Multi-source rainfall data fusion and hydrological applicability analysis method

Abstract

The invention belongs to the field of hydrology, and particularly provides a multi-source rainfall data fusion and hydrological applicability analysis method, which comprises the following steps: preprocessing multi-source rainfall data; constructing a multi-source data point-surface fusion model to obtain a fused precipitation product; drought identification application analysis of various rainfall products; hydrological simulation application analysis of various rainfall products; uncertainty calculation of hydrological simulation of various rainfall products; and evaluating the hydrological applicability of the rainfall product by adopting the hydrological applicability index. According to the method, multi-source precipitation data are fused, specific quantitative evaluation indexes of precipitation products are provided, and important reference is provided for guiding effective utilization of the multi-source precipitation products in different regions and different hydrological applications.

Description

Technical Field

[0001] The present invention belongs to the field of hydrology. Specifically, it relates to a method for fusing multi-source precipitation data and analyzing its hydrological applicability. Background Art

[0002] Multi-source precipitation data is indispensable in hydrological simulation and related flood and drought disaster management. In the past few decades, researchers have developed many widely used precipitation products, mainly including interpolation products based on station measurement data, satellite precipitation products, satellite products corrected by stations, and climate model outputs (reanalysis data). With the development of technology, the popularization of multi-source precipitation data and its fusion products in hydrological applications is an inevitable trend.

[0003] The existing technologies mainly focus on improving the accuracy of multi-source precipitation data. For example, CN202110557021.3 proposes a method and device for fusing ground station and remote sensing observed precipitation data to obtain high-precision precipitation data through error correction; CN202310092569.4 uses a high-precision earth surface modeling method to solve the error problem and multi-scale problem in multi-source precipitation fusion. The existing technologies lack the quantification of the applicability of multi-source precipitation products and their fusion products, and pay less attention to the application effects of multi-source precipitation data and fused precipitation data.

[0004] However, each precipitation product has its own limitations. In-situ observation station data is restricted by terrain, station density, cost, etc., while radar estimation and satellite inversion data are restricted by algorithms and sensors. All precipitation data has certain errors, and its fusion products will also transmit the original errors. This kind of error is one of the main sources of hydrological simulation errors. Therefore, different from the inherent cognition that the hydrological simulation accuracy of fusion products will necessarily improve, the effect of fusion products in simulating hydrological variables is not necessarily better than that of the original products. The influence mechanism of the uncertainty of precipitation products on their applications in hydrology is not clear, which hinders the applications of precipitation products in precipitation forecasting, water resource management, disaster warning, etc. Therefore, it is necessary to quantify the hydrological applicability of multi-source precipitation products to guide the applications of precipitation products with different performance. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for fusing multi-source precipitation data and analyzing its hydrological applicability, which fuses multi-source precipitation data and gives specific quantitative evaluation indicators for precipitation products, providing an important reference for guiding the effective utilization of multi-source precipitation products in different regions and different hydrological applications.

[0006] To solve the above technical problem, the technical solution adopted by the present invention is: a method for fusing multi-source precipitation data and analyzing its hydrological applicability, comprising the following steps: Step 1. Preprocessing of multi-source precipitation data; Step 2: Construct a multi-source data point-surface fusion model to obtain a fused precipitation product; Step 3: Application analysis of drought identification for various precipitation products; Step 4: Application analysis of hydrological simulation for various precipitation products; Step 5: Uncertainty calculation of hydrological simulation for various precipitation products.

[0007] In the preferred solution, in Step 1, the multi-source precipitation data includes in-situ observation station precipitation and multi-source precipitation products. The in-situ observation station precipitation is obtained from rain gauges, and the multi-source precipitation products are remote sensing precipitation products obtained through different radars and satellites.

[0008] In the preferred solution, in Step 1, the operation method for preprocessing the multi-source precipitation data is as follows: Select a reference resolution and time scale, register the multi-source precipitation data, unify the resolution through interpolation to obtain preprocessed raster-type data, and each raster contains three dimensions: longitude, latitude, and date.

[0009] In the preferred solution, in Step 2, the method for constructing a multi-source data point-surface fusion model includes the following steps: S201: Select several factors that affect precipitation estimation as feature variables, process them into a feature matrix. The feature matrix and multi-source precipitation data are used as a sample set. The factors that affect precipitation estimation include environmental factors and meteorological factors. The environmental factors include slope, aspect, longitude, latitude, DEM, station, and altitude. The meteorological factors include temperature and wind speed; S202: Select a machine learning algorithm as the multi-source data point-surface fusion model, train the sample set data, and obtain a fused precipitation product.

[0010] In the preferred solution, in Step S202, the machine learning algorithm uses classification and regression in two steps to obtain a fused precipitation product: Use the in-situ station measured data as the supervision label for training, and determine whether precipitation occurs based on the measured data; On the basis of precipitation occurrence, perform fusion of precipitation data. Use the precipitation raster obtained by station interpolation as the dependent variable, and the remaining original multi-source precipitation products and the feature variables of the corresponding raster as independent variables for regression training. The default precipitation amount on non-precipitation days is 0, and combine the fused precipitation amount with non-precipitation days to obtain the final fused precipitation product.

[0011] In the preferred solution, it further includes Step S203. Based on the original multi-source precipitation products and the fused precipitation product, calculate the correlation coefficient or / and determination coefficient or / and false alarm rate or / and success index using the in-situ observation station data as the benchmark as the accuracy index I1 of the precipitation product.

[0012] In the preferred solution, in the third step, a drought index based on the original multi-source precipitation products and the fused precipitation product is constructed for drought feature analysis. The drought index selects the standardized precipitation index. The drought feature analysis includes drought identification ability and drought event identification ability. Among them, the drought identification ability is represented by the proportion of each grid of the precipitation product in the same region identifying drought, and the drought event identification ability is represented by the hit rate of the droughts identified by each precipitation product and the actual drought occurrence. The drought identification performance index I2 is the arithmetic mean of the drought identification ability and the drought event identification ability.

[0013] In the preferred solution, in the fourth step, the original multi-source precipitation product and the fused precipitation product are used to drive the hydrological model respectively, and the simulation effect of hydrological variables of different precipitation products in the selected typical watershed is evaluated as the hydrological simulation performance index I3 to verify whether the accuracy of the fused precipitation product in hydrological simulation has been improved; the hydrological simulation performance index includes runoff simulation performance and / or other hydrological parameter simulation performance. The runoff simulation accuracy is evaluated by the Nash coefficient NSE, and other indexes are evaluated by the coefficient of determination R2.

[0014] In the preferred solution, in the fifth step, the ratio of the normalized central root mean square error of the precipitation product to the normalized central root mean square error of the hydrological variable driven by the precipitation product simulation is used as the uncertainty index of the precipitation product. The calculation formula is as follows: ; ; Among them, is the normalized central root mean square error, is the variable value based on other reference data, is the hydrological simulation variable driven by the precipitation product, is the number of samples, is the normalized central root mean square error of the hydrological variable, is the normalized central root mean square error of the precipitation product, represents the NCRMSE error metric ratio of the hydrological variable simulated by the hydrological model and the precipitation product, and is used to measure the uncertainty index I4 of the precipitation product.

[0015] In the preferred solution, it further includes the sixth step of evaluating the hydrological applicability of the precipitation product by using the hydrological applicability index. The hydrological applicability index is calculated through the precipitation product accuracy index, drought identification performance index, hydrological simulation performance index and uncertainty index constructed in the second to fifth steps. The calculation formula is: ; Among them, PA is the hydrological applicability index, i represents the i-th precipitation product, and j represents the j-th hydrological variable.

[0016] A method for fusing multi-source precipitation data and analyzing its hydrological applicability provided by the present invention has the following beneficial effects: 1. By preprocessing multi-source precipitation data, constructing a point-to-area fusion model for multi-source data, and using machine learning algorithms to train in two steps of classification and regression, a fused precipitation product is obtained, which can effectively integrate precipitation data from different sources, improve the accuracy and integrity of precipitation data, and provide a more reliable data basis for subsequent analysis.

[0017] 2. It effectively reveals the advantages and disadvantages of various precipitation data in hydrological simulation, gives specific quantitative evaluation indicators for precipitation products, and provides an important reference for guiding the effective utilization of multi-source precipitation products in different regions and different hydrological applications.

[0018] 3. By calculating hydrological applicability indicators through the constructed accuracy indicators, drought identification performance indicators, hydrological simulation performance indicators, and uncertainty indicators of precipitation products, the hydrological applicability of precipitation products can be comprehensively evaluated, providing an important reference for guiding the effective utilization of multi-source precipitation products in different regions and different hydrological applications, helping users select appropriate precipitation products according to specific needs, and improving the accuracy and reliability of hydrological applications.

[0019] 4. The fused precipitation product of the present invention can better capture the intra-annual and inter-annual variations of runoff, shows robustness in estimating low-flow and peak-flow discharges, and has the ability to improve the effect of hydrological simulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The present invention will be further described below with reference to the drawings and embodiments: Figure 1 is a flow chart of the present invention; Figure 2 is a flow chart of multi-source precipitation data fusion; Figure 3 is a bar chart of the contribution degree of feature variables of the fusion model; Figure 4 is a comparison chart of correlation coefficients and root mean square errors before and after precipitation product fusion; Figure 5 is a comparison chart of hit rate, false alarm rate, and success index before and after precipitation product fusion; Figure 6 is a performance comparison chart of MSFPPs; Figure 7 is the drought identification ability of each grid cell in Hubei region; Figure 8 is an mHM runoff simulation chart driven by multi-source precipitation products and MSFPPs; Figure 9 is a comparison of the simulation effects of the original multi-source precipitation products and the fused precipitation products; Figure 10This is a comparison chart of the simulation runoff errors driven by RF-PP and ERA5L. Specific implementation manners

[0021] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0022] A method for fusing multi-source precipitation data and analyzing its hydrological applicability, as Figure 1 shown, includes the following steps: Step 1: Preprocessing of multi-source precipitation data.

[0023] The multi-source precipitation data includes in-situ observation station precipitation and multi-source precipitation products. The in-situ observation station precipitation is obtained from rain gauges, and the multi-source precipitation products are remote sensing precipitation products obtained through different radars and satellites.

[0024] For the in-situ observation station data, the stations containing missing measurement periods are screened out, and the precipitation observation data with complete time coverage is selected as the basis for precipitation accuracy verification and data fusion. The corresponding periods are selected for the remote sensing data. The in-situ observation station data has undergone quality control such as extreme value, daily average value, internal consistency, spatio-temporal consistency inspection, and manual verification and correction.

[0025] For the source precipitation products, mature precipitation products that are widely used, updated in a timely manner, have a large spatio-temporal coverage range, and high accuracy among various precipitation products are selected.

[0026] The selected reference resolution and time scale are used to register the multi-source precipitation data. Since there are slightly differences in the spatio-temporal benchmarks of different precipitation product data, it is necessary to register the multi-source precipitation data to align the longitude, latitude, and date of different product grids.

[0027] For example: for some precipitation products, the longitude is represented as -180~180, some are represented as 0~360, and some are represented as -179.5~179.5 (the midpoint of two meridians). They are all converted into the same representation, for example, all represented as 0-360, to facilitate the alignment and comparison of data with the same coordinates; For some precipitation products, the data is hourly, and some are daily. The hourly data needs to be accumulated into daily data before calculation to unify the time scale.

[0028] Through interpolation, raster type data after interpolation resampling is obtained. The main purpose of interpolation is to unify the resolution. For some products, the raster is 0.1 degree, that is, a raster covers an area of 10km * 10km, and for some it is 0.25 degree. By interpolation, raster data with a unified resolution is obtained, and each raster contains three dimensions: longitude, latitude, and date. For example, bilinear interpolation or Kriging interpolation can be used for interpolation operations.

[0029] Interpolate the in-situ observation station data to obtain raster data with the same resolution as the reference resolution. If the time scales do not match, the time scales also need to be matched, such as accumulating hourly precipitation into daily precipitation.

[0030] Step 2: Construct a multi-source data point-to-area fusion model to obtain a fused precipitation product, as Figure 2 shown, which specifically includes the following steps: S201: Select several factors that affect precipitation estimation as characteristic variables and process them into a characteristic matrix. The characteristic matrix contains multiple characteristic variables, and each characteristic variable is raster data of the same size as the multi-source precipitation data, serving as the input quantity of physical factors for model calculation.

[0031] The characteristic matrix and multi-source precipitation data are used as a sample set. The factors that affect precipitation estimation include environmental factors and meteorological factors. Environmental factors include slope, aspect, longitude, latitude, DEM, station, altitude, and meteorological factors include air temperature and wind speed; S202: Select a machine learning algorithm as the multi-source data point-to-area fusion model, train the sample set data, and obtain a fused precipitation product.

[0032] The multi-source data point-to-area fusion model divides the preprocessed data into a training set and a test set, and uses grid search and cross-validation to determine the optimal parameters.

[0033] The machine learning algorithm uses classification and regression in two steps to obtain a fused precipitation product: train with the in-situ station measured data as the supervision label to determine whether precipitation occurs based on the measured data; on the basis of precipitation occurrence, perform fusion of precipitation data. Use the precipitation raster interpolated from the station as the dependent variable, and the remaining original multi-source precipitation products and the characteristic variables of the corresponding raster as independent variables for regression training. The default precipitation amount for non-precipitation days is 0, and combine the fused precipitation amount with non-precipitation days to obtain the final fused precipitation product.

[0034] The multi-source data point-to-area fusion model, in order to reduce the calculation amount while maintaining the credibility of the training results and avoid overfitting caused by feature variables with too low contribution to the model participating in training, uses pre-training for trial calculation to determine the contribution degree ranking of feature variables, and eliminates feature variables with low influence on the results based on the ranking, and only selects some feature variables with higher rankings as the classification sample set of the fusion model.

[0035] The multi-source data point-to-surface fusion model can select various machine learning models to train sample data, compare and select the model with the best classification effect for precipitation data regression simulation, and then select the model result with the best simulation effect as the fused precipitation data.

[0036] S203. Based on the original multi-source precipitation products and the fused precipitation products, calculate the correlation coefficient or / and determination coefficient or / and false alarm rate or / and success index using the in-situ observation station data as the benchmark as the accuracy index I1 of the precipitation products.

[0037] Step 3. Analyze the application of various precipitation products in drought identification, explore the performance of multi-source precipitation products in drought identification, and analyze the applicability of the drought index constructed based on multi-source precipitation products in drought feature identification.

[0038] Construct a drought index based on the original multi-source precipitation products and the fused precipitation products for drought feature analysis. The drought index selects the standardized precipitation index. The drought feature analysis includes drought identification ability and drought event identification ability. Among them, the drought identification ability is represented by the proportion of each grid of precipitation products in the same region identifying drought, and the drought event identification ability is represented by the hit rate of the droughts identified by each precipitation product and the actual droughts. The drought identification performance index I2 is the arithmetic mean of the drought identification ability and the drought event identification ability.

[0039] The drought event identification rate can be based on the meteorological disaster yearbook of the study area as the evaluation basis, and the drought features can be based on independent drought indicators as a reference, such as based on the standardized precipitation-evapotranspiration index.

[0040] Step 4. Analyze the application of various precipitation products in hydrological simulation.

[0041] Use the original multi-source precipitation products and the fused precipitation products to drive the hydrological model respectively. Evaluate the hydrological variable simulation effects of different precipitation products in the selected typical basins as the hydrological simulation performance index I3, and verify whether the accuracy of the fused precipitation products in hydrological simulation has been improved.

[0042] The hydrological variable simulation effects are based on measured data or widely used high-precision data products as the verification benchmark. For example, runoff is verified by comparing with measured data, and soil moisture content, evapotranspiration, and water storage anomalies are compared with developed and mature data products.

[0043] When the hydrological model selects different precipitation products as inputs, keep other input data and model parameters consistent. The model parameters are calibrated through the measured flow stations in the basin.

[0044] The hydrological simulation performance indicators include runoff simulation performance and other hydrological parameter simulation performances. The runoff simulation accuracy is evaluated by the Nash coefficient NSE, and other indicators are evaluated by the coefficient of determination R2.

[0045] Step Five: Calculate the uncertainty of hydrological simulation of various precipitation products. Considering the uncertainty introduced by precipitation products, establish a quantitative index to discuss the error caused by precipitation input.

[0046] Take the ratio of the normalized central root mean square error of the precipitation product to the normalized central root mean square error of the hydrological variable simulated by the precipitation product drive as the uncertainty index of the precipitation product. The calculation formula is as follows: ; ; Among them, is the normalized central root mean square error, is the variable value based on other reference data, is the hydrological simulation variable driven by the precipitation product, is the number of samples, is the normalized central root mean square error of the hydrological variable, is the normalized central root mean square error of the precipitation product, represents the NCRMSE error metric ratio of the hydrological variable simulated by the hydrological model to the precipitation product, and is used to measure the uncertainty index I4 of the precipitation product.

[0047] The normalized central root mean square error (NCRMSE) combines the root mean square error and the variance of the target variable to better evaluate the uncertainty of the precipitation product and the hydrological variable driven by it.

[0048] Step Six: Evaluate the hydrological applicability of various precipitation products.

[0049] Use the hydrological applicability index to evaluate the hydrological applicability of precipitation products. The hydrological applicability index is calculated through the precipitation product accuracy index, drought identification performance index, hydrological simulation performance index, and uncertainty index constructed in Steps Two to Five. Since the precipitation product accuracy index, drought identification performance index, and hydrological simulation performance index are all positive indicators, while the uncertainty index is a reverse indicator, the uncertainty index I4 is made positive. The calculation formula is: ; Among them, PA is the hydrological applicability index, i represents the i-th precipitation product, and j represents the j-th hydrological variable.

[0050] The larger PA is, the better the hydrological applicability of the precipitation product in simulating the hydrological variable; PA iThe calculation method can be selected independently. In addition, according to the emphasis on different application aspects of precipitation products, some comprehensive evaluation methods can be adopted to assign different weights to the four indicators.

[0051] Specific implementation cases: In this embodiment, the daily value dataset V3.0 of basic meteorological elements of Chinese national ground meteorological stations, the fifth-generation land reanalysis dataset of the European Centre for Medium-Range Weather Forecasts (ECMWF ReAnalysis 5th Generation Land Component, ERA5-Land), the integrated multi-satellite retrievals for precipitation data of the Global Precipitation Measurement program (IMERG), and the Tropical Rainfall Measuring Mission (TRMM) are used for fusion calculation and hydrological applicability analysis. Among them, the station data has completed quality control such as extreme value, daily average value, internal consistency, spatio-temporal consistency check, and manual verification and correction, and the precipitation remote sensing products have been unified in spatio-temporal resolution.

[0052] Step 1: Extract the corresponding precipitation grids of the remote sensing precipitation products according to the longitude and latitude coordinates of each rain gauge station. The daily-scale accuracy of the three single-source precipitation products is evaluated based on the measured precipitation at the stations, as shown in Table 1. Among them, CC is the correlation coefficient, RMSE is the root mean square error, SD is the standard deviation, POD is the hit rate, FAR is the false alarm rate, and CSI is the success index. The results show that the accuracy of the remote sensing precipitation is relatively low.

[0053]

[0054] Step 2: Process the station precipitation and characteristic variables into grids of the same size as the remote sensing precipitation products for convenient input into model training, and establish Figure 2 the multi-source fusion precipitation model shown. The random forest (RF) algorithm, gradient boosting decision tree (GBDT) algorithm, and extreme gradient boosting tree (XGBoost) algorithm are respectively selected for supervised training. To avoid overfitting problems in model training and reduce the computational burden at the same time, pre-training is carried out. In addition to the three precipitation data required for fusion calculation, the top seven characteristic variables with the highest contribution are selected as characteristic variables, and the relative contribution of the characteristic variables is as Figure 3 shown.

[0055] The model classification results are shown in Table 2. The results of the three decision tree models show that the random forest model has the best classification effect. Therefore, the subsequent precipitation fusion is based on the classification results of the random forest.

[0056]

[0057] The merged product and the single-source precipitation product are grids of the same size. Therefore, for the identified non-precipitation days, the precipitation amounts corresponding to the grids at the corresponding times are directly filled with 0. According to the results of pre-training, the data samples are adjusted for formal training, and the comprehensive performance of the multi-source merged precipitation products (MSFPPs) is evaluated. Among them, the merged products trained by three machine learning methods are named "algorithm + PP". Figure 4 The CC and RMSE before and after merging are shown. Figure 5 The POD, FAR, and CSI before and after merging are shown. The SD of the RF product is slightly smaller than that of the XGBoost product, and its performance is the best. See Figure 6 . The comprehensive evaluation of the multi-source merged precipitation products shows that the merging algorithm can significantly improve the accuracy of precipitation products. The merged precipitation products, especially XGBoost-PP and RF-PP, show high correlation and reliability. Compared with the multi-source precipitation products, all three merged products have greatly improved. Among them, the CC of GBDT-PP has increased by 134% - 329.6%, the RMSE has decreased by 38.1% - 49.2%, and the CSI has increased by 49.3% - 88.7%; the CC of XGBoost-PP has increased by 161.8% - 380.4%, the RMSE has decreased by 67.3% - 70.8%, and the CSI has increased by 50.2% - 130.4%; the CC of RF-PP has increased by 171.2% - 397.8% at most, the RMSE has decreased by 67.4% - 73.6%, and the CSI has increased by 81.9% - 131.3%. The POD of the multi-source merged precipitation products is almost close to 1, and compared with the original precipitation products, the FAR has decreased by 12.2% - 65.9%, greatly reducing the false alarm rate of precipitation events in the multi-source precipitation products.

[0058] Step 3: As a direct application of precipitation products, the standardized precipitation index can better reflect the differences in precipitation products in drought feature identification. Taking the ERA5-Land product as a representative of the single-source precipitation product and comparing it with the merged product, the standardized precipitation index (SPI) is calculated. With -0.5 as the threshold, during a drought period, if the drought index drops to -0.5 or below, it is considered that the index successfully identifies the drought event. Taking the scope of Hubei Province as the application area and the historical drought records in the "China Meteorological Disaster Yearbook" as the measured drought events that occurred, the comparison results are shown in Table 3. Figure 7 The drought identification results of each grid unit in the Hubei area reflect that the SPI constructed based on the multi-source merged precipitation products has a great improvement in grid drought identification ability compared with the single precipitation product.

[0059]

[0060] Step 4. Examine the simulation effect of the hydrological model driven by the fused precipitation product. Taking the simulation of runoff by the mesoscale hydrological model as an example, the runoff is verified using the daily measured data of the flow station, and other indicators are verified using widely used mainstream data products. To avoid the influence of model parameters on the comparison results, the same model parameters are used and different precipitation products are used to drive the simulation of runoff. The results are as Figure 8 , Figure 9 shown. The multi-source fused precipitation product can better capture the intra-annual and inter-annual variations of runoff, and shows robustness in estimating the low-flow and peak-flow discharges, and has the ability to improve the hydrological simulation effect.

[0061] Step 5. Take the ratio of the normalized central root mean square error of the precipitation product to the normalized central root mean square error of the hydrological variable simulated by the product as the uncertainty index of the precipitation product. Still taking runoff as an example for calculation, compare the ERA5L and RF-PP with the best performance of the single-source precipitation product and the multi-source fused precipitation product respectively, as shown in Figure 10 . Compared with the single-source precipitation product ERA5L, the error degree of the fused precipitation product RF-PP decreases by about 21.3%, and the error degree of its simulated runoff decreases by about 35.0%, which proves that the multi-source fusion algorithm constructed in Step 2 significantly reduces the error of precipitation. The NCRMSE metric ratio of the ERA5L product for simulating runoff is 1.065, which proves that the error of the single-source precipitation product will be slightly amplified and transmitted to the runoff simulation process; while the NCRMSE metric ratio of the RF-PP product is 0.881, and the error of the fused precipitation product will be suppressed, and the NCRMSE metric ratio is less than 1, which corresponds to the decrease in the simulated runoff error of the fused precipitation.

[0062] Step 6. Finally, comprehensively consider the accuracy of the precipitation product, drought identification ability, hydrological simulation ability, and uncertainty indicators to assign scores to the precipitation product. Since the precipitation product accuracy index, drought identification performance index, and hydrological simulation performance index are all positive indicators, while the uncertainty index is a reverse indicator, the uncertainty index I4 is made positive. In this example, the precipitation product accuracy I1 takes the success index CSI; the drought identification performance I2 takes the arithmetic mean of the grid drought identification ability and the drought event identification ability; the hydrological simulation performance takes the arithmetic square root of the sum of the squares of the Nash coefficient (NSE) and the Kling-Gupta efficiency coefficient (KGE); I4 takes the reciprocal.

[0063]

[0064] It is easy for those skilled in the art to understand that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for the fusion of multi-source precipitation data and its hydrological applicability analysis, characterized in that It includes the following steps: Step 1, preprocessing of multi-source precipitation data; Step 2, constructing a multi-source data point-to-area fusion model to obtain a fused precipitation product; Step 3, application analysis of drought identification for various precipitation products; Step 4, application analysis of hydrological simulation for various precipitation products; Step 5, calculation of the uncertainty of hydrological simulation for various precipitation products.

2. The method for fusing multi-source precipitation data and analyzing its hydrological applicability according to claim 1, wherein In the said Step 1, the multi-source precipitation data includes in-situ observation station precipitation and multi-source precipitation products. The in-situ observation station precipitation is obtained from rain gauges, and the multi-source precipitation products are remote sensing precipitation products obtained through different radars and satellites.

3. The method for the fusion of multi-source precipitation data and its hydrological applicability analysis according to claim 2, wherein In the said Step 1, the operation method for preprocessing the multi-source precipitation data is: selecting a reference resolution and time scale, registering the multi-source precipitation data, and performing interpolation to unify the resolution to obtain preprocessed raster-type data. Each raster contains three dimensions: longitude, latitude, and date.

4. A method for the fusion of multi-source precipitation data and its hydrological applicability analysis according to claim 2, characterized in that, In the said Step 2, the method for constructing a multi-source data point-to-area fusion model includes the following steps: S201, selecting several factors that affect precipitation estimation as characteristic variables, processing them into a characteristic matrix. The characteristic matrix and the multi-source precipitation data are used as a sample set. The factors that affect precipitation estimation include environmental factors and meteorological factors. The environmental factors include slope, aspect, longitude, latitude, DEM, station, altitude, and the meteorological factors include air temperature and wind speed; S202, selecting a machine learning algorithm as the multi-source data point-to-area fusion model, training the sample set data to obtain a fused precipitation product.

5. A method for the fusion of multi-source precipitation data and its hydrological applicability analysis according to claim 4, characterized in that, In the said Step S202, the machine learning algorithm uses classification and regression in two steps to obtain a fused precipitation product: training with the in-situ station measured data as the supervision label to determine whether precipitation occurs based on the measured data; on the basis of precipitation occurring, performing fusion of precipitation data. Using the precipitation raster obtained by station interpolation as the dependent variable, and the remaining original multi-source precipitation products and the characteristic variables of the corresponding raster as independent variables for regression training. The default precipitation amount on non-precipitation days is 0, and the fused precipitation amount is combined with non-precipitation days to obtain the final fused precipitation product.

6. The method for the fusion of multi-source precipitation data and its hydrological applicability analysis according to claim 5, characterized in that, It also includes Step S203, calculating the correlation coefficient or / and determination coefficient or / and false alarm rate or / and success index based on the original multi-source precipitation products and the fused precipitation product, with the in-situ observation station data as the benchmark, as the precipitation product accuracy index I1.

7. A method for the fusion of multi-source precipitation data and its hydrological applicability analysis according to claim 1, characterized in that, In the said Step 3, constructing a drought index based on the original multi-source precipitation products and the fused precipitation product for drought characteristic analysis. The drought index selects the standardized precipitation index. The drought characteristic analysis includes drought identification ability and drought event identification ability. Among them, the drought identification ability is represented by the proportion of each raster of the precipitation product in the same region identifying drought, and the drought event identification ability is represented by the hit rate of each precipitation product identifying the occurred drought and the actual occurred drought situation. The drought identification performance index I2 is the arithmetic mean of the drought identification ability and the drought event identification ability.

8. A method for the fusion of multi-source precipitation data and its hydrological applicability analysis according to claim 1, characterized in that, In Step 4, the original multi-source precipitation product and the merged precipitation product are used to drive the hydrological model respectively, and the simulation effect of different precipitation products on hydrological variables in the selected typical basins is evaluated as the hydrological simulation performance index I3 to verify whether the accuracy of the merged precipitation product in hydrological simulation has been improved; the hydrological simulation performance index includes runoff simulation performance and / or other hydrological parameter simulation performance. The runoff simulation accuracy is evaluated by the Nash coefficient NSE, and other indexes are evaluated by the coefficient of determination R2.

9. A method for the fusion of multi-source precipitation data and its hydrological applicability analysis according to claim 1, characterized in that In Step 5, the ratio of the normalized central root mean square error of the precipitation product to the normalized central root mean square error of the hydrological variable driven and simulated by the precipitation product is used as the uncertainty index of the precipitation product, and the calculation formula is as follows: ； ； Among them, is the normalized central root mean square error, is the variable value based on other reference data, is the hydrological simulation variable driven by precipitation products, is the number of samples, is the normalized central root mean square error of hydrological variables, is the normalized central root mean square error of precipitation products, represents the NCRMSE error metric ratio between the hydrological variables simulated by the hydrological model and the precipitation products, and is used to measure the uncertainty index I4 of the precipitation products.

10. A method for the fusion of multi-source precipitation data and its hydrological applicability analysis according to claim 1, characterized in that, It further includes Step 6, in which the hydrological applicability index is used to evaluate the hydrological applicability of the precipitation product. The hydrological applicability index is calculated through the precipitation product accuracy index, drought identification performance index, hydrological simulation performance index and uncertainty index constructed in Steps 2 to 5, and the calculation formula is: ； where PA is the hydrological applicability index, i represents the i-th precipitation product, and j represents the j-th hydrological variable.

Images