A snow cover acquisition method based on multi-source data fusion of Fengyun satellites

Abstract

The present invention belongs to the technical field of satellite snow remote sensing product processing, and specifically provides a method for obtaining snow cover by fusing multi-source data mainly from Fengyun satellites, including a Fengyun satellite data snow cover identification product S FY ; Get MODIS morning star and afternoon star snow cover composite product S MODIS ; Get the pre-processed snow product S IMS 、S RE 、S MW This method generates a preliminary fused snow cover product, a temporally filtered snow cover product, a spatially filtered snow cover product, a snow cover product corrected based on land surface observations, a snowline correction product for snow-covered areas, and a completely cloud-free daily snow cover product. This method fuses multi-source data to provide more snow cover information and remove cloud contamination from snow observation images, enabling high-precision, cloud-free, dynamic snow cover monitoring on a daily basis.

Description

Technical Field

[0001] The present invention belongs to the technical field of satellite snow remote sensing product processing, and in particular relates to a method for obtaining snow cover by fusing multi-source data based on Fengyun satellites. Background Art

[0002] Snow is a major component of the cryosphere, influencing processes such as global climate change, regional hydrological cycles, and the surface radiation budget. Snow can have long-lasting effects on atmospheric circulation and weather and climate systems by altering heat flux, thermodynamic processes, and vertical atmospheric motion. It can also modulate the spatiotemporal distribution of precipitation by affecting the strength of the monsoon, thereby affecting the distribution of water resources. In addition, snowmelt, as an important form of river runoff replenishment, can provide water for more than one-sixth of the world's population. Against the backdrop of global warming leading to the accelerated retreat of mountain glaciers and snowmelt, dynamic monitoring of snow distribution is becoming increasingly important for water resources planning and management, people's livelihoods and production, and socioeconomic development.

[0003] Currently, monitoring snow cover distribution relies primarily on meteorological observation stations, microwave remote sensing, and optical satellite remote sensing. Ground-based meteorological stations, through manual snow monitoring, can typically obtain relatively accurate snow depth data. However, this data is sparse, making observations difficult in complex terrain and inaccessible areas. Passive microwave remote sensing has the potential to penetrate clouds to observe snow cover, but its spatial resolution is coarse, making detailed observations difficult. In contrast, optical remote sensing satellites, with their higher temporal and spatial resolution, are the primary means of monitoring snow cover, accurately identifying snow-covered areas in clear weather. However, optical satellite remote sensing observations cannot penetrate clouds, resulting in loss of snow cover information due to cloud obscuration or detector saturation. Currently, optical remote sensing satellite snow cover monitoring methods primarily leverage the characteristics of optical imagery, such as those from MODIS sensors. Temporal filtering, spatial filtering, and elevation-based filtering are employed to remove cloud contamination from snow cover observation images, improve snow cover monitoring accuracy, and produce cloud-free snow cover products on a daily basis.

[0004] However, current snow cover monitoring methods still have some problems; the data source used is relatively single, usually based on optical images from MODIS sensors, and some methods add IMS snow and ice data as a supplement. However, when the observed image has a large area of ​​cloud pollution, there are still a lot of missing data and the spatial resolution decreases after cloud removal. The snow monitoring effect is still not ideal. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for obtaining snow cover by fusing multi-source data based on Fengyun satellites, which fuses multi-source data to provide more snow information and removes cloud pollution in snow observation images to achieve high-precision, cloudless dynamic snow cover monitoring on a daily basis.

[0006] To solve the above technical problems, the present invention adopts a technical solution: a method for obtaining snow cover by fusing multi-source data mainly from Fengyun satellites, comprising the following steps:

[0007] Step 1: Obtain the snow cover identification product S from Fengyun satellite data FY ;

[0008] Step 2: Select the meteorological observation stations in the region based on the quality of observation data and spatial representativeness, perform quality control on the snow cover observation data of the meteorological observation stations according to the quality control requirements of meteorological data business, and obtain the snow cover dataset S of the meteorological observation stations. station ;

[0009] Step 3: Obtain the MODIS morning and afternoon snow cover composite product S MODIS ;

[0010] Step 4: Obtain IMS snow and ice data, reanalysis datasets, and microwave data to obtain the preprocessed snow cover product S IMS 、S RE 、S MW ;

[0011] Step 5: Snow cover identification product S based on Fengyun satellite data FY Fusion of the Moderate Resolution Imaging Spectrometer's morning and afternoon star snow cover composite product S MODIS Generate preliminary snow cover fusion product S1;

[0012] Step 6: Given a time window range Δt, synthesize the preliminary snow cover fusion product S1 within the time window to generate the snow cover time filtered product S2;

[0013] Step 7: Given the spatial window ranges Δs1 and Δs2, the snow cover temporal filter product S2 is synthesized according to the spatial window synthesis rule to generate the snow cover spatial filter product S3;

[0014] Step 8: Given a time series range Δtm, according to the snow cover dataset S of the meteorological observation station station Estimate the conditional probability of snow cover in pixels and the total probability of snow / non-snow cover in the snow cover spatial filtering product S3, and generate the snow cover product S4 based on land surface observation corrections;

[0015] Step 9: Generate the snow cover regional snow line correction product S5 according to the regional snow line synthesis rules based on the snow cover product S4 corrected based on land surface observations;

[0016] Step 10: Compare the snow line correction product S5 and the snow cover product S IMS 、S RE 、SMW Fusion obtains a completely cloudless daily snow cover product S6.

[0017] In a preferred embodiment, in step 1, the method for obtaining the snow cover identification product from the Fengyun satellite data is as follows: first, spectral observation data from the Fengyun satellite medium-resolution spectral imager is obtained, a Fengyun satellite snow cover identification algorithm is constructed, and image pixels are preliminarily identified as cloud, land, snow, water, and sea / lake ice;

[0018] The identification rules of the Fengyun satellite snow identification algorithm are as follows:

[0019] 1) Cloud discrimination: Cloud pixel categories include cloud, cloud shadow, missing value, uncertain value, detector saturation, and fill value. The discrimination method is:

[0020] Cloud identification method:

[0021] ;

[0022] Cloud shadow identification method:

[0023] ;

[0024] Missing values, uncertain values, detector saturation, filled values, and other pixels that do not meet the following 2), 3), 4), and 5) identification categories are classified as cloud pixels;

[0025] 2) Land identification method:

[0026] ;

[0027] 3) Snow accumulation determination method:

[0028] ;

[0029] 4) Water body identification method:

[0030] ;

[0031] ;

[0032] 5) Sea / lake ice identification method:

[0033] ;

[0034] ;

[0035] ;

[0036] in, These are the image channel data of the near-infrared band, red light band, short-wave infrared band and far-infrared band of Fengyun satellites; The vegetation index NDVI and snow cover index NDSI calculated from the image data respectively; 、 、 、 、 、 、 、 、 、 、 are the empirical threshold parameters of the Fengyun satellite snow identification algorithm. The values ​​in brackets (0.85, 1.15, 0.3, 0.205, 0.05, -0.05, 0.2, 0.25, 0.1, 244 K, and 0.01) are commonly used empirical thresholds in the corresponding cloud, land, snow, water, and sea / lake ice identification methods, which are adjusted according to the actual conditions of satellite images and the study area.

[0037] In the preferred solution, in step 2, the snow accumulation dataset S of the meteorological observation station is obtained. station The following steps are involved:

[0038] Step S2.1: Filter meteorological observation stations within the region based on observation data quality and spatial representativeness. Extract multi-year snow cover, surface cover, terrain characteristics, snow environment characteristics, and the spatial representativeness evaluation level of snow cover observations at the meteorological stations within their surrounding windows. Select meteorological observation stations with high spatial representativeness within the region and obtain snow cover observation data.

[0039] Step S2.2: Perform quality control on the snow cover observation data of the meteorological observation station in accordance with the meteorological data business quality control requirements, including: checking the snow depth limit value and checking the internal consistency of snow depth, snow pressure, and minimum temperature;

[0040] Step S2.3: Define the snow depth observation record of the meteorological observation station as 999999 as missing, 0 as no snow, 999990 or snow depth less than 0.5 cm as partial snow, and snow depth greater than or equal to 0.5 cm as snow. Obtain the snow cover dataset S of the meteorological observation station. station .

[0041] In the preferred solution, in step 3, the MODIS morning star and afternoon star snow cover composite product S is obtained. MODIS , including the following steps:

[0042] Step S3.1, obtain the MODIS snow cover product afternoon star MYD10A1 and morning star MOD10A1, and convert The pixels are identified as snow. The pixels of inland water and ocean are reclassified as water; the pixels of cloud, missing data, no accuracy, undetermined, night and detector saturation are reclassified as cloud;

[0043] Step S3.2: The snow cover products of the afternoon star MYD10A1 and the morning star MOD10A1 are combined to generate a composite image of the morning star and the afternoon star according to the composite rule of the morning and afternoon stars. The composite rule of the morning star and the afternoon star is as follows: if a pixel in either MYD10A1 or MOD10A1 image contains snow, the pixel is identified as snow; if a pixel is a cloud in one image but not a snow pixel in the other image, the pixel is identified as a non-snow pixel; if one image product is missing on a given day, the other product is used alone;

[0044] Step S3.3: Use the majority algorithm to resample the composite image of the morning and afternoon stars to the spatial resolution of the Fengyun satellites to obtain the MODIS morning and afternoon star snow cover composite product S MODIS .

[0045] In a preferred embodiment, the step 4 comprises the following steps:

[0046] S4.1. Obtain IMS snow and ice data, reanalysis datasets, and microwave data, and use the majority algorithm to resample the snow cover product of the IMS snow and ice data to match the spatial resolution of the Fengyun satellites to generate a new snow cover product S. IMS ;

[0047] S4.2. Use interpolation algorithms to resample the snow depth and snow water equivalent of the reanalysis dataset or microwave data to be consistent with the spatial resolution of the Fengyun satellites. If the pixel snow depth or snow water equivalent meets , then the pixel is reclassified as snow; if , the pixel is reclassified as land, where Represent the pixel snow depth and snow water equivalent, are the corresponding empirical threshold parameters. The 0 in the bracket is the commonly used empirical threshold, which needs to be adjusted according to the actual situation. The snow cover product S is generated from the reanalysis dataset and microwave data respectively. RE and S MW .

[0048] In the preferred solution, in step 5, the regional snow environment characteristics are analyzed, and the snow cover identification product S is used based on the Fengyun satellite data. FY Based on the data, according to the multi-source satellite synthesis rules, the MODIS morning satellite and afternoon satellite snow cover synthesis product S MODIS , remove S FY The cloud pixels in the image are used to generate the preliminary fusion product S1;

[0049] The multi-source satellite synthesis rule is as follows: select a satellite image as the base data; if a pixel in any multi-source satellite image has snow, then the pixel is identified as snow; if a pixel is cloud in one image and non-snow in another image, then the pixel is identified as non-snow; other pixels are consistent with the base data; if the base data is missing a product on that day, only other satellite images are used. The expression formula is:

[0050] ;

[0051] ;

[0052] ;

[0053] The priority of pixel identification in the multi-source satellite synthesis rule follows:

[0054] ;

[0055] in, It's time Date, Location Pixel at It is a snow pixel; Is it a non-snow pixel value or a snow pixel value; is the cloud pixel value.

[0056] In the preferred solution, in step 6, the synthesis rule of the preliminary snow cover fusion product S1 within the time window is: if the current day is t, then select the snow cover images S1 of t-1 and t+1. If a pixel on day t is a cloud, and both t-1 and t+1 are non-snow pixels or snow pixels, then the pixel on day t is identified as a non-snow pixel or snow pixel; and so on, select t-1 and t+2, t-2 and t+1... until t-Δt1 and t+Δt2, where the time window size Δt=Δt1+Δt2. According to the above rules, the cloud pixels of t are reclassified to generate the snow cover time filtered product S2, which is expressed as:

[0057] ;

[0058] in, 、 、 The locations are The pixels at time t, t-Δt1, and t+Δt2.

[0059] In the preferred solution, in step seven, the spatial window synthesis rule is as follows: with the cloud pixel of the image as the central pixel, if snow pixels account for more than half of the pixels in the nearby spatial window Δs1, the central pixel is identified as a snow pixel; if non-snow pixels account for more than half of the pixels, the central pixel is identified as a non-snow pixel; with the remaining cloud pixel as the central pixel, if there is a snow pixel in the nearby spatial window Δs2 with an altitude lower than that of the central pixel, the central pixel is identified as a snow pixel; otherwise, the central pixel remains a cloud pixel. The expression is:

[0060] ;

[0061] ;

[0062] in, 、 For location 、 The altitude of It's location The pixel at time t.

[0063] In a preferred solution, in step eight, within the time Δtm, the conditional probability and total probability of the pixel are calculated as follows:

[0064] ;

[0065] ;

[0066] in, Indicates time ,Location Pixel at , and satisfy the time Within the selected time range Δtm; Indicates time , No. Weather station observations, and meeting the time Within the selected time range Δtm; Is it a non-snow pixel value or a snow pixel value; It's location The total number of valid pixels at is the conditional probability between the pixel and the site; is the total probability of the pixel;

[0067] According to the conditional probability and total probability of cloud pixels, the pixel is judged to be snow or non-snow. The judgment rule expression is: .

[0068] In a preferred embodiment, in step nine, the method for generating the snow line correction product S5 for the snow-covered area is as follows: dividing the space into multiple sub-areas according to watershed and topographic factors, calculating the average elevation demS of all snow pixels and the average elevation deml of all land pixels in each sub-area of ​​the snow cover product S4 corrected based on land surface observations on a daily basis; if the cloud cover in the sub-area is greater than Cp or the snow cover is less than Sp, where Cp and Sp are the area percentages of clouds and snow in the watershed, respectively, demS and deml are not retained; otherwise, demS and deml are retained; calculating the demS and deml of each sub-area each year, performing time series interpolation on them, and synthesizing the snow line correction product S5 for the snow-covered area according to the regional snow line synthesis rule;

[0069] The regional snowline synthesis rule is: if the elevation of the cloud pixel in the sub-region is greater than demS and the slope is If the elevation of the cloud pixel in the sub-region is less than deml, the cloud pixel is identified as snow. If the elevation of the cloud pixel in the sub-region is less than deml, the cloud pixel is identified as land. The remaining pixels remain unchanged. The expression is:

[0070] ;

[0071] .

[0072] In the preferred solution, in step 10, the specific operation method is: if at the cloud pixel of S5, S IMS 、S RE 、S MW If there is snow in any product, the cloud pixel is identified as snow, otherwise it is identified as a non-snow pixel. The expression is:

[0073] .

[0074] The present invention provides a method for obtaining snow cover by fusing multi-source data based on Fengyun satellites, which has the following beneficial effects:

[0075] 1. This paper uses Fengyun meteorological satellites as the main data source, integrating meteorological observation data, reanalysis data, MODIS, microwave datasets and other multi-source data to extract snow cover. It fully taps the potential of Fengyun satellites in snow cover monitoring, integrates rich multi-source data to provide more snow cover information, removes cloud contamination in snow observation images, and realizes high-precision, cloud-free and dynamic snow cover monitoring on a daily basis.

[0076] 2. This paper leverages information from multi-source remote sensing data, including snow cover products from domestic Fengyun satellites, as well as measured snow depth data from meteorological stations and reanalysis snow cover datasets. Based on the spatiotemporal characteristics of snow cover, a multi-source data fusion method based on Fengyun satellites is constructed. By calculating the probability of snow / no snow conditions from pixels to meteorological stations and estimating regional snowline elevations over time series, this method reclassifies cloud cover or missing data in snow cover products, removing cloud effects and filling in missing remote sensing information. This method generates high-quality, cloud-free snow cover products and enables high-precision dynamic snow cover monitoring.

[0077] 3. The daily cloudless snow cover dataset of this invention, which is mainly based on domestic Fengyun satellite data, has important reference value in the development and application of independent equipment. It can provide snow data support for scientific research such as climate analysis and runoff prediction, and provide a scientific basis for government work such as comprehensive utilization of water resources and disaster prevention and mitigation. It has significant scientific value and economic benefits.

[0078] 4. This invention establishes a land surface observation correction method based on snow cover data from meteorological stations to improve snow cover products. Based on the environmental characteristics of meteorological stations within a region and their surrounding window areas, meteorological stations with high data quality and spatial representativeness are selected. The snow cover characteristics of pixels and stations are associated by calculating the snow / no snow conditional probability from the pixel to the meteorological station. Using the measured snow cover data at the station and the total snow cover probability of the pixel, cloud pixels in the image are reclassified, reducing cloud cover and improving data quality.

[0079] 5. This invention has constructed a snow cover product fusion technology based on multi-source data, mainly Fengyun satellite data. Relying on multi-source remote sensing data, it integrates technical methods such as spatiotemporal filtering, land surface observation correction, and regional snow line elevation to further improve the quality of domestic Fengyun satellite data and the ability to apply remote sensing monitoring of water resources such as snow. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] The present invention will be further described below with reference to the accompanying drawings and examples:

[0081] Figure 1 Flowchart of the present invention;

[0082] Figure 2 This is a schematic diagram of the judgment logic of step 1 of the present invention;

[0083] Figure 3 It is a logic block diagram of the present invention;

[0084] Figure 4 This is a snow cover product image obtained by fusion of multi-source data on October 13, 2019, according to an embodiment of the present invention;

[0085] Figure 5This is the MODIS snow product image on October 13, 2019, according to an embodiment of the present invention;

[0086] Figure 6 This is the IMS snow and ice data chart for October 13, 2019, according to an embodiment of the present invention;

[0087] Figure 7 This is a map of Landsat8, multi-source fusion snow cover product, MODIS snow cover product, and IMS snow and ice data on October 13, 2019, according to an embodiment of the present invention;

[0088] Figure 8 This is a snow cover product image obtained by fusion of multi-source data on January 20, 2019, according to an embodiment of the present invention;

[0089] Figure 9 This is the MODIS snow product image on January 20, 2019, according to an embodiment of the present invention;

[0090] Figure 10 This is the IMS snow and ice data chart for January 20, 2019, according to an embodiment of the present invention;

[0091] Figure 11 This is a map of Landsat8, multi-source fusion snow cover product, MODIS snow cover product and IMS snow and ice data on January 20, 2019, according to an embodiment of the present invention. DETAILED DESCRIPTION

[0092] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0093] refer to Figures 1 to 3 As shown in the figure, a snow cover acquisition method based on multi-source data fusion of Fengyun satellites is used to remove the cloud from the snow cover product of the upper Jinsha River basin, and finally obtain a daily cloud-free product. Specifically, it includes:

[0094] Step 1: Obtain MERSI spectral observation data from FY-3D in the upper Jinsha River basin. According to the Fengyun satellite snow cover identification algorithm and combined with land cover classification data, the image pixels are identified as land (code 25), snow (code 200), water (code 37), ocean (code 39), sea / lake ice (code 100), and cloud (code 50). Pixels with missing information (code 0), uncertain value (code 1), detector saturation (code 254), and fill value (code 255) are all classified as clouds, and the Fengyun satellite data snow cover identification product S is obtained. FY .

[0095] Step 2: Screen the meteorological observation stations in the study area based on the quality and spatial representativeness of the observation data. A total of 11 meteorological observation stations in the upper reaches of the Jinsha River were selected. The snow observation data of the meteorological observation stations were quality controlled according to the meteorological data business quality control requirements, including snow depth limit value inspection and internal consistency inspection of snow depth, snow pressure and minimum temperature. The snow depth observation record of the meteorological observation station with a value of 999999 is defined as missing, 0 as no snow (land), 99990 or snow depth less than 0.5 cm as partial snow, and snow depth greater than 0.5 cm as snow. The snow data set S of the meteorological observation station was obtained. station .

[0096] Step 3: Obtain the MODIS (Moderate-resolution Imaging Spectroradiometer) snow cover products MYD10A1 and MOD10A1, identify pixels as snow (coded 0), land (coded 25), water (coded 37), ocean (coded 39), sea / lake ice (coded 100), and cloud (coded 50), and then generate a MODIS morning and afternoon star composite image according to the morning and afternoon star composite rules.

[0097] Step 4: Use the majority algorithm to resample the snow cover product of the IMS (interactive multisensors snow and ice mapping system) snow and ice data to match the spatial resolution of the Fengyun satellites, and generate a new IMS snow cover product.

[0098] The mode algorithm is a simple algorithm that selects the value that occurs most frequently. The mode is the value that occurs most frequently in a set of data. In different types of statistical analysis, the mode may be used to describe the central tendency of data or the asymmetry of the distribution.

[0099] Step 5: Snow cover identification product S based on Fengyun satellite data FY The MODIS Morning and Afternoon Snow Cover Product is fused according to multi-source satellite synthesis rules, removing cloud pixels from the Wind and Cloud Snow Cover Product. If a pixel in any image contains snow, the pixel is identified as snow. If a pixel is cloud in the Fengyun satellite imagery but not snow in the MODIS Morning and Afternoon Snow Cover Product, and the non-snow pixel is ground, water, or sea / lake ice, the pixel is also identified as non-snow. If the base data is missing for the day, only the MODIS Morning and Afternoon Snow Cover Product is used. This generates the preliminary snow cover fusion product S1.

[0100] Step 6: Synthesize the preliminary snow cover fusion product S1 within 3 days to generate the snow cover time filtered product S2. The rules are as follows.

[0101] If the current day is t, select the snow cover images from t-1 and t+1. If a pixel on day t is cloud, but both t-1 and t+1 are non-snow or snow-covered, then the pixel on day t is classified as non-snow or snow-covered. Then, select the products from t-1 and t+2, and t-2 and t+1, and continue reclassifying the cloud pixels from t according to the above rules to generate the snow cover temporal filtered product S2.

[0102] Step 7: Generate the snow cover spatial filtering product S3 by combining the snow cover temporal filtering product S2 according to the following spatial synthesis rules.

[0103] Taking the cloud pixel of the image as the central pixel, if any category of pixels (snow, land, water, sea / lake ice) accounts for more than half of the pixels in its four neighboring areas, the central pixel is identified as a pixel of that category; taking the remaining cloud pixel as the central pixel, if there is a snow pixel in its eight neighboring areas with an altitude lower than the central pixel, the central pixel is identified as a snow pixel.

[0104] Step 8: Generate snow cover product S4 based on land surface observation correction.

[0105] Every month, the snow cover product S4, which is corrected based on land surface observations, is generated based on the conditional probability of snow cover in the pixel of the snow cover time filter product S3 at the meteorological observation station and the total probability of the pixel being covered by snow / not snow.

[0106] Step 9: Generate snow line correction product S5 for snow-covered areas.

[0107] The upper Jinsha River is divided into four subregions based on factors such as watershed and topography. Within each subregion, the mean elevation demS of all snow pixels and the mean elevation demL of all land pixels are calculated daily for the snow cover product S4, which is corrected based on land surface observations. If the cloud cover in a subregion is greater than Cp = 70% or the snow cover is less than Sp = 5%, demS and deml are discarded; otherwise, they are retained. demS and deml are calculated for each subregion each year, and the time series are interpolated to create the snow cover regional snow line correction product S5 according to the regional snow line synthesis rules.

[0108] Step 10: Generate daily cloudless snow cover product S6.

[0109] The snowline correction product S5 for snow-covered areas is fused with the IMS snow cover product. If there is snow in the IMS snow cover product at a cloud pixel in S5, the cloud pixel is identified as snow; otherwise, it is identified as a non-snow pixel. The final result is the completely cloud-free daily snow cover product S6.

[0110] In order to evaluate the accuracy of the snow cover acquisition method based on multi-source data fusion of Fengyun satellites proposed in this paper, and to test the accuracy of the snow cover product fused with multi-source data, the snow area accuracy index was adopted, and the 30m resolution snow image of the Landsat8 satellite was used as the reference data. The fused product, MODIS snow cover product and IMS product were compared respectively.

[0111] Accuracy indicators include overall classification accuracy (Accuracy), recall rate (Recall), and snow classification accuracy (Precision).

[0112]

[0113] Total classification accuracy (Accuracy) = (TP + TN) / (TP + FN + FP + TN);

[0114] Recall = TP / (TP+FN);

[0115] Snow classification accuracy (precision) = TP / (TP+FP);

[0116] MODIS snow cover products are processed according to the morning and afternoon star composite, time filter composite (time window is 3 days), spatial filter composite (spatial windows are 4 neighborhoods and 8 neighborhoods respectively) and regional snowline composite rules (no time series interpolation is performed, only single-day images are processed).

[0117] The verification data used includes 79 Landsat 8 images of the upper Jinsha River basin in 2019. The final verification results are shown in the table below, where the average cloud cover and day percentage are calculated based on the product image results for the entire year of 2019.

[0118]

[0119] Overall, a multi-source data fusion method for obtaining snow cover using Fengyun satellites as the primary source can completely remove cloud contamination from snow cover products while maintaining good snow area accuracy. From the initial fused product S1 to the final cloud-free snow cover product S6, the snow area accuracy remains largely stable after each processing step, with overall classification accuracy, snow classification accuracy, and recall rates around 0.91, 0.84, and 0.91, respectively. Cloud cover gradually decreases, with the temporal filtering and regional snowline correction steps contributing the most to cloud removal, resulting in a reduction of over 16% in average cloud cover and over 30% in the number of days with cloud cover greater than 20%.

[0120] The multi-source data fusion snow cover product S6 shows similar overall classification accuracy, snow classification accuracy, and recall compared to the cloud-covered MODIS snow cover product. Compared to the cloudless IMS snow and ice product, the multi-source data fusion snow cover product shows improvements of 0.10, 0.16, and 0.04 in overall classification accuracy, recall, and snow classification accuracy, respectively. This demonstrates that the multi-source data fusion snow cover product offers significant advantages over the MODIS snow cover and IMS snow and ice products, providing a highly accurate cloud-free snow cover product for dynamic monitoring of snow cover distribution and snowmelt runoff forecasting.

[0121] like Figures 4-6 As shown in Figure 2, the spatial distribution of the multi-source data fusion snow cover product and the MODIS snow cover product is relatively consistent, but the MODIS snow cover product is obscured by clouds by about 13.3%, and the snow area identified by the IMS snow and ice data is significantly smaller. Figure 7 As shown, this is a comparison of the Landsat 8 image at the source of the Jinsha River with the multi-source fusion snow cover product, the MODIS snow cover product, and the IMS snow and ice data. It can be found that the IMS snow and ice data has poor details in the spatial distribution of snow cover, a high local missed detection rate, and poor accuracy.

[0122] like Figures 8-10 The following table shows the snow cover product on January 20, 2019. Due to good observation conditions, the cloud cover of the MODIS snow cover product is relatively low, and the spatial distribution of snow cover is not much different from the multi-source data fusion snow cover product. However, the IMS snow and ice product has significantly more snow in the middle and lower reaches of the implementation area. Figure 11 In the local detail map, the multi-source data fusion snow cover product, the MODIS snow cover product, and the Landsat 8 NDSI image are relatively consistent, while the IMS product differs significantly from Landsat 8. This indicates that the IMS product's snow cover identification is relatively unstable, while the multi-source data fusion snow cover product's snow cover identification is more stable and reliable.

[0123] In summary, the present invention proposes a snow cover acquisition method based on multi-source data fusion of Fengyun satellites, which comprehensively considers the spatiotemporal distribution characteristics of snow from ground meteorological observation stations and reanalysis datasets. While maintaining the temporal resolution of the snow product at 1 day and the spatial resolution of 1 km, it achieves the complete removal of cloud contamination in the snow product and ensures high product accuracy, effectively making up for the shortcomings of binary snow products, ensuring product reliability and availability, and improving the accuracy of snow monitoring.

[0124] The contents not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A method for obtaining snow cover by fusion of multi-source data mainly from Fengyun satellites, characterized in that: The following steps are involved: Step 1: Obtain the snow cover identification product S from Fengyun satellite data FY ; Step 2: Select the meteorological observation stations in the region based on the quality of observation data and spatial representativeness, perform quality control on the snow cover observation data of the meteorological observation stations according to the quality control requirements of meteorological data business, and obtain the snow cover dataset S of the meteorological observation stations. station ; Step 3: Obtain the MODIS morning and afternoon snow cover composite product S MODIS ; Step 4: Obtain IMS snow and ice data, reanalysis datasets, and microwave data to obtain the preprocessed snow cover product S IMS 、S RE 、S MW ; Step 5: Snow cover identification product S based on Fengyun satellite data FY Fusion of the Moderate Resolution Imaging Spectrometer's morning and afternoon star snow cover composite product S MODIS Generate preliminary snow cover fusion product S1; Step 6: Given a time window range Δt, synthesize the preliminary snow cover fusion product S1 within the time window to generate the snow cover time filtered product S2; Step 7: Given the spatial window ranges Δs1 and Δs2, the snow cover temporal filter product S2 is synthesized according to the spatial window synthesis rule to generate the snow cover spatial filter product S3; Step 8: Given a time series range Δtm, according to the snow cover dataset S of the meteorological observation station station Estimate the conditional probability of snow cover in pixels and the total probability of snow / non-snow cover in the snow cover spatial filtering product S3, and generate the snow cover product S4 based on land surface observation corrections; Step 9: Generate the snow cover regional snow line correction product S5 according to the regional snow line synthesis rules based on the snow cover product S4 corrected based on land surface observations; Step 10: Compare the snow line correction product S5 and the snow cover product S IMS 、S RE 、S MW Fusion obtains a completely cloudless daily snow cover product S6.

2. The method for obtaining snow cover by fusion of multi-source data based on Fengyun satellites according to claim 1 is characterized in that: In step 1, the method for obtaining the snow cover identification product from the Fengyun satellite data is as follows: first, spectral observation data from the Fengyun satellite medium-resolution spectral imager is obtained, a Fengyun satellite snow cover identification algorithm is constructed, and image pixels are preliminarily identified as cloud, land, snow, water, and sea / lake ice; The identification rules of the Fengyun satellite snow identification algorithm are as follows: 1) Cloud discrimination: Cloud pixel categories include cloud, cloud shadow, missing value, uncertain value, detector saturation, and fill value. The discrimination method is: Cloud identification method: ； Cloud shadow identification method: ； Missing values, uncertain values, detector saturation, filled values, and other pixels that do not meet the following 2), 3), 4), and 5) identification categories are classified as cloud pixels; 2) Land identification method: ； 3) Snow accumulation determination method: ； 4) Water body identification method: ； ； 5) Sea / lake ice identification method: ； ； ； in, These are the image channel data of the near-infrared band, red light band, short-wave infrared band and far-infrared band of Fengyun satellites; The vegetation index NDVI and snow cover index NDSI calculated from the image data respectively; 、 、 、 、 、 、 、 、 、 、 are the empirical threshold parameters of the Fengyun satellite snow identification algorithm. The values ​​in brackets (0.85, 1.15, 0.3, 0.205, 0.05, -0.05, 0.2, 0.25, 0.1, 244 K, and 0.01) are commonly used empirical thresholds in the corresponding cloud, land, snow, water, and sea / lake ice identification methods, which are adjusted according to the actual conditions of satellite images and the study area.

3. The method for obtaining snow cover by fusion of multi-source data based on Fengyun satellites according to claim 1 is characterized in that: In step 2, obtain the snow data set S of the meteorological observation station station The following steps are involved: Step S2.1: Filter meteorological observation stations within the region based on observation data quality and spatial representativeness. Extract multi-year snow cover, surface cover, terrain characteristics, snow environment characteristics, and the spatial representativeness evaluation level of snow cover observations at the meteorological stations within their surrounding windows. Select meteorological observation stations with high spatial representativeness within the region and obtain snow cover observation data. Step S2.2: Perform quality control on the snow cover observation data of the meteorological observation station in accordance with the meteorological data business quality control requirements, including: checking the snow depth limit value and checking the internal consistency of snow depth, snow pressure, and minimum temperature; Step S2.3: Define the snow depth observation record of the meteorological observation station as 999999 as missing, 0 as no snow, 999990 or snow depth less than 0.5 cm as partial snow, and snow depth greater than or equal to 0.5 cm as snow. Obtain the snow cover dataset S of the meteorological observation station. station .

4. The method for obtaining snow cover by fusion of multi-source data based on Fengyun satellites according to claim 1 is characterized in that: In step 3, the MODIS morning and afternoon snow cover composite product S is obtained. MODIS , including the following steps: Step S3.1, obtain the MODIS snow cover product afternoon star MYD10A1 and morning star MOD10A1, and convert The pixels are identified as snow. The pixels of inland water and ocean are reclassified as water; the pixels of cloud, missing data, no accuracy, undetermined, night and detector saturation are reclassified as cloud; Step S3.2: The snow cover products of the afternoon star MYD10A1 and the morning star MOD10A1 are combined to generate a composite image of the morning star and the afternoon star according to the composite rule of the morning and afternoon stars. The composite rule of the morning star and the afternoon star is as follows: if a pixel in either MYD10A1 or MOD10A1 image contains snow, the pixel is identified as snow; if a pixel is a cloud in one image but not a snow pixel in the other image, the pixel is identified as a non-snow pixel; if one image product is missing on a given day, the other product is used alone; Step S3.3: Use the majority algorithm to resample the composite images of the morning and afternoon stars to the spatial resolution of the Fengyun satellites to obtain the MODIS morning and afternoon star snow cover composite product S MODIS .

5. The method for obtaining snow cover by fusion of multi-source data based on Fengyun satellites according to claim 1 is characterized in that: The step 4 comprises the following steps: S4.

1. Obtain IMS snow and ice data, reanalysis datasets, and microwave data, and use the majority algorithm to resample the snow cover product of the IMS snow and ice data to match the spatial resolution of the Fengyun satellites to generate a new snow cover product S. IMS ; S4.

2. Use interpolation algorithms to resample the snow depth and snow water equivalent of the reanalysis dataset or microwave data to be consistent with the spatial resolution of the Fengyun satellites. If the pixel snow depth or snow water equivalent meets , then the pixel is reclassified as snow; if , the pixel is reclassified as land, where Represent the pixel snow depth and snow water equivalent, are the corresponding empirical threshold parameters. The 0 in the bracket is the commonly used empirical threshold, which needs to be adjusted according to the actual situation. The snow cover product S is generated from the reanalysis dataset and microwave data respectively. RE and S MW .

6. The method for obtaining snow cover by fusion of multi-source data based on Fengyun satellites according to claim 1, characterized in that: In step 5, the regional snow environment characteristics are analyzed and the snow cover identification product S is used to identify the snow cover of the Fengyun satellite data. FY Based on the data, according to the multi-source satellite synthesis rules, the MODIS morning satellite and afternoon satellite snow cover synthesis product S MODIS , remove S FY The cloud pixels in the image are used to generate the preliminary fusion product S1; The multi-source satellite synthesis rule is as follows: select a satellite image as the base data; if a pixel in any multi-source satellite image has snow, then the pixel is identified as snow; if a pixel is cloud in one image and non-snow in another image, then the pixel is identified as non-snow; other pixels are consistent with the base data; if the base data is missing a product on that day, only other satellite images are used. The expression formula is: ； ； ； The priority of pixel identification in the multi-source satellite synthesis rule follows: ； in, It's time Date, Location Pixel at It is a snow pixel; Is it a non-snow pixel value or a snow pixel value; is the cloud pixel value.

7. The method for obtaining snow cover by fusion of multi-source data based on Fengyun satellites according to claim 6, characterized in that: In step 6, the synthesis rule of the preliminary snow cover fusion product S1 within the time window is as follows: if the current day is t, then select the snow cover images S1 of t-1 and t+1. If a pixel on day t is a cloud, and both t-1 and t+1 are non-snow pixels or snow pixels, then the pixel on day t is identified as a non-snow pixel or snow pixel; and so on, select t-1 and t+2, t-2 and t+1, and so on until t-Δt1 and t+Δt2, where the time window size Δt = Δt1 + Δt2. According to the above rules, the cloud pixels of t are reclassified to generate the snow cover time filtered product S2, which is expressed as: ； in, 、 、 The locations are Place, time t day, t-Δt1 day, t+Δt2 Pixel of the day.

8. The method for obtaining snow cover by fusion of multi-source data based on Fengyun satellites according to claim 7, characterized in that: In step 7, the spatial window synthesis rule is as follows: take the cloud pixel of the image as the central pixel, and if the snow pixels account for more than half in the spatial window Δs1 near it, then the central pixel is identified as a snow pixel; if the non-snow pixels account for more than half, then the central pixel is identified as a non-snow pixel; take the remaining cloud pixel as the central pixel, and if there is a snow pixel with a lower altitude than the central pixel in the spatial window Δs2 near it, then the central pixel is identified as a snow pixel; otherwise, the central pixel remains a cloud pixel. The expression is: ； ； in, 、 For location 、 The altitude of It's location The pixel at time t.

9. The method for obtaining snow cover by fusion of multi-source data based on Fengyun satellites according to claim 8, characterized in that: In step eight, within the time Δtm, the conditional probability and total probability of the pixel are calculated as follows: ； ； in, Indicates time ,Location Pixel at , and satisfy the time Within the selected time range Δtm; Indicates time , No. Weather station observations, and meeting the time Within the selected time range Δtm, Is it a non-snow pixel value or a snow pixel value; It's location The total number of valid pixels at is the conditional probability between the pixel and the site; is the total probability of the pixel; According to the conditional probability and total probability of cloud pixels, the pixel is judged to be snow or non-snow. The judgment rule expression is: ; in, It's time ,Location Pixel at For location Pixel at Relative to the Observations at weather stations The conditional probability of Is it a non-snow pixel value or a snow pixel value; Indicates time is t Time, Observations from weather stations; Indicates location The total probability that a pixel is snowy or non-snowy within the time range Δtm.

10. The method for obtaining snow cover by fusion of multi-source data based on Fengyun satellites according to claim 9, characterized in that: In step nine, the operation method for generating the snow line correction product S5 for the snow-covered area is as follows: dividing the space into multiple sub-areas according to watershed and topographic factors, calculating the average elevation demS of all snow pixels and the average elevation deml of all land pixels in each sub-area of ​​the snow cover product S4 corrected based on land surface observations on a daily basis, if the cloud cover in the sub-area is greater than Cp or the snow cover is less than Sp, where Cp and Sp are the area percentages of clouds and snow in the watershed, respectively, then demS and deml are not retained; otherwise, demS and deml are retained; demS and deml are calculated for each sub-area each year, time series interpolation is performed on them, and the snow line correction product S5 for the snow-covered area is synthesized according to the regional snow line synthesis rule; Among them, the regional snow line synthesis rule is: if the elevation of the cloud pixel in the sub-region is greater than demS and the slope If the elevation of the cloud pixel in the sub-region is less than deml, the cloud pixel is identified as snow. If the elevation of the cloud pixel in the sub-region is less than deml, the cloud pixel is identified as land. The remaining pixels remain unchanged. The expression is: ； 。 11. The method for obtaining snow cover by fusion of multi-source data based on Fengyun satellites according to claim 10, characterized in that: In the step 10, the specific operation method is: if at the cloud pixel of S5, S IMS 、S RE 、S MW If there is snow in any of the products, the cloud pixel is identified as snow, otherwise it is identified as a non-snow pixel.

Images