GLASS-MODIS product correction value and FY3D MERS-II data-based earth surface average net radiation estimation method

Abstract

The invention provides an earth surface daily average net radiation estimation method based on a GLASS-MODIS product correction value and FY3D MERS-II data, and the method comprises the steps: carrying out the correction of GLASS-MODIS net radiation products under different conditions based on the actual measurement data of a site and a random forest, and obtaining correction models under different conditions; correcting the GLASS-MODIS net radiation product at the time corresponding to the FY3D MERS-II by using the correction model, taking the corrected data as a pseudo-truth value, and performing abnormal sample elimination on the pseudo-truth value corresponding to a sample point generated by a Poisson disk sampling method, the FY3D MERS-II and auxiliary data to obtain a global sample library; based on a global sample library and a multi-layer perceptron method, different LRD condition models are constructed, and high-resolution and high-precision earth surface net radiation estimation is achieved under the condition that FY3D MERS-II satellite remote sensing data are in shortage.

Description

Technical Field

[0001] The invention relates to the technical field of remote sensing quantitative inversion, and in particular to a method for estimating surface sky-averaged net radiation based on GLASS-MODIS product correction values ​​and FY3D MERSI-II data. Background Art

[0002] Net surface radiation represents the energy balance of the earth's surface, plays a decisive role in the energy exchange and redistribution process between the earth's underlying surface and the atmosphere, and drives important geophysical and chemical processes such as global climate change and vegetation growth. Therefore, net surface radiation has become a necessary input parameter for surface process models. Accurate estimation of net surface radiation plays an irreplaceable role in the analysis, simulation and prediction of climate and ecosystems.

[0003] Net surface radiation can be obtained from ground-based observation networks (such as the Baseline Surface Radiation Network (BSRN), AmeriFlux, and Lathuile) through direct observation or the addition of four-component observations. Although ground-based measured data are widely regarded as the "true value", due to the uneven distribution of observation sites around the world, especially the scarcity of observation sites in polar regions, bare land, and deserts, high maintenance costs, and limited spatial representativeness, the measured data cannot meet the actual needs. Currently, there are many products that can provide net radiation data, mainly including reanalysis and remote sensing products. Reanalysis products are obtained by assimilating multi-source data based on numerical weather forecast models. They have high temporal resolution and long time span, and can provide long-term gridded products, such as ERA5 (European Centre for Medium-Range Weather Forecasts Reanalysis5) and ERA5-land (Fifth Generation of the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis for Land Surface). However, the spatial resolution of reanalysis products is generally coarse (hundreds of kilometers), and there are problems such as long-term trends being too flat. Remote sensing products are generated from satellite data (such as CERES_SYN1deg_Ed4A (referred to as CERES4A)). Studies have shown that the estimation accuracy of remote sensing radiation products is generally better than that of reanalysis data, and they can more accurately describe the temporal and spatial changes of parameter trends. Although there are remote sensing radiation products with spatial resolutions up to kilometers (such as Global Land Surface Satellite products (GLASS)), due to the limited service life of satellites, how to obtain long-term and spatially consistent medium- and high-resolution net radiation remote sensing products is still an urgent problem to be solved.

[0004] Polar-orbiting satellite observation data is one of the most popular data sources for radiation inversion due to its high spatial resolution and global coverage, such as the Moderate Resolution Imaging Spectroradiometer (MODIS) data carried on the Terra and Aqua polar-orbiting satellites. Among the existing algorithms for estimating surface radiation remote sensing, the newly developed algorithm for directly estimating surface radiation parameters based on top-of-atmosphere (TOA) satellite observations has received widespread attention in recent years. This method is based on a large amount of extensive surface observations or model simulation data, combined with other auxiliary information, and uses classical statistical methods or machine learning methods to directly establish an empirical model of TOA satellite observations and surface radiation parameters. Studies have pointed out that this algorithm has more advantages than traditional methods in the inversion of radiation parameters because it avoids the error propagation and accumulation caused by the complex atmospheric transmission process and the multi-parameter input of other algorithms. The algorithm has also been successfully applied to the estimation of net surface radiation, such as MODIS and AVHRR (Advanced Very High Resolution Radiometer) TOA observations.

[0005] However, the Aqua and Terra satellites are about to be retired, and the satellites that can serve as the continuation and replacement of MODIS have a short operating time (such as FY3D MERSI-Ⅱ). The relevant research on radiation parameters, especially net radiation inversion, has not been fully carried out. Therefore, how to use the scarce new satellite remote sensing data to achieve high-resolution and high-precision surface net radiation estimation has become an urgent problem to be solved. Summary of the invention

[0006] The present invention provides a method for estimating the average net radiation of the earth's surface sky based on the correction value of GLASS-MODIS products and FY3D MERSI-Ⅱ data, so as to overcome the above problems.

[0007] The present invention provides a method for estimating the average net radiation of the earth's surface based on the correction value of the GLASS-MODIS product and the FY3D MERSI-Ⅱ data, comprising the following steps: Step 110: Based on the measured data of the site and the random forest model, the GLASS-MODIS net radiation products under different conditions are corrected respectively to obtain the correction models under different conditions; Step 120: Use the correction model to correct the GLASS-MODIS net radiation product at the time corresponding to FY3D MERSI-Ⅱ, use the corrected GLASS-MODIS net radiation data as the pseudo-true value, use the Poisson disk sampling method to randomly and evenly select three sample points in each 1-degree grid to form a sample point set, remove abnormal samples of the GLASS-MODIS net radiation pseudo-true value, FY3D MERSI-Ⅱ remote sensing data and other auxiliary data corresponding to the sample point set, and obtain a global sample library; Step 130: Use a multi-layer perceptron to construct a conditional model for different proportions of daytime duration to the whole day, train the conditional model based on the global sample library, use the 25-band top of atmosphere observations of FY3D MERSI-Ⅱ and its observation geometry information and other auxiliary information as inputs of the conditional model, and estimate the global surface average net radiation based on the conditional model.

[0008] According to a method for estimating the average net radiation of the surface sky based on the GLASS-MODIS product correction value and FY3D MERSI-Ⅱ data provided by the present invention, step 110 specifically includes: Different correction conditions are divided based on the station's daytime duration, latitude, and GLASS NDVI and Albedo products; The calibration conditions include: The absolute value of latitude is greater than 58 degrees as condition 1, NDVI is less than 0.2 or Albedo is greater than 0.7 as condition 2, daytime is 9 to 16 hours or Albedo is 0 to 0.25 or NDVI is greater than or equal to 0.2 as condition 3, Albedo is greater than 0.25 and less than 0.7 as condition 4, and these four conditions are used for classification; When condition 1 is met, the sample belongs to case 5; when condition 1 and condition 2 are not met, the sample belongs to case 1; when condition 1 is not met but condition 2 and condition 3 are met, the sample belongs to case 2; when condition 1 and condition 3 are not met but condition 2 and condition 4 are met, the sample belongs to condition 3; when condition 2 is met and other conditions are not met, the sample belongs to case 4; Based on the divided cases 1-5, random forests are used to correct the GLASS-MODIS net radiation products. The correction models are expressed as follows:

[0009] in, is the average daily net radiation measured at the site, is the GLASS-MODIS net radiation, is the length of daylight, is the relative distance between the sun and the earth, NDVI is the normalized vegetation index, and Albedo is the surface albedo; in, dt and dr It is obtained by calculation, and the calculation formula is as follows:

[0010]

[0011]

[0012]

[0013]

[0014] in, doy Indicates the day of the year. is the solar hour angle, and Represents the sunrise and sunset times, It refers to latitude.

[0015] According to a method for estimating the average net radiation of the surface sky based on the GLASS-MODIS product correction value and FY3D MERSI-Ⅱ data provided by the present invention, the auxiliary data are ERA5-land net radiation data, relative sun-earth distance dr and relative azimuth RAA ; The abnormal sample removal refers to extracting the GLASS-MODIS net radiation pseudo-true value, FY3D MERSI-Ⅱ remote sensing data and ERA5-land net radiation data at the corresponding position of the sample point set, and calculating the relative sun-earth distance dr and relative azimuth RAA, And remove abnormal samples; the FY3D MERSI-Ⅱ remote sensing data includes L1 products and L1_GEO products, and the L1 products and L1_GEO products contain 25-band top-of-atmosphere observations of FY3D MERSI-Ⅱ and their observation geometry information.

[0016] According to a method for estimating the average net radiation of the surface sky based on the GLASS-MODIS product correction value and FY3D MERSI-Ⅱ data provided by the present invention, step 130 specifically includes: Based on the global sample library, the 25-band top-of-atmosphere observations of FY3D MERSI-Ⅱ and its observation geometry information and other auxiliary information are used as inputs of the conditional model; Using the multi-layer perceptron method, an LRD conditional model with different proportions of daytime duration to the whole day was constructed to estimate the global surface daily average net radiation. The classification attribute of the area to be tested was determined according to the LRD and latitude. If the LRD was less than or equal to 0.3, the classification attribute of the area to be tested was determined to be an area with short daytime duration; if the LRD was greater than or equal to 0.75 and the absolute value of the latitude was greater than 60°, the classification attribute of the area to be tested was determined to be a high-latitude area with long daytime duration; otherwise, the classification attribute of the area to be tested was determined to be other areas; the area with short daytime duration was Case 1, the high-latitude area with long daytime duration was Case 2, and the other areas were Case 3; For Case 1, shortwave radiation is not used to estimate the net radiation of the surface. The model expression is as follows:

[0017] in, is the long-wave data of MERSI, SZA, VZA, RAA (RAA=VAA-SAA), height, and lat are the solar zenith angle, observation zenith angle, relative azimuth, altitude, and latitude of MERSI respectively. It is the pseudo-true value of GLASS-MODIS net radiation after correction; For Case 2, all bands are used to estimate the net radiation of the surface, and the model expression is as follows:

[0018] For Case 3, the ERA5-land net radiation is also added as auxiliary information, and the model expression is as follows:

[0019] in, Refers to ERA5-land net radiation data; The MERSI data are all daytime transit data, but for Case 1, if there are nighttime data, use the nighttime data for modeling; The formula of LRD is: LRD = dt / twenty four.

[0020] The method for estimating the average net radiation of the surface sky based on the GLASS-MODIS product correction value and FY3D MERSI-Ⅱ data provided by the present invention uses the corrected GLASS-MODIS net radiation as a pseudo-true value and obtains a large number of sample points through global uniform random sampling, and then pre-processes, controls the quality and matches the GLASS-MODIS net radiation "pseudo-true value", FY3D MERSI-Ⅱ remote sensing data and other auxiliary data to build a global sample library, thereby greatly increasing the sample volume; secondly, LRD and latitude are used to divide different classification attributes, and more suitable model input is used for different categories to improve the model accuracy, so as to achieve high-resolution and high-precision surface net radiation estimation results under the condition of shortage of FY3DMERSI-Ⅱ satellite remote sensing data. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 It is a flow chart of a method for estimating the average net radiation of the surface sky based on the GLASS-MODIS product correction value and FY3D MERSI-Ⅱ data provided by the present invention.

[0023] Figure 2 It is a schematic diagram of the GLASS-MODIS net radiation correction model provided by the present invention.

[0024] Figure 3 It is a schematic diagram of the surface net radiation estimation model provided by the present invention.

[0025] Figure 4 A schematic diagram of the locations of land surface measurement sites provided by the present invention.

[0026] Figure 5 It is a schematic diagram of the overall accuracy of the surface sky average net radiation estimation method based on GLASS-MODIS product correction values ​​and FY3D MERSI-Ⅱ data provided by the present invention.

[0027] Figure 6 It is a schematic diagram of different estimation accuracies of the surface sky average net radiation estimation method based on GLASS-MODIS product correction values ​​and FY3D MERSI-Ⅱ data provided by the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] The terms "first", "second", etc. in the present invention are used to distinguish similar objects, rather than to describe a specific order or precedence. It should be understood that the terms used in this way can be interchangeable where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of the same type.

[0030] Figure 1 FIG. 1 is a flow chart of a method for estimating the average net radiation of the surface sky based on the GLASS-MODIS product correction value and FY3D MERSI-Ⅱ data provided by the present invention, such as Figure 1 As shown, the method includes step 110 , step 120 and step 130 .

[0031] Step 110: Based on the measured data of the site and the random forest model, the GLASS-MODIS net radiation products under different conditions are corrected respectively to obtain the correction models under different conditions; Specifically, Figure 2 is a schematic diagram of the GLASS-MODIS net radiation correction model provided by the present invention, such as Figure 2 As shown, different correction conditions are divided based on the station daytime duration, latitude, GLASS NDVI and Albedo products; The calibration conditions include: The absolute value of latitude is greater than 58 degrees as condition 1, NDVI is less than 0.2 or Albedo is greater than 0.7 as condition 2, daytime is 9 to 16 hours or Albedo is 0 to 0.25 or NDVI is greater than or equal to 0.2 as condition 3, Albedo is greater than 0.25 and less than 0.7 as condition 4, and these four conditions are used for classification; When condition 1 is met, the sample belongs to case 5; when condition 1 and condition 2 are not met, the sample belongs to case 1; when condition 1 is not met but condition 2 and condition 3 are met, the sample belongs to case 2; when condition 1 and condition 3 are not met but condition 2 and condition 4 are met, the sample belongs to condition 3; when condition 2 is met and other conditions are not met, the sample belongs to case 4; Based on the divided cases 1-5, random forests are used to correct the GLASS-MODIS net radiation products. The correction models are expressed as follows:

[0032] in, is the average daily net radiation measured at the site, is the GLASS-MODIS net radiation, is the length of daylight, is the relative distance between the sun and the earth, NDVI is the normalized vegetation index, and Albedo is the surface albedo; in, dt and dr It is obtained by calculation, and the calculation formula is as follows:

[0033]

[0034]

[0035]

[0036]

[0037] in, doy Indicates the day of the year. is the solar hour angle, and Represents the sunrise and sunset times, It refers to latitude.

[0038] Here, the site measured data is the true value of the site's measured daily average net radiation value.

[0039] Step 120: Use the correction model to correct the GLASS-MODIS net radiation product at the time corresponding to FY3D MERSI-Ⅱ, use the corrected GLASS-MODIS net radiation data as the pseudo-true value, use the Poisson disk sampling method to randomly and evenly select three sample points in each 1-degree grid to form a sample point set, and remove abnormal samples of the GLASS-MODIS net radiation pseudo-true value, FY3D MERSI-Ⅱ remote sensing data and other auxiliary data corresponding to the sample point set to obtain a global sample library.

[0040] Specifically, the auxiliary data are ERA5-land net radiation data, relative sun-earth distance dr and relative azimuth RAA ; Abnormal sample removal refers to extracting the GLASS-MODIS net radiation pseudo-true value, FY3D MERSI-Ⅱ remote sensing data and ERA5-land net radiation data at the corresponding position of the sample point set, and calculating the relative sun-earth distancedr and relative azimuth RAA, And remove abnormal samples; FY3D MERSI-Ⅱ remote sensing data includes L1 product and L1_GEO product. Table 1 provides the parameters of FY3D MERSI-Ⅱ product used in the embodiment of the present invention.

[0041] Table 1 FY3D MERSI-Ⅱ products used in the study

[0042] The embodiment of the present invention uses the corrected GLASS-MODIS net radiation as the "pseudo-true value" and obtains a large number of sample points through global uniform random sampling, and then pre-processes, quality controls and matches the GLASS-MODIS net radiation pseudo-true value, FY3D MERSI-Ⅱ remote sensing data and other auxiliary data to build a global sample library, greatly increasing the sample size.

[0043] Step 130: Use a multi-layer perceptron to construct a conditional model for different proportions of daytime duration to the whole day, train the conditional model based on the global sample library, use the 25-band top of atmosphere observations of FY3D MERSI-Ⅱ and its observation geometry information and other auxiliary information as inputs of the conditional model, and estimate the global surface average net radiation based on the conditional model.

[0044] Specifically, based on the global sample library, the top-of-atmosphere observations of 25 bands of FY3D MERSI-Ⅱ and its observation geometry information and other auxiliary information are used as inputs of the conditional model; Figure 3 is a schematic diagram of the surface net radiation estimation model provided by the present invention, such as Figure 3 As shown in the figure, the multi-layer perceptron (MLP) method is used to construct a conditional model of LRD (LRD = dt / 24) with different proportions of daytime duration to the whole day to estimate the global surface daily average net radiation. The classification attribute of the area to be tested is determined according to LRD and latitude. If LRD is less than or equal to 0.3, the classification attribute of the area to be tested is determined to be a region with short daytime duration; if LRD is greater than or equal to 0.75 and the absolute value of latitude is greater than 60°, the classification attribute of the area to be tested is determined to be a high-latitude region with long daytime duration; otherwise, the classification attribute of the area to be tested is determined to be other regions; the region with short daytime duration is Case 1, the high-latitude region with long daytime duration is Case 2, and other regions are Case 3; It should be noted that other areas are mainly concentrated in mid- and low-latitude areas.

[0045] For Case 1, this situation often occurs in high-latitude areas and long-wave radiation plays a dominant role. Short-wave radiation is not used to estimate the net radiation of the surface. The model expression is as follows:

[0046] in, is the long-wave data of MERSI, SZA, VZA, RAA (RAA=VAA-SAA), height, and lat are the solar zenith angle, observation zenith angle, relative azimuth, altitude, and latitude of MERSI respectively. It is the pseudo-true value of GLASS-MODIS net radiation after correction; For Case 2, all bands are used for surface net radiation estimation, and the model expression is as follows:

[0047] For Case 3, the ERA5-land net radiation is also added as auxiliary information, and the model expression is as follows:

[0048] in, Refers to ERA5-land net radiation data; MERSI data are all daytime transit data, but for Case 1, if nighttime data is available, use the nighttime data for modeling.

[0049] Here, the observation geometry information may include the solar zenith angle, the observation zenith angle, the relative azimuth angle, the altitude and the latitude, which is not specifically limited in the embodiment of the present invention.

[0050] The solar zenith angle refers to the angle between the incident direction of the sun's rays and the zenith direction, and its calculation formula is 90° minus the solar altitude angle. When the solar altitude angle is 90°, that is, the sun is at the zenith, the solar zenith angle is 0°.

[0051] The observation zenith angle is the angle between the line of sight from the observation point to the target (such as a satellite or a star) and the local zenith. The relative azimuth refers to the relative position of two objects on the horizontal plane.

[0052] The band top-of-atmosphere observations of FY3D MERSI-Ⅱ include top-of-atmosphere observation data of 25 bands.

[0053] The method provided by the embodiment of the present invention uses the corrected GLASS-MODIS net radiation as the pseudo-true value and obtains a large number of sample points through global uniform random sampling, and then pre-processes, controls the quality and matches the GLASS-MODIS net radiation "pseudo-true value", FY3D MERSI-Ⅱ remote sensing data and other auxiliary data to build a global sample library, which greatly increases the sample size; secondly, LRD and latitude are used to divide different classification attributes, and more suitable model inputs are used for different categories to improve the model accuracy, thereby achieving high-resolution and high-precision surface net radiation estimation results under the condition of shortage of FY3D MERSI-Ⅱ satellite remote sensing data.

[0054] Based on any of the above embodiments, since the currently available FY3D MERSI-Ⅱ data has a limited spatiotemporal range, in order to make the final constructed model have better spatiotemporal generalization, the embodiment of the present invention first constructs a global sample library for the period when MERSI data is available (20 days in 2020) based on the site measured data and GLASS-MODIS net radiation products, and develops a MERSI net radiation inversion model applicable to the world on this basis. The specific introduction is as follows: Data and Preprocessing The embodiment of the present invention collects measured data of land surface stations distributed globally, FY3D MERSI-II remote sensing data, ERA5-land reanalysis data and GLASS-MODIS net radiation products, and performs a series of processing and quality control.

[0055] (1) Measured data from land surface sites Figure 4 A schematic diagram of the locations of the land surface measurement sites provided by the present invention, such as Figure 4 As shown in Figure 2, radiation observation data from 490 stations in 12 observation networks around the world from 2000 to 2021 are collected. The specific information of the stations is shown in Table 2. These stations are located at latitudes from 75.1°S to 79.91°N and cover a variety of major land cover types, such as farmland, bare soil, forest, shrub, ice, etc. The spatial location distribution of the stations is as follows: Figure 4 As shown. These stations are distributed in various climate zones between the North and South Poles. Therefore, the net radiation observation data collected in this embodiment is comprehensive, can comprehensively reflect different surface types and climatic conditions, and has good temporal and spatial representativeness. Since the observation frequencies and data control standards of various observation networks are inconsistent, it is necessary to uniformly preprocess the measured data, mainly including: i. Eliminate outliers and missing values ​​of station data, and retain high-quality data (screened according to data quality codes); ii. Convert all high-quality data to local time and process them into daily averages; iii. Finally, determine the purpose of the data based on information such as the geographical location of the station and the type of surface cover.

[0056] Table 2 Ground observation network information

[0057] (2) FY3D MERSI-Ⅱ remote sensing data Fengyun-3 (FY-3) meteorological satellite is my country's second-generation polar-orbiting meteorological satellite, which is mainly used for medium-term numerical weather forecasting, monitoring ecological environment and large-scale natural disasters. The embodiment of the present invention intends to collect 1KM products provided by FY3D MERSI-Ⅱ carried by FY3D from the 26th to the 30th of January, April, July and October 2020. The detailed information is shown in Table 2. Secondly, the required remote sensing data can be extracted according to the longitude and latitude and the observation time. Converting UTC to local time can match the corresponding measured data or net radiation products.

[0058] (3) ERA5-land reanalysis data ERA5-Land (Fifth Generation of the European Centre for Medium-RangeWeather Forecasts (ECMWF) Reanalysis for Land Surface) is the fifth generation of land surface reanalysis dataset developed by the European Centre for Medium-Range Weather Forecasts (ECMWF). This product has been optimized specifically for surface applications. Compared with ERA5 and the older ERA-Interim, ERA5-Land has a higher spatial resolution (0.1°×0.1°), and the temporal resolution is once an hour, the same as ERA5. Since the radiation data of ERA5-land is a cumulative value, it needs to be converted into instantaneous values ​​first, then the data is converted to local time and the daily average is calculated, and finally the corresponding measured data or net radiation remote sensing products can be matched according to the longitude, latitude and time.

[0059] (4) GLASS series products Table 3 lists the GLASS series products used in the embodiments of the present invention, mainly including GLASS-MO-DIS net radiation, wide-band albedo and normalized difference vegetation index (NDVI).

[0060] Table 3 GLASS series products used in this example

[0061] In this embodiment, the GLASS-MODIS net radiation, Albedo and NDVI data from 2000 to 2021 are extracted according to the site location and observation time for correction of the GLASS-MODIS net radiation product (NDVI and Albedo use the data corresponding to the closest days). The corrected GLASS-MODIS net radiation product is used as the "pseudo-true value" for the subsequent construction of the global average daily net radiation model.

[0062] (5) Other auxiliary data calculations, The calculation formulas for other auxiliary data are as follows:

[0063]

[0064]

[0065]

[0066]

[0067]

[0068] in, doy Indicates the day of the year. is the solar hour angle (in radians), and Represents the sunrise and sunset times, It refers to latitude.

[0069] Method (1) GLASS-MODIS net radiation correction model The embodiment of the present invention is first based on the measured data of the site collected from 2000 to 2021, and refers to the GLASS-MODIS model condition setting, and is divided into case 1-5 based on the daytime length, latitude, GLASS NDVI and Albedo products of the site (division conditions are as follows Figure 3 The training and validation sets are divided into 7:3. The samples under different conditions are shown in Table 4.

[0070] Table 4 GLASS-MODIS correction model and sample conditions used

[0071] The expressions of different case models are as follows:

[0072] in, is the average daily net radiation measured at the site, is the GLASS-MODIS net radiation, is the length of daylight, is the relative distance between the sun and the earth, NDVI is the normalized vegetation index, and Albedo is the surface albedo. Random forest is used to build the model.

[0073] (2) Estimation of average net radiation on the Earth’s surface The trained correction model is used to correct the GLASS-MODIS net radiation product corresponding to the FY3D MERSI-Ⅱ, and the corrected GLASS-MODIS net radiation data is used as the "pseudo-true value". The Poisson disk sampling method is used to randomly and evenly select three sample points in each 1-degree grid to form a sample point set. The GLASS-MODIS net radiation "pseudo-true value", FY3D MERSI-Ⅱ remote sensing data and other auxiliary data are further preprocessed, quality controlled and matched to build a global sample library; auxiliary data refers to ERA5-land net radiation data, dr and relative azimuth ( RAA ); The preprocessing, quality control and matching process refers to extracting the GLASS-MODIS net radiation "pseudo-true value", FY3D MERSI-Ⅱ remote sensing data and ERA5-land net radiation data at the corresponding position of the sample point set, and calculating dr and RAA, And remove abnormal samples.

[0074] Based on the sample library, the 25-band top-of-atmosphere observations of FY3D MERSI-Ⅱ and their observation geometry information and other auxiliary information were used as model inputs (the wind and cloud products used and their parameters are shown in Table 1). The multi-layer perceptron method was used to construct different LRDs (the proportion of daytime duration to the whole day, LRD = dt / 24) Conditional model estimates the global surface average net radiation; Figure 3 As shown, the classification attributes of the area to be tested are judged according to LRD and latitude (lat), and are divided into three categories: areas with short daytime (Case 1), high-latitude areas with long daytime (Case 2), and other areas (Case 3). Finally, this embodiment collected a total of 1,341,732 samples for model training and 574,226 samples for model verification. In addition, the measured data of 250 stations on the 26th to 30th day of January, April, July and October 2020 were introduced as independent verification data to evaluate the model more objectively. The sample situation is detailed in Table 5.

[0075] Table 5. Sample set used to construct the average daily net radiation on the surface

[0076] For Case 1, shortwave radiation is not used to estimate the net radiation of the surface. The model expression is as follows:

[0077] in, is the long-wave data of MERSI, SZA, VZA, RAA (RAA=VAA-SAA), height, and lat are the solar zenith angle, observation zenith angle, relative azimuth, and altitude latitude of MERSI respectively. It is the “pseudo-true value” of GLASS-MODIS net radiation after correction; For Case 2, all bands are used for surface net radiation estimation, and the model expression is as follows:

[0078] For Case 3, the ERA5-land net radiation is also added as auxiliary information, and the model expression is as follows:

[0079] in, Refers to ERA5-land net radiation data; It should be noted that the MERSI data used in this embodiment are all daytime transit data.

[0080] Results Analysis In the embodiments of the present invention, Figure 5 : is a schematic diagram of the overall accuracy of the surface sky average net radiation estimation method based on the GLASS-MODIS product correction value and FY3D MERSI-Ⅱ data provided by the present invention, such as Figure 5 As shown in the figure, the global model developed based on the constructed sample library has similar training and validation accuracy, and the accuracy RMSE is between 13 and 15 W / m². The results verified based on the 20-day site measured data in 2020 are worse than the sample-based verification results, with an independent verification accuracy RMSE of 24.664 W / m² and a Bias of -4.950 W / m². However, considering the spatial distribution limitations of the site data, this result is acceptable.

[0081] Figure 6 Schematic diagram of different estimation accuracies of the surface sky average net radiation estimation method based on GLASS-MODIS product correction value and FY3D MERSI-Ⅱ data provided by the present invention, such as Figure 6As shown in the figure, the accuracy of independent verification of the model sites in case 1 and 2 is worse than that in case 3, which is related to the fact that they represent winter and summer in high latitudes, the performance of the GLASS-MODIS net radiation product itself, and the model estimation error. Case 3 represents most cases, so the site verification samples are the largest, and its accuracy is close to the overall accuracy. However, the site data used for independent verification is still relatively small, and more MERSI data and site measured data need to be collected for model evaluation and assessment.

Claims

1. A method for estimating the average net radiation of the surface sky based on the GLASS-MODIS product correction value and FY3D MERSI-Ⅱ data, characterized in that: include: Step 110: Based on the measured data of the site and the random forest model, the GLASS-MODIS net radiation products under different conditions are corrected respectively to obtain the correction models under different conditions; Step 120: Use the correction model to correct the GLASS-MODIS net radiation product at the time corresponding to FY3D MERSI-Ⅱ, use the corrected GLASS-MODIS net radiation data as the pseudo-true value, use the Poisson disk sampling method to randomly and evenly select three sample points in each 1-degree grid to form a sample point set, remove abnormal samples of the GLASS-MODIS net radiation pseudo-true value, FY3D MERSI-Ⅱ remote sensing data and other auxiliary data corresponding to the sample point set, and obtain a global sample library; Step 130: Use a multi-layer perceptron to construct a conditional model for different proportions of daytime duration to the whole day, train the conditional model based on the global sample library, use the 25-band top of atmosphere observations of FY3D MERSI-Ⅱ and its observation geometry information and other auxiliary information as inputs of the conditional model, and estimate the global surface average net radiation based on the conditional model.

2. The method for estimating the average net radiation of the surface sky based on the GLASS-MODIS product correction value and FY3D MERSI-Ⅱ data according to claim 1, characterized in that: The step 110 specifically includes: Different correction conditions are divided based on the station's daytime duration, latitude, and GLASS NDVI and Albedo products; The calibration conditions include: The absolute value of latitude is greater than 58 degrees as condition 1, NDVI is less than 0.2 or Albedo is greater than 0.7 as condition 2, daytime is 9 to 16 hours or Albedo is 0 to 0.25 or NDVI is greater than or equal to 0.2 as condition 3, Albedo is greater than 0.25 and less than 0.7 as condition 4, and these four conditions are used for classification; When condition 1 is met, the sample belongs to case 5; when condition 1 and condition 2 are not met, the sample belongs to case 1; when condition 1 is not met but condition 2 and condition 3 are met, the sample belongs to case 2; when condition 1 and condition 3 are not met but condition 2 and condition 4 are met, the sample belongs to condition 3; when condition 2 is met and other conditions are not met, the sample belongs to case 4; Based on the divided cases 1-5, random forests are used to correct the GLASS-MODIS net radiation products. The correction models are expressed as follows: ; in, is the average daily net radiation measured at the site, is the GLASS-MODIS net radiation, is the length of daylight, is the relative distance between the sun and the earth, NDVI is the normalized vegetation index, and Albedo is the surface albedo; in, dt and dr It is obtained by calculation, and the calculation formula is as follows: ； ； ； ； ； in, doy Indicates the day of the year. is the solar hour angle, and Represents the sunrise and sunset times, It refers to latitude.

3. The method for estimating the average net radiation of the surface sky based on the GLASS-MODIS product correction value and FY3D MERSI-Ⅱ data according to claim 1, characterized in that: The auxiliary data are ERA5-land net radiation data, relative sun-earth distance dr and relative azimuth RAA ; The abnormal sample removal refers to extracting the GLASS-MODIS net radiation pseudo-true value, FY3D MERSI-Ⅱ remote sensing data and ERA5-land net radiation data at the corresponding position of the sample point set, and calculating the relative sun-earth distance dr and relative azimuth RAA, And remove abnormal samples; the FY3D MERSI-Ⅱ remote sensing data includes L1 products and L1_GEO products, and the L1 products and L1_GEO products contain 25-band top-of-atmosphere observations of FY3D MERSI-Ⅱ and their observation geometry information.

4. The method for estimating the average net radiation of the surface sky based on the GLASS-MODIS product correction value and FY3D MERSI-Ⅱ data according to claim 1, characterized in that: The step 130 specifically includes: Based on the global sample library, the 25-band top-of-atmosphere observations of FY3D MERSI-Ⅱ and its observation geometry information and other auxiliary information are used as inputs of the conditional model; Using the multi-layer perceptron method, an LRD conditional model with different proportions of daytime duration to the whole day was constructed to estimate the global surface daily average net radiation. The classification attribute of the area to be tested was determined according to the LRD and latitude. If the LRD was less than or equal to 0.3, the classification attribute of the area to be tested was determined to be an area with short daytime duration; if the LRD was greater than or equal to 0.75 and the absolute value of the latitude was greater than 60°, the classification attribute of the area to be tested was determined to be a high-latitude area with long daytime duration; otherwise, the classification attribute of the area to be tested was determined to be other areas; the area with short daytime duration was Case 1, the high-latitude area with long daytime duration was Case 2, and the other areas were Case 3; For Case 1, shortwave radiation is not used to estimate the net radiation of the surface. The model expression is as follows: ； in, is the long-wave data of MERSI, SZA, VZA, RAA (RAA=VAA-SAA), height, and lat are the solar zenith angle, observation zenith angle, relative azimuth, altitude, and latitude of MERSI respectively. It is the pseudo-true value of GLASS-MODIS net radiation after correction; For Case 2, all bands are used to estimate the net radiation of the surface, and the model expression is as follows: ； For Case 3, the ERA5-land net radiation is also added as auxiliary information, and the model expression is as follows: ； in, Refers to ERA5-land net radiation data; The MERSI data are all daytime transit data, but for Case 1, if there are nighttime data, use the nighttime data for modeling; The formula of LRD is: LRD = dt / twenty four.

Images