Forecasting rainfall deviation correction method and system based on quantile mapping

Abstract

The invention relates to the technical field of rainfall forecast correction, and provides a forecast rainfall deviation correction method and system based on quantile mapping, and the method comprises the following steps: obtaining historical observation rainfall and air temperature data, historical forecast rainfall and air temperature data, and to-be-corrected future forecast rainfall data, and carrying out the preprocessing of the data, obtaining a data set with aligned time dimensions in the target space-time range; performing fitting distribution on the data set to obtain a distribution function model; distributing functions and inverse functions of different data are quantized on the basis of the distributing function model, a mapping function for mapping forecast rainfall to observation rainfall is constructed on the basis of quantile mapping, and the mapping function is used for correcting future univariate forecast rainfall data to obtain corrected univariate forecast rainfall data; and a matrix operation method is used to correct the correlation between the forecast rainfall and the air temperature data variables, and corrected multivariable forecast rainfall data is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of precipitation forecast correction, and more specifically, to a method and system for correcting forecast precipitation deviation based on quantile mapping. Background Art

[0002] As climate change research deepens, more and more impact assessment studies, such as those in hydrology, environmental science or economics, have begun to use simulation results from global climate models (GCM) and regional climate models (RCM) as input data. This trend reflects the increasing importance of climate models in scientific research, and also highlights the critical impact of the quality of climate simulation data on interdisciplinary research. However, despite the continuous scientific progress in the field of climate modeling, there are still deviations between climate simulation results and observed data. This deviation is reflected in statistical characteristics such as mean, variance, extreme values, and even the dependency structure between multiple variables and / or multiple locations, which may be significantly different from the results calculated based on historical records. The existence of this deviation highlights the inconsistency between climate model output data and actual observations, which is an important issue for impact assessment studies that rely on these data.

[0003] At present, it has been proposed to use methods such as quantile mapping method, joint probability distribution method, Bayesian joint probability model and ensemble pattern output statistics to correct forecast precipitation errors. However, due to the characteristics of forecast precipitation such as containing a large number of zero values, obeying skewed distribution, and having heteroscedastic errors, it is difficult to obtain high-accuracy correction results using only the above methods. Summary of the invention

[0004] In order to overcome the defect that it is difficult to obtain a high-accuracy forecast precipitation deviation correction result in the above-mentioned prior art, the present invention provides a forecast precipitation deviation correction method and system based on quantile mapping.

[0005] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0006] A method for correcting forecast precipitation deviation based on quantile mapping comprises the following steps:

[0007] Obtain historical observed precipitation and temperature data, historical forecast precipitation and temperature data, and future forecast precipitation data to be revised, and preprocess them to obtain a data set aligned in time dimension within the target time and space range;

[0008] Fitting distribution to the data set to obtain a distribution function model;

[0009] quantifying the distribution function and inverse function of different data based on the distribution function model, and constructing a mapping function from predicted precipitation to observed precipitation based on quantile mapping, so as to correct the future predicted precipitation data of the single variable and obtain the corrected predicted precipitation data of the single variable;

[0010] The matrix operation method is used to correct the correlation between the forecast precipitation and temperature data variables, and the revised multivariate forecast precipitation data is obtained.

[0011] Furthermore, the present invention also proposes a forecast precipitation deviation correction system based on quantile mapping, and applies the forecast precipitation deviation correction method proposed in the present invention. The system comprises:

[0012] The data acquisition module is used to obtain historical observed precipitation and temperature data, historical forecast precipitation and temperature data, and future forecast precipitation data to be revised, and preprocess them to obtain a data set aligned with the time dimension within the target time and space range;

[0013] A fitting module, used to fit the distribution of the data set to obtain a distribution function model;

[0014] A single variable correction module, used for quantifying the distribution function and its inverse function of different data based on the distribution function model, constructing a mapping function from forecast precipitation to observed precipitation based on quantile mapping, and correcting the future forecast precipitation data of the single variable to obtain the corrected single variable forecast precipitation data;

[0015] The multivariate correction module is used to correct the correlation between the forecast precipitation and temperature data variables using matrix operation methods to obtain the corrected multivariate forecast precipitation data.

[0016] Furthermore, the present invention also proposes a device, including a memory and a processor, wherein the memory stores computer-readable instructions, wherein when the computer-readable instructions are executed by the processor, the processor executes all or part of the steps of the forecast precipitation deviation correction method as described in the present invention.

[0017] Furthermore, the present invention also proposes a storage medium on which computer-readable instructions are stored, wherein the computer-readable instructions, when executed by a processor, implement all or part of the steps of the forecast precipitation deviation correction method as described in the present invention.

[0018] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0019] The present invention constructs a mapping function from predicted precipitation to observed precipitation based on quantile mapping, which is used to correct the systematic error between predicted data and observed data, thereby effectively improving the forecast accuracy. The present invention utilizes a matrix operation method to correct the correlation between variables in the predicted precipitation data, thereby realizing the correction of single-variable data and multi-variable data, and effectively improving the accuracy of the correction results of multi-variable predicted precipitation deviations. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 The present invention is a flowchart of a method for correcting forecast precipitation deviation according to an embodiment of the present invention.

[0021] Figure 2 The figure is a flow chart of a method for correcting forecast precipitation deviation according to an embodiment of the present invention.

[0022] Figure 3 It is a schematic diagram of a univariate correction result according to an embodiment of the present invention.

[0023] Figure 4 It is a schematic diagram of correlation correction results according to an embodiment of the present invention.

[0024] Figure 5 FIG. 4 is a time series diagram of univariate correction results according to an embodiment of the present invention.

[0025] Figure 6 The diagram is a structural diagram of a system for correcting forecast precipitation deviation according to an embodiment of the present invention. DETAILED DESCRIPTION

[0026] 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.

[0027] 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.

[0028] 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".

[0029] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0030] Example 1

[0031] This embodiment proposes a forecast precipitation deviation correction method based on quantile mapping, such as Figure 1 , 2 As shown, it is a flow chart of the forecast precipitation deviation correction method of this embodiment.

[0032] The forecast precipitation deviation correction method based on quantile mapping proposed in this embodiment includes the following steps:

[0033] S1. Obtain historical observed precipitation and temperature data, historical forecast precipitation and temperature data, and future forecast precipitation data to be revised, and preprocess them to obtain a data set aligned in time dimension within the target time and space range;

[0034] S2, fitting the distribution of the data set to obtain a distribution function model;

[0035] S3, quantifying the distribution function and its inverse function of different data based on the distribution function model, constructing a mapping function from the forecast precipitation to the observed precipitation based on the quantile mapping, and correcting the future forecast precipitation data of the single variable to obtain the corrected forecast precipitation data of the single variable;

[0036] S4. Using matrix operation method, the correlation between the predicted precipitation and temperature data variables is corrected to obtain the revised multivariate predicted precipitation data.

[0037] In this embodiment, a mapping function from forecast data to observed data is constructed based on quantile mapping to correct the systematic error between forecast data and observed data, thereby effectively improving the forecast accuracy. Matrix operation methods are used to correct the correlation between forecast precipitation and temperature data variables, thereby realizing the correction of single-variable and multi-variable data and effectively improving the accuracy of the correction results of multi-variable forecast precipitation deviations.

[0038] In an optional embodiment, in step S1, the historical observed precipitation data, the historical forecast precipitation data and the future forecast precipitation data to be corrected are preprocessed, including the following steps:

[0039] Dividing the historical observed precipitation and temperature data, the historical forecast precipitation and temperature data, and the future forecast precipitation data into extreme data and normal data;

[0040] According to the preset time range and longitude and latitude range, the data is clipped and aligned in the time dimension.

[0041] Exemplarily, the time period of "year-month-day" to "year-month-day" input in the form of a character string and the spatial range of ["longitude range", "latitude range"] determine the target time and space range for which forecast correction is required.

[0042] Exemplarily, methods such as quantiles, indexes or thresholds are used to divide data into extreme data and normal data.

[0043] In an optional embodiment, in step S2, fitting distribution to the data set includes the following steps: fitting distribution using one or more of empirical distribution, gamma distribution, normal distribution, lognormal distribution, and kappa distribution to construct a distribution function model.

[0044] Further optionally, before fitting the distribution to the data, simulation of historical observed precipitation distribution is also included.

[0045] The process of selecting all data according to the preset time range and longitude and latitude range is as follows:

[0046] Read all historical observation data using vector express: Among them, i represents the value of the target observation, and p is the total length of the target data matrix in time;

[0047] Read the observation information within the input time range, using vector express: Among them, n and m are the starting time point and ending time point of the target time period respectively;

[0048] Select historical observation data that is both within the target data and within the target time range, and use vector express:

[0049] Among them, historical observation data include historical observation precipitation data and historical observation temperature data.

[0050] Similarly, read the corresponding historical forecast data to obtain the forecast data within the target data and the target time range, using vector express:

[0051] Among them, historical forecast data include historical forecast precipitation data and historical forecast temperature data.

[0052] Similarly, read the future forecast precipitation data to be corrected, and obtain the forecast precipitation within the target data and the target time range, using vector express.

[0053] Among them, the target data includes all observation and forecast data used.

[0054] Further optionally, for precipitation data, the gamma distribution is selected to construct the distribution function model; for temperature data, the normal distribution is selected to construct the distribution function model; for precipitation data with a high degree of heavy tail, the Kappa distribution is selected to construct the distribution function model.

[0055] In an optional embodiment, the method of constructing a mapping function from predicted precipitation to observed precipitation based on quantile mapping includes: constructing a mapping function using one of the classical quantile mapping method, the quantile increment mapping method, the detrended quantile mapping method, the equidistant CDF matching method, and the CDF transformation method.

[0056] Further optionally, when the future forecast precipitation data is corrected based on the classical quantile mapping (QM) method, the following steps are included:

[0057] Calculate the quantile of future forecast precipitation data corresponding to the historical forecast precipitation data, map it to the historical observed precipitation data, establish the mapping relationship between quantiles, and then convert the future forecast precipitation data to be corrected into the corresponding observed data quantile; its expression is:

[0058]

[0059] in, It represents the revised future forecast precipitation data; Represents the inverse function of the distribution function of historical observed precipitation data; CDF m,h (·) represents the distribution function of historical forecast precipitation data; f t Indicates future forecast precipitation data to be revised.

[0060] The QM method is a simple statistical interpolation method used to correct the forecast data to be corrected output by the climate model to the observed data. The principle is to calculate the quantiles of the observed data and the forecast data to be corrected, establish a mapping relationship between the quantiles, and then convert the forecast data to be corrected to the corresponding quantiles of the observed data to achieve the purpose of correction or standardization. It is simple and easy to implement and better retains the overall distribution characteristics of the data.

[0061] Further optionally, when the future forecast precipitation data is corrected based on the Quantile Delta Mapping (QDM), the following steps are included:

[0062] Estimate the empirical distribution function of the forecast precipitation at time t in the future period, and calculate the position of the data point to be corrected in the future empirical distribution function:

[0063]

[0064] Then, the ratio between the observed precipitation and the forecast precipitation is calculated based on the location as the scaling factor:

[0065]

[0066] Finally, the scaling factor is combined with the traditional quantile mapping model; its expression is:

[0067]

[0068] Among them, f m,p (t) represents the future forecast precipitation data at time t; τ m,p (t) represents the position of the data point to be corrected in the future empirical distribution function; Represents the inverse function of the distribution function of historical forecast precipitation data.

[0069] When making corrections based on QDM, the future forecast precipitation bias is corrected by combining the distribution relationship between historical observation data and future forecast precipitation data, and retaining the internal trend of future forecast precipitation. While correcting the future forecast precipitation bias, the QDM method also retains the relative changes in historical and future precipitation medians. This means that if the forecast precipitation has a trend of increasing or decreasing in the future, this trend will be retained in the corrected data.

[0070] Further optionally, when the future forecast precipitation data is corrected based on the detrend quantile mapping method (Detrend Quantile Mapping, DQM), the following steps are included:

[0071] The future forecast precipitation data is detrended, and then a quantile mapping relationship is established between the detrended future forecast precipitation data and the historical observed precipitation data, and then restored to the observed distribution according to the trend; its expression is:

[0072]

[0073] in, is the mean of historical forecast precipitation data, Represents the mean of future forecast precipitation data.

[0074] Optionally, when detrending, multiplication is generally used for precipitation and temperature, and addition can be used for some interval variables. This can reduce extrapolation due to trends. If the forecast value exceeds the historical range, some form of extrapolation is required, such as using parameter distribution method or constant correction method.

[0075] Furthermore, one way to avoid frequent extrapolation is to explicitly account for changes in the forecast values, for example by first removing the model trend from the long-term mean before plotting the magnitudes, which will shift the future distribution to be within the support of the historical distribution, and then re-applying the trend.

[0076] The DQM method is a commonly used univariate post-processing method under climate change conditions, which aims to improve the shortcomings of the traditional QM method, especially considering the relative changes in the mean of model-simulated precipitation. The basic idea of ​​the DQM method is to remove the long-term trend (such as linear trend) from the forecast precipitation before quantile mapping, and then perform quantile mapping, and then restore it to the observed distribution according to the trend. Among them, through detrending processing, the DQM method can better retain the deviation of model-simulated precipitation relative to the observed value, while reducing the impact of systematic deviations (such as trend deviation) on the mapping results. Since the DQM method can more accurately reflect the relationship between model-simulated precipitation and observed precipitation, it can improve the accuracy of prediction.

[0077] Further optionally, when the future forecast precipitation data is corrected based on the equidistant CDF matching method, the following steps are included:

[0078] Select several equidistant points on the distribution function model of future forecast precipitation data, and map the cumulative probability values ​​corresponding to the equidistant points to the corresponding forecast precipitation data values, and adjust each value in the forecast precipitation data according to the mapping results; its expression is:

[0079]

[0080] Among them, CDF m,p (·) represents the distribution function of future forecast precipitation data; f m,p Represents the forecast precipitation data value.

[0081] This embodiment takes into account the ability of traditional QM methods to adjust all moments of the observed distribution, with the basic assumption that the climate distribution does not change much over time, that is, its variance and skewness in the distribution are fixed, and only the mean changes. For a given percentile, we assume that the difference between the model and the observations during training also applies to future periods, which means that the adjustment function remains unchanged and only a linear shift is required.

[0082] The equidistant CDF matching method can significantly reduce the differences between observed precipitation data and forecast precipitation due to scale and location changes, thereby more accurately comparing their distributions.

[0083] Further optionally, when the future forecast precipitation data is corrected based on the CDF transformation method (CDF-t), the following steps are included:

[0084] The distribution function of future forecast precipitation data is mapped to the distribution function of historical forecast precipitation data, and quantile mapping is performed on each variable separately to obtain the corrected forecast precipitation data; its expression is:

[0085]

[0086] Among them, CDF m,p (f m,p ) represents the CDF function of future forecast precipitation data.

[0087] The CDF-t method is mainly based on the characteristics of the cumulative distribution function (CDF). It corrects the forecast precipitation data to be revised by linearly or nonlinearly transforming the cumulative probability distribution between the historical forecast precipitation and the forecast precipitation to be revised.

[0088] Specifically, the method relates the cumulative distribution function of the variable of interest in the model simulation to the CDF of the reference data set by applying a univariate transformation function T. CDF-t assumes that the transformation equation T is still valid under climate conditions different from the calibration period, thereby generating a new CDF of the variable with bias correction for the forecast period. Subsequently, the bias-corrected data are obtained by performing a quantile-quantile mapping method between the new (reference) CDF and the CDF of the model simulation for the forecast period. This two-step process allows the potential changes in the univariate distribution between the calibration period and the forecast period to be taken into account in the correction process, and is an extension of the quantile mapping method.

[0089] Specifically, suppose that quantile mapping is performed on each variable separately:

[0090] Assume that there exists a transformation T such that the prediction f m,p can be transformed into observation h :T(CDF o,h (x))=CDF m,h (x);

[0091] The first step to complete the transformation is to estimate the transformation T:

[0092] Assume that the above formula still holds true in the future:

[0093] T(CDFo,p (f))=CDF m,p (f)

[0094]

[0095] Then in CDF s,f After (·) is estimated, the correction result can be calculated according to the following formula:

[0096]

[0097] In the calculation, the empirical estimate is Estimating CDF m,p (·).

[0098] In an optional embodiment, in step S4, the correlation between the forecast precipitation and temperature data variables is corrected using a matrix operation method, including: using the MBCn method, the TSQM method, the R 2 D 2 One of the methods, EC-BC method and other methods, adjusts the correlation between variables in the forecast precipitation data.

[0099] Further optionally, the correlation relationship is adjusted by using an MBCn (Multivariate Bias Correction with N-dimensional probability density function transform) method, including the following steps:

[0100] The marginal distribution of each variable is corrected based on quantile mapping;

[0101] In each iteration, an orthogonal rotation matrix is ​​generated, and the orthogonal rotation matrix is ​​randomly combined with each variable. The marginal distribution of the rotated variables is corrected by the quantile increment mapping method, and the adjusted forecast precipitation data is obtained based on the quantile mapping. The correlation analysis is performed on the adjusted forecast precipitation data until the similarity of the internal correlation with the historical observation data is lower than the preset threshold, then the iteration is stopped and the corrected multivariate forecast precipitation data is output.

[0102] Among them, the MBCn method is a multivariate bias correction technique based on probability density function transformation, which is used to correct multivariate biases in climate models or other complex system simulations. This method aims to simultaneously consider the dependencies and marginal distributions between multiple variables, such as correlation and covariance, to provide more accurate and comprehensive simulation results.

[0103] Specifically, the MBCn algorithm is used to construct multivariate internal correlations through random matrices, where for each iteration j, an orthogonal rotation matrix R can be generated[j] :

[0104]

[0105] The orthogonal rotation matrix can be used to obtain random combinations of initial variables. This algorithm can converge under the background of multivariate Gaussian distribution and random vector iteration. For the rotated variables, the marginal distribution is corrected by the QDM method, and then the j+1th iteration data is obtained by inverse transformation:

[0106]

[0107] After each iteration, the results can be checked for Spearman or Pearson correlation. If there is no significant difference between the internal correlation of the observed data, it can be considered that a relatively good correction result has been obtained.

[0108] Further optionally, a TSQM (Two-Stage Quantile Mapping) method is used to adjust the correlation relationship, including the following steps:

[0109] The historical forecast precipitation data and the revised univariate forecast precipitation data are constructed as m×n matrices respectively. and Calculating the Matrix and The rank of each column in is calculated, and its normal score is calculated, that is, its van der Waerden scores matrix is ​​normalized, and the test statistic is calculated based on this; its expression is:

[0110] [E m,h ]=[Φ -1 {i / (n+1)}]

[0111] Among them, W m,h represents the normalized historical forecast matrix; Φ represents the CDF function of the standard normal distribution; i represents The sorting size of each element in this column; n represents the number of elements in each column;

[0112] Construct a correlation coefficient covariance matrix, and calculate a correction matrix based on the correlation coefficient covariance matrix. and After rearrangement, the corrected multivariate forecast precipitation data is obtained; its expression is:

[0113]

[0114] in, and represents the van der Waalden score matrix of the corrected sorted historical and future forecasts; [W m,h ] and [W m,p ] represents the van der Waalden score matrix of historical forecasts and future forecasts before correction sorting; [C m,h ] and [C o,f ] are the correlation coefficient covariance matrices of historical forecasts and historical observations respectively; represents the inverse matrix of the Cholesky decomposition result of the correlation coefficient matrix of precipitation and temperature in historical forecasts; [P o,h ] represents the Cholesky decomposition result of the correlation coefficient matrix of precipitation and temperature in historical observations; Represents the inverse matrix of the Cholesky decomposition result of the correlation coefficient matrix between precipitation and temperature in future forecasts.

[0115] In the TSQM method adopted in this embodiment, the forecast is first corrected by a single variable method, and then the internal relationship is corrected by shuffle to improve the prediction accuracy of future climate change.

[0116] Alternatively, R 2 D 2 The Rank Resampling For Distributions and Dependences method is used to adjust the correlation relationship, including the following steps:

[0117] The marginal distribution of each climate variable is adjusted based on the CDF-t method:

[0118]

[0119] Calculate the ranks of all dimensions in the revised future forecast precipitation data and historical observed precipitation data at time t; its expression is:

[0120]

[0121] and [R o,h ] represent the rank sets of the revised future forecast precipitation data and historical observed precipitation data, respectively. Represents the revised future forecast precipitation data matrix; Represents the historical observed precipitation data matrix; d = 1, 2, ..., D, where D is the total number of dimensions;

[0122] For any dimension, select a dimension d that does not change before and after correction as the reference dimension, and for each predicted time t, find a time t * Make Get the reconstruction matrix:

[0123]

[0124] Among them, R o,h (t * ) represents the matrix observation rank set matrix after adjusting the order; represents time t in dimension d * The sorting of historical observations in ; represents the order of future forecasts at time t in dimension d; the reconstruction matrix R o,h (t * ) represents the rank matrix of the precipitation-temperature relationship during the introduced observation period;

[0125] The future forecast precipitation data is rearranged according to the reconstruction matrix to obtain the revised multivariate forecast precipitation data.

[0126] Among them, R 2 D 2 The method is a multivariate bias correction method for high-dimensional climate simulations that aims to adjust not only the distribution of a single variable but also the statistical dependencies between different variables and locations, thereby providing a more comprehensive and accurate correction. 2 D 2 The method is based on the Schaake Shuffle reordering technique, which reorders the samples so that their rank structure corresponds to the rank structure of the reference samples, thereby reconstructing the multivariate dependency structure.

[0127] Further optionally, the correlation relationship is adjusted by using an EC-BC (Empirical Copula-Bias Correction) method, including the following steps:

[0128] The Schaake Shuffle method is used to reconstruct the relationship between variables. The data after univariate bias correction are reordered to make their time series the same as the time series of historical forecast precipitation data, and the corrected multivariate forecast precipitation data are obtained.

[0129] The principle of the EC-BC method is as follows: First, a forecast correction is performed on each variable. Second, the univariate bias-corrected data are reordered so that their time series are the same as the time series of the reference sample. This univariate bias correction for each variable allows us to reproduce the time, location and variable dependencies of the reference data.

[0130] Furthermore, in an optional embodiment, the following steps are also included: drawing spatial maps and diagnostic maps based on the revised multivariate forecast precipitation data and visualizing them.

[0131] Example 2

[0132] This embodiment applies the forecast precipitation deviation correction method proposed in Embodiment 1 to the open source Python language platform to form a forecast precipitation deviation correction method for NC files. The steps include:

[0133] (1) The Dataset function of the Python open source third-party library netcdf and the load_mfdataset of xarray were used to read the precipitation data and temperature data in the .nc file to extract the corresponding precipitation measurement data and spatiotemporal dimensions (time, longitude and latitude) and other information.

[0134] (2) Determine the spatiotemporal area where forecast correction is required based on the time period of "year-month-day" to "year-month-day" input in the form of a character string and the spatial range of ["longitude range", "latitude range"].

[0135] (3) The Python third-party libraries xarray and dask were used for spatiotemporal analysis. The main contents are as follows: the dimensional information of precipitation and temperature was extracted, and the spatial dimension of precipitation and temperature data was clipped according to the longitude and latitude of the selected spatial area; the temporal dimension of precipitation and temperature data was clipped according to the input time range; and the precipitation and temperature data were aligned.

[0136] (4) Using Python’s numpy and scipy libraries, the precipitation data and temperature data are segmented and fitted into specified statistical distributions.

[0137] (5) Using the numpy and scipy libraries, the CDF function and its inverse function of different data are quantified, and then a function mapping from forecast to observation is formed according to the mapping relationship between the specified variable distributions, and finally a corrected univariate forecast is formed.

[0138] (6) Using the numpy-based matrix operation method, the correlation between variables is corrected to obtain the revised multivariate forecast precipitation data.

[0139] (7) Use Python third-party libraries geopandas and cartopy to visualize the attribution result data.

[0140] When drawing a spatial map, use the plt.subplots function in Matplotlib to set the appropriate sub-canvas size; use the ccrs.PlateCarree() function in cartopy to call the geoaxes object of matplotlib to control its projection type; use the ax.pcolormesh method to draw the spatial map. Use the fig.add_axes method to add a sub-map to control the colorbar position, use the fig.colorbar method to set the colorbar position, use the cbar.set_ticks method to set the tick position, and use the tick label to display the content. Finally, use the savefig method of matplotlib.figure.Figure to save the drawn spatial map to the specified path.

[0141] Exemplarily, the forecast precipitation deviation correction method proposed in Example 1 is applied to the correction of precipitation and temperature of the CMIP6 model in the Yangtze River Basin from 1901 to 2020.

[0142] Specifically, first store the precipitation observation data (NetCDF file) and temperature observation data (NetCDF file) required for drawing in the selected folder, and define the variables path_precipitation and path_temperature respectively and store the corresponding paths, and read and merge these files through xarray.open_mfdataset and geopandas.read_file functions.

[0143] Input the target time and space range in the form of a string, cut the time and space range through Python third-party libraries xarray and numpy, and use Python's numpy and scipy libraries to correct the precipitation and temperature data:

[0144] 1) Use xarray.dataset to extract precipitation data and observation data. Use the broadcast interface of numpy function in xarray to filter out precipitation and observation data in set S through .sel method;

[0145] 2) Use numpy and scipy methods to divide the data into extreme events and non-extreme events according to the specified method (quantile, index or threshold);

[0146] 3) Extract the trimmed temperature and precipitation data and correct them:

[0147] First, the distribution_transform method is used to correct the relationship between variables based on the selected model type and the fitted distribution parameters. Secondly, the relationship between variables is corrected by the method based on N-pdf or shuffle. According to the above two results, the correction effect is quantified, and the following is obtained: Figure 3 , 4 The schematic diagram of the single variable correction result and the schematic diagram of the correlation correction result are shown;

[0148] 4) Use the apply_func broadcast method in xarray and the for loop in python to repeat the above process for data in different spatial locations to obtain data for visualization.

[0149] Furthermore, the Python third-party library Mpl_toolkits and Matplotlib are used to draw spatial maps and diagnostic maps:

[0150] 1) Set the map projection mode to PlateCarree through the cartopy.crs.ccrs method of cartopy;

[0151] 2) Use the plt.subplots function in the third-party visualization library Matplotlib to set the canvas size to be drawn, and use the cartopy.set_extent method to set the longitude and latitude space area to be drawn;

[0152] 3) Use the ax.pcolormesh method to draw the spatial map corresponding to the classification;

[0153] 4) Use the fig.add_axes method to add a sub-figure to control the colorbar position, use the cbar.set_ticks method to set the tick position, and use the tick labels to display the content;

[0154] 5) Use the savefig method of matplotlib.figure.Figure to save the drawn precipitation data visualization graph to the specified path, such as Figure 5 As shown, this is the time series diagram of the univariate correction results.

[0155] Example 3

[0156] This embodiment applies the forecast precipitation deviation correction method proposed in Embodiment 1, and proposes a forecast precipitation deviation correction system based on quantile mapping, such as Figure 6 As shown, it is the architecture diagram of the forecast precipitation deviation correction system based on quantile mapping of this embodiment.

[0157] The forecast precipitation deviation correction system based on quantile mapping proposed in this embodiment includes:

[0158] The data acquisition module is used to obtain historical observed precipitation and temperature data, historical forecast precipitation and temperature data, and future forecast precipitation data to be revised, and preprocess them to obtain a data set aligned with the time dimension within the target time and space range;

[0159] A fitting module, used to fit the distribution of the data set to obtain a distribution function model;

[0160] A single variable correction module, used for quantifying the distribution function and its inverse function of different data based on the distribution function model, constructing a mapping function from forecast precipitation to observed precipitation based on quantile mapping, and correcting the future forecast precipitation data of the single variable to obtain the corrected single variable forecast precipitation data;

[0161] The multivariate correction module is used to correct the correlation between the forecast precipitation and temperature data variables using matrix operation methods to obtain the corrected multivariate forecast precipitation data.

[0162] 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 will not be described repeatedly here.

[0163] Example 4

[0164] 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 forecast precipitation deviation correction method proposed in Example 1.

[0165] Example 5

[0166] This embodiment proposes a storage medium on which computer-readable instructions are stored, wherein the computer-readable instructions, when executed by a processor, implement all or part of the steps of the forecast precipitation deviation correction method proposed in Example 1.

[0167] 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.

[0168] Exemplarily, the instructions, programs, code sets or instruction sets may be implemented using conventional programming languages.

[0169] 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 forecast precipitation deviation correction method described in Example 1.

[0170] 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.

[0171] 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 method for correcting forecast precipitation deviation based on quantile mapping, characterized in that: The following steps are involved: Obtain historical observed precipitation and temperature data, historical forecast precipitation and temperature data, and future forecast precipitation data to be revised, and preprocess them to obtain a data set aligned in time dimension within the target time and space range; Fitting distribution to the data set to obtain a distribution function model; quantifying the distribution function and inverse function of different data based on the distribution function model, and constructing a mapping function from predicted precipitation to observed precipitation based on quantile mapping, so as to correct the future predicted precipitation data of the single variable and obtain the corrected predicted precipitation data of the single variable; The matrix operation method is used to correct the correlation between the forecast precipitation and temperature data variables, and the revised multivariate forecast precipitation data is obtained.

2. The method for correcting the forecast precipitation deviation according to claim 1, characterized in that: Preprocessing the historical observed precipitation and temperature data, the historical forecast precipitation and temperature data, and the future forecast precipitation data to be revised includes the following steps: Dividing the historical observed precipitation and temperature data, the historical forecast precipitation and temperature data, and the future forecast precipitation data into extreme data and normal data; According to the preset time range and longitude and latitude range, the data is clipped and aligned in the time dimension.

3. The method for correcting the forecast precipitation deviation according to claim 1, characterized in that: Fitting the distribution of the data set comprises the following steps: One or more of the empirical distribution, gamma distribution, normal distribution, lognormal distribution, and Kappa distribution are used to fit the distribution and construct a distribution function model.

4. The method for correcting the forecast precipitation deviation according to any one of claims 1 to 3, characterized in that: The method of constructing a mapping function from predicted precipitation to observed precipitation based on quantile mapping includes: The mapping function is constructed by using one of the classic quantile mapping method, quantile increment mapping method, detrended quantile mapping method, equidistant CDF matching method and CDF transformation method.

5. The method for correcting the forecast precipitation deviation according to claim 4, characterized in that: The step of constructing a mapping function from the forecast precipitation to the observed precipitation based on the quantile mapping to obtain the corrected univariate forecast precipitation data comprises any of the following steps: (1) Correcting the future forecast precipitation data based on the classical quantile mapping method, including: Calculate the quantile of future forecast precipitation data corresponding to the historical forecast precipitation data, map it to the historical observed precipitation data, establish the mapping relationship between quantiles, and then convert the future forecast precipitation data to be corrected into the corresponding observed data quantile; its expression is: in, It represents the revised future forecast precipitation data; Represents the inverse function of the distribution function of historical observed precipitation data; CDF m,h (·) represents the distribution function of historical forecast precipitation data; f t Indicates the future forecast precipitation data to be revised; (2) Correcting the future forecast precipitation data based on the quantile increment mapping method, including: Estimate the empirical distribution function of the forecast precipitation at time t in the future period, calculate the position of the data point to be corrected in the future empirical distribution function, and then calculate the scaling factor between the observed precipitation and the forecast precipitation based on the position, and finally combine the scaling factor with the traditional quantile mapping model; its expression is: Among them, f m,p (t) represents the future forecast precipitation data at time t; τ m,p (t) represents the position of the data point to be corrected in the future empirical distribution function; Represents the inverse function of the distribution function of historical forecast precipitation data; (3) Correcting the future forecast precipitation data based on the detrended quantile mapping method, including: The future forecast precipitation data is detrended, and then a quantile mapping relationship is established between the detrended future forecast precipitation data and the historical observed precipitation data, and then restored to the observed distribution according to the trend; its expression is: in, is the mean of historical forecast precipitation data, Represents the mean of future forecast precipitation data; (4) Correcting the future forecast precipitation data based on the equidistant CDF matching method, including: Select several equidistant points on the distribution function model of future forecast precipitation data, and map the cumulative probability values ​​corresponding to the equidistant points to the corresponding forecast precipitation data values, and adjust each value in the forecast precipitation data according to the mapping results; its expression is: Among them, CDF m,p (·) represents the distribution function of future forecast precipitation data; f m,p Indicates the forecast precipitation data value; (5) Correcting the future forecast precipitation data based on the CDF transformation method, including: The distribution function of future forecast precipitation data is mapped to the distribution function of historical forecast precipitation data, and quantile mapping is performed on each variable separately to obtain the corrected forecast precipitation data; its expression is: Among them, CDF m,p (f m,p ) represents the CDF function of future forecast precipitation data.

6. The method for correcting the forecast precipitation deviation according to any one of claims 1 to 3, characterized in that: The matrix operation method is used to correct the correlation between the forecast precipitation and temperature data variables, including: Using MBCn method, TSQM method, R 2 D 2 One of the methods, EC-BC method and other methods, adjusts the correlation between variables in the forecast precipitation data.

7. The method for correcting the forecast precipitation deviation according to claim 6, characterized in that: The method of using a matrix operation method to correct the correlation between the forecast precipitation and temperature data variables includes any of the following steps: (1) Adopting the MBCn method to adjust the correlation relationship, including: The marginal distribution of each variable is corrected based on quantile mapping; In each iteration, an orthogonal rotation matrix is ​​generated, the orthogonal rotation matrix is ​​randomly combined with each variable, the marginal distribution of the rotated variables is corrected by the quantile increment mapping method, and the adjusted forecast precipitation data is obtained based on the quantile mapping; the correlation analysis is performed on the adjusted forecast precipitation data until the similarity of the internal correlation with the historical observation data is lower than the preset threshold, the iteration is stopped and the corrected multivariate forecast precipitation data is output; (2) Use the TSQM method to adjust the correlation relationship, including: The historical forecast precipitation data and the revised univariate forecast precipitation data are constructed as m×n matrices respectively. and Calculating the Matrix and The rank of each column in is normalized to obtain its van der Waalden score matrix; its expression is: [W m,h ]=[Φ -1 {i / (n+1)}] Among them, W m,h represents the normalized historical forecast matrix; Φ represents the CDF function of the standard normal distribution; i represents the matrix The sorting size of each element in this column; b represents the number of elements in each column; Construct a correlation coefficient covariance matrix, and calculate a correction matrix based on the correlation coefficient covariance matrix. and After rearrangement, the corrected multivariate forecast precipitation data is obtained; its expression is: in, and represents the van der Waalden score matrix of the corrected sorted historical and future forecasts; [W m,h ] and [W m,p ] represents the van der Waalden score matrix of historical forecasts and future forecasts before correction sorting; [C m,h ] and [C o,f ] are the correlation coefficient covariance matrices of historical forecasts and historical observations respectively; represents the inverse matrix of the Cholesky decomposition result of the correlation coefficient matrix of precipitation and temperature in historical forecasts; [P o,h ] represents the Cholesky decomposition result of the correlation coefficient matrix of precipitation and temperature in historical observations; The inverse matrix represents the Cholesky decomposition result of the correlation coefficient matrix of precipitation and temperature in future forecasts; (3) Using R 2 D 2 Methods for correlation adjustment include: The marginal distribution of each climate variable is adjusted based on the CDF-t method; Calculate the ranks of all dimensions in the revised future forecast precipitation data and historical observed precipitation data at time t; its expression is: in, and [R o,h ] represent the rank sets of the revised future forecast precipitation data and historical observed precipitation data, respectively. Represents the revised future forecast precipitation data matrix; Represents the historical observed precipitation data matrix; d = 1, 2, ..., D, where D is the total number of dimensions; For any dimension, select a dimension d that does not change before and after correction as the reference dimension, and for each predicted time t, find a time t * Make Get the reconstruction matrix: Among them, R o,h (t * ) represents the matrix observation rank set matrix after adjusting the order; represents time t in dimension d * The sorting of historical observations; represents the order of future forecasts at time t in dimension d; the reconstruction matrix R o,h (t * ) represents the rank matrix of the precipitation-temperature relationship during the introduced observation period; Rearranging the future forecast precipitation data according to the reconstruction matrix to obtain the revised multivariate forecast precipitation data; (4) Adopting the EC-BC method to adjust the correlation relationship, including: The Schaake Shuffle method is used to reconstruct the relationship between variables. The data after univariate bias correction are reordered to make their time series the same as the time series of historical forecast precipitation data, and the corrected multivariate forecast precipitation data are obtained.

8. A precipitation forecast deviation correction system based on quantile mapping, using the precipitation forecast deviation correction method according to any one of claims 1 to 7, characterized in that: include: The data acquisition module is used to obtain historical observed precipitation and temperature data, historical forecast precipitation and temperature data, and future forecast precipitation data to be revised, and preprocess them to obtain a data set aligned with the time dimension within the target time and space range; A fitting module, used to fit the distribution of the data set to obtain a distribution function model; A single variable correction module, used for quantifying the distribution function and its inverse function of different data based on the distribution function model, constructing a mapping function from forecast precipitation to observed precipitation based on quantile mapping, and correcting the future forecast precipitation data of the single variable to obtain the corrected single variable forecast precipitation data; The multivariate correction module is used to correct the correlation between the forecast precipitation and temperature data variables using matrix operation methods to obtain the corrected multivariate forecast precipitation data.

9. A device comprising a memory and a processor, wherein the memory stores computer-readable instructions, characterized in that: When the computer-readable instructions are executed by the processor, the processor executes all or part of the steps of the forecast precipitation deviation correction method according to any one of claims 1 to 7.

10. A storage medium having computer-readable instructions stored thereon, characterized in that: When the computer-readable instructions are executed by a processor, all or part of the steps of the forecast precipitation deviation correction method according to any one of claims 1 to 7 are implemented.

Images