CN112285808B - Method for reducing scale of APHRODITE precipitation data - Google Patents

Method for reducing scale of APHRODITE precipitation data Download PDF

Info

Publication number
CN112285808B
CN112285808B CN202011169231.7A CN202011169231A CN112285808B CN 112285808 B CN112285808 B CN 112285808B CN 202011169231 A CN202011169231 A CN 202011169231A CN 112285808 B CN112285808 B CN 112285808B
Authority
CN
China
Prior art keywords
resolution
data set
aphrodite
gpm
imerg
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202011169231.7A
Other languages
Chinese (zh)
Other versions
CN112285808A (en
Inventor
闵肖肖
许金涛
史舟
李丹璐
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN202011169231.7A priority Critical patent/CN112285808B/en
Publication of CN112285808A publication Critical patent/CN112285808A/en
Application granted granted Critical
Publication of CN112285808B publication Critical patent/CN112285808B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01W—METEOROLOGY
    • G01W1/00—Meteorology
    • G01W1/10—Devices for predicting weather conditions
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01W—METEOROLOGY
    • G01W1/00—Meteorology
    • G01W1/14—Rainfall or precipitation gauges
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting

Abstract

The invention discloses a method for downscaling APHRODITE precipitation data. The method comprises the following steps: acquiring a GPM-IMERG data set, an SM2RAIN-ASCAT data set and an APHRODITE daily precipitation data set which are used for describing the same object; acquiring an environmental factor auxiliary data set which is the same as the three data set description objects; respectively matching the four data sets based on different observation dimensions, and performing total amount control processing on the APHRODITE data set; based on the space dimension, taking the low-resolution GPM-IMERG data set, the SM2RAIN-ASCAT data set and the auxiliary data set obtained from S22 as independent variables, and taking the APHRODITE data set obtained from S24 as dependent variables, and establishing a regression model; and (3) inputting the high-resolution GPM-IMERG data set obtained from S21, the high-resolution SM2RAIN-ASCAT data set obtained from S23 and the high-resolution auxiliary data set obtained from S22 as input variables into a regression model based on the spatial dimension to obtain a high-resolution APHRODITE daily precipitation data set. The precipitation data obtained by the method has high result precision and quality, and has certain theoretical and practical significance and popularization and application values.

Description

Method for reducing scale of APHRODITE precipitation data
Technical Field
The invention relates to a downscaling algorithm for meteorological satellite observation of precipitation data and ground variable inversion of precipitation data, in particular to a downscaling method for APHRODIE precipitation data.
Background
Precipitation is an important component of global material and energy circulation, plays an important role in earth surface models and weather prediction models, influences regional weather changes and global climate formation, and is a back driving force of hydrological processes such as floods, storms and the like. High-precision precipitation data plays an important role in the fields of climate, weather, hydrology, agriculture and the like. The ground monitoring station is the most direct acquisition means and source of rainfall information, can accurately represent the rainfall amount of the peripheral area of the station, but cannot accurately describe the large-area rainfall distribution condition, and in addition, the ground station cannot cover oceans, terrain complex areas and unmanned areas due to the limitation of natural environment and economic conditions, so that the acquisition and the use of rainfall data are limited. The satellite remote sensing has the characteristics of wide coverage space range and high observation time-space frequency, can well make up the defects of sparse sites, uneven distribution and the like, and gradually becomes an important data source for rainfall monitoring.
A Global Precipitation planning observation satellite (GPM) is provided with a radar and microwave radiometer system, rainfall information of a lower layer is monitored from an upper layer space based on a Top-down strategy, the observation range is wide, the space-time resolution is high, the produced IMERG Precipitation product has good correlation with ground observation data, the spatial distribution rule of Precipitation can be well reflected, the performance is poor in certain areas due to indirect observation and limited by a built-in model and an algorithm, high-intensity rainfall events are difficult to capture, the product quality still has a great progress space, and the observation precision still needs to be improved. The SM2RAIN-ASCAT product is a precipitation product derived by applying the SM2RAIN algorithm to ASCAT (the Advanced SCATTEROMER) satellite inverted soil moisture data. The product is based on a bottom-up strategy, precipitation information in the atmosphere is reversely deduced by soil moisture, and the precision of the product is limited due to certain errors and defects of soil moisture in satellite inversion, so that the product still needs to be further improved in practical application. The APHRODITE is high-quality long-time sequence precipitation data suitable for Asian areas, is formed by dense ground station interpolation, is high in data precision and reliability, but is relatively low in spatial resolution and cannot meet the small-scale application requirements of the areas.
The APHRODITE high-precision low-spatial-resolution ground observation precipitation data is subjected to scale reduction operation by using GPM-IMERG meteorological satellite observation precipitation data with high spatial resolution and low precision and SM2RAIN-ASCAT soil moisture inversion precipitation data, the advantages of three types of data can be fully fused, and the method is an effective method for obtaining the precipitation data with high precision and high spatial resolution.
Disclosure of Invention
The invention aims to solve the problems in the prior art, and provides a downscaling method of APHRODITE precipitation data based on GPM-IMERG satellite observation precipitation data with high spatial resolution and SM2RIAN-ASCAT soil moisture inversion precipitation data of an SVR-TB model.
The specific technical scheme of the invention is as follows:
a method for downscaling APHRODITE precipitation data comprises the following specific steps:
s1: acquiring a satellite remote sensing precipitation data set GPM-IMERG, a ground variable inversion precipitation data set SM2RAIN-ASCAT and a ground observation daily precipitation data set APHRODITE for describing the same object; acquiring an auxiliary data set of environmental factors which are the same as the three data set description objects, wherein the spatial resolution of the auxiliary data set is higher than that of APHRODITE, and the auxiliary data set comprises an SRTM DEM digital elevation model, a Slope, a Slope Aspect, a longitude Lon and a latitude Lat;
s2: respectively matching the four data sets based on different observation dimensions, and performing total amount control processing on the APHRODITE, wherein the following steps are specifically performed in sequence:
s21: processing the resolution of the GPM-IMERG to be the same as the resolution of the APHRODITE by a data matching method based on the time dimension;
s22: based on spatial dimension, aggregating the resolutions of GPM-IMERG and SM2RAIN-ASCAT to be the same as that of APHRODITE by a data matching method to obtain low-resolution GPM-IMERG and SM2 RAIN-ASCAT; aggregating the resolution of the auxiliary data set to be the same as the resolution of the GPM-IMERG to obtain a high-resolution auxiliary data set;
s23: based on the spatial dimension, the resolution of the SM2RAIN-ASCAT is degraded to be the same as that of the GPM-IMERG by a data matching method, and the SM2RAIN-ASCAT with high resolution is obtained;
s24: selecting nearby adjacent grid pixels for assignment of null pixels in the APHRODITE appearing in the sea-land boundary area to obtain APHRODITE daily precipitation data after total amount control processing;
s3: based on the space dimension, taking the GPM-IMERG, SM2RAIN-ASCAT and the auxiliary data set with low resolution obtained in S22 as independent variables, and taking APHRODITE day-by-day rainfall data after total amount control processing in S24 as dependent variables to establish a regression model;
s4: and on the basis of the spatial dimension, inputting the high-resolution SM2RAIN-ASCAT obtained in S23, the GPM-IMERG obtained in S21 and the high-resolution auxiliary data set obtained in S22 as input variables into the regression model to obtain the high-resolution APHRODITE.
Preferably, the GPM-IMERG in S1 has a temporal resolution and a spatial resolution of 30min and 0.1 DEG, the SM2RAIN-ASCAT has a temporal resolution and a spatial resolution of 1 day and 12.5km, and the APHRODITE has a temporal resolution and a spatial resolution of 1 day and 0.25 DEG, respectively; the auxiliary data sets all belong to static environment factor data sets, and the spatial resolution is 90 m.
Preferably, the longitudes Lon and latitudes Lat of the auxiliary data set in S1 are obtained by calculating central longitude values and latitude values of grid pixels in the SRTM DEM digital elevation model, respectively.
Preferably, the data matching method in S21 is an accumulation method, which specifically includes:
and determining the number of GPM-IMERGs contained in the APHRODITE based on the time dimension, and accumulating the GPM-IMERGs successively according to the number to obtain the GPM-IMERG with the same resolution as the APHRODITE.
Preferably, the data matching method in S22 is a moving window method, which specifically includes:
respectively processing the GPM-IMERG, the SM2RAIN-ASCAT and the auxiliary data set, establishing a moving window of a by taking each pixel in the data set to be aggregated as a center, respectively calculating the covered area proportion of each pixel in a low-resolution grid mesh aggregated under the coverage of the moving window, and taking the pixel with the largest covered proportion as the center pixel of the moving window; taking the average pixel value of a x a data set pixels forming the moving window as the pixel value of the central pixel, and assigning values to all pixels in the low-resolution grid network obtained by aggregation; where a is the integer value resulting from dividing the resolution value of the low resolution grid mesh by the original resolution value of the data set and rounding up.
Preferably, the data matching method in S23 is a nearest neighbor interpolation method, which includes:
and respectively reassigning each pixel element in the GPM-IMERG, the SM2RAIN-ASCAT and the auxiliary data set according to the adjacent pixel element in the APHRODIE based on the spatial dimension.
Preferably, the specific method of S3 is as follows:
based on the spatial dimension, the low-resolution GPM-IMERG, SM2RAIN-ASCAT and the auxiliary data set obtained in S22 are used as independent variables, APHRODITE day-by-day rainfall data after the total amount control processing in S24 is used as dependent variables, and a regression model based on SVR-TB is established for each grid point, specifically as follows:
randomly generating a training set and a testing set, wherein the sample number ratio of the training set to the testing set is 4: 1; training set and test set are processed to [ -1,1 ] respectively]The interval is normalized, and the processing principle is that y is 2 (x-x)min)/(xmax-xmin) -1; wherein y is the normalized data, x is the data to be normalized, x ismaxAnd xminThe maximum and minimum values of x, respectively; if the data in x are all the same, i.e. xmax＝xminIf so, the data y is not changed;
setting a cross validation mode of a training set as ten-fold cross validation, setting a built-in type of the model as epsilon-SVR, and setting a kernel Function type of the model as Radial Basis Function (RBF); searching for an optimal penalty coefficient c and a kernel parameter g by using a cross validation grid search method, setting the search ranges of the c and g parameters to be [ -10,10 ], setting the step length to be 0.5, constructing and solving an optimal problem, and obtaining the optimal c and g parameters; the hyper-parameters of the obtained model are adjusted through the feedback of the test set, so that the overfitting of the model is avoided; and according to the obtained optimal c and g parameters, constructing a regression model for each grid point in the APHRODITE daily precipitation data after the total amount control processing in S24.
Compared with the prior art, the invention has the following beneficial effects:
the invention provides a method for reducing the scale of a ground observation precipitation data APHRODITE day-by-day precipitation data set based on a satellite remote sensing precipitation data GPM-IMERG data set and a ground variable inversion precipitation data SM2RAIN-ASCAT data set with high spatial resolution, and finally obtains a precipitation data result with high spatial resolution and high quality, wherein the time resolution is 1 day, and the spatial resolution is 0.1 degrees multiplied by 0.1 degrees. After the method disclosed by the invention is used for reducing the scale of the APHRODITE daily precipitation data set, the obtained precipitation data result is high in precision and quality, and certain theoretical and practical significance and popularization and application values are achieved.
Drawings
Fig. 1 is a box-type diagram of a false alarm index (a) and a critical success index (b) of precipitation data in zhejiang province in 2015, wherein an original APH and a Downscaled APH are verification results of APHRODITE precipitation data relative to a ground station before and after the scale reduction respectively;
fig. 2 shows the behavior of original APHRODITE precipitation data (a) and downscaled APHRODITE precipitation data (b) in the "bright" typhoon with super strong in 2015;
fig. 3 shows interpolation (a) of ground station precipitation data, original APHRODITE precipitation data (b) and downscaled APHRODITE precipitation data (c) in the "bright" super typhoon in 2015.
Detailed Description
The invention will be further elucidated and described with reference to the drawings and the detailed description. The technical features of the embodiments of the present invention can be combined correspondingly without mutual conflict.
The invention provides a downscaling method of APHRODITE precipitation data, which comprises the following specific steps:
s1: the method comprises the steps of obtaining a GPM-IMERG data set of satellite remote sensing precipitation data based on a 'Top-down' strategy, a SM2RAIN-ASCAT data set of ground variable inversion precipitation data based on a 'Bottom-up' strategy and an APHRODITE daily precipitation data set of ground observation precipitation data, wherein the GPM-IMERG data set is used for describing the same object (namely the same research area). Wherein the accuracy of the APHRODITE dataset is higher than that of the GPM-IMERG dataset and the SM2RAIN-ASCAT dataset, the spatial resolution is lower than that of the two datasets, the spatial resolution of the SM2RAIN-ASCAT dataset is lower than that of the GPM-IMERG dataset, and the temporal resolution of the GPM-IMERG dataset is higher than that of the other precipitation datasets; and acquiring an environment factor auxiliary data set which is the same as the three data set description objects, wherein the spatial resolution of the auxiliary data set is higher than that of the GPM-IMERG data set, and the auxiliary data set comprises an SRTM DEM digital elevation model, a Slope, a Slope Aspect, a longitude Lon and a latitude Lat. The longitude Lon and the latitude Lat are respectively obtained by calculating the central longitude value and the latitude value of the grid pixel in the SRTM DEM digital elevation model.
Wherein, the time resolution and the spatial resolution of the GPM-IMERG data set are respectively 30min and 0.1 degree multiplied by 0.1 degree, the time resolution and the spatial resolution of the SM2RAIN-ASCAT data set are respectively 1 day and 12.5km, and the time resolution and the spatial resolution of the APHRODITE daily precipitation data set are respectively 1 day and 0.25 degree multiplied by 0.25 degree. The auxiliary data sets all belong to static environment factor data sets, and the spatial resolution is 90 m.
S2: respectively matching four data sets of GPM-IMERG, SM2RAIN-ASCAT, APHRODITE and an auxiliary data set based on different observation dimensions, and performing total amount control processing on the APHRODITE, wherein the total amount control processing method comprises the following specific steps:
s21: and on the basis of the time dimension, processing the resolution of the GPM-IMERG data set to be the same as that of the APHRODITE daily precipitation data set by a data matching method. The time resolution of the GPM-IMERG precipitation data is processed to be 1 day, and the time resolution of the GPM-IMERG, the SM2RAIN-ASCAT and the APHRODITE is 1 day at the moment.
The data matching method in this step is an accumulative method, which specifically comprises the following steps:
and determining the number of GPM-IMERG data sets contained in the APHRODITE daily precipitation data set based on the time dimension, and accumulating the GPM-IMERG data sets successively according to the number to obtain the GPM-IMERG data set with the same resolution as the APHRODITE daily precipitation data set. The daily resolution of the GPM-IMERG data is generated by adding all the pixel data of the day after removing null values.
S22: on the basis of S21, based on spatial dimension, aggregating the resolutions of the GPM-IMERG data set, the SM2RAIN-ASCAT data set and the auxiliary data set to be the same as the resolution of the APHRODITE daily precipitation data set through a data matching method, and obtaining a GPM-IMERG data set, an SM2RAIN-ASCAT data set and an auxiliary data set with low resolution; and aggregating the resolution of the auxiliary data set to be the same as that of the GPM-IMERG data set to obtain the high-resolution auxiliary data set. I.e. the resolution of the GPM-IMERG dataset, the SM2RAIN-ASCAT dataset and the auxiliary dataset are aggregated to 0.25 ° (low spatial resolution) while the resolution of the auxiliary dataset is aggregated to 0.1 ° (high spatial resolution).
The data matching method in this step is a moving window method, which specifically comprises the following steps:
establishing a moving window of a x a by taking each pixel in a data set to be aggregated as a center, respectively calculating the covered area proportion of each pixel in a low-resolution grid mesh obtained by aggregation under the coverage of the moving window, and taking the pixel with the largest covered proportion as the center pixel of the moving window; taking the average pixel value of a x a data set pixels forming the moving window as the pixel value of the central pixel, and assigning values to all pixels in the low-resolution grid network obtained by aggregation; where a is the integer value resulting from dividing the resolution value of the low resolution grid mesh by the original resolution value of the data set and rounding up. The moving window sizes used for aggregating 90m auxiliary data to 0.1 ° and 0.25 ° are 112 × 112(a ═ 112) and 278 × 278(a ═ 278), respectively, and the moving window size used for aggregating 0.1 ° GPM-IMERG data to 0.25 ° is 3 × 3(a ═ 3); the moving window size used to aggregate 12.5km of SM2RAIN-ASCAT data to 0.25 ° was 2 × 2(a ═ 2).
S23: based on the spatial dimension, the resolution of the SM2RAIN-ASCAT data set is degraded to be the same as that of the GPM-IMERG data set through a data matching method, and the SM2RAIN-ASCAT data set with high resolution is obtained.
The data matching method in this step is a nearest neighbor interpolation method, which specifically includes the following steps:
based on the spatial dimension, assigning the value of the SM2RAIN-ASCAT image element nearest to a certain image element in the 0.1-degree grid to the image element;
s24: and on the basis of S21, selecting nearby adjacent grid pixels for assignment of null pixels in the APHRODITE day-by-day rainfall data set appearing in the sea-land boundary area to obtain APHRODITE day-by-day rainfall data after total amount control processing.
S3: based on the spatial dimension, the GPM-IMERG data set, the SM2RAIN-ASCAT data set and the auxiliary data set with low resolution obtained in S22 are used as independent variables, and APHRODITE daily precipitation data after the total amount control processing in S24 are used as dependent variables to establish a regression model.
The specific method comprises the following steps:
based on the spatial dimension, the low-resolution GPM-IMERG data set, the SM2RAIN-ASCAT data set and the auxiliary data set obtained in S22 are used as independent variables, the APHRODITE day-by-day rainfall data after the total amount control processing in S24 is used as dependent variables, and a regression model based on SVR-TB is established for each grid point, specifically as follows:
to ensure the stability of the regression model, training sets and test sets were randomly generated with a 4:1 ratio of sample numbers in the training sets to the test sets. To accelerate the convergence rate of the program operation, the training set and the test set are respectively processed to [ -1,1 [ -1 ] ]]The interval is normalized, and the processing principle is that y is 2 (x-x)min)/(xmax-xmin) -1. Wherein y is the normalized data, x is the data to be normalized, x ismaxAnd xminRespectively, the maximum and minimum values of x. If the data in x are all the same, i.e. xmax＝xminThen data y is not changed at this time.
Setting a cross validation mode of a training set as ten-fold cross validation, setting a built-in type of a model as epsilon-SVR, and setting a kernel Function type of the model as Radial Basis Function (RBF). And searching for an optimal penalty coefficient c and a kernel parameter g by using a cross validation grid search method, setting the search ranges of the c and g parameters to be [ -10,10 ], setting the step length to be 0.5, constructing and solving an optimal problem, and obtaining the optimal c and g parameters to ensure that the model prediction precision is highest. And the hyper-parameters of the obtained model are adjusted through the feedback of the test set, so that the overfitting of the model is avoided. And according to the acquired optimal c and g parameters, constructing an SVR-TB model for each grid point in the APHRODITE grid data acquired in S24.
S4: and inputting the high-resolution SM2RAIN-ASCAT data set obtained in the S23, the GPM-IMERG data set obtained in the S21 and the high-resolution auxiliary data set obtained in the S22 into a regression model as input variables to obtain a high-resolution APHRODITE daily precipitation data set.
Examples
Selecting Zhejiang province (118 degrees to 123 degrees to 08'E, 27 degrees to 31 degrees to 10' N) as a research area, and carrying out scale reduction on the APHRODITE daily precipitation dataset in the same period by utilizing a GPM-IMERG and SM2RAIN-ASCAT precipitation dataset in 2015 and an auxiliary dataset comprising SRTM DEM, Slope, Aspect, Lon and Lat to finally obtain APHRODITE precipitation data with the time resolution of 1 day and the spatial resolution of 0.1 degree multiplied by 0.1 degree. The dimension reduction method specifically comprises the following steps:
step 1) data acquisition: and acquiring a GPM-IMERG data set, an SM2RAIN-ASCAT data set and an APHRODITE daily precipitation data set of the research area. And acquiring an environment factor auxiliary data set in the same research area, wherein the spatial resolution of the auxiliary data set is higher than that of an APHRODITE daily precipitation data set, and the auxiliary data set comprises an SRTM DEM digital elevation model, a Slope, a Slope Aspect, a longitude Lon and a latitude Lat. The longitude Lon and the latitude Lat are respectively obtained by calculating the central longitude value and the latitude value of the grid pixel in the SRTM DEM digital elevation model.
The GPM-IMERG data is an L3-level product of Final-Run V06 version, the IMERG data of the level achieves scientific research precision, the spatial resolution is 0.1 degree multiplied by 0.1 degree, the time resolution is 30min, the spatial range covers the whole world, the time range is from 2000 to 6 months, and all the data can be freely downloaded on the national aeronautics and astronautics administration (NASA) precipitation measurement task office network. The SM2RAIN-ASCAT data is global precipitation data obtained by soil moisture inversion based on ASCAT satellite observation, the spatial resolution is 12.5km, the time resolution is 1 day, and the data can be freely downloaded in the official website. The APHRODITE data is V1801_ R1 high resolution daily precipitation product (mm/day) with spatial resolution of 0.25 ° x 0.25 ° and temporal resolution of 1 day, and can be downloaded for free in APHRODITE official website for 1998 2015 year precipitation data, and the three data formats are NetCDF format.
Step 2) data preprocessing: reading the data in the NetCDF format by utilizing matlab software, converting the data into a tiff format and outputting the tiff format; the time resolution of the GPM-IMERG product is 30min, so 48 grids of the GPM-IMERG can be obtained every day, and the non-empty grid pixel values corresponding to the 48 grids are accumulated to be used as the pixel values of the day resolution.
Processing the time resolution of the GPM-IMERG precipitation data obtained in the step 1) into days; converting a coordinate system of SM2RAIN-ASCAT data into a GCS _ WGS _1984 geographical coordinate system by using a coordinate conversion tool provided by an Arcpy package in python, selecting a sampling method as a nearest neighbor method by using a resampling tool in ArcGIS 10.3 software, and resampling the converted SM2RAIN-ASCAT data to 0.1 degrees; writing a moving window method program by using matlab software, taking each pixel in a GPM-IMERG grid with daily resolution as the center, establishing a moving window with the size of 3 x 3, calculating the covered area proportion of each 0.25-degree grid pixel covered by the moving window, taking the covered 0.25-degree grid pixel with the largest proportion as the center pixel of the moving window, taking the average pixel value of the 3 x 3 GPM-IMERG grid pixels forming the moving window as the pixel value of the center pixel, taking all assigned 0.25-degree grids as the aggregated 0.25-degree GPM-IMERG data, aggregating an SM2RAIN-ASCAT data set after coordinate conversion into 0.25 degrees by the same method, and taking the used moving window with the size of 2 x 2; extracting gradient (Slope) and Aspect (Aspect) factors from DEM image data by using a terrain analysis tool in ArcGIS 10.3 software, and then aggregating 90m auxiliary data of DEM, Slope and Aspect into 0.1 degrees and 0.25 degrees by using the moving window program in matlab software, wherein the sizes of the moving windows are respectively set to be 112 x 112 and 278 x 278; converting a grid of 0.1 degree and a grid of 0.25 degree in Zhejiang province into point data by using a grid point conversion tool in ArcGIS 10.3 software, and calculating the longitude and latitude values of each point as an auxiliary variable data set; aiming at the situation that APHRODITE pixels appearing near sea-land boundaries and in partial areas are null values, matlab software is used for screening out the null values, and adjacent grid pixels are selected nearby for assignment.
Step 3) carrying out regression modeling: using the APHRODITE precipitation data of 0.25 degrees in the research area processed in the step 2) as a dependent variable, and using GPM-IMERG and SM2RAIN-ASCAT precipitation data with the spatial resolution of 0.25 degrees and 5 auxiliary data including DEM, Slope, Aspect, Lon and LatThe auxiliary data set is used as an independent variable, and an SVR-TB regression model is established for each grid point by using a support vector machine (libsvm) tool box provided in matlab software; the model is characterized in that: (1) in order to ensure the stability of the regression model, a training set and a test set are randomly generated, and the sample number ratio of the training set to the test set is 4: 1; (2) to speed up the convergence of the program run, the test set and training set are [ -1,1 [ -1]The interval is normalized, and the processing principle is that y is 2 (x-x)min)/(xmax-xmin) -1; where y is the normalized data, x is the data to be normalized, x ismaxAnd xminAre the maximum and minimum values of x, respectively, and if the data in x are all the same, the divisor is 0 (x)max＝xmin) If the data is not changed, the data is not changed; (3) setting a training set cross validation mode as 10-fold cross validation, setting a built-in type of an SVR-TB model as epsilon-SVR, setting a kernel function type of the model as RBF (radial Basis function), searching for an optimal penalty coefficient c and a kernel parameter g by using a cross validation grid search method, and setting search ranges of the c and g parameters as [ -10,10]Step length is 0.5, an optimal problem is constructed and solved, optimal c and g parameters are obtained, the prediction accuracy of the model is enabled to be highest, and overfitting of the model is avoided by adjusting hyper-parameters of the model through test set feedback; (4) and constructing an SVR-TB model for each grid point in 0.25-degree APHRODITE grid data according to the acquired optimal c and g parameters.
Step 4) downscaling prediction: based on the regression rule determined in the step 3), pixel values of each grid of GPM-IMERG and SM2RAIN-ASCAT precipitation data and DEM, Slope, Aspect, Lon and Lat environmental auxiliary factor data at 0.1 degree are used as input variables, the input variables are input into the established SVR-TB model, and APHRODITE precipitation product grid data sets with high precision and high spatial resolution in Zhejiang 2015 are calculated and output.
The Zhejiang province is selected as a verification area, as shown in figure 1, the 2015-year rate false alarm probability (FAR) and the Critical Success Index (CSI) are used as indexes, and the results show that the precipitation data after the scale reduction is obviously improved in precipitation event capturing capacity and the false alarm rate is reduced compared with GPM-IMERG, SM2RAIN-ASCAT and original APHRODITE precipitation data. Selecting 'brilliant' super typhoon of 7 months in 2015 as a verification event, as shown in fig. 2, wherein the result shows that the correlation coefficient R of the reduced-scale precipitation data is improved by 0.013 compared with the original APHRODITE precipitation data; the relative deviation bias is reduced by 0.053; the unbiased root mean square error is reduced by 1.709 millimeters; the root mean square error is reduced by 1.710 mm, and all indexes are obviously improved. As shown in fig. 3, in the event of "brilliant", the downscaled product not only inherits the capability of the original APHRODITE product to capture the spatial distribution characteristics (high northeast and low southwest) of the actual precipitation, but also has higher spatial resolution.
The above-described embodiments are merely preferred embodiments of the present invention, which should not be construed as limiting the invention. Various changes and modifications may be made by one of ordinary skill in the pertinent art without departing from the spirit and scope of the present invention. Therefore, the technical scheme obtained by adopting the mode of equivalent replacement or equivalent transformation is within the protection scope of the invention.

Claims (7)

1. A downscaling method for APHRODITE precipitation data is characterized by comprising the following steps:
s1: acquiring an original satellite remote sensing precipitation data set GPM-IMERG, an original ground variable inversion precipitation data set SM2RAIN-ASCAT and an original ground observation daily precipitation data set APHRODITE which are used for describing the same object; acquiring an original environment factor auxiliary data set which is the same as the three data set description objects, wherein the spatial resolution of the auxiliary data set is higher than that of APHRODITE, and the auxiliary data set comprises an SRTM DEM digital elevation model, a Slope, an Aspect Slope, a longitude Lon and a latitude Lat;
s2: respectively matching the four original data sets obtained in the step S1 based on different observation dimensions, and performing total amount control processing on the original APHRODITE, specifically sequentially performing the following steps:
s21: processing the resolution of the original GPM-IMERG to be the same as the resolution of the original APHRODITE by a data matching method based on the time dimension;
s22: based on spatial dimension, aggregating the resolutions of the GPM-IMERG, the original SM2RAIN-ASCAT and the original auxiliary data set obtained by S21 into the resolution which is the same as the original APHRODITE by a data matching method to obtain a low-resolution GPM-IMERG, a low-resolution SM2RAIN-ASCAT and a low-resolution auxiliary data set; aggregating the resolution of the original auxiliary data set to be the same as the resolution of the original GPM-IMERG to obtain a high-resolution auxiliary data set;
s23: based on the spatial dimension, degrading the resolution of the original SM2RAIN-ASCAT to be the same as the resolution of the original GPM-IMERG by a data matching method to obtain high-resolution SM2 RAIN-ASCAT;
s24: selecting nearby adjacent grid pixels for assignment of null value pixels in original APHRODITE appearing in sea-land boundary regions to obtain APHRODITE daily precipitation data after total amount control processing;
s3: based on the spatial dimension, taking the low-resolution GPM-IMERG, the low-resolution SM2RAIN-ASCAT and the low-resolution auxiliary data set obtained in S22 as independent variables, and taking APHRODITE daily precipitation data after the total amount control processing in S24 as dependent variables to establish a regression model;
s4: and on the basis of the spatial dimension, inputting the high-resolution SM2RAIN-ASCAT obtained in S23, the GPM-IMERG obtained in S21 and the high-resolution auxiliary data set obtained in S22 as input variables into the regression model to obtain the APHRODITE with high resolution.
2. The downscaling method of claim 1, wherein the GPM-IMERG in S1 has a temporal resolution and a spatial resolution of 30min and 0.1 °, respectively, the SM2RAIN-ASCAT has a temporal resolution and a spatial resolution of 1 day and 12.5km, respectively, and the APHRODITE has a temporal resolution and a spatial resolution of 1 day and 0.25 °, respectively; the auxiliary data sets all belong to static environment factor data sets, and the spatial resolution is 90 m.
3. The downscaling method of claim 1, wherein the longitudes Lon and latitudinal Lat of the auxiliary data set in S1 are obtained by calculating central longitude values and latitude values, respectively, of grid mesh pixels in the SRTM DEM digital elevation model.
4. The downscaling method of claim 1, wherein the data matching method in S21 is an addition method, and specifically includes the following steps:
and determining the number of GPM-IMERGs contained in the APHRODITE based on the time dimension, and accumulating the GPM-IMERGs successively according to the number to obtain the GPM-IMERG with the same resolution as the APHRODITE.
5. The downscaling method of claim 1, wherein the data matching method in S22 is a moving window method, and specifically includes the following steps:
respectively processing the GPM-IMERG, the SM2RAIN-ASCAT and the auxiliary data set, establishing a moving window of a by taking each pixel in the data set to be aggregated as a center, respectively calculating the covered area proportion of each pixel in a low-resolution grid mesh aggregated under the coverage of the moving window, and taking the pixel with the largest covered proportion as the center pixel of the moving window; taking the average pixel value of a x a data set pixels forming the moving window as the pixel value of the central pixel, and assigning values to all pixels in the low-resolution grid network obtained by aggregation; where a is the integer value resulting from dividing the resolution value of the low resolution grid mesh by the original resolution value of the data set and rounding up.
6. The downscaling method of claim 1, wherein the data matching method in S23 is a nearest neighbor interpolation method, which is as follows:
and respectively reassigning each pixel element in the GPM-IMERG, the SM2RAIN-ASCAT and the auxiliary data set according to the adjacent pixel element in the APHRODIE based on the spatial dimension.
7. The downscaling method according to claim 1, wherein the specific method of S3 is as follows:
based on the spatial dimension, the low-resolution GPM-IMERG, SM2RAIN-ASCAT and the auxiliary data set obtained in S22 are used as independent variables, APHRODITE day-by-day rainfall data after the total amount control processing in S24 is used as dependent variables, and a regression model based on SVR-TB is established for each grid point, specifically as follows:
randomly generating a training set and a testing set, wherein the sample number ratio of the training set to the testing set is 4: 1; training set and test set are processed to [ -1,1 ] respectively]The interval is normalized, and the processing principle is that y is 2 (x-x)min)/(xmax-xmin) -1; wherein y is the normalized data, x is the data to be normalized, x ismaxAnd xminThe maximum and minimum values of x, respectively; if the data in x are all the same, i.e. xmax＝xminWhen y is equal to xmax＝xmin；
Setting a cross validation mode of a training set as ten-fold cross validation, setting a built-in type of the model as epsilon-SVR, and setting a kernel Function type of the model as Radial Basis Function (RBF); searching for an optimal penalty coefficient c and a kernel parameter g by using a cross validation grid search method, setting the search ranges of the c and g parameters to be [ -10,10 ], setting the step length to be 0.5, constructing and solving an optimal problem, and obtaining the optimal c and g parameters; the hyper-parameters of the obtained model are adjusted through the feedback of the test set, so that the overfitting of the model is avoided; and according to the obtained optimal c and g parameters, constructing a regression model for each grid point in the APHRODITE daily precipitation data after the total amount control processing in S24.
CN202011169231.7A 2020-10-28 2020-10-28 Method for reducing scale of APHRODITE precipitation data Active CN112285808B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011169231.7A CN112285808B (en) 2020-10-28 2020-10-28 Method for reducing scale of APHRODITE precipitation data

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011169231.7A CN112285808B (en) 2020-10-28 2020-10-28 Method for reducing scale of APHRODITE precipitation data

Publications (2)

Publication Number Publication Date
CN112285808A CN112285808A (en) 2021-01-29
CN112285808B true CN112285808B (en) 2021-08-06

Family

ID=74372824

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011169231.7A Active CN112285808B (en) 2020-10-28 2020-10-28 Method for reducing scale of APHRODITE precipitation data

Country Status (1)

Country Link
CN (1) CN112285808B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112989286B (en) * 2021-03-22 2021-09-07 自然资源部国土卫星遥感应用中心 Space-time information fused microwave remote sensing soil moisture product downscaling method
CN117409335B (en) * 2023-12-14 2024-02-23 成都远望探测技术有限公司 Meteorological radar precipitation rate downscaling method based on visible light image

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105160192A (en) * 2015-09-17 2015-12-16 浙江大学 TRMM (Tropical Rainfall Measuring Mission) satellite rainfall data downscaling method based on M5-Local
CN111695088A (en) * 2020-06-12 2020-09-22 水利部交通运输部国家能源局南京水利科学研究院 Method and device for screening optimal regression window of satellite precipitation space downscaling

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106021868B (en) * 2016-05-10 2018-05-29 浙江大学 A kind of remotely-sensed data NO emissions reduction method based on more rules algorithm
EP3639055A4 (en) * 2017-06-14 2021-03-10 BAE SYSTEMS Information and Electronic Systems Integration Inc. Satellite tomography of rain and motion via synthetic aperture
US11205028B2 (en) * 2018-09-06 2021-12-21 Terrafuse, Inc. Estimating physical parameters of a physical system based on a spatial-temporal emulator
CN111797366A (en) * 2020-07-06 2020-10-20 丁思哲 Method for improving spatial resolution of precipitation data

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105160192A (en) * 2015-09-17 2015-12-16 浙江大学 TRMM (Tropical Rainfall Measuring Mission) satellite rainfall data downscaling method based on M5-Local
CN111695088A (en) * 2020-06-12 2020-09-22 水利部交通运输部国家能源局南京水利科学研究院 Method and device for screening optimal regression window of satellite precipitation space downscaling

Also Published As

Publication number Publication date
CN112285808A (en) 2021-01-29

Similar Documents

Publication Publication Date Title
US11333796B2 (en) Spatial autocorrelation machine learning-based downscaling method and system of satellite precipitation data
Marjanović et al. Landslide susceptibility assessment using SVM machine learning algorithm
Kilibarda et al. Spatio‐temporal interpolation of daily temperatures for global land areas at 1 km resolution
Li et al. Effects of DEM sources on hydrologic applications
CN110516816B (en) All-weather earth surface temperature generation method and device based on machine learning
Wilson Digital terrain modeling
CN109840553B (en) Extraction method and system of cultivated land crop type, storage medium and electronic equipment
CN103810376A (en) Ground daily rainfall predicting method based on satellite remote sensing and regression Kriging
Song et al. An improved surface soil moisture downscaling approach over cloudy areas based on geographically weighted regression
CN112016052A (en) Near-surface daily maximum air temperature estimation method, system and terminal based on multi-source data
EP3794386B1 (en) Determination of location-specific weather information for agronomic decision support
CN112285808B (en) Method for reducing scale of APHRODITE precipitation data
Meddens et al. Developing 5 m resolution canopy height and digital terrain models from WorldView and ArcticDEM data
Yoo et al. Spatial downscaling of MODIS land surface temperature: Recent research trends, challenges, and future directions
CN112329265A (en) Satellite remote sensing rainfall refinement space estimation method and system
Singh et al. Deep learning for improved global precipitation in numerical weather prediction systems
Huang et al. A novel fusion method for generating surface soil moisture data with high accuracy, high spatial resolution, and high spatio‐temporal Continuity
Mouatadid et al. A machine learning approach to non-uniform spatial downscaling of climate variables
Fan et al. A comparative study of four merging approaches for regional precipitation estimation
Dieng et al. Sea level anomalies using altimetry, model and tide gauges along the African coasts in the Eastern Tropical Atlantic Ocean: Inter-comparison and temporal variability
CN114170524A (en) Single-day tidal flat digital terrain construction method based on high-time-space satellite data fusion
Gao et al. A novel dynamic interpolation method based on both temporal and spatial correlations
CN114461983B (en) Satellite precipitation product space downscaling method based on water balance principle
CN112052720B (en) High-space-time normalization vegetation index NDVI fusion model based on histogram clustering
Shah et al. A Comparative Study of Spatial Interpolation Methods for CMIP6 Monthly Historical and Future Hydro-climatic Datasets for Indian Region

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant