into a training set and a test set, wherein the training set is used to train the sample data of the future short-term meteorological enhanced forecast model, and the test set is used to verify the sample data of the performance of the future short-term meteorological enhanced forecast model; The training set data is used to train the future short-term meteorological enhanced forecast model, and the mean square error (MSE) indicator is used to evaluate the performance of the future short-term meteorological enhanced forecast model. During the model training process, the random forest algorithm constructs multiple decision trees by randomly selecting features and data subsets, and outputs the final prediction results by voting; the hyperparameters are adjusted according to the performance of the model to obtain the trained future short-term meteorological enhanced forecast model.

[0015] Optionally, in step (9), the latest hot start file is obtained, and the trained future short-term weather enhanced forecast model is used, and the weather forecast data for the next week is used as input, and the future weather enhanced forecast result, i.e., the weather correction value, is output. The weather correction value plus the value after the predicted rainfall will be used as the meteorological boundary condition of the small watershed hydrological and water quality early warning model; the hot start file and the meteorological boundary condition are used to simulate and calculate the water volume and pollution load of each time step through the small watershed hydrological and water quality early warning model; The small watershed hydrological and water quality early warning model will be calculated on a rolling basis every day in the next week. The results of each simulation on the first day will be used as the hot start file before the assimilation of the small watershed hydrological and water quality early warning model on the second day; the flow statistics are based on the daily scale to calculate the flow of the region; the pollution statistics are to calculate the time series of pollutant concentration changes in different water bodies in the region and the pollution load of different indicators at the basin outlet. Different water bodies refer to rivers, wetlands or reservoirs; Record and store the output results of the small watershed hydrological and water quality early warning model, including flow and pollution production, and store the output results in time series; visualize the statistical results of flow and pollution production, and use charts or maps to show the water volume and pollution load in different regions; use time series graphs or heat maps to show the dynamic changes in hydrology and water quality; generate statistical reports on regional flow and pollution production, and analyze hydrological and water quality changes under different scenarios. The report content includes the flow and pollution production statistics and early warning status for the next week. The early warning status means that the water volume or pollution load in the region exceeds the predetermined threshold; if the flow or pollution production exceeds the predetermined threshold, a water quality or hydrological early warning is triggered.

[0016] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: (1) The present invention performs fine segmentation and reconstruction of the hydrological response unit HRU of the watershed, and carefully describes the microscopic characteristics such as land use type, crop planting structure, reservoir / wetland, etc., to generate a deep base model of the small watershed, thereby improving the sensitivity of the deep base model of the small watershed to the rapid response characteristics of small watersheds; (2) The present invention constructs a reservoir / wetland simulation module to dynamically characterize the hydrological and water quality regulation role of reservoirs / wetlands in small watersheds, and adopts a tight coupling method to tightly couple the small watershed deep base model with the reservoir / wetland simulation module, thereby achieving efficient and accurate joint simulation; (3) Based on the data assimilation algorithm, the present invention generates the model hot start file on a daily basis, uses the real-time monitoring data to continuously optimize the model's initial state, improves the model's early warning capability, and enhances the real-time and flexibility of the model's early warning; (4) Based on the random forest algorithm, the present invention constructs a high-resolution future short-term weather enhanced forecast model, corrects the input weather forecast data, and improves the reliability of weather boundary conditions; (5) The present invention can significantly improve the accuracy, timeliness and response capability of hydrological and water quality predictions, provide a reliable basis for regional water resources management, pollution prevention and control, disaster warning, etc., and has significant social, economic and technical effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the process of the present invention; Figure 2 It is a schematic diagram for comparing the simulated and measured values ​​of the wetland outlet outflow of the present invention and the comparative example respectively; Figure 3 It is a schematic diagram for comparing the simulated and measured total phosphorus values ​​of the wetland outlet water quality of the present invention and the comparative example respectively. DETAILED DESCRIPTION

[0018] 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 specific 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.

[0019] It should be understood that the present invention can be implemented in different forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, providing these embodiments will make the disclosure thorough and complete and fully convey the scope of the present invention to those skilled in the art. In the accompanying drawings, the sizes and relative sizes of the components may be exaggerated for clarity. The same reference numerals throughout represent the same components.

[0020] Example 1: Figure 1 As shown, the present invention discloses a method for refining short-term prediction of hydrological runoff and pollution load in a small watershed, comprising the following steps: (1) Collect data from the study area, perform preprocessing, and construct a modeling dataset; The data of the study area include static data and dynamic data. The static data include soil data, elevation DEM data, crop data, land use type data and reservoir / wetland distribution data, among which soil data include soil type and soil physical and chemical properties; elevation DEM data is used to extract the watershed range and calculate terrain characteristics; crop data include crop planting structure and growing season; land use type data include agricultural land, construction land and unused land; reservoir / wetland distribution data mark the range and location of reservoirs / wetlands; dynamic data include historical meteorological data, historical hydrological and water quality monitoring data and soil moisture data. Historical meteorological data include rainfall, temperature, humidity, wind direction and speed, air pressure, evaporation, cloud cover, sunshine and solar radiation; historical hydrological and water quality monitoring data include river water level flow, reservoir / wetland water level and pollutant concentration, and soil moisture data include soil moisture sensor data; static data and dynamic data are spatially superimposed, unified coordinate system, and format converted to construct a modeling data set.

[0021] (2) Construct a small watershed foundation model based on the modeling data set; Select a hydrological and water quality simulation computing platform suitable for small watersheds as the small watershed basic base model. The hydrological and water quality simulation computing platform is the SWAT model, HSPF model, LSPC model or MIKE SHE model; convert the elevation DEM data into slope space vector data, and fuse them based on the acquired land use type data, soil data, and slope space vector data to construct a hydrological response unit HRU; simulate macroscopic hydrological processes and macroscopic water quality processes, where simulating macroscopic hydrological processes refers to simulating the hydrological cycle of the small watershed. Simulating macroscopic hydrological processes includes simulating rainfall interception, infiltration, evaporation and transpiration, surface runoff, and runoff and confluence processes within the region; simulating macroscopic water quality processes includes simulating pollutant accumulation, scouring, degradation and transport processes; verify the preliminary operation results of the small watershed basic base model based on the regional runoff coefficient and the annual average pollutant concentration survey data to ensure the rationality of the small watershed basic base model simulation.

[0022] (3) Refine and segment the hydrological response unit HRU and reconstruct the basic base model of the small watershed to form a deep base model of the small watershed; The hydrological response units (HRUs) were segmented and refined according to land use types, crop planting structures, and reservoir / wetland distribution data, and different HRU segmentation schemes were generated. The small watershed foundation model was run for each segmentation scheme, and the simulation results before and after the segmentation were recorded and compared. First, the acquired land use type data is used as the first-level classification, which includes cultivated land, construction land, forest land and water body. The subdivided land types of the first-level classification after investigation are combined as the second-level classification, for example, the second-level classification of cultivated land includes paddy field, irrigated land and dry land, to segment the hydrological response unit HRU; the segmentation method adopts the cyclic iteration method, that is, the first-level classification and the second-level classification of land use types are gradually combined to generate different hydrological response unit HRU segmentation schemes; the small watershed foundation base model is run for each segmentation scheme, and the simulation results before and after the segmentation are recorded. The simulation results include the runoff and pollution load of the most downstream outlet of the watershed, and the differences in the simulation results before and after the segmentation are compared; The hydrological response units HRU are segmented and refined according to the crop planting structure. First, the crop planting structure in the acquired crop data is used to combine and segment the hydrological response units HRU, that is, different crop planting combinations are divided into separate hydrological response units HRU, and the planting combinations include intercropping or rotation; a variety of hydrological response unit HRU segmentation schemes are generated for different planting combinations, and the small watershed base model is run for each segmentation scheme, and the simulation results of the small watershed base model are recorded. The simulation results include the runoff and pollution load of the most downstream outlet of the watershed, and the differences in simulation results before and after segmentation are compared; The hydrological response units HRU were refined and segmented according to the reservoir / wetland distribution data, and the reservoir / wetland areas in the basin were separated from the original hydrological response units HRU and divided into independent hydrological response units HRU to improve the expression accuracy of these special areas; the small watershed basic base model was run for each segmentation scheme, and the simulation results of the small watershed basic base model were recorded, including the runoff and pollution load at the downstream outlet of the basin, and the differences in simulation results before and after segmentation were compared; The model significance, model stability, model physical parameters and model calculation efficiency are compared and analyzed in turn to select the small watershed foundation base model corresponding to the most preferred segmentation scheme and generate the small watershed deep base model. Model significance analysis refers to comparing the differences in the mean and variance of the output variables of the simulation results before and after segmentation. The output variables include the runoff and pollution load at the most downstream outlet of the basin. Statistical test methods are used to analyze whether the differences are significant. The statistical test methods are t-test and variance analysis ANOVA. When the statistical difference is significant p<0.05, model stability analysis is performed. The standard of model stability analysis is that the standard deviation of the output variables of the simulation results of the segmented model shall not exceed twice that of the original model; model physical parameter analysis refers to verifying whether the physical parameters of the segmented model are consistent with the range of actual observations or literature reports. The physical parameters include evaporation and runoff coefficients; model calculation efficiency analysis refers to recording the running time of different hydrological response unit HRU segmentation schemes, and finding the small basin base model corresponding to the segmentation scheme with the shortest calculation time for 1 week of simulation.

[0023] (4) Construct a reservoir / wetland simulation module to simulate the interception and degradation process of reservoirs / wetlands, and dynamically characterize the impact of reservoirs / wetlands on the microscopic hydrological and water quality processes of small-scale watersheds; The simulation calculation of the reservoir / wetland simulation module includes interception process calculation and degradation process calculation, among which the interception process calculation includes water inflow process calculation, rainfall, evaporation and infiltration process calculation and water outflow process calculation; The calculation of the water inflow process includes the calculation of the water production upstream of the reservoir / wetland. The water production upstream of the reservoir / wetland is output by the deep base model. The water production upstream of the reservoir / wetland is the water inflow of the reservoir / wetland. ; The calculation of rainfall, evaporation and infiltration process refers to the calculation of rainfall, evaporation and infiltration in reservoirs / wetlands. The calculation formula of rainfall in reservoirs / wetlands is: , in, is the length of the reservoir / wetland, is the width of the reservoir / wetland, is the rainfall depth in the reservoir / wetland over a period of time, To adjust the coefficient, the reservoir / wetland calculation time step is converted to the deep base model output time step; The calculation formula for evaporation from reservoirs / wetlands is: , in, is the potential evapotranspiration depth over a period of time, is the adjustment coefficient used to adjust potential evapotranspiration to actual evaporation; the calculation formula for infiltration of reservoirs / wetlands is: , in, is the infiltration volume per unit area of ​​the reservoir / wetland. The outflow process calculation refers to the calculation of the outflow volume of the reservoir / wetland. The calculation formula for the outflow volume of the reservoir / wetland is: , in is an empirical parameter; The width of the weir at the outlet of the reservoir / wetland; represents the acceleration due to gravity; The water depth of the reservoir / wetland exceeds the height of the weir. When the water depth of the reservoir / wetland does not exceed the height of the weir The value of is 0; The calculation formula is: , in The water storage capacity of reservoirs / wetlands, is the area of ​​reservoir / wetland, Height of reservoir / wetland; Water storage capacity of reservoir / wetland The calculation formula is calculated by the water balance relationship between inflow and outflow and stored water: , is the water storage capacity of the reservoir / wetland, is the water storage capacity of the reservoir / wetland at the initial moment, The amount of water added to the reservoir / wetland from the first moment to the last moment. , is the reduction in water volume in the reservoir / wetland from the initial moment to the final moment, ; The degradation process calculation refers to the use of the small watershed deep base model to calculate the regional flow and load of the upstream catchment area of ​​the reservoir / wetland. According to the regional flow and load of the upstream catchment area over a period of time and the load retained in the reservoir / wetland over a period of time, the load of the reservoir / wetland degradation over a period of time is calculated through a first-order degradation process; at the beginning of the iteration, the initial water volume and concentration are set to calculate the initial value of the load retained in the reservoir / wetland. During the iteration process, the load retained in the reservoir / wetland is the load retained in the reservoir / wetland after the previous iteration.

[0024] (5) Coupling the small watershed depth base model with the reservoir / wetland simulation module to form a small watershed hydrological and water quality early warning model; The water yield and load of the upstream area of ​​the reservoir / wetland are calculated using the small watershed deep base model as the water inflow of the reservoir / wetland simulation module. The reservoir / wetland simulation module calculates the water output of the reservoir / wetland and the load. The water inflow and load of the reservoir / wetland downstream area of ​​the small watershed deep base model are used as the water inflow and load of the reservoir / wetland downstream area of ​​the small watershed deep base model, and then the small watershed deep base model is used to calculate the final water outflow and load of the watershed.

[0025] (6) Based on the study area data obtained in step (1), the small watershed hydrological and water quality early warning model is calibrated to obtain an optimized small watershed hydrological and water quality early warning model; The acquired historical hydrological and water quality monitoring data are divided into training data sets and verification data sets, and the model parameters of the small watershed hydrological and water quality early warning model are encoded and processed. Based on the training data set of historical hydrological and water quality monitoring data, the nanotime efficiency coefficient is introduced as the core fitness index, and genetic operations are performed using genetic algorithms. Genetic operations include selection, crossover and mutation steps. Through iteration and convergence mechanisms, the optimal model parameter combination is gradually screened out; finally, the verification data set of historical hydrological and water quality monitoring data is used for verification to obtain the optimized small watershed hydrological and water quality early warning model; The specific operation steps of genetic algorithm are: (A) Determine the model parameters that need to be optimized, including flow, ammonia nitrogen, total nitrogen, total phosphorus and chemical oxygen demand; use binary coding to encode each parameter into one or more genes; (B) Introduction of nanohour efficiency coefficient , , in express t Actual monitoring data at all times, It indicates the average value of actual monitoring data within a period of time. Represents the simulated forecast data at time t; (C) Using the nanotime efficiency coefficient as the core indicator, a fitness function is constructed. The higher the value of the fitness function, the better the model parameters. (D) Randomly generate a set of initial model parameter combinations as the initial population of the genetic algorithm; use the roulette wheel selection strategy to select a portion of excellent individuals from the population as parents according to the value of the fitness function; use one-point crossover, two-point crossover and uniform crossover strategies to perform crossover operations on the selected parent individuals to generate new offspring individuals; randomly mutate the genes of the offspring individuals and introduce new genes; (E) Repeat the selection, crossover and mutation operations to generate a new population; each generation of population should be better than the previous generation, that is, the value of the fitness function is higher; (F) Setting a convergence condition, which includes reaching the maximum number of iterations or the average value or maximum value of the population fitness function reaching a set threshold. When the convergence condition is met, the iteration is stopped and the optimal parameter combination is output; (G) The optimal parameter combination is verified using a validation dataset of historical hydrological and water quality monitoring data.

[0026] (7) Based on the simulation results and real-time automatic monitoring data of the small watershed hydrological and water quality early warning model, a data assimilation algorithm is used to generate a hot start file for short-term prediction on a daily basis, providing dynamic adjustment initial conditions for the small watershed hydrological and water quality early warning model; (7.1) Obtain regional soil moisture data, reservoir / wetland water level and water quality data, and river water level, flow and water quality data, and perform data preprocessing; The regional soil moisture data refers to the soil moisture distribution data in the study area obtained based on soil moisture sensor monitoring; the reservoir / wetland water level and water quality data refers to the water level monitoring data and pollutant concentration data of the reservoir or wetland, and the pollutant concentration data includes multiple concentration data of chemical oxygen demand, nitrogen, phosphorus, and biochemical oxygen demand; the river water level flow and water quality data refers to the hydrological monitoring data of the river and the pollutant concentration monitoring data of the river section; The specific steps of data preprocessing are: unify the data time step, such as hourly level; use linear interpolation to fill in missing data; perform spatial interpolation, that is, convert point data into complete data corresponding to the model sub-basin and sub-river space, and use Kriging or IDW interpolation methods; finally, perform unit and coordinate system one to ensure that all data formats are consistent and match the input requirements of the small watershed hydrological and water quality early warning model; (7.2) The data assimilation algorithm selects the Ensemble Kalman Filter (EnKF), which is suitable for processing nonlinear systems; it can optimize multiple variables simultaneously, including soil moisture, water level and pollutant concentration; and it uses the data assimilation algorithm to perform data assimilation; The specific steps of data assimilation using data assimilation algorithm are as follows: (7.2.1) Initial state setting: Extract the initial state from the results of the last model run. The initial state includes soil moisture distribution, reservoir water level, reservoir water quality, river flow and river water quality; (7.2.2) Assimilation of observational data: Compare the observational data with the model prediction state and calculate the residual; the observational data include the latest soil moisture, reservoir water level, reservoir water quality, river flow and river water quality; (7.2.3) State update: Apply the data assimilation algorithm to fuse the observed data with the model state and update the model state: , in, is the updated model status, i.e., the analysis status; is the model prediction device, i.e., the prior state; is the observed data, is the observation operator, which maps the model state to the observation space; is the Kalman gain, which determines the weight distribution between the observed data and the model prediction; in, and The calculation method is: Observation Operator Used to map the state variables of the model to the observation space. In practical applications, The form and specific value of the value are usually determined by the following factors: Linear system: If the relationship between the model and the observation is linear, the observation operator is a constant matrix calculated from the known relationship between the model's state variables and the observed data; the relationship between the water level observations and the water level variables in the model is linear; Nonlinear system: For nonlinear relationships, the observation operator is locally linearized, that is, Taylor expansion is performed on the state of a certain time step; for the nonlinear relationship between soil moisture and water level, the local linearization is used to obtain matrix; The recommended value range and construction method are: Water level observation: The relationship between water level observation and water level variable in the model is linear. Assuming the model state is , and the observation is water level data , then set is the unit matrix, that is , for the water level observation variable; Soil moisture observation: If the observed soil moisture is obtained through sensors, and the model state is the soil moisture at each grid point ,So It is an interpolation operator from a grid point to an observation point. The specific interpolation method can be nearest neighbor interpolation or bilinear interpolation. Nonlinear systems: For nonlinear mappings, local linearization is recommended, that is, through Taylor expansion, for example: , in is the nonlinear relationship from the model state to the observed data and is the current state of the analysis; Kalman Gain Used to determine the weighted ratio between observed data and predicted data, Can be set to 0; (7.3) Extract the updated model state and generate the corresponding hot start file format according to the model requirements. Dynamically generate the hot start file based on the rolling updated observation data and model operation results to ensure that the latest initial conditions are used for each model operation; The updated model states are extracted including soil moisture, reservoir / wetland water level and water quality, and river water level, flow and water quality. The assimilated soil moisture distribution is used as the initial condition of the small watershed hydrological and water quality early warning model. The updated water level and pollutant concentration are used to adjust the initial conditions of the reservoir / wetland simulation module in the small watershed hydrological and water quality early warning model. The initial state of the river in the small watershed hydrological and water quality early warning model is updated. The initial state of the river includes flow, cross-section water level and water quality distribution.

[0027] (8) Construct a future short-term meteorological enhanced forecast model, map the difference between historical rainfall forecast data and measured rainfall data to the correction amount of future meteorological forecast, and output the future meteorological enhanced forecast result; the goal of the future short-term meteorological enhanced forecast model is to improve the accuracy of future meteorological forecast through the difference between historical forecast data and measured rainfall data, and use the random forest algorithm to correct the data; (8.1) Obtain historical rainfall forecast data, measured rainfall data and field monitoring station information, perform data alignment and spatiotemporal synchronization, and perform data cleaning and preprocessing; the historical rainfall forecast data includes the rainfall forecast information at the location of the field monitoring station at each time step, and the measured rainfall data comes from the historical rainfall measured data of the field monitoring station. The measured rainfall data is time series data with actual observation values; the field monitoring station information includes location and terrain; Match historical rainfall forecast data with measured rainfall data according to timestamps, and perform interpolation or adjustments to align historical rainfall forecast data with measured rainfall data in time and space; use interpolation to fill missing values, check and process abnormal data points, and finally standardize or normalize the data to ensure that the input data are at the same magnitude; (8.2) Calculate the difference between historical rainfall forecast data and measured rainfall data. Construct input variables based on the data in step (8.1), including forecast time series characteristics, forecast meteorological characteristics, and terrain characteristics of each field monitoring station. The output variable is the difference characteristics of forecast and measured rainfall. The calculation formula of the difference characteristics is: , in is the difference between the historical rainfall forecast data and the measured rainfall data at time t, is the historical rainfall forecast data at time t, is the measured rainfall data at time t; The forecast time series features include the month, day and hour of the timestamp, the forecast meteorological features include the predicted rainfall value, the terrain features include the height and terrain slope, and the difference feature is the difference between the historical rainfall forecast data and the measured rainfall data; (8.3) Construct a future short-term meteorological enhanced forecast model based on the random forest algorithm, use input variables and output variables to construct a model for training and verifying the future short-term meteorological enhanced forecast model, and realize the mapping of predicted rainfall to corrected rainfall; map the difference between historical rainfall forecast data and measured rainfall data to the correction amount of future meteorological forecast, and output the future meteorological enhanced forecast result; divide the input variables and output variables into a training set and a test set, where the training set is used to train the sample data of the future short-term meteorological enhanced forecast model, and the test set is used to verify the sample data of the performance of the future short-term meteorological enhanced forecast model, using 80% of the data as the training set and 20% of the data as the test set; The training set data is used to train the future short-term meteorological enhanced forecast model, and the mean square error (MSE) indicator is used to evaluate the performance of the future short-term meteorological enhanced forecast model. During the model training process, the random forest algorithm constructs multiple decision trees by randomly selecting features and data subsets, and outputs the final prediction results by voting. The hyperparameters are adjusted according to the performance of the model to obtain the trained future short-term meteorological enhanced forecast model. The hyperparameters include the number of trees n_estimators, the maximum depth of each tree max_depth, and the minimum sample leaf node min_samples_leaf.

[0028] (9) Based on the latest hot start file and the future meteorological enhanced forecast results of the future short-term meteorological enhanced forecast model, by constructing meteorological boundary conditions and driving the operation of the small watershed hydrological and water quality early warning model, the future water volume and load generation are calculated in a rolling manner, and the hydrological and water quality results and visualization are output to achieve the prediction and statistics of regional water volume and pollution load; Get the latest hot start file, use the trained future short-term meteorological enhanced forecast model, take the meteorological forecast data for the next week as input, and output the future meteorological enhanced forecast result, i.e., the meteorological correction. The meteorological correction plus the value after the predicted rainfall will be used as the meteorological boundary condition of the small watershed hydrological and water quality early warning model; use the hot start file and meteorological boundary conditions to simulate and calculate the water volume and pollution load of each time step through the small watershed hydrological and water quality early warning model; The small watershed hydrological and water quality early warning model will be calculated on a rolling basis every day in the next week. The results of each simulation on the first day will be used as the hot start file before the assimilation of the small watershed hydrological and water quality early warning model on the second day; the flow statistics are based on the daily scale to calculate the flow of the region; the pollution statistics are to calculate the time series of pollutant concentration changes in different water bodies in the region and the pollution load of different indicators at the basin outlet. Different water bodies refer to rivers, wetlands or reservoirs; Record and store the output results of the small watershed hydrological and water quality early warning model, including flow and pollution production, and store the output results in CSV files or database records in time series; visualize the statistical results of flow and pollution production, and use charts or maps to show the water volume and pollution load in different regions; use time series graphs or heat maps to show the dynamic changes in hydrology and water quality; generate statistical reports on regional flow and pollution production, and analyze the changes in hydrology and water quality under different scenarios. The report content includes the flow and pollution production statistics and early warning status for the next week. The early warning status means that the water volume or pollution load in the region exceeds the predetermined threshold; if the flow or pollution production exceeds the predetermined threshold, a water quality or hydrological early warning is triggered.

[0029] (10) Automatically collect actual monitoring data every day, compare the output results of the small watershed hydrological and water quality early warning model with the actual monitoring data, calculate the error statistics, and if the error rate exceeds the preset 30% threshold, trigger the automatic calibration module. This automatic calibration module uses the sensitive parameters of the small watershed hydrological and water quality early warning model as the key decision variables, sets the error index as the optimization objective function, and optimizes the parameter combination with the minimum error through the genetic algorithm. The parameters of the small watershed hydrological and water quality early warning model are updated to obtain the updated small watershed hydrological and water quality early warning model; The specific operation steps of the genetic algorithm are: (a) Determine the parameters that need to be optimized, such as lzsn, infilt, uzsn, cepsc, acqop, and wsqop parameters in the LSPC model; use binary coding to encode each parameter as one or more genes; (b) The error index is set as the optimization objective function, the error index is the mean square error (MSE) or the mean absolute error (MAE), the inverse or negative value of the error index is used as the fitness value, and a fitness function is constructed. The higher the value of the fitness function, the better the model parameters are; (c) Randomly generate a set of initial parameter combinations as the initial population of the genetic algorithm; use the roulette wheel selection strategy to select a portion of excellent individuals from the population as parents according to the value of the fitness function; use one-point crossover, two-point crossover, uniform crossover and other strategies to perform crossover operations on the selected parent individuals to generate new offspring individuals; randomly mutate the genes of the offspring individuals and introduce new genes; (d) Repeat the selection, crossover and mutation operations to generate a new population; each generation of population should be better than the previous generation, that is, the value of the fitness function should be higher; (e) Set a convergence condition, such as reaching the maximum number of iterations, the average value or maximum value of the population fitness function reaching a certain threshold, etc. When the convergence condition is met, stop the iteration and output the optimal parameter combination; (f) Use actual monitoring data to verify the optimal parameter combination to ensure its effectiveness in practical applications; (h) Apply the optimal parameter combination to the small watershed hydrological and water quality early warning model and update the model parameters.

[0030] The present invention uses Example 1 with a reservoir / wetland simulation module to predict runoff and pollution load for a small watershed, and uses a comparative example without a reservoir / wetland simulation module to predict runoff and pollution load for a small watershed, wherein the difference between the comparative example and Example 1 is only the reservoir / wetland simulation module, wherein the outflow simulation schematic diagram is as follows Figure 2 As shown in the figure, the total phosphorus simulation schematic diagram is as follows: Figure 3 As shown, after adding the reservoir / wetland simulation module, in terms of water volume, the outflow simulation is more stable, the peak value is lower, and it is closer to the measured results; after adding the reservoir / wetland simulation module, in terms of water quality, taking total phosphorus as an example, the simulated fluctuation of total phosphorus concentration is more gentle, the peak value is lower, the low value is higher, and the overall level is closer to the measured results.

Claims

1. A refined method for short-term prediction of hydrological runoff and pollution load in a small watershed, characterized in that: The steps include: (1) Collect static and dynamic data of the study area, perform preprocessing, and construct a modeling data set; static data include soil data, elevation DEM data, crop data, land use type data, and reservoir / wetland distribution data; dynamic data include historical meteorological data, historical hydrological and water quality monitoring data, and soil moisture data; (2) Construct a small watershed foundation model based on the modeling data set; (3) The hydrological response units are segmented according to land use type, crop planting structure and reservoir / wetland distribution data, and different hydrological response unit segmentation schemes are generated. The small watershed foundation base model is run for each segmentation scheme, and the simulation results before and after the segmentation are recorded and compared. The model significance, model stability, model physical parameters and model calculation efficiency are analyzed in turn to select the small watershed foundation base model corresponding to the most preferred segmentation scheme as the small watershed deep base model. (4) Construct a reservoir / wetland simulation module to simulate the interception and degradation processes of reservoirs / wetlands, and dynamically characterize the impact of reservoirs / wetlands on micro-hydrological and water quality processes in small-scale watersheds; (5) Couple the small watershed deep base model with the reservoir / wetland simulation module to form a small watershed hydrological and water quality early warning model; use the small watershed deep base model to calculate the water yield and load of the upstream area of ​​the reservoir / wetland as the water inflow and load of the reservoir / wetland simulation module; the reservoir / wetland simulation module calculates the water outflow and load of the reservoir / wetland as the water inflow and load of the downstream area of ​​the reservoir / wetland of the small watershed deep base model; then use the small watershed deep base model to calculate the final water outflow and load of the watershed; (6) The acquired historical hydrological and water quality monitoring data are divided into a training data set and a verification data set, and the model parameters of the small watershed hydrological and water quality early warning model are encoded and processed. Based on the training data set of the historical hydrological and water quality monitoring data, the genetic algorithm is used to gradually screen out the optimal model parameter combination; finally, the verification data set of the historical hydrological and water quality monitoring data is used for verification to obtain the optimized small watershed hydrological and water quality early warning model; (7) Based on the simulation results and real-time automatic monitoring data of the small watershed hydrological and water quality early warning model, a data assimilation algorithm is used to generate a hot start file for short-term prediction on a daily basis, providing dynamic adjustment initial conditions for the small watershed hydrological and water quality early warning model; (8) Construct a future short-term meteorological enhanced forecast model, map the difference between historical rainfall forecast data and measured rainfall data to the correction amount of future meteorological forecast, and output the future meteorological enhanced forecast results; (9) Based on the latest hot start file and future meteorological enhanced forecast results, the meteorological boundary conditions are constructed, and the operation of the small watershed hydrological and water quality early warning model is driven. The future water volume and load generation are calculated in a rolling manner, and the hydrological and water quality results are output and visualized; (10) Automatically collect actual monitoring data every day, compare the output results of the small watershed hydrological and water quality early warning model with the actual monitoring data, and calculate the error rate. If the error rate exceeds the preset threshold, the automatic calibration module is triggered. The automatic calibration module uses the sensitive parameters of the small watershed hydrological and water quality early warning model as the key decision variables, sets the error index as the optimization objective function, and optimizes the parameter combination with the minimum error through the genetic algorithm. The parameters of the small watershed hydrological and water quality early warning model are updated to obtain the updated small watershed hydrological and water quality early warning model, which is then fed back to step (9).

2. According to claim 1, a method for refining short-term prediction of hydrological runoff and pollution load in a small watershed is characterized by: In the step (1), the soil data includes soil type and soil physical and chemical properties; the elevation DEM data is used to extract the watershed range and calculate the terrain characteristics; the crop data includes the crop planting structure and growing season; the land use type data includes agricultural land, construction land and unused land; the reservoir / wetland distribution data marks the range and location of the reservoir / wetland; the historical meteorological data includes rainfall, temperature, humidity, wind direction, wind speed, air pressure, evaporation, cloud cover, sunshine and solar radiation; the historical hydrological and water quality monitoring data includes river water level flow, reservoir / wetland water level and pollutant concentration; the soil moisture data includes soil moisture sensor data; the static data and the dynamic data are spatially superimposed, unified in coordinate system, and format converted to construct a modeling data set.

3. According to claim 1, a method for refining short-term prediction of hydrological runoff and pollution load in a small watershed is characterized by: In the step (2), a hydrological and water quality simulation computing platform suitable for a small watershed is selected as a small watershed basic base model, the elevation DEM data is converted into slope space vector data, and the acquired land use type data, soil data, and slope space vector data are integrated to construct a hydrological response unit; Simulate macro-hydrological processes and macro-water quality processes, where simulation of macro-hydrological processes refers to simulation of the hydrological cycle of a small watershed. Simulation of macro-hydrological processes includes simulation of rainfall interception, infiltration, evaporation and transpiration, surface runoff, and runoff generation and confluence processes within the region; simulation of macro-water quality processes includes simulation of pollutant accumulation, scouring, degradation and transport processes; and verify the preliminary operating results of the small watershed foundation base model based on regional runoff coefficients and annual average pollutant concentration survey data.

4. According to claim 1, a method for refining short-term prediction of hydrological runoff and pollution load in a small watershed is characterized by: In the step (3), the acquired land use type data is used as the primary classification, and the subdivided land types of the primary classification after investigation are combined as the secondary classification to segment the hydrological response units; The segmentation method adopts a cyclic iteration method, that is, the primary classification and secondary classification of land use types are gradually combined to generate different hydrological response unit segmentation schemes; the small watershed foundation base model is run for each segmentation scheme, and the simulation results before and after the segmentation are recorded. The simulation results include the runoff and pollution load of the most downstream outlet of the watershed, and the differences in the simulation results before and after the segmentation are compared; The hydrological response units are segmented and refined according to the crop planting structure. First, the hydrological response units are segmented and combined using the planting structure in the acquired crop data, that is, different crop planting combinations are divided into separate hydrological response units. The planting combinations include intercropping or rotation. A variety of hydrological response unit segmentation schemes are generated for different planting combinations. The small watershed foundation base model is run for each segmentation scheme, and the simulation results of the small watershed foundation base model are recorded. The simulation results include the runoff and pollution load at the downstream outlet of the watershed, and the differences in simulation results before and after segmentation are compared. The hydrological response units are refined and segmented according to the reservoir / wetland distribution data, and the reservoir / wetland areas in the basin are separated from the original hydrological response units and divided into independent hydrological response units. The small watershed foundation base model is run for each segmentation scheme, and the simulation results of the small watershed foundation base model are recorded. The simulation results include the runoff and pollution load at the downstream outlet of the basin, and the differences in simulation results before and after the segmentation are compared.

5. According to claim 1, a method for refining short-term prediction of hydrological runoff and pollution load in a small watershed is characterized by: The model significance analysis in step (3) refers to comparing the differences in the mean and variance of the output variables of the simulation results before and after the segmentation, the output variables include the runoff and pollution load at the most downstream outlet of the basin, and using statistical test methods to analyze whether the differences are significant. When the statistical difference is significant p<0.05, the model stability analysis is performed. The standard of the model stability analysis is that the standard deviation of the output variables of the simulation results of the segmented model shall not exceed twice that of the original model; the model physical parameter analysis refers to verifying whether the physical parameters of the model after segmentation are consistent with the range of actual observations or literature reports, and the physical parameters include evaporation and runoff coefficients; the model calculation efficiency analysis refers to recording the running time of different hydrological response unit segmentation schemes, and finding the small basin base model corresponding to the segmentation scheme with the shortest calculation time for 1 week of simulation.

6. The method for short-term prediction of hydrological runoff and pollution load in a small watershed according to claim 1 is characterized in that: The interception process simulation calculation in step (4) includes water inflow process calculation, rainfall, evaporation and infiltration process calculation and water outflow process calculation; The calculation of the water inflow process includes the calculation of the water production upstream of the reservoir / wetland. The water production upstream of the reservoir / wetland is output by the small watershed depth base model. The water production upstream of the reservoir / wetland is the water inflow of the reservoir / wetland. ; The calculation of rainfall, evaporation and infiltration process refers to the calculation of rainfall, evaporation and infiltration in reservoirs / wetlands. The calculation formula of rainfall in reservoirs / wetlands is: ， in, is the length of the reservoir / wetland, is the width of the reservoir / wetland, is the rainfall depth in the reservoir / wetland over a period of time, To adjust the coefficient, the reservoir / wetland calculation time step is converted to the deep base model output time step; The calculation formula for evaporation from reservoirs / wetlands is: , in, is the potential evapotranspiration depth over a period of time, is the adjustment factor used to adjust potential evapotranspiration to actual evaporation; The calculation formula for the infiltration rate of reservoirs / wetlands is: ， in, is the infiltration volume per unit area of ​​reservoir / wetland; The calculation of the water discharge process refers to the calculation of the water discharge of the reservoir / wetland. The calculation formula for the water discharge of the reservoir / wetland is: ， in is an empirical parameter; The width of the weir at the outlet of the reservoir / wetland; represents the acceleration due to gravity; The water depth of the reservoir / wetland exceeds the height of the weir. When the water depth of the reservoir / wetland does not exceed the height of the weir The value of is 0; The calculation formula is: , in The amount of water stored in the reservoir / wetland. is the area of ​​reservoir / wetland, Height of reservoir / wetland; Water storage capacity of reservoir / wetland The calculation formula is calculated by the water balance relationship between inflow and outflow and stored water: ， is the water storage capacity of the reservoir / wetland, is the water storage capacity of the reservoir / wetland at the initial moment, The amount of water added to the reservoir / wetland from the first moment to the last moment. , is the reduction in water volume in the reservoir / wetland from the initial moment to the final moment, .

7. The method for refining short-term prediction of hydrological runoff and pollution load in a small watershed according to claim 1 is characterized in that: The degradation process simulation calculation in step (4) refers to calculating the regional flow and load of the upstream catchment area of ​​the reservoir / wetland using the small watershed deep base model, and calculating the load of the reservoir / wetland degradation over a period of time through a first-order degradation process based on the regional flow and load of the upstream catchment area over a period of time and the load retained in the reservoir / wetland over a period of time; wherein the initial water volume and concentration are set at the beginning of the iteration to calculate the initial value of the load retained in the reservoir / wetland, and the load retained in the reservoir / wetland during the iteration process is the load retained in the reservoir / wetland after the previous iteration.

8. The method for short-term prediction of hydrological runoff and pollution load in a small watershed according to claim 1 is characterized in that: The step (7) specifically includes the following steps: (7.1) Obtain regional soil moisture data, reservoir / wetland water level and water quality data, and river water level, flow and water quality data, and perform data preprocessing; Among them, regional soil moisture data refers to the soil moisture distribution data in the study area obtained based on soil moisture sensor monitoring; reservoir / wetland water level and water quality data refers to the water level monitoring data and pollutant concentration data of reservoirs or wetlands, and the pollutant concentration data includes multiple concentration data of chemical oxygen demand, nitrogen, phosphorus, and biochemical oxygen demand; river water level flow and water quality data refers to the hydrological monitoring data of rivers and pollutant concentration monitoring data of river sections; the specific steps of data preprocessing are: unify the data time step; use linear interpolation to fill in missing data; perform spatial interpolation, that is, convert point data into complete data corresponding to the space of model sub-basins and sub-rivers; finally, perform unit and coordinate system one to ensure that all data formats are consistent and match the input requirements of the small watershed hydrological and water quality early warning model; (7.2) The data assimilation algorithm selects the ensemble Kalman filter to optimize multiple variables at the same time. The multiple variables include soil moisture, water level and pollutant concentration. The data assimilation algorithm is used to assimilate the data and update the model status. (7.3) Extract the updated model state, generate the corresponding hot start file format according to the model requirements, and dynamically generate the hot start file based on the rolling updated observation data and model operation results, so that each model operation uses the latest initial conditions; The updated model states are extracted including soil moisture, reservoir / wetland water level and water quality, and river water level, flow and water quality. The assimilated soil moisture distribution is used as the initial condition of the small watershed hydrological and water quality early warning model. The updated water level and pollutant concentration are used to adjust the initial conditions of the reservoir / wetland simulation module in the small watershed hydrological and water quality early warning model. The initial state of the river in the small watershed hydrological and water quality early warning model is updated. The initial state of the river includes flow, cross-section water level and water quality.

9. The method for short-term prediction of hydrological runoff and pollution load in a small watershed according to claim 1 is characterized in that: The step (8) specifically includes the following steps: (8.1) Obtain historical rainfall forecast data, measured rainfall data and field monitoring station information, perform data alignment and spatiotemporal synchronization, and perform data cleaning and preprocessing; the historical rainfall forecast data includes the rainfall forecast information at the location of the field monitoring station at each time step, the measured rainfall data comes from the historical rainfall measured data of the field monitoring station, and the measured rainfall data is time series data; the field monitoring station information includes location and terrain; Match historical rainfall forecast data with measured rainfall data according to timestamps, so that historical rainfall forecast data and measured rainfall data are aligned in time and space; use interpolation to fill missing values, check and process abnormal data points, and finally standardize or normalize the data to ensure that the input data are at the same magnitude; (8.2) Calculate the difference between the historical rainfall forecast data and the measured rainfall data, and construct input variables based on the data in step (8.1), the input variables include forecast time series characteristics, forecast meteorological characteristics and terrain characteristics of each field monitoring station; The output variable is the difference characteristic of the predicted and measured rainfall; the calculation formula of the difference characteristic is: ， in is the difference between the historical rainfall forecast data and the measured rainfall data at time t, is the historical rainfall forecast data at time t, is the measured rainfall data at time t; The forecast time series features include the month, day and hour of the timestamp, the forecast meteorological features include the predicted rainfall value, the terrain features include the height and terrain slope, and the difference feature is the difference between the historical rainfall forecast data and the measured rainfall data; (8.3) Construct a future short-term meteorological enhanced forecast model based on the random forest algorithm, use input variables and output variables to train and verify the future short-term meteorological enhanced forecast model, and realize the mapping from predicted rainfall to corrected rainfall; The input variables and output variables are divided into a training set and a test set, wherein the training set is used to train the sample data of the future short-term meteorological enhanced forecast model, and the test set is used to verify the sample data of the performance of the future short-term meteorological enhanced forecast model; The training set data is used to train the future short-term meteorological enhanced forecast model, and the mean square error (MSE) indicator is used to evaluate the performance of the future short-term meteorological enhanced forecast model. During the model training process, the random forest algorithm constructs multiple decision trees by randomly selecting features and data subsets, and outputs the final prediction results by voting; the hyperparameters are adjusted according to the performance of the model to obtain the trained future short-term meteorological enhanced forecast model.

10. The method for short-term prediction of hydrological runoff and pollution load in a small watershed according to claim 1 is characterized in that: In the step (9), the latest hot start file is obtained, and the trained future short-term weather enhanced forecast model is used. The weather forecast data for the next week is used as input, and the future weather enhanced forecast result, i.e., the weather correction value, is output. The weather correction value plus the value after the predicted rainfall will be used as the meteorological boundary condition of the small watershed hydrological and water quality early warning model; the hot start file and the meteorological boundary condition are used to simulate and calculate the water volume and pollution load of each time step through the small watershed hydrological and water quality early warning model; The small watershed hydrological and water quality early warning model will be calculated on a rolling basis every day in the next week. The results of the first day of each simulation will be used as the hot start file before the small watershed hydrological and water quality early warning model is assimilated on the second day. Flow statistics are based on the day as the statistical unit to calculate the flow of the region; Pollution production statistics are to calculate the time series of pollutant concentration changes in different water bodies in the region and the pollution load of different indicators at the basin outlet. Different water bodies refer to rivers, wetlands or reservoirs. Record and store the output results of the small watershed hydrological and water quality early warning model, including runoff and pollution production, and the output results are stored in time series; Visualize the statistical results of runoff and pollution production, and use charts or maps to show the water volume and pollution load in different areas; use time series graphs or heat maps to show the dynamic changes of hydrology and water quality; Generate statistical reports on regional runoff and pollution output, and analyze hydrological and water quality changes under different scenarios. The report includes statistics on runoff and pollution output for the next week and early warning conditions. Early warning conditions refer to when the water volume or pollution load in the region exceeds a predetermined threshold. If the flow or pollution output exceeds the predetermined threshold, a water quality or hydrological warning will be triggered.

Images