A method and device for determining global snow cover based on geostationary satellites

Abstract

The present application discloses a method for distinguishing global snow accumulation based on geostationary orbit satellites, comprising the following steps: performing data preprocessing, including reprojecting and resampling geostationary orbit satellite data according to the same standard, and calculating the reflectivity and brightness temperature of each band; performing maximum brightness temperature fusion on multi-phase data, and the 10μm band brightness temperature corresponding to the pixel value of the synthesized image is the maximum value of the 10μm band brightness temperature in the non-night phase in multiple phases of the day; selecting snow accumulation discrimination factors, and determining the discrimination threshold of each discrimination factor based on the pixel value; processing the pixel value of the geostationary orbit satellite according to the selected snow accumulation discrimination factor and the threshold corresponding to each discrimination factor, and extracting the snow accumulation range. The present application also includes a device for implementing the method. The present invention solves the problem of low recognition accuracy in the prior art.

Description

Technical Field

[0001] The present application relates to the field of remote sensing technology, and in particular to a global snow cover identification method and device based on a geostationary orbit satellite. Background Art

[0002] Satellite remote sensing has the advantages of easy data acquisition, low cost, large range, multi-spectral bands, and multi-temporal phases, so it has become a powerful means of snow detection. Remote sensing snow monitoring has received more and more attention. It can obtain large-scale and even global snow cover change information more timely and effectively than conventional land observations. For example, the most widely used MODIS series of snow products, due to its high spatial resolution, rich band information and high snow identification accuracy, plays an important role in snow monitoring. In addition, remote sensing snow monitoring has the ability to extract more types of snow cover information, such as snow depth, snow water equivalent, whether the snow is melting, and the surface state under the snow cover, which makes up for the shortcomings of conventional observation data such as low spatial resolution and large investment.

[0003] Geostationary orbit satellites have the characteristics of wide imaging range and high temporal resolution. The characteristics of cloud changes over time can effectively reduce the cloud cover in the observation area. The geostationary orbit satellites currently in orbit mainly include GOES-R (Geostationary Operational Environmental Satellite) launched by the United States, MSG (Meteosat Second Generation) launched by the European Meteorological Satellite Organization, Himawari-8 launched by Japan, and Fengyun series geostationary meteorological satellites (FY-2 and FY-4) launched by my country. Except for MSG-1 launched in 2002, the other geostationary meteorological satellites were developed, launched, received data and provided data services in the second generation after 2017. Although the new generation of geostationary meteorological satellites are still only equipped with optical sensor systems, compared with the first generation of geostationary meteorological satellites, the new generation of geostationary meteorological satellites have richer bands and higher temporal resolution, which increases the frequency of observing real surface pixels during the day. For the global land area, the imaging range of polar orbit satellites is small, which will cause the disadvantages of huge data volume and inconvenience in processing. At the same time, the temporal resolution of polar-orbiting satellites is low. Therefore, for areas covered by clouds during imaging, it is impossible to effectively identify the true surface information under the cloud cover. However, these defects can be effectively eliminated by taking advantage of the advantages of geostationary satellites. Therefore, geostationary meteorological satellite data is of great significance for cloud removal and snow identification.

[0004] Remote sensing technology was used to monitor snow cover in foreign countries at an early stage. Hüsler F et al. used data obtained by AVHRR sensors to study the spatiotemporal changes of snow cover in the European Alps. They found that the downward trend of snowfall in the high mountain areas of the Alps has slowed down over the past 27 years and proposed a new set of cloud-free satellite snow cover products. Hall et al. proposed the SNOMAP algorithm based on Landsat5 TM images, which uses the difference in reflectivity of snow and other surfaces to distinguish them. The MODIS snow products MOD10A1 and MYD10A1 generated by the National Aeronautics and Space Administration (NASA) based on this algorithm have been widely used. However, the algorithm has low recognition accuracy in situations such as cirrus clouds, forests, and mountain shadows. Romanov et al. used visible light (VIS), mid-infrared (MIR), and infrared (IR) measurement data from the Geosynchronous Operational Environmental Satellite (GOES) and microwave observation data from SSM / I, and developed an automatic snow mapping system by comprehensively utilizing GOES's multiple observations and SSM / I's ability to penetrate clouds and observe the ground. SSM / I data are used to fill in the cloudy areas on the GOES snow products, and the resulting product accuracy is close to or even higher than that of the IMS products. Surer et al. verified the MSG-SEVIRI snow product, using six channels of reflectivity (0.6, 0.8, 1.6, 3.9, 10.8, and 12.0 μm) and three channels of brightness temperature (3.9, 10.8, 12.0 μm) in flat areas, and using the Snow Index (SI) to extract snow in mountainous areas. The verification results of the actual data of meteorological stations and MODIS snow products showed that the product can greatly reduce cloud cover, but it will underestimate snow in some areas.

[0005] In my country, early snow monitoring was mainly based on data from ground meteorological stations. The use of remote sensing technology to monitor snow started late but has developed rapidly. Li Sanmei et al. used FY-2C data to identify snow cover in China. Their algorithm added an auxiliary factor function on the basis of the multi-threshold snow cover identification algorithm, and used ground station data and NOAA-17 snow cover data to verify and analyze the accuracy of the identification results. It was found that the algorithm well solved the problem of identifying snow on different underlying surface types in FY-2C full-disk data; Yang Juntao et al. used GOES-13 satellite data to add ice cloud and snow differentiation factors (T2-T4) on the basis of the Romanov snow cover identification process, and combined SI, R1, R2, T4 and other channel data to identify snow, clouds, and snow-free surfaces. For cloudy areas, AMSR-E snow cover products were used for fusion, and finally a cloudless snow product was obtained; Han Chenhui used Himawari-8 and FY-4A data to improve the FY-4A global snow identification business algorithm developed by the National Satellite Meteorological Center, focusing on distinguishing cirrus clouds, ice clouds and desert snow. Finally, the improved algorithm has a high accuracy in snow cover identification.

[0006] In summary, most of the snow cover products currently used are polar-orbiting satellite data products, and there is no algorithm for developing global snow cover products using the new generation of geostationary satellites. Polar-orbiting satellites have low temporal resolution, so for areas covered by clouds during imaging, it is impossible to effectively identify the true surface information under the cloud cover. However, these defects can be effectively eliminated by using the advantages of geostationary satellites. Summary of the invention

[0007] The present application proposes a method and device for identifying global snow cover based on geostationary orbit satellites to solve the problem of low recognition accuracy in the prior art.

[0008] The present application embodiment provides a method for determining global snow cover based on a geostationary orbit satellite, comprising the following steps:

[0009] Perform data preprocessing, including reprojecting and resampling geostationary satellite data according to the same standards, and calculating the reflectivity and brightness temperature of each band;

[0010] The maximum brightness temperature fusion is performed on the multi-temporal data. The brightness temperature of the 10μm band corresponding to the pixel value of the synthesized image is the maximum value of the brightness temperature of the 10μm band in the non-night phase among multiple phases of the day.

[0011] Select snow cover discrimination factors, and determine discrimination thresholds of the discrimination factors based on the pixel values;

[0012] According to the selected snow cover discrimination factors and the threshold values ​​corresponding to each discrimination factor, the pixel values ​​of the geostationary orbit satellite are processed to extract the snow cover range.

[0013] Preferably, the step of selecting snow accumulation discriminant factors and determining the discrimination threshold of each discriminant factor based on the pixel value further comprises:

[0014] Select training samples, preferably, use MODIS images, and determine the snow range through visual translation; train snow discriminant factors, including selecting pixel values ​​used as snow discriminant factors, drawing a ground feature separation map, and determining a threshold range; use the selected snow discriminant factors and thresholds to construct a discriminant for snow range recognition using a multi-threshold decision tree.

[0015] Preferably, the snow discrimination factor of the present application includes at least one combination of the brightness temperature near the 10μm band, the normalized difference snow index NDSI, the brightness temperature near the 12μm band, the brightness temperature near the 3μm band, the reflectivity near the 1.36μm band and the reflectivity near the 1.56μm band.

[0016] Preferably, in the step of extracting the snow coverage range by processing the pixel values ​​of the geostationary orbit satellite according to the selected snow coverage discriminant factors and the threshold values ​​corresponding to each discriminant factor, for the FY-4A data mainly covering Asia and Oceania, the snow coverage range is extracted according to the discriminant factors and threshold values ​​of the FY-4A snow area extraction method in Table 4 of the embodiment of the present application; for the Meteosat-8 data mainly covering Europe and Africa, the snow coverage range is extracted according to the discriminant factors and threshold values ​​of the MSG snow coverage range extraction method in Table 5 of the embodiment of the present application; for the GOES-16 data mainly covering North America and South America, the snow coverage range is extracted according to the discriminant factors and threshold values ​​of the GOES-16 snow coverage range extraction method in Table 6 of the embodiment of the present application.

[0017] Further preferably, the data preprocessing step of the present application also includes geometrically correcting the number of geostationary orbit satellites.

[0018] Further preferably, the data preprocessing step of the present application also includes performing angle effect correction on the geostationary orbit satellite data.

[0019] Furthermore, the global snow cover discrimination method based on geostationary orbit satellites also includes the step of verifying the accuracy of the extracted snow cover range. The consistency check includes the overall accuracy (OA) when there is no cloud and the overall accuracy (QA) when there is cloud. When using MODIS data for accuracy verification, the accuracy verification under cloudless conditions is considered. The snow cover inconsistency check includes the underestimation error (IU) of the snow cover and the overestimation error (IO) of the snow cover.

[0020] The present application also proposes a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method described in any embodiment of the present application is implemented.

[0021] Furthermore, the present application also proposes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in any embodiment of the present application when executing the computer program.

[0022] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects:

[0023] Currently, mainstream snow area products are extracted from polar-orbiting satellite data. Polar-orbiting satellites pass by 1-2 times a day. Therefore, high-altitude cloudy areas and areas covered by instantaneous clouds have a greater impact on snow extraction. This patent utilizes the advantages of high temporal resolution of geostationary orbit satellites and the principle of "clouds move but snow does not", which can remove more than 30% of the cloud impact. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0025] Figure 1 This is a flow chart of an embodiment of the method of this application;

[0026] Figure 2 Schematic diagram of the principle of the maximum brightness temperature synthesis step;

[0027] Figure 3 This is the maximum brightness temperature synthesis cloud removal effect picture;

[0028] Figure 4 A flowchart of an embodiment of a multi-threshold decision tree construction process;

[0029] Figure 5 This is a snow cover discrimination effect diagram based on geostationary orbit satellites;

[0030] Figure 6 Global regional accuracy verification timing diagram, where

[0031] (a) is the overall accuracy time series diagram of the global region;

[0032] (b) is the time series diagram of the global regional overestimation error and underestimation error;

[0033] Figure 7 is a timing diagram of the accuracy verification of different geostationary orbit satellites, where

[0034] (a) is the overall accuracy time series diagram of different geostationary orbit satellites;

[0035] (b) is the time series diagram of overestimation error of different geostationary orbit satellites;

[0036] (c) is the timing diagram of the underestimation error of different geostationary orbit satellites;

[0037] Figure 8 is a timing diagram of accuracy verification for each continent, where:

[0038] (a) Overall accuracy time series diagram for each continent;

[0039] (b) Time series of overestimation errors in each continent;

[0040] (c) Time series of underestimation errors for each continent. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0042] In view of the problem that the snow cover area products of polar-orbiting satellites are greatly affected by clouds, the present invention proposes to use the multi-frequency surface observations of geostationary orbit satellites within a day to reduce the impact of clouds and obtain a more realistic surface snow state. The present invention is oriented to the global land area (60°N-60°S, 180°W-180°E), using FY-4A data, MSG data and GOES-16 data in the snow season, and proposes a global geostationary orbit satellite snow cover range extraction method. The method realizes the intra-day data fusion of geostationary orbit satellites based on the maximum brightness temperature synthesis algorithm, introduces a new snow cover identification factor to improve the decision tree snow cover identification algorithm of Romanov and Wildt, and optimizes and adjusts the parameters for the different spectral characteristics and monitoring areas of the three geostationary orbit satellite data, and finally forms a new global snow cover identification algorithm. The proposed method was quantitatively analyzed using the MOD10A1 (V6 version) snow cover area product, and the accuracy and cloud removal effect of the geostationary orbit satellite snow cover range extraction method were verified.

[0043] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0044] Figure 1 This is a flow chart of an embodiment of the method of this application.

[0045] Step 10: Perform data preprocessing, including reprojecting and resampling the geostationary satellite data according to the same standard, and calculating the reflectivity and brightness temperature of each band.

[0046] For example, the projection methods of the three geostationary orbit satellite FY-4A data, Meteosat-8 data and GOES-16 data are all nominal projections. Since different data or even different bands of the same data have different spatial resolutions, in order to facilitate comparison and analysis with other data products, the three geostationary orbit satellite data are uniformly reprojected and resampled to 0.04° (which is approximately consistent with the spatial resolution of FY-4A).

[0047] Preferably, the number of geostationary satellites is geometrically corrected. For example, FY-4A data and MSG data are geometrically corrected using the official longitude and latitude lookup table, and GOES-16 data are geometrically corrected using python GDAL.

[0048] The calculation of reflectivity and brightness temperature involves the following bands: The data used in the snow range extraction method include the reflectivity of the visible light band (the band near 0.56μm), the reflectivity of the near-infrared band (the band near 1.56μm), the brightness temperature of two far-infrared bands (the band near 10μm and 12μm) and the brightness temperature of the mid-infrared band (the band near 3μm).

[0049] For example, the FY-4A data contains the lookup table data of the reflectivity and brightness temperature corresponding to each band. The reflectivity and brightness temperature data of the required band can be obtained by associating the original gray value data with the lookup table. The reflectivity and brightness temperature of MSG data and GOES-16 data can be calculated according to the conversion formula and parameters in the data description to obtain the corresponding reflectivity and brightness temperature.

[0050] Preferably, the geostationary satellite data is corrected for angle effects. Geostationary satellites have the characteristics of high temporal resolution, that is, the same ground object can be observed frequently within a day. However, since the imaging time of each scene is different and the single imaging range is wide, the solar zenith angle, satellite zenith angle and satellite azimuth angle of the same pixel vary greatly over time, causing the reflectivity of the same ground object to change continuously within a day. This phenomenon has a more significant impact in the visible light band, which has a great impact on the threshold selection for completing snow recognition. Therefore, it is necessary to correct the angle effect of geostationary satellites.

[0051] In the best embodiment of the present application, a semi-empirical model combining the advantages of empirical statistical models and physical models is used to correct the angle effect based on the semi-empirical kernel-driven model method proposed by Roujean et al. The model consists of two kernel functions, two kernel coefficients and a constant term. The first kernel is a typical radiation transfer type volume scattering kernel, which is an approximation of the single scattering of vegetation canopy radiation transfer theory, also known as Ross or RossThick kernel; the second kernel is a geometric optics kernel, which mainly considers the shadow caused by the protrusion of the Lambertian surface; the constant term represents the isotropic scattering effect, see model formula (1):

[0052]

[0053] Among them, α 0 is a constant term representing isotropic scattering, Kvol is a typical radiation transfer type volume scattering kernel, which mainly considers the scattering and absorption of light by a random surface, Kgeo is a geometric optics kernel, which mainly considers the effects of lighting, shadows and terrain, and α 1 and α 2 are the kernel coefficients respectively. Formula (2) and Formula (3) are the Kvol and Kgeo kernel functions respectively:

[0054]

[0055]

[0056] where θ i is the solar zenith angle, θ v is the satellite observation zenith angle, φ is the relative azimuth angle between the sun and the satellite, and g is the phase angle. The phase angle needs to be calculated through the known θ i ,θ v and are obtained using formulas (4) and (5).

[0057]

[0058] g=arccosg (5)

[0059] For snow area monitoring, the solar zenith angle and satellite zenith angle of the target pixel are corrected to 45 degrees, and the relative azimuth angle is corrected to 30 degrees. The solar zenith angle, satellite zenith angle, relative azimuth angle and visible light reflectance are statistically analyzed, and the parameters of the Roujean model are fitted using the least squares regression method. Substituting the corrected angle data into formula (1) to obtain the parameter R 0 ; Substitute the satellite angle parameter of the pixel to be corrected into formula (1) to obtain the parameter R 1 , R s is the reflectivity of visible light received by the satellite, R r is the corrected visible light reflectance, as shown in formula (6).

[0060]

[0061] After the angle effect correction, taking the reflectivity data of the first band of FY-4A in eight phases on November 15, 2019 as an example, the reflectivity of the pixels with snow and the reflectivity of the pixels without snow were selected for comparison before and after correction. The reflectivity variance of the corrected data was reduced compared with that of the data before correction, as shown in Tables 1 and 2, which reduced the error caused by the angle effect.

[0062] Table 1 Comparison of reflectivity of FY-4A snow-free pixels before and after correction

[0063]

[0064] Table 2 Comparison of reflectivity of snow pixels in FY-4A before and after correction

[0065]

[0066] Step 20, perform maximum brightness temperature fusion on the multi-temporal data, and the 10μm band brightness temperature corresponding to the pixel value of the synthesized image is the maximum value of the 10μm band brightness temperature in the non-night phase among multiple phases of the day, that is, the pixel value of the synthesized image is the value under the phase condition when the 10μm band brightness temperature is at the maximum value in the non-night phase of the day, including the brightness temperature of other bands and the reflectivity of each band at this time.

[0067] The biggest feature of geostationary orbit satellites is their high temporal resolution, which can almost achieve real-time observation of ground objects. The temporal resolution of the three geostationary orbit satellites is basically within 60 minutes, and the data available every day can reach 8 to 20 scenes. Generally speaking, clouds are fluid and move all the time, and the advantage of the high temporal resolution of geostationary satellites can be used to reduce the impact of clouds through the characteristics of cloud flow. According to the reflectivity spectrum curve of the ground objects, it can be known that the brightness temperature of clouds in the far (thermal) infrared band is lower than that of other ground objects. The present invention uses this feature and uses the maximum brightness temperature synthesis algorithm to fuse multi-phase data, thereby minimizing the cloud cover of the day.

[0068] like Figure 2The figure shows the principle diagram of the maximum brightness temperature synthesis algorithm. The purpose of the algorithm is to make the brightness temperature of the 10μm band corresponding to the pixel value of the synthesized image the maximum value of the brightness temperature of the 10μm band in multiple phases of the day in non-night phases. For example, if you need to fill a value in the pixel numbered 1, by comparing the brightness temperatures of multiple phases corresponding to pixel number 1, take the value of pixel number 1 of the image corresponding to the maximum brightness temperature, and finally assign the value to the fused image. If the ground object corresponding to a certain pixel is a cloud in all images, this method cannot completely remove the cloud, such as the southeastern part of the Qinghai-Tibet Plateau and high-latitude areas in China are covered by clouds for a long time.

[0069] After using the maximum brightness temperature synthesis algorithm, it can be found that the cloud cover has been significantly reduced, such as Figure 3 Taking the fused data of FY-4A on November 15, 2019 as an example, through intuitive comparison, the cloud cover in some areas has decreased due to the synthesis of multi-temporal data, but the declouding effect is not obvious for the situation covered by clouds all day. As shown in Table 4, the declouding effect of the maximum brightness temperature synthesis algorithm is evaluated by quantitative calculation method. The final fused image removes 35.11% of the cloud cover in the original multi-temporal image on average, and the remaining cloud cover accounts for only 19.72% of the entire disk. The same processing was done on MSG data and GOES-R data, and it was found that the average declouding rate reached 32.86% and 34.33%. Therefore, in theory, using this algorithm to reduce cloud cover is of great significance for monitoring snow accumulation.

[0070] Table 3 Evaluation table of cloud removal effect of FY-4A maximum brightness temperature synthesis

[0071]

[0072]

[0073] Step 30: Select snow accumulation discrimination factors, and determine discrimination thresholds of the discrimination factors based on the pixel values.

[0074] The introduced snow identification factors include: brightness temperature near the 10μm band, normalized difference snow index NDSI, brightness temperature near the 12μm band, brightness temperature near the 3μm band, reflectivity near the 1.36μm band and reflectivity near the 1.56μm band.

[0075] The present invention improves the snow cover identification factor in the original model and further determines the threshold value according to the characteristics of the three geostationary orbit satellite data and the areas monitored.

[0076] Preferably, the present invention provides a method for selecting a snow accumulation discrimination factor and a threshold thereof, such as Figure 4 As shown,

[0077] Step 31, select training samples, preferably use MODIS images, and determine the snow cover range through visual translation;

[0078] Step 32, training a snow accumulation discriminant factor, including selecting a pixel value used as a snow accumulation discriminant factor, drawing a ground feature separation map, and determining a threshold range;

[0079] Step 33: Use the selected snow discriminant factor and threshold to construct a discriminant for snow range recognition using a multi-threshold decision tree.

[0080] It should be noted that different decision tree identification algorithms can be designed for different satellites. Because the idea of ​​establishing a decision tree is the same, all of them select high-resolution cloud-free data as the true value, manually select training samples, and establish a multi-threshold decision tree snow identification method, so the idea is the same. Decision tree is a classification method used in many classification systems, but decision trees are built in a targeted manner. Our method is to build a training data set of satellite data for each of the three selected satellites and calculate the decision tree threshold for each satellite.

[0081] A global snow range extraction method based on geostationary satellites, according to the reflectivity spectrum characteristics of ground objects, the reflectivity of snow in the visible light band of 0.4μm to 0.8μm is significantly higher than the reflectivity of ground objects such as vegetation and soil in the same band range, which can well distinguish snow from other ground objects except clouds. In the visible light band range, the reflectivity of clouds is similar to the reflectivity spectrum characteristics of snow in overall characteristics, which has a great impact on snow identification. However, in the near-infrared band near 1.36μm, there is a large difference in the reflectivity of snow and clouds, so the reflectivity information of the band within the range of 1.36μm to 1.39μm can be used to achieve the purpose of distinguishing clouds from snow. However, for MSG data, there is no band of this wavelength. Based on the brightness temperature difference between 12.0μm and 3.9μm, the present invention introduces the normalized difference snow index (NDSI) to effectively distinguish clouds and snow, and establishes a global snow identification model suitable for geostationary satellite data.

[0082] The specific calculation formula of NDSI is:

[0083]

[0084] In the formula, Green and SWIR1 represent the reflectivity of the visible light band (band near 0.56μm) and the near-infrared band (band near 1.56μm) of geostationary satellite data, respectively. For snow-free surfaces, the brightness temperature of the 10μm band can be used for effective distinction. For water bodies, on the one hand, according to the reflectivity spectrum curve of surface characteristics, the reflectivity of water bodies near the 1.5μm band is low, and this feature can be used to distinguish water from snow. On the other hand, there is also a large difference in the brightness temperature near the 10μm band between water bodies and snow.

[0085] Step 40: Process the pixel values ​​of the geostationary orbit satellite according to the selected snow accumulation discriminant factors and the threshold values ​​corresponding to the discriminant factors to extract the snow accumulation range.

[0086] According to the spectral characteristics and basic model analysis of FY-4A satellite data, factors are selected for the snow cover identification algorithm of FY-4A satellite data. A total of 60,989 samples were selected using FY-4A data in November 2019, including snow samples (underlying surface is non-vegetated), snow samples (underlying surface is vegetated), cloud samples, water samples and snow-free surface samples. According to the statistical analysis results, the factors in the FY-4A snow cover identification algorithm are as follows: T10 is the brightness temperature of the 10.8μm band, (T12-T3) is the difference between the brightness temperatures of the 12.3μm band and the 3.8μm band, the normalized snow index NDSI, B4 is the reflectivity of the 1.36-1.39μm band, and B5 is the reflectivity of the 1.56-1.65μm band. The snow cover identification algorithm of FY-4A satellite data is shown in Table 4.

[0087] Table 4 Discriminant formula for extracting snow area of ​​FY-4A

[0088]

[0089]

[0090] The MSG data is different from the other two data in band setting, and it does not set the 1.36-1.39μm band. Therefore, the FY-4A snow cover identification model needs to be improved when distinguishing between clouds and snow. By selecting 83,134 samples of various types and performing statistical analysis, the factors in the MSG snow cover identification algorithm are: The factors in the MSG snow cover identification algorithm include: T10 is the brightness temperature of the 10.8μm band, (T12-T3) is the difference between the brightness temperatures of the 12.3μm band and the 3.8μm band, the normalized snow index NDSI, and B3 is the reflectivity of the 1.56-1.65μm band. The snow cover identification algorithm of MSG satellite data is shown in Table 5.

[0091] Table 5 Discriminant formula for extracting MSG snow range

[0092]

[0093] For GOES-16 satellite data, although the types of band settings are basically consistent with those of FY-4A satellite data, a large amount of sample data has verified that GOES-16 is sensitive to snow after calculating NDSI, that is, various types of snow can be identified through NDSI and some auxiliary identification factors. Therefore, as shown in Table 7, the snow cover identification model does not distinguish the underlying surface type when it is established. The factors in the snow cover identification algorithm of GOES-R satellite data are: brightness temperature T10 of the 10.8μm band, the difference T12-T3 between the brightness temperatures of the 12.3μm band and the 3.8μm band, the normalized snow cover index NDSI, the reflectivity B4 of the 1.36-1.39μm band, and the reflectivity B5 of the 1.56-1.65μm band. The snow cover identification algorithm of GOES-16 satellite data is shown in Table 6.

[0094] Table 6 Discriminant formula for extracting snow cover extent from GOES-16

[0095]

[0096]

[0097] According to the above snow cover identification criteria, daily global snow cover products can be produced. Figure 5 This is a global snow cover map as of November 15, 2019.

[0098] It should be noted that the band labels B1 to 5 are represented according to the band labels of each satellite. For example, in FY-4A and GOES16, the 1.56 to 1.65 um band is band 5 B5, but in MSG, the 1.56 to 1.65 um band is band 3 B3. Because the subsequent decision trees are processed for different satellites, the band symbols in the discriminants of Tables 4 to 6 are subject to the definitions of the corresponding satellites.

[0099] Step 50: Verify the accuracy of the extracted snow range

[0100] Meteorological station data is often used as an important data source for product accuracy verification. However, since the type of meteorological station data is point data, using station data to verify the accuracy of geostationary orbit satellite data may have problems such as excessive scale effect and inaccurate verification. Therefore, the present invention uses the MOD10A1 snow product as the true value to quantitatively verify the accuracy of the snow recognition algorithm.

[0101] Step 51: Accuracy verification method based on MOD10A1 data under clear sky conditions

[0102] The snow cover recognition accuracy test generally includes consistency test and inconsistency test. The consistency test refers to the similarity between the test result and the true value, that is, the similarity between the snow cover identification result of the geostationary satellite and the snow cover identification result of the MOD10A1 data. The consistency check includes the overall accuracy (OA) when there is no cloud and the overall accuracy (QA) when there is cloud. In general, when using MODIS data for accuracy verification, only the accuracy verification under cloudless conditions is considered, as shown in formula (8). The snow cover inconsistency test includes the underestimated error (IU) of snow cover and the overestimated error (IO) of snow cover. The underestimated error (IU) indicates the proportion of snow pixels identified as snow-free pixels by the geostationary satellite snow cover identification results, as shown in formula (9). The overestimated error (IO) indicates the proportion of snow-free pixels identified as snow pixels by the geostationary satellite snow cover identification results, as shown in formula (10).

[0103]

[0104]

[0105]

[0106] Where S1 is the number of pixels where both the MOD10A1 data snow cover identification result and the geostationary satellite snow cover identification result are snow covered, S2 is the number of pixels where both the MOD10A1 data snow cover identification result and the geostationary satellite snow cover identification result are not snow covered, D1 is the number of pixels where the MOD10A1 data snow cover identification result is not snow covered but the geostationary satellite snow cover identification result is snow covered, D2 is the number of pixels where the MOD10A1 data snow cover identification result is snow covered but the geostationary satellite snow cover identification result is not snow covered, and CA is the sum of the number of pixels involved in the calculation.

[0107] Step 52: Global area accuracy verification

[0108] Figure 6 shows the batch processing results of the accuracy verification of the experimental data using the method of this application. In the absence of clouds, the average overall accuracy is 93.96%, the overestimation error is 3.36%, and the underestimation error is 2.68%. The average values ​​of the overestimation error and the underestimation error are not much different, indicating that the application of the snow recognition algorithm is relatively stable. As can be seen from Figure 6(b), the errors have a certain regularity. In general, the overestimation error is greater than the underestimation error. The reason for this phenomenon is that, on the one hand, some snow pixels under cloud cover are identified through cloud flow, and on the other hand, it is caused by the misjudgment of clouds and snow. However, in some cases, the underestimation error is greater than the overestimation error, such as on December 8, 2019. This may be mainly due to the algorithm's misidentification of clouds and snow, which leads to misjudgment of snow pixels.

[0109] When calculating the global overall accuracy, the area value is used as the weight for weighted averaging. The Southern Hemisphere occupies 40% of the total area, but there is snow in the Southern Hemisphere except for a small part of South America. There is basically no snow in other areas. This method is easier to identify snow-free surfaces, which may lead to an overestimation of the overall accuracy. Therefore, the accuracy of the snow recognition algorithm is verified for each satellite and each satellite monitoring different continents.

[0110] Step 53: Accuracy verification of different geostationary orbit satellites

[0111] The present invention uses three different geostationary orbit satellites to extract the global snow cover range, and different sensors have different effects on the accuracy of snow mapping. The influence of different satellites on the accuracy of snow recognition is analyzed here, and the error percentage (IU, IO) and overall accuracy (OA) of the three geostationary orbit satellites under clear sky conditions are calculated respectively. Figure 7 shows the overall accuracy, overestimation error and underestimation error of snow recognition of different geostationary satellites under clear sky conditions. The average overall accuracy of the three geostationary orbit satellites is 94.995% (GOES-16), 94.134% (MSG) and 93.017% (FY-4A) respectively. Among the three sets of data, the overall accuracy of snow recognition of GOES-16 satellite is the best, and the overall accuracy of snow recognition of FY-4A satellite is the worst. However, it can be clearly seen from the overall accuracy verification diagram that the overall accuracy of GOES-16 data and FY-4A data is more stable than the overall accuracy of MSG data.

[0112] Figure 8 shows the overall accuracy, overestimation error, and underestimation error of different continents under clear sky conditions. In general, on the one hand, in the selected research data, the overall accuracy of the southern hemisphere is higher than that of the northern hemisphere. Among the continents in the southern hemisphere, except for South America, which has a small amount of snow, other regions are basically snow-free surfaces. However, the snow recognition algorithm is easier to identify snow-free pixels, which easily leads to an overestimation of the overall accuracy. Although the snow recognition accuracy in the northern hemisphere is lower, the overall average accuracy still reaches 90.96%, which proves the reliability of the algorithm. On the other hand, for the continents in the northern hemisphere, the average overall accuracy of snow recognition is 92.96% (North America), 91.69% (Asia), and 81.75% (Europe). Obviously, the snow recognition accuracy in Europe is the lowest, which is contrary to the previous results. This is mainly due to the fact that the overall accuracy in Africa is too high, which leads to an erroneous estimation of the MSG snow recognition accuracy.

[0113] Through the above analysis, the snow identification accuracy of the three geostationary orbit satellites is from high to low, namely GOES-16, FY-4A and MSG. Since the MSG satellite has a small number of bands, it does not have a near-infrared band specifically for distinguishing cirrus clouds, and chooses to use other factors instead of it in the algorithm to distinguish between clouds and snow, resulting in the lowest overall accuracy of the MSG data monitoring area.

[0114] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0115] Therefore, the present application also proposes a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method described in any embodiment of the present application is implemented.

[0116] Furthermore, the present application also proposes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in any embodiment of the present application when executing the computer program.

[0117] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0118] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0119] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0120] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0121] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0122] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0123] The present invention develops snow cover area products for the global land area (60°N-60°S, 180°W-180°E), and proposes a global geostationary orbit satellite snow cover range extraction method using the new generation geostationary orbit satellite FY-4A data, MSG data and GOES-16 data. The fusion algorithm of geostationary orbit satellite data is implemented based on the maximum brightness temperature synthesis algorithm, and a new snow cover identification factor is introduced. A global snow cover identification decision tree is established for the different spectral characteristics and monitoring areas of the three types of data, and the MOD10A1 (V6 version) data is used as the true value to verify the effectiveness of the method.

[0124] It should be noted that the present invention utilizes the characteristic that the brightness temperature of clouds in the far (thermal) infrared band is lower than that of other landforms, and proposes a maximum brightness temperature synthesis algorithm to fuse multi-temporal data, thereby minimizing the cloud cover of the day. At present, there is no global snow cover product from geostationary orbit satellites, even if it is reduced to a region, there is no snow cover product using FY4A and GOES16. The present invention designs different decision tree snow recognition algorithms for different satellites and produces a global snow cover map.

[0125] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0126] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for determining global snow cover based on geostationary satellites, characterized in that: The following steps are involved: Perform data preprocessing, including reprojecting and resampling geostationary satellite data according to the same standards, and calculating the reflectivity and brightness temperature of each band; The maximum brightness temperature fusion is performed on the multi-temporal data. The brightness temperature of the 10μm band corresponding to the pixel value of the synthesized image is the maximum value of the brightness temperature of the 10μm band in the non-night phase among multiple phases of the day. Select snow cover discriminant factors of FY-4A, MSG or GOES-16 satellites, and determine the discrimination thresholds of each discriminant factor based on the pixel values; The snow cover discrimination factors of FY-4A or GOES-16 satellites include: brightness temperature of 10.8μm band, difference between brightness temperature of 12.3μm band and 3.8μm band, normalized snow cover index, reflectivity of 1.36~1.39μm band, reflectivity of 1.56~1.65μm band; The snow cover discrimination factors of the MSG satellite include: brightness temperature of the 10.8μm band, the difference between the brightness temperatures of the 12.3μm band and the 3.8μm band, the normalized snow cover index, and the reflectivity of the 1.56~1.65μm band; According to the selected snow cover discrimination factors and the threshold values ​​corresponding to each discrimination factor, the pixel values ​​of the geostationary orbit satellite are processed to extract the snow cover range.

2. The method for determining global snow cover based on geostationary satellites as claimed in claim 1, characterized in that: The step of selecting snow accumulation discriminant factors and determining the discrimination threshold of each discriminant factor based on the pixel value further comprises: Select training samples, use MODIS images, and determine the snow range through visual translation; train snow discriminant factors, including selecting pixel values ​​used as snow discriminant factors, drawing ground feature separation maps, and determining threshold ranges; use the selected snow discriminant factors and thresholds to construct a discriminant for snow range identification using a multi-threshold decision tree.

3. The method for determining global snow cover based on geostationary satellites as claimed in claim 1, characterized in that: The reflectivity information of the band within the range of 1.36μm~1.39μm is used to distinguish between clouds and snow; or, based on the difference in brightness temperature between 12.0μm and 3.9μm, the normalized difference snow index is introduced to distinguish between clouds and snow.

4. The method for determining global snow cover based on geostationary orbit satellites as claimed in claim 1, characterized in that: In the step of processing the pixel values ​​of the geostationary satellite according to the selected snow accumulation discriminant factors and the threshold values ​​corresponding to the discriminant factors to extract the snow accumulation range, the discriminant formula for the FY-4A data is: Among them, T10 is the brightness temperature of the 10.8μm band, T12-T3 is the difference between the brightness temperatures of the 12.3μm band and the 3.8μm band, the normalized snow index NDSI, B4 is the reflectivity of the 1.36~1.39μm band, and B5 is the reflectivity of the 1.56~1.65μm band.

5. The method for determining global snow cover based on geostationary satellites as claimed in claim 1, characterized in that: In the step of processing the pixel values ​​of the geostationary orbit satellite according to the selected snow cover discriminant factors and the threshold values ​​corresponding to the discriminant factors to extract the snow cover range, the discriminant formula for Meteosat-8 data is: Among them, T10 is the brightness temperature of the 10.8μm band, T12-T3 is the difference between the brightness temperatures of the 12.3μm band and the 3.8μm band, the normalized snow cover index NDSI, and B3 is the reflectivity of the 1.56-1.65μm band.

6. The method for determining global snow cover based on geostationary satellites as claimed in claim 1, characterized in that: In the step of processing the pixel values ​​of the geostationary orbit satellite according to the selected snow cover discriminant factors and the threshold values ​​corresponding to the discriminant factors to extract the snow cover range, for the GOES-16 data, the discriminant formula is: Among them, the brightness temperature of the 10.8μm band is T10, the difference between the brightness temperatures of the 12.3μm band and the 3.8μm band is T12-T3, the normalized snow index NDSI, the reflectivity of the 1.36~1.39μm band is B4, and the reflectivity of the 1.56~1.65μm band is B5.

7. The method for determining global snow cover based on geostationary satellites as claimed in claim 1, characterized in that: The data preprocessing step also includes a step of geometrically correcting the number of geostationary orbit satellites, and / or, further includes correcting the angle effect of the geostationary orbit satellite data.

8. The method for determining global snow cover based on geostationary orbit satellites as claimed in claim 1, characterized in that: It also includes steps to verify the accuracy of the extracted snow cover extent: The consistency check includes the overall accuracy when there is no cloud and the overall accuracy when there is cloud; When using MODIS data for accuracy verification, consider accuracy verification under cloud-free conditions; Snow cover inconsistency tests include underestimation errors and overestimation errors of snow cover.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

10. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 8 is implemented.

Images