Runoff reanalysis data error analysis method and system based on panel regression analysis

Abstract

The invention relates to the technical field of hydrological data analysis, and provides a runoff reanalysis data error analysis method and system based on panel regression analysis, and the method comprises the following steps: collecting a watershed hydro meteorological data set, watershed runoff reanalysis data and meteorological reanalysis data, extracting drainage basin runoff reanalysis time series data and drainage basin meteorological reanalysis time series data; constructing a fixed effect panel regression model of the meteorological input deviation and the runoff simulation deviation based on the deviation between the reanalysis data and the observation value; the method comprises the following steps of: clustering all drainage basins according to drainage basin site attribute data to obtain spatial clusters of the drainage basins and attribute characteristics of each cluster, resampling each cluster through a bootstrap method to form first panel data, performing fixed effect panel regression on the first panel data to obtain a corresponding regression coefficient, and performing fixed effect panel regression on the first panel data to obtain a regression result; a first coefficient distribution for measuring spatial heterogeneity effects of meteorological deviations is generated.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrological data analysis, and more specifically, to a runoff reanalysis data error analysis method and system based on panel regression analysis. Background Art

[0002] Global runoff reanalysis data is the output of global hydrological models driven by meteorological reanalysis data. It can provide continuous time-series runoff data and is widely used to provide a basis for water resource adaptability assessment and management decisions. Precipitation and evaporation are the main drivers of global runoff changes, but climate models are prone to bias in simulating precipitation and potential evapotranspiration variables. When meteorological reanalysis data are biased, they are input into global hydrological models (GHMs) for runoff simulation, resulting in systematic biases in global runoff reanalysis data. Due to the heterogeneity of land surfaces and the complexity of the interaction between climate inputs and watershed characteristics, it is currently difficult to conduct effective and highly accurate diagnostic analysis of the heterogeneous effects of meteorological errors on the annual scale of systematic errors in runoff reanalysis in large sample basins. Summary of the invention

[0003] In order to overcome the defect that the above-mentioned prior art is difficult to effectively diagnose the heterogeneous effect of meteorological errors on the annual scale of systematic errors in runoff reanalysis in a large sample basin, the present invention provides a runoff reanalysis data error analysis method and system based on panel regression analysis.

[0004] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0005] A runoff reanalysis data error analysis method based on panel regression analysis includes the following steps:

[0006] Collect basin hydrological and meteorological data sets, basin runoff reanalysis data and meteorological reanalysis data, and extract basin runoff reanalysis time series data and basin meteorological reanalysis time series data;

[0007] Based on the deviation between reanalysis data and observations, a fixed effect panel regression model of meteorological input deviation and runoff simulation deviation was constructed.

[0008] All river basins are clustered according to the attribute data of river basin stations to obtain the spatial clustering of the river basins and the attribute characteristics of each cluster. The clusters are resampled through the bootstrap method to form the first panel data. The fixed effect panel regression is performed on the first panel data to obtain the corresponding regression coefficients, generating the first coefficient distribution for measuring the spatial heterogeneity effect of meteorological deviation.

[0009] Furthermore, the present invention also proposes a runoff reanalysis data error analysis system based on panel regression analysis, and applies the runoff reanalysis data error analysis method proposed in the present invention. The system includes:

[0010] The data processing module is used to collect the basin hydrological and meteorological data set, basin runoff reanalysis data and meteorological reanalysis data, and extract the basin runoff reanalysis time series data and basin meteorological reanalysis time series data;

[0011] The first sampling module is used to cluster all watersheds according to the watershed site attribute data, obtain the spatial clustering of the watershed and the attribute characteristics of each cluster, and resample each cluster through the bootstrap method to form the first panel data;

[0012] The fixed effect panel regression module is equipped with a fixed effect panel regression model of meteorological input deviation and runoff simulation deviation, which is used to perform fixed effect panel regression on panel data, obtain the corresponding regression coefficients, and generate the first coefficient distribution for measuring the spatial heterogeneity effect of meteorological deviation.

[0013] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0014] The present invention uses a fixed effect panel regression model to analyze the runoff reanalysis data errors, introduces fixed effects of individual dimension and time dimension, captures the differences between individuals in the basin and / or between times through the fixed effect panel regression model, reduces the bias caused by omitted variables, controls the heterogeneity of unobservable variables between different basins and in time, and effectively diagnoses the heterogeneity effect of meteorological errors on the annual scale of runoff reanalysis in large sample basins;

[0015] The present invention uses a fixed-effect panel regression model to analyze the impact of meteorological input errors on the systematic error of runoff simulation on the annual scale, which solves the problem that traditional statistical analysis methods are easily affected by omitted variables, cross-basin heterogeneity and temporal changes, and facilitates the systematic error evaluation and influencing factor diagnosis of runoff reanalysis data in large sample basins. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The present invention is a flowchart of a method for error analysis of runoff reanalysis data based on panel regression analysis according to one embodiment of the present invention.

[0017] Figure 2 The annual average flow of runoff reanalysis data and the CAMELS watershed site map are shown according to one embodiment of the present invention.

[0018] Figure 3 It is a schematic diagram showing the annual variation of the deviation between the reanalysis data and the observation data of the CAMELS basin according to one embodiment of the present invention.

[0019] Figure 4 The figure is a scatter plot of annual average reanalysis data and observation data of the CAMELS basin according to one embodiment of the present invention.

[0020] Figure 5 It is a schematic diagram of clustering results of CAMELS watershed based on watershed attributes according to one embodiment of the present invention.

[0021] Figure 6 It is a schematic diagram of watershed attributes of each cluster in the CAMELS watershed according to an embodiment of the present invention.

[0022] Figure 7 The figure is a schematic diagram showing the sensitivity of runoff quantile deviation to meteorological deviation under different watershed clustering according to an embodiment of the present invention.

[0023] Figure 8 The figure is a schematic diagram showing the sensitivity of seasonal runoff quantile deviation to meteorological deviation according to an embodiment of the present invention.

[0024] Fig. 9 The figure is a schematic diagram showing the sensitivity of the runoff quantile deviation of each watershed cluster to the precipitation deviation according to an embodiment of the present invention.

[0025] Fig.10 FIG. 4 is a schematic diagram showing the sensitivity of the runoff quantile deviation of each watershed cluster to the potential evapotranspiration deviation according to an embodiment of the present invention.

[0026] Fig.11 This is a schematic diagram showing the influence of watershed attributes on precipitation deviation effect according to an embodiment of the present invention.

[0027] Fig.12 This is a schematic diagram showing the influence of watershed attributes on potential evapotranspiration deviation effect according to an embodiment of the present invention.

[0028] Fig.13 The present invention is a schematic diagram of an error analysis system for runoff reanalysis data based on panel regression analysis according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0030] The terms used in the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "the" and "the" used in the present invention and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0031] It should be understood that although the terms first, second, third, etc. may be used in the present invention to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present invention, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0032] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0033] Example 1

[0034] This example proposes a runoff reanalysis data error analysis method based on panel regression analysis. Figure 1 , which is a flow chart of the runoff reanalysis data error analysis method of this embodiment.

[0035] The runoff reanalysis data error analysis method proposed in this embodiment includes the following steps:

[0036] S1. Collect basin hydrological and meteorological data sets, basin runoff reanalysis data and meteorological reanalysis data, and extract basin runoff reanalysis time series data and basin meteorological reanalysis time series data;

[0037] S2. Based on the deviation between reanalysis data and observations, a fixed effect panel regression model of meteorological input deviation and runoff simulation deviation was constructed;

[0038] S3. Cluster all the river basins according to the attribute data of the river basin stations to obtain the spatial clustering of the river basins and the attribute characteristics of each cluster, and resample each cluster through the bootstrap method to form the first panel data, perform fixed effect panel regression on the first panel data to obtain the corresponding regression coefficients, and generate the first coefficient distribution for measuring the spatial heterogeneity effect of meteorological deviations.

[0039] The Fixed Effects Panel Regression Model used in this example is a statistical model used to analyze panel data, which refers to observations of the same group of individuals (i.e., watersheds) at multiple time points. This model is particularly suitable for studying the impact of individual effects on the relationship between variables, and for controlling individual effects when they may be related to explanatory variables.

[0040] This embodiment uses a fixed effect panel regression model to analyze the errors of runoff reanalysis data, introduces fixed effects of individual dimension and time dimension, captures the differences between individuals in the basin and / or between times through the fixed effect panel regression model, reduces the bias caused by omitted variables, and controls the heterogeneity of unobservable variables between different basins and over time, thereby improving the accuracy of error analysis.

[0041] This embodiment uses a fixed-effect panel regression model to analyze the impact of meteorological input errors on the systematic error of runoff simulation at the annual scale. It solves the problem that traditional statistical analysis methods are easily affected by omitted variables, cross-basin heterogeneity, and temporal changes, and facilitates the systematic error assessment and influencing factor diagnosis of runoff reanalysis data in large sample basins.

[0042] In an optional embodiment, in step S1, extracting basin runoff reanalysis time series data and basin meteorological reanalysis time series data includes the following steps:

[0043] For the watershed runoff reanalysis data, the watershed runoff reanalysis time series data are extracted from the watershed runoff reanalysis data according to the latitude and longitude attributes of the watershed outlet hydrological station;

[0044] For the watershed meteorological reanalysis data, the raster data in the meteorological reanalysis data is clipped according to the vector boundary of the watershed, and only the raster information inside the watershed is retained;

[0045] For the clipped raster data, calculate the actual area of ​​each raster within the watershed;

[0046] Traversing the clipped raster data, extracting the meteorological value of each pixel and its corresponding area weight; the meteorological value includes precipitation, potential evapotranspiration, and temperature;

[0047] The area-weighted average calculation is performed based on the meteorological value and area weight of each pixel to generate the time series data of basin meteorological reanalysis.

[0048] Among them, the expression for area weighted average calculation is:

[0049]

[0050] For each watershed, the above calculations are repeated to eventually generate the time series data of watershed meteorological reanalysis.

[0051] It should be noted that the basin hydrological and meteorological dataset collected in step S1 is observation data, and the basin runoff reanalysis data and meteorological reanalysis data are simulation data.

[0052] Among them, there are multiple grids in each watershed unit, the values ​​between the grids are different, and some grids do not completely fall within the watershed vector boundary, so they need to be extracted.

[0053] In this embodiment, the area weighted average method is selected to extract the meteorological reanalysis data to obtain time series data with the basin as the unit. Its functions include:

[0054] 1) Area-weighted average takes into account the relative size and position of each grid in the basin, which can more accurately reflect the true situation of meteorological variables in the basin and effectively improve the accuracy of error analysis;

[0055] 2) Area-weighted average can effectively integrate the spatial heterogeneity of meteorological conditions within the basin;

[0056] 3) Use area-weighted average to eliminate the deviation caused by the excessive or insufficient impact of meteorological data from certain small areas on the basin average, ensuring the rationality of the results;

[0057] 4) The average value of each river basin is calculated based on the same weighting principle, which can reasonably compare the meteorological data of different river basins or different time periods to ensure the comparability of the data.

[0058] In an optional embodiment, in step S2, based on the deviation between the reanalysis data and the observation value, a fixed effect panel regression model of meteorological input deviation and runoff simulation deviation is constructed, including the following steps:

[0059] Construct a panel data set; the panel data set includes the runoff quantile deviation Q at the annual scale of each basin bias and annual scale meteorological data deviation P bias , PET bias and T bias ; Its expression is:

[0060]

[0061] Among them, the subscript bias represents the bias of the basin, the subscript reanalysis represents the area-weighted time series of the reanalysis data of the basin, and the subscript obs represents the time series of the observation data of the basin; specifically, P bias represents the annual precipitation deviation in the basin, P reanalysisrepresents the annual precipitation under the basin runoff reanalysis time series data, P obs Represents the annual precipitation under the basin observation data time series data; PET bias The PET is the annual average potential evapotranspiration deviation of the basin. reanalysis represents the annual average potential evapotranspiration under the basin runoff reanalysis time series data, PET obs represents the annual average potential evapotranspiration under the time series data of basin observation data; T bias represents the annual mean temperature deviation of the basin, T reanalysis represents the annual average temperature under the basin runoff reanalysis time series data, T obs represents the annual average temperature under the time series data of the basin observation data; Q bias represents the runoff quantile deviation at the annual scale of the basin, Q reanalysis represents the runoff quantile under the basin runoff reanalysis time series data, Q obs It represents the runoff quantile under the time series data of the basin observation data;

[0062] Based on the panel data set, a fixed effect panel regression model of meteorological input deviation and runoff simulation deviation is constructed; its expression is:

[0063]

[0064] in, represents the systematic deviation of the runoff percentile q of catchment i at time period t, P i,t represents the precipitation deviation in basin i during time period t, PET i,t represents the average potential evapotranspiration deviation of basin i in time period t, T i,t represents the average temperature deviation of basin i in time period t; β 1 and β 2 are the regression coefficients of precipitation bias and potential evapotranspiration bias, β 3 is the regression coefficient of the coupling effect of temperature and precipitation deviation; μ i is the fixed effect of basin i, γ t is the time fixed effect for time period t, ε i,t is the error term for basin i in time period t.

[0065] In this embodiment, the deviation between the reanalysis data and the observed value refers to the deviation value obtained by subtracting the runoff reanalysis data and meteorological reanalysis data used as simulation data in the same time period in any basin from the observed data extracted from the hydrological and meteorological data set, and the runoff quantile deviation and annual scale meteorological data deviation of any basin are obtained by calculation. The expression is:

[0066] bias i=sim i -obs i

[0067] Among them, sim i is the reanalysis time series of runoff in basin i, obs i is the observed data of basin i, bias i is the deviation value of basin i. The annual precipitation deviation of basin i is bias , basin annual average potential evapotranspiration deviation PET bias and the annual mean temperature deviation of the basin T bias Composition of annual scale meteorological data deviation.

[0068] In an optional embodiment, in step S3, resampling each cluster by a bootstrap method to form first panel data includes the following steps:

[0069] According to the watershed attribute values ​​of each watershed, the watershed is classified using the self-organizing map clustering method to obtain the spatial clustering results of the watershed and the attribute characteristics of each cluster;

[0070] The watershed subsets of each cluster are sampled with replacement, where the number of samples drawn each time is the number of watersheds in the cluster, and a panel data set of the watershed subsets is constructed to obtain the first panel data.

[0071] In this embodiment, the watersheds are first classified according to their attributes, and the watersheds with similar attribute characteristics are grouped into one category, thereby obtaining the spatial clustering of the watersheds and the attribute characteristics of each cluster.

[0072] Furthermore, Bootstrap is used for sampling with replacement, and the number of samples drawn each time is the number of watersheds in the cluster. Then, a panel data set of the subset is constructed, and then a fixed effect panel regression model is used to obtain the corresponding regression coefficient. This step is repeated several times to obtain the coefficient distribution interval of the cluster. Repeat this process for all clusters to obtain the coefficient distribution of each cluster, and thus obtain the spatial effect distribution of meteorological deviation on the systematic deviation of different runoff quantiles.

[0073] For example, in a specific implementation process, Bootstrap sampling with replacement is performed on a subset of the watersheds of a cluster to construct a first panel data set, and the corresponding regression coefficients are obtained through a fixed effect panel regression model. This process is repeated 1000 times to obtain the coefficient distribution interval of the cluster. This process is repeated for each cluster to obtain the coefficient distribution of each cluster, and the spatial effect distribution of meteorological deviation on the systematic deviation of different runoff quantiles is obtained.

[0074] Furthermore, in an optional embodiment, step S3 further includes the following steps:

[0075] The basin hydrological and meteorological data set is resampled through the bootstrap method to form the second panel data, and the second panel data is subjected to fixed effect panel regression to obtain the corresponding regression coefficients, thereby generating a second coefficient distribution for measuring the uncertainty of the regression coefficients.

[0076] In this embodiment, the bootstrap method is used to quantify the uncertainty of the regression coefficient. Specifically, for each bootstrap, a sample is drawn with replacement from the large sample basin data set, and the number of samples is the number of basin samples. Some basins may be drawn multiple times, while others are not drawn at all. Each bootstrap forms a new panel data set, and the fixed effect panel regression is performed on the new panel data to obtain the corresponding regression coefficient. This step is repeated several times to obtain the statistics of the coefficient. Furthermore, a preset confidence interval is taken for each coefficient to obtain a coefficient distribution confidence interval used to measure the uncertainty of the response coefficient.

[0077] Exemplarily, in a specific implementation process, a second panel data is obtained by sampling with replacement from a large sample watershed data set, and a fixed effect panel regression is performed on the second panel data to obtain the corresponding regression coefficient. This process is repeated 1000 times to obtain the statistics of the corresponding coefficient, and then a 95% confidence interval is taken for each coefficient to obtain a second coefficient distribution for measuring the uncertainty of the regression coefficient.

[0078] Furthermore, in an optional embodiment, step S3 further includes the following steps:

[0079] All watersheds are divided into three categories according to the attribute values ​​of each watershed attribute;

[0080] The attributes of each river basin are resampled through the bootstrap method to form the third panel data, and the third panel data is subjected to fixed effect panel regression to obtain the corresponding regression coefficients, thereby generating a third coefficient distribution for measuring the degree of influence of each river basin attribute value on the meteorological deviation effect.

[0081] In this embodiment, the influence of the watershed attribute value on the meteorological deviation effect is analyzed by classifying the watershed attributes according to their attribute value and then performing resampling and fixed effect panel regression analysis.

[0082] For example, in a specific implementation process, all basins are divided into three categories: small, medium, and large according to the attribute values ​​of the basin attributes, and Bootstrap resampling is performed on these three categories to obtain the third panel data. Fixed effect panel regression is performed on the third panel data to obtain the corresponding regression coefficients. This process is repeated 1000 times to obtain the distribution of the corresponding coefficients. The Mann-Whitney U test is then performed on the three categories of each attribute value to obtain whether the attribute values ​​with different sizes have a statistically significant impact on the meteorological deviation.

[0083] Among them, the basin attributes include basin area, average basin slope, precipitation-snow-water ratio, forest coverage and dryness.

[0084] Example 2

[0085] This embodiment applies the runoff reanalysis data error analysis method proposed in Embodiment 1 to analyze the sources of systematic errors in the GloFAS runoff reanalysis data in the CAMELS large sample basin in the continental United States.

[0086] Step 1: Collect the basin hydrological and meteorological data set, basin runoff reanalysis data and meteorological reanalysis data, and extract the basin runoff reanalysis time series data and basin meteorological reanalysis time series data.

[0087] In a specific implementation process, the third-party libraries Pandas, Xarray and Geopandas packages in Python are used to read the runoff reanalysis data (.nc) and meteorological reanalysis data (.nc), which are stored in variables named Q_reanalysis, P_reanalysis, PET_reanalysis, and T_reanalysis, respectively. Pandas is used to read the basin hydrological and meteorological observation data, which are stored in variables named Q_obs, P_obs, PET_obs, and T_obs, respectively, and the basin attributes are stored in the attribute attribute.

[0088] For example, Figure 2 As shown, the multi-year average flow of the runoff reanalysis data of this embodiment and the location map of the CAMELS watershed stations.

[0089] Then, the function nc_remapper() of the Python third-party library EASYMORE is combined with the basin attribute vector boundary file .shp file to extract the area-weighted average time series of meteorological data reanalysis at the basin scale, and the runoff reanalysis time series of the basin site is extracted from the runoff reanalysis grid data through the longitude and latitude information of the hydrological station at the basin outlet.

[0090] Step 2: Compare the meteorological reanalysis and runoff reanalysis time series with the observed data to obtain the deviations, and construct a fixed-effect panel regression model of meteorological input deviation and runoff simulation deviation.

[0091] In a specific implementation process, Pandas is used to add timestamps to the reanalysis data time series and the observation data time series, and the daily scale mean and the daily scale mean deviation are calculated to obtain the deviation between the reanalysis data and the observation value.

[0092] For example, Figure 3As shown in the figure, the annual changes of the reanalysis data and observation data of 671 basins in the CAMELS dataset. It can be seen from the figure that the variability of runoff reanalysis is smaller than the annual variability of measured runoff, and the daily changes of precipitation reanalysis and observation data are large within the year, and the mean daily precipitation deviation fluctuates between -0.5 and 0.5 mm. The potential evapotranspiration reanalysis data is lower than the observed value when the evapotranspiration is high in summer, and higher than the observed value when the evapotranspiration is low in spring and winter, and its deviation has a strong seasonality. Compared with other hydrological variables, the temperature reanalysis data is in good agreement with the measured values, and the temperature reanalysis data is greater than the observed value within the year. As shown in the figure, the annual variability of runoff reanalysis and observation data is smaller than the observed value. Figure 4 The figure shows the annual average scatter plot of reanalysis data and observation data for 671 basins in the CAMELS dataset, showing the annual average observation and reanalysis scatter values ​​of runoff, precipitation, potential evapotranspiration, and air temperature.

[0093] Then, the panel regression equation is established using the PanelOLS function in the Python third-party library linearmodels.

[0094] Step 3: Classify the large sample basins according to basin attributes and resample them to obtain panel data.

[0095] In a specific implementation process, the watershed attributes are clustered using the third-party library minisom and MiniSom and scipy.cluster.hierarchy in scipy to obtain the cluster number of each watershed, and the watersheds are divided into 10 categories.

[0096] For example, Figure 5 As shown in the figure, it is a schematic diagram of the clustering results of the CAMELS basin based on basin attributes, where the CAMELS basin dataset is divided into 10 clusters. Cluster 1 is located in the northern part of the western Pacific coast and has a temperate marine climate; Clusters 2 and 3 are located in the northeastern coastal areas and the Great Lakes region and have a temperate continental humid climate; Cluster 4 is located in the northern part of the Appalachian Mountains; Cluster 5 is located in the southern part of the Appalachian Mountains and has a subtropical climate; Cluster 6 is located in the western Rocky Mountains and has a plateau climate; Cluster 7 is located in the southern part of the western Pacific coast and the southwestern arid desert and has a Mediterranean and desert climate; Cluster 8 is located along the Gulf of Mexico and has a tropical and subtropical climate; Cluster 9 is mainly located in the central Great Plains and has a temperate continental climate; Cluster 10 is located in the upper Mississippi River basin. Figure 6 As shown in Figure 1, it is a schematic diagram of the watershed attributes of each cluster in the CAMELS watershed. Figure 5 , 6It can be seen that cluster 1 has a relatively small basin area, a relatively large basin slope, precipitation is concentrated in winter, there are relatively few heavy precipitation events, and the basin runoff is relatively large; cluster 2 has a relatively large basin area, thick soil, gentle terrain, precipitation is concentrated in summer, and the basin base flow is relatively large; cluster 3 has relatively small precipitation seasonality, and the basin is wet all year round; cluster 4 has a relatively small basin area, high soil permeability, small underground porosity, large forest coverage, and relatively large runoff variability; cluster 5 has small precipitation seasonality, a small precipitation-snow-water ratio, low soil permeability, and large forest coverage; cluster 6 has a large basin slope, high dryness, and is greatly affected by snowfall; cluster 7 has a high dryness, a large slope, a small forest coverage rate, and a long duration of low flow; cluster 8 has a flat basin terrain, thick soil, high permeability, and large forest coverage; cluster 9 has a flat terrain, a large basin area, a dry climate, and large precipitation seasonality; cluster 10 has thick basin soil, little forest coverage, is affected by snowfall in winter, and is dry in summer.

[0097] A panel dataset is constructed by using the annual meteorological reanalysis bias and runoff reanalysis bias of all watersheds.

[0098] Step 4: Perform fixed effect panel regression analysis to obtain error analysis results.

[0099] In a specific implementation process, for each type of data, the Python third-party library Numpy is used to perform Bootstrap sampling with replacement on the panel data set to obtain a new panel data set. This process is repeated 1000 times to obtain 1000 panel data subsets, and then the fixed effect panel regression equation function is called for each subset to obtain the corresponding regression coefficient.

[0100] For the panel data set obtained by clustering and Bootstrap resampling based on basin station attributes, the distribution of regression coefficients obtained after fixed effect panel regression can be used to measure the spatial heterogeneity effect of meteorological deviation and visualize it as the result of error analysis.

[0101] For example, Figure 7 , 8 Shown are schematic diagrams of the sensitivity of runoff quantile deviations to meteorological deviations under different watershed clustering, and schematic diagrams of the sensitivity of runoff quantile deviations to meteorological deviations at the seasonal scale. Figure 7 It can be seen that the runoff deviation becomes sensitive to the deviation of meteorological elements starting from Q50. As the runoff quantile increases, the sensitivity increases nonlinearly and the confidence interval becomes larger. Figure 8 The seasonality of the meteorological bias effect on the runoff quantile bias in different seasons is shown. The seasonal difference of the precipitation bias effect is smaller than that of the potential evapotranspiration bias effect.

[0102] Furthermore, based on the watershed clusters obtained in step 3, each cluster imports the fixed effect model and resampling function in step 4 in Python to calculate the sensitivity of the runoff quantile deviation of different watershed clusters to the precipitation deviation and the sensitivity to the potential evapotranspiration deviation, and obtains the following: Fig. 9 , 10 Shown are schematic diagrams of the sensitivity of runoff quantile deviations of each basin cluster to precipitation deviations, and schematic diagrams of the sensitivity of runoff quantile deviations of each basin cluster to potential evapotranspiration deviations. Fig. 9 It can be seen that for low flow, clusters 2, 3, 4, 5, and 10 are more sensitive to low flow than other clusters. For high flow, it can be seen that clusters 8, 9, and 10 have greater effects and greater uncertainty. The basins of these clusters have flat terrain and small basin slopes. Clusters 2 and 8 have large uncertainties in their effects on low and high flows, respectively. The soil thickness of both clusters is large, the underground porosity is large, and the water storage and release of water in the basins are large and affected by groundwater. Clusters 6 and 7 have relatively small uncertainties in their sensitivity to precipitation deviations, whether high or low flow. The basins of these clusters are dry, precipitation is concentrated in winter, the slope is large, the uncertainty of runoff generation on the land surface is small, and the uncertainty of the confluence process is also small. Cluster 6 is a snow-dominated basin. Precipitation enters the basin first as snow accumulation and snow cover, which has a buffering effect on the conversion of precipitation into runoff. Fig.10 It can be seen that the potential evapotranspiration bias of clusters 1, 2, 3, 4, and 8 has a negative median low flow effect for these clusters, indicating that the potential evapotranspiration bias will make the runoff reanalysis data more accurate in simulating low flows. The main reason is Figure 3 It is shown in the figure that the characteristics of the global runoff reanalysis data are that high flows are lower than observed values, and low flows are higher than observed values. The effect uncertainty of cluster 4 at high flows will be negative because the soil permeability of cluster 4 is high, and groundwater can replenish the soil water content caused by evaporation. The low-flow bias effect of potential evapotranspiration deviation in clusters 6, 7, 9, and 10 is near 0. The forest coverage rate and dryness of the basins in these clusters are low, and the maximum leaf area index is smaller than that of other clusters, indicating that the seasonal changes in crop transpiration in these basins are not large. Among them, the effect uncertainty of clusters 6 and 7 is near 0 for all runoff quantiles. Clusters 2 and 3 have large effects and uncertainties at high flows, the basins are moist and the maximum leaf area index is large, and the transpiration of the forest increases the sensitivity and uncertainty of the deviation.

[0103] For the panel data formed by resampling the basin hydrological and meteorological data set, the regression coefficient distribution obtained after fixed effect panel regression can be used to measure the uncertainty of the regression coefficient and visualize it as the error analysis result.

[0104] Step 5: Diagnose the impact of the size of the watershed attribute value.

[0105] Specifically, basin attributes such as basin area, basin average slope, precipitation-snow-water ratio, forest coverage and dryness are selected, and the stations are divided into three categories: small, medium and large according to the size of the basin attribute value. By calling the resampling function in step 4, Bootstrap resampling is performed on each class to obtain panel data on the size of the basin attribute value. After the fixed effect panel regression is performed, the obtained regression coefficient distribution can be used to measure the degree of influence of each basin attribute value on the meteorological deviation effect. Furthermore, the Mann-Whitney U test is performed on the three categories of each attribute value to obtain whether the attribute values ​​with different sizes have a statistically significant impact on the meteorological deviation.

[0106] For example, Fig.11 , 12 As shown, it is a schematic diagram of the influence of basin attributes on the precipitation bias effect, and a schematic diagram of the influence of basin attributes on the potential evapotranspiration bias effect. In this embodiment, the CAMELS large sample basin is divided into low attribute value, medium attribute value, and high attribute value basins by the size of the specific basin attribute value. Fig.11 The influence of specific watershed attributes on the precipitation deviation effect on low flow (Q5), medium flow (Q50), and high flow (Q95) runoff quantiles is analyzed. Overall, watershed attributes will affect the size and uncertainty of the precipitation deviation effect, and it can be found that the change trend of the watershed attribute value on the low, medium, and high flow effects is consistent. It can be seen from the figure that the larger the watershed area, soil thickness, and underground porosity, the greater the precipitation deviation effect. The larger the watershed slope and the precipitation-snow-water ratio, the smaller the effect and the smaller the uncertainty. In addition, Fig.11 It can also be seen that the difference in the impact of basin attributes on low flows is smaller than that on high flows. The greater the slope, the smaller the effect of precipitation and the smaller the uncertainty, especially for high flows, because the surface runoff convergence time becomes shorter. Precipitation seasonality also shows that the greater the seasonality, the greater the deviation effect and the greater the uncertainty. The more concentrated the precipitation, the greater the deviation of the runoff convergence pattern and the process described by the hydrological model. The soil depth can be seen as the regulating role of soil water. For low flows, the medium attribute value does not change with the soil depth effect, but the high flow continues to change with depth.

[0107] and Fig.12The figure shows that the trend of the effect of basin attribute values ​​on potential evapotranspiration is inconsistent in low, medium and high flow. For low flow, when the dryness is small or the ratio of forest share to maximum leaf area index is medium or high, the potential evapotranspiration bias effect has a negative value. For medium flow, the larger the forest share attribute value, the greater the impact on the potential evapotranspiration bias effect. The greater the soil thickness and permeability, the smaller the potential evapotranspiration bias. Among them, the effect of the basin with high basin attribute of maximum leaf area index is negative. The trend of the forest share attribute affecting the potential evapotranspiration bias effect shows that as the attribute value increases, the effect on medium and high flow increases, and the effect on low flow decreases. The effect of forest share on low and high flow is different. Low flow occurs in autumn. The leaf area index of forest in autumn varies greatly, so the uncertainty of basin evaporation variation is large. The greater the underground permeability, the smaller the effect and uncertainty on medium and high flow. The higher the underground permeability, the groundwater can supplement evaporation and play a role in reducing the potential evapotranspiration bias effect.

[0108] Optionally, the above steps can be encapsulated into a function by defining a function def(), and the code can be saved as a .py file. When a class function needs to be called, the developed function can be used after being called using the import function.

[0109] Example 3

[0110] This embodiment applies the runoff reanalysis data error analysis method proposed in Embodiment 1, and proposes a runoff reanalysis data error analysis system based on panel regression analysis, such as Fig.13 , which is a schematic diagram of the runoff reanalysis data error analysis system of this embodiment.

[0111] The runoff reanalysis data error analysis system proposed in this embodiment includes:

[0112] The data processing module is used to collect the basin hydrological and meteorological data set, basin runoff reanalysis data and meteorological reanalysis data, and extract the basin runoff reanalysis time series data and basin meteorological reanalysis time series data;

[0113] The first sampling module is used to cluster all watersheds according to the watershed site attribute data, obtain the spatial clustering of the watershed and the attribute characteristics of each cluster, and resample each cluster through the bootstrap method to form the first panel data;

[0114] The fixed effect panel regression module is equipped with a fixed effect panel regression model of meteorological input deviation and runoff simulation deviation, which is used to perform fixed effect panel regression on panel data, obtain the corresponding regression coefficients, and generate the first coefficient distribution for measuring the spatial heterogeneity effect of meteorological deviation.

[0115] Further optionally, the system further includes:

[0116] The second sampling module is used to resample the basin hydrological and meteorological data set to form the second panel data through the bootstrap method;

[0117] The fixed effect panel regression module is also used to perform a fixed effect panel regression on the second panel data to obtain corresponding regression coefficients and generate a second coefficient distribution for measuring the uncertainty of the regression coefficients.

[0118] Further optionally, the system further includes:

[0119] The third sampling module is used to divide all the watersheds into three categories according to the attribute values ​​of each watershed attribute; the attributes of each watershed are resampled by the bootstrap method to form the third panel data;

[0120] The fixed effect panel regression module is also used to perform fixed effect panel regression on the third panel data to obtain corresponding regression coefficients, and generate a third coefficient distribution for measuring the degree of influence of each watershed attribute value on the meteorological deviation effect.

[0121] It can be understood that the system of this embodiment corresponds to the method of the above-mentioned embodiment 1, and the options in the above-mentioned embodiment 1 are also applicable to this embodiment, so they are not described repeatedly here.

[0122] Example 4

[0123] This embodiment proposes a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes all or part of the steps of the runoff reanalysis data error analysis method based on panel regression analysis proposed in Example 1.

[0124] Example 5

[0125] This embodiment proposes a storage medium having computer-readable instructions stored thereon, wherein the computer-readable instructions, when executed by a processor, implement all or part of the steps of the runoff reanalysis data error analysis method based on panel regression analysis proposed in Embodiment 1.

[0126] Exemplarily, the storage medium includes, but is not limited to, a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0127] Exemplarily, the instructions, programs, code sets or instruction sets may be implemented using conventional programming languages.

[0128] Exemplarily, the processor includes but is not limited to a smart phone, a personal computer, a server, a network device, etc., and is used to execute all or part of the steps of the runoff reanalysis data error analysis method based on panel regression analysis described in Example 1.

[0129] Each embodiment of the present invention is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device embodiment described above is merely exemplary, in which the modules described as separate components may or may not be physically separated, and the functions of each module can be implemented in the same or one or more software and / or hardware when implementing the scheme of the present invention. It is also possible to select some or all of the modules according to actual needs to achieve the purpose of the scheme of this embodiment.

[0130] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A runoff reanalysis data error analysis method based on panel regression analysis, characterized in that: The following steps are involved: Collect basin hydrological and meteorological data sets, basin runoff reanalysis data and meteorological reanalysis data, and extract basin runoff reanalysis time series data and basin meteorological reanalysis time series data; Based on the deviation between reanalysis data and observations, a fixed effect panel regression model of meteorological input deviation and runoff simulation deviation was constructed. All river basins are clustered according to the attribute data of river basin stations to obtain the spatial clustering of the river basins and the attribute characteristics of each cluster. The clusters are resampled through the bootstrap method to form the first panel data. The fixed effect panel regression is performed on the first panel data to obtain the corresponding regression coefficients, generating the first coefficient distribution for measuring the spatial heterogeneity effect of meteorological deviation.

2. The runoff reanalysis data error analysis method according to claim 1, characterized in that: The method of extracting the watershed runoff reanalysis time series data and the watershed meteorological reanalysis time series data comprises the following steps: Extracting basin runoff reanalysis time series data from the basin runoff reanalysis data according to the latitude and longitude attributes of the basin outlet hydrological station; Clipping the raster data in the meteorological reanalysis data according to the vector boundary of the watershed; For the clipped raster data, calculate the actual area of ​​each raster within the watershed; Traverse the clipped raster data to extract the meteorological value of each pixel and its corresponding area weight; The area-weighted average calculation is performed based on the meteorological value and area weight of each pixel to generate the time series data of basin meteorological reanalysis.

3. The runoff reanalysis data error analysis method according to claim 1, characterized in that: The method of constructing a fixed effect panel regression model of meteorological input deviation and runoff simulation deviation based on the deviation between reanalysis data and observations includes the following steps: Construct a panel data set; the panel data set includes the runoff quantile deviation Q at the annual scale of each basin bias and annual scale meteorological data deviation; its expression is: Among them, P bias represents the annual precipitation deviation in the basin, P reanalysis represents the annual precipitation under the basin runoff reanalysis time series data, P obs Represents the annual precipitation under the basin observation data time series data; PET bias The PET is the annual average potential evapotranspiration deviation of the basin. reanalysis represents the annual average potential evapotranspiration under the basin runoff reanalysis time series data, PET obs represents the annual average potential evapotranspiration under the time series data of basin observation data; T bias represents the annual mean temperature deviation of the basin, T reanalysis represents the annual average temperature under the basin runoff reanalysis time series data, T obs represents the annual average temperature under the time series data of the basin observation data; Q bias represents the runoff quantile deviation at the annual scale of the basin, Q reanalysis represents the runoff quantile under the basin runoff reanalysis time series data, Q obs It represents the runoff quantile under the time series data of the basin observation data; Based on the panel data set, a fixed effect panel regression model of meteorological input deviation and runoff simulation deviation is constructed; its expression is: in, represents the systematic deviation of the runoff percentile q of catchment i at time period t, P i,t represents the precipitation deviation in basin i during time period t, PET i,t represents the average potential evapotranspiration deviation of basin i in time period t, T i,t represents the average temperature deviation of basin i in time period t; β1 and β2 are the regression coefficients of precipitation deviation and potential evapotranspiration deviation, respectively; β3 is the regression coefficient of the coupled effect of temperature and precipitation deviation; μ i is the fixed effect of basin i, γ t is the time fixed effect for time period t, ε i,t is the error term for basin i in time period t.

4. The runoff reanalysis data error analysis method according to claim 1, characterized in that: The step of resampling each cluster by the bootstrap method to form the first panel data comprises the following steps: According to the watershed attribute values ​​of each watershed, the watershed is classified using the self-organizing map clustering method to obtain the spatial clustering results of the watershed and the attribute characteristics of each cluster; The watershed subsets of each cluster are sampled with replacement, where the number of samples drawn each time is the number of watersheds in the cluster, and a panel data set of the watershed subsets is constructed to obtain the first panel data.

5. The runoff reanalysis data error analysis method according to any one of claims 1 to 4, characterized in that: The method further comprises: The basin hydrological and meteorological data set is resampled through the bootstrap method to form the second panel data, and the second panel data is subjected to fixed effect panel regression to obtain the corresponding regression coefficients, thereby generating a second coefficient distribution for measuring the uncertainty of the regression coefficients.

6. The runoff reanalysis data error analysis method according to any one of claims 1 to 4, characterized in that: The method further comprises: All watersheds are divided into three categories according to the attribute values ​​of each watershed attribute; The attributes of each river basin are resampled through the bootstrap method to form the third panel data, and the third panel data is subjected to fixed effect panel regression to obtain the corresponding regression coefficients, thereby generating a third coefficient distribution for measuring the degree of influence of each river basin attribute value on the meteorological deviation effect.

7. The runoff reanalysis data error analysis method according to claim 6, characterized in that: The method further comprises: The Mann-Whitney U test was performed on the three categories of each watershed attribute to obtain the significance analysis results of the influence of watershed attribute values ​​on the meteorological deviation effect.

8. A runoff reanalysis data error analysis system based on panel regression analysis, applying the runoff reanalysis data error analysis method according to any one of claims 1 to 7, characterized in that: include: The data processing module is used to collect the basin hydrological and meteorological data set, basin runoff reanalysis data and meteorological reanalysis data, and extract the basin runoff reanalysis time series data and basin meteorological reanalysis time series data; The first sampling module is used to cluster all watersheds according to the watershed site attribute data, obtain the spatial clustering of the watershed and the attribute characteristics of each cluster, and resample each cluster through the bootstrap method to form the first panel data; The fixed effect panel regression module is equipped with a fixed effect panel regression model of meteorological input deviation and runoff simulation deviation, which is used to perform fixed effect panel regression on panel data, obtain the corresponding regression coefficients, and generate the first coefficient distribution for measuring the spatial heterogeneity effect of meteorological deviation.

9. The runoff reanalysis data error analysis system according to claim 8, characterized in that: The system further comprises: The second sampling module is used to resample the basin hydrological and meteorological data set to form the second panel data through the bootstrap method; The fixed effect panel regression module is also used to perform a fixed effect panel regression on the second panel data to obtain corresponding regression coefficients and generate a second coefficient distribution for measuring the uncertainty of the regression coefficients.

10. The runoff reanalysis data error analysis system according to claim 8, characterized in that: The system further comprises: The third sampling module is used to divide all the watersheds into three categories according to the attribute values ​​of each watershed attribute; the attributes of each watershed are resampled by the bootstrap method to form the third panel data; The fixed effect panel regression module is also used to perform fixed effect panel regression on the third panel data to obtain corresponding regression coefficients, and generate a third coefficient distribution for measuring the degree of influence of each watershed attribute value on the meteorological deviation effect.

Images