A kilometer-level soil moisture inversion method based on massive big data in the Qinghai-Tibet Plateau

Abstract

The present invention proposes a kilometer-level soil moisture inversion working method for the Qinghai-Tibet Plateau based on massive big data, comprising the following steps: S1, acquiring and collecting multi-source remote sensing observations and soil moisture big data; S2, optimizing soil and vegetation parameters on a spatial scale according to the soil moisture big data acquired in step S1 to obtain spatial soil parameters; S3, selecting different vegetation proxy indicators from one or any combination of optical, active and passive microwave data for screening to obtain vegetation optical thickness; S4, correcting and acquiring vegetation moisture content suitable for alpine grassland areas according to the soil moisture data of sites in vegetation coverage areas.

Description

Technical Field

[0001] The present invention relates to the technical field of soil moisture, and in particular to a kilometer-level soil moisture inversion method for the Qinghai-Tibet Plateau based on massive big data. Background Art

[0002] The Qinghai-Tibet Plateau is not only the source of major rivers such as the Yangtze River, the Yellow River, the Yarlung Zangbo River and the Lancang River, but also a major gathering place for lakes and wetlands. It is known as the "Asian Water Tower". At the same time, the Qinghai-Tibet Plateau has an average altitude of more than 4,000 meters, making it the highest plateau in the world and the most widely distributed permafrost in mid- and low-latitude regions. Under the background of climate warming, the permafrost on the Qinghai-Tibet Plateau is degrading to varying degrees, mainly manifested in the increase of permafrost temperature, the thickening of the active layer, and the gradual melting of the shallow permafrost and the ground ice in it into liquid water, which in turn affects the formation of surface runoff in cold regions and the migration process and distribution pattern of groundwater. Soil moisture is not only a key surface parameter that connects surface water, groundwater, atmospheric water and vegetation water in the Qinghai-Tibet Plateau, but also redistributes land energy and water through processes such as soil heat capacity and surface evaporation, thereby changing the water and heat flux from the surface to the atmosphere. Therefore, how to obtain accurate and high-resolution soil moisture over a large area in the Qinghai-Tibet Plateau is important for understanding the climate and energy and water cycle in the region and even the world. Summary of the invention

[0003] The present invention aims to at least solve the technical problems existing in the prior art, and in particular innovatively proposes a kilometer-level soil moisture inversion method for the Qinghai-Tibet Plateau based on massive big data.

[0004] In order to achieve the above-mentioned object of the present invention, the present invention provides a kilometer-level soil moisture inversion method of the Qinghai-Tibet Plateau based on massive big data, comprising the following steps:

[0005] S1, obtains the collected multi-source remote sensing observations and soil moisture big data;

[0006] S2, optimizing soil and vegetation parameters from a spatial scale according to the soil moisture big data collected in step S1 to obtain spatial soil parameters;

[0007] S3, selecting different vegetation proxy indicators from one or any combination of optical, active and passive microwave data to obtain vegetation optical thickness;

[0008] S4, based on the soil moisture data of the sites in the vegetation coverage area, calibrate and obtain the vegetation moisture content suitable for the alpine grassland area.

[0009] In a preferred embodiment of the present invention, the method for collecting soil moisture big data in step S1 includes field observation data, soil moisture product data set, Sentinel-1 backscatter data, one of other auxiliary data or any combination thereof.

[0010] In a preferred embodiment of the present invention, spatial downscaling and active microwave remote sensing are used in step S2 to achieve soil and vegetation parameter optimization.

[0011] In a preferred embodiment of the present invention, in step S4, the water content is calculated as follows:

[0012]

[0013] in, represents the reference incident angle θ ref Normalized backscatter;

[0014] represents the backscattering observed at the local incident angle θ;

[0015] θ ref represents the reference incident angle;

[0016] θ represents the local angle of incidence;

[0017]

[0018] in, represents the total backscatter received by the radar;

[0019] represents vegetation backscatter;

[0020] γ 2 is the vegetation penetration coefficient;

[0021] represents soil backscatter;

[0022]

[0023] in, represents vegetation backscatter;

[0024] A represents the backscatter of the vegetation canopy under full vegetation cover;

[0025] V1 represents vegetation parameters;

[0026] θ ref is the reference incident angle;

[0027] γ 2 is the vegetation penetration coefficient;

[0028]

[0029] Among them, γ 2 is the vegetation penetration coefficient;

[0030] e is the natural base;

[0031] τ is the vegetation optical thickness;

[0032] θ ref is the reference incident angle;

[0033] τ=B×VWC (2.13)

[0034] Where, τ is the vegetation optical thickness;

[0035] B is an empirical parameter that depends on the vegetation canopy type;

[0036] VWC is vegetation water content;

[0037]

[0038] in, represents soil backscatter;

[0039] C is the soil parameter, representing the backscattering under dry soil conditions;

[0040] D is a soil parameter representing the sensitivity of backscattering to soil moisture;

[0041] SM represents soil moisture parameter;

[0042]

[0043] Among them, SM represents soil moisture parameter;

[0044] represents the total backscatter received by the radar;

[0045] represents vegetation backscatter;

[0046] γ 2 is the vegetation penetration coefficient;

[0047] C and D are soil parameters.

[0048] In a preferred embodiment of the present invention, the correction method of soil parameters C and D is:

[0049] Screening of pixels of bare land and sparse vegetation areas;

[0050] Calculate soil parameters for filtered cells.

[0051] In a preferred embodiment of the present invention, the soil parameters of the entire study area are also corrected.

[0052] In a preferred embodiment of the present invention, the correction method of the vegetation parameter A is:

[0053]

[0054] in, represents the total backscatter received by the radar;

[0055] represents vegetation backscatter;

[0056] A dense(t,i,j) Vegetation parameters indicating dense vegetation period;

[0057] θ ref is the reference incident angle;

[0058]

[0059] Among them, A (t) represents all A on the dense vegetation pixel on the tth day dense(t,i,j) The spatial average of the values;

[0060] A dense(t,i,j) Vegetation parameters indicating dense vegetation period;

[0061] ∑A dense(t,i,j) represents all A on the dense vegetation pixel on the tth day dense(t,i,j) The sum of values;

[0062] A represents the backscatter of the vegetation canopy under full vegetation cover.

[0063] In a preferred embodiment of the present invention, it also includes an evaluation index, which includes one or any combination of mean deviation, Pearson correlation coefficient, root mean square error, and unbiased root mean square error;

[0064] The average deviation is calculated as:

[0065]

[0066] Among them, Bias represents the average deviation;

[0067] N represents the total number of data;

[0068] Represents product soil moisture data;

[0069] represents observed soil moisture data;

[0070] The Pearson correlation coefficient calculation formula is:

[0071]

[0072] Where R represents the Pearson correlation coefficient;

[0073] N represents the total number of data;

[0074] Represents product soil moisture data;

[0075] Represents the average value of product soil moisture data;

[0076] represents observed soil moisture data;

[0077] Represents the average value of observed soil moisture data;

[0078] The formula for calculating the root mean square error is:

[0079]

[0080] Among them, RMSE means root mean square error;

[0081] N represents the total number of data;

[0082] Represents product soil moisture data;

[0083] represents observed soil moisture data;

[0084] The unbiased root mean square error calculation formula is:

[0085]

[0086] Among them, ubRMSE represents the unbiased root mean square error;

[0087] RMSE stands for root mean square error;

[0088] Bias represents the average deviation.

[0089] In a preferred embodiment of the present invention, the correction method of vegetation water content VWC is one of the following:

[0090]

[0091] Among them, VWC is vegetation water content;

[0092] a1, a2, b1, and b2 represent the fitting coefficients of the vegetation moisture content formula;

[0093] NDVI stands for Normalized Difference Vegetation Index;

[0094] VWC=a×NDVI 2 –b×NDVI+c×(((NDVImax-NDVImin) / (1-NDVImin))),

[0095] Among them, VWC is vegetation water content;

[0096] a, b, and c represent the fitting coefficients of the vegetation moisture content formula;

[0097] NDVI stands for Normalized Difference Vegetation Index;

[0098] NDVImax represents the maximum value of the normalized difference vegetation index of the pixel;

[0099] NDVImin represents the minimum value of the normalized difference vegetation index of the pixel.

[0100] VWC=exp(LAI / a3)-b3

[0101] Among them, VWC is vegetation water content;

[0102] Among them, LAI stands for leaf area index;

[0103] a3 and b3 represent the fitting coefficients of the vegetation moisture content formula;

[0104] VWC=a4*NDWI+b4

[0105] Among them, VWC is vegetation water content;

[0106] a4 and b4 represent the fitting coefficients of the vegetation moisture content formula;

[0107] NDWI stands for Normalized Difference Water Index;

[0108] VWC=a5*NDWI 2 +b5*NDWI+c5

[0109] Among them, VWC is vegetation water content;

[0110] a5, b5, and c5 represent the fitting coefficients of the vegetation water content formula;

[0111] NDWI stands for Normalized Difference Water Index;

[0112] VWC=a6*RVI+b6

[0113]

[0114] Among them, VWC is vegetation water content;

[0115] a6, b6, and c6 represent the fitting coefficients of the vegetation moisture content formula;

[0116] represents VH polarized backscattering;

[0117] represents VV polarized backscatter;

[0118] SMAP-L3 DCA VOD represents the vegetation optical depth product data calculated by the DCA algorithm of the SMAP L3 product.

[0119] In a preferred embodiment of the present invention, the correction method of vegetation water content VWC is:

[0120] VWC=a×NDVI 2 –b×NDVI+c×(((NDVImax-NDVImin) / (1-NDVImin))), where VWC is vegetation water content;

[0121] a, b, and c represent the fitting coefficients of the vegetation moisture content formula;

[0122] NDVI stands for Normalized Difference Vegetation Index;

[0123] NDVImax represents the maximum value of the normalized difference vegetation index of the pixel;

[0124] NDVImin represents the minimum value of the normalized difference vegetation index of the pixel.

[0125] In a preferred embodiment of the present invention, in order to ensure the security of the kilometer-level soil moisture inversion working method and prevent the leakage of the kilometer-level soil moisture inversion working method during transmission, the following steps are also included:

[0126] S5, encrypting the kilometer-level soil moisture inversion working method in steps S1 to S4, and transmitting the encrypted data via a USB flash drive, a mobile hard disk or a network;

[0127] S6, when the kilometer-level soil moisture inversion working method is needed, the encrypted kilometer-level soil moisture inversion working method is decrypted for use.

[0128] In summary, due to the adoption of the above technical solution, the present invention can realize accurate high-resolution identification of soil moisture content over a large area in the Qinghai-Tibet Plateau.

[0129] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0130] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0131] Figure 1 It is a schematic diagram of the Sentinel-1 soil moisture inversion process of the present invention.

[0132] Figure 2 Schematic diagram of the calculation results of soil parameters C and D of bare soil and sparse vegetation area pixels screened by the present invention, where:

[0133] (a)–(b) are schematic diagrams of the spatial distribution of the estimated values ​​of C and D, and (c)–(d) are histograms of the estimated values ​​of C and D, respectively.

[0134] Figure 3 The present invention Figure 2 Schematic diagram of variable importance ranking of the random forest model for predicting C in (a) and D in (b).

[0135] Figure 4 Schematic diagram of spatial distribution of the prediction results of C and D parameters in the Qinghai-Tibet Plateau study area of ​​the present invention, where:

[0136] (a)–(b) show the scatter plots between the estimated and predicted values ​​of C and D, and the C RF and D RF Schematic diagram of the model’s prediction accuracy. (c)–(d) show the spatial distribution of the predicted values ​​of C and D, respectively.

[0137] Figure 5 It is a schematic diagram of the time series changes of the daily average value of the A0 parameter space from 2017 to 2021 of the present invention.

[0138] Figure 6 Schematic diagram of violins showing the statistical indicators ((a) R and (b) ubRMSD) between soil moisture and observed data calculated based on different vegetation index formulas of the present invention.

[0139] Figure 7 Schematic diagram of the spatial distribution of the mean values ​​of (a) NDVI and (b) LAI time series in 2017–2018 of the present invention.

[0140] Figure 8 The different soil moisture products (a) SM of the present invention are S-1 ,(b)ESA CCI,(c)SM Chai ,(d)SM Li ,(e)SM Song ,(f)SM Sg ,(g)SM Han and (h)SM SMAPSchematic diagram of the spatial distribution of mean soil moisture during the melt season from 2017 to 2021.

[0141] Fig. 9 It is a schematic diagram of the distribution of grassland types (AD: alpine desert; AS: alpine steppe; AM: alpine meadow) in the Qinghai-Tibet Plateau study area of ​​the present invention.

[0142] Fig.10 It is a schematic diagram of the mean moisture value of each soil moisture product based on statistics of different grassland types in the present invention (the error bar is the standard deviation of the soil moisture value of different grassland types).

[0143] Fig.11 This invention is the 2017–2018 SM S-1 ,SM Chai ,SM Li ,SM Song ,SM Sg ,and SM Han Boxplots of (a) R, (b) ubRMSD, (c) RMSD and (d) Bias between soil moisture products and observations.

[0144] Fig.12 Schematic diagram of the present invention, wherein:

[0145] (a)–(e) respectively show the time series schematic diagrams of three soil moisture product data, backscatter data and NDVI values ​​of a single station in five different observation networks from 2017 to 2021; (a1)–(e1) respectively show the scatter point schematic diagrams between soil moisture product data and observation data.

[0146] Fig.13 The SM shown in the present invention Song ,SM Sg and SM Han Product time series diagram.

[0147] Fig.14 The SM shown in the present invention SMAP And SMESA CCI product time series diagram.

[0148] Fig.15 It is a box-and-line diagram of the performance indicators between different soil moisture products (a) R, (b) ubRMSD, (c) RMSD, and (d) Bias and the observed data in the Naqu observation network in 2019–2021 of the present invention.

[0149] Fig.16It is a box-line diagram of the performance indicators between the soil moisture data inverted from vegetation water content based on different coefficients (1 represents the coefficient based on the default in the literature, and 2 represents the coefficient corrected in this study) and the soil moisture observation values ​​of the Maqu observation network from 2017 to 2018. DETAILED DESCRIPTION

[0150] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0151] Chapter 1 Introduction to Patent Application

[0152] In view of the generally low spatial resolution of mainstream soil moisture products, an active microwave 1km near-surface soil moisture inversion algorithm for the melting season was developed. First, based on the semi-empirical model, the soil and vegetation parameters were optimized from the spatial scale, and the spatial soil parameters were obtained; then, different vegetation proxy indicators were selected from the optical, active and passive microwave data for screening, and the optimal vegetation optical thickness was obtained. At the same time, based on the station soil moisture data in the vegetation coverage area, the vegetation moisture content calculation formula suitable for alpine grassland areas was corrected and obtained. Finally, the 1km soil moisture data of the Qinghai-Tibet Plateau study area from 2017 to 2021 was inverted, and its performance was evaluated and compared based on the station observation data of 5 observation networks in the study area and 6 sets of published 1km soil moisture product data covering the Qinghai-Tibet Plateau. The results show that the inversion results are better than most 1km downscaled soil moisture products, with an R of 0.57 and a ubRMSD of 0.064m 3 / m 3 , Bias is -0.042m 3 / m 3 , and presents a reasonable spatial distribution pattern and reflects more spatial details. For details, see Figures 1 to 16 shown.

[0153] 1.1 Background and significance of the topic

[0154] The Qinghai-Tibet Plateau is not only the source of major rivers such as the Yangtze River, the Yellow River, the Yarlung Zangbo River and the Lancang River, but also a major gathering place for lakes and wetlands. It is known as the "Asian Water Tower". At the same time, the Qinghai-Tibet Plateau has an average altitude of more than 4,000 meters, making it the highest plateau in the world and the most widely distributed permafrost in mid- and low-latitude regions. Under the background of climate warming, the permafrost on the Qinghai-Tibet Plateau is undergoing varying degrees of degradation, mainly manifested in the increase in permafrost temperature, the thickening of the active layer, and the gradual melting of the shallow permafrost and the ground ice therein into liquid water, which in turn affects the formation of surface runoff in cold regions and the migration process and distribution pattern of groundwater. Soil moisture is not only a key surface parameter that connects surface water, groundwater, atmospheric water and vegetation water in the Qinghai-Tibet Plateau, but also redistributes land energy and water through processes such as soil heat capacity and surface evaporation, thereby changing the water and heat flux from the surface to the atmosphere. Therefore, obtaining accurate and high-resolution soil moisture over a large area in the Qinghai-Tibet Plateau is important for understanding the climate and energy and water cycle in the region and even the world.

[0155] At present, there are three main ways to obtain soil moisture data in the Qinghai-Tibet Plateau, including site observation, remote sensing inversion and numerical simulation. In the Qinghai-Tibet Plateau, due to limited sampling points and high maintenance costs, traditional point-based ground measurement methods are difficult to achieve large-scale continuous observation. However, the soil moisture distribution in different regions is affected by factors such as diverse surface cover types, large rainfall differences, soil texture and topography, and has great spatial heterogeneity, making it difficult for traditional ground measurements to accurately capture the spatiotemporal distribution of soil moisture in a large spatial range. The rise and development of remote sensing and numerical simulation technology provides a good opportunity to obtain global and regional scale soil moisture data. Remote sensing satellites can conduct large-scale continuous observations, especially microwave remote sensing satellites, which have the advantages of high temporal resolution, wide observation range, and all-day and all-weather monitoring. Microwave remote sensing uses the dielectric properties of soil as a medium to build a relationship between microwave signals and soil moisture, making microwave remote sensing technology the most promising method for quantitatively monitoring global and regional surface soil moisture. In addition, by combining numerical models, remote sensing data and field observation data using an assimilation system, global-scale reanalysis soil moisture products can also be obtained. In recent decades, global satellite near-surface soil moisture datasets produced based on different satellite-borne microwave sensor monitoring data and reanalysis soil moisture products at different soil depths have been continuously updated and released.

[0156] The spatial resolution of the current mainstream reanalysis and remote sensing microwave soil moisture products is generally low (~25km), which is more suitable for global scale application research. Therefore, directly applying coarse-resolution soil moisture products to regional or fine-scale ecohydrological research is one of the reasons why some current research results have large uncertainties. Considering the strong spatiotemporal heterogeneity of soil moisture, high spatiotemporal resolution and high-precision soil moisture products have become an urgent need for regional scale applications in many fields such as climate change, hydrological simulation, drought monitoring, crop yield estimation, and food security in the earth system. In order to obtain high-resolution soil moisture data at regional or even global scales, spatial downscaling methods have gradually become one of the hot topics in passive microwave remote sensing soil moisture research. The core of the spatial downscaling method is to use the rich spatial heterogeneity of high-resolution downscaling factors such as optical or thermal infrared to improve the spatial resolution of soil moisture products through downscaling relationship models. However, the accuracy of the downscaling results depends to a certain extent on the direct or indirect relationship between the downscaling factors and the soil moisture values. Therefore, for the Qinghai-Tibet Plateau region with complex terrain, the relationship between soil moisture and downscaling factors is relatively complex, which limits the performance of the downscaling algorithm and makes it difficult to achieve strong applicability at the regional scale. Synthetic Aperture Radar (SAR) in active microwave remote sensing improves the spatial resolution of images through pulse compression and synthetic aperture technology, providing a new idea for solving the above problems. The Sentinel-1 satellite carries a C-band synthetic aperture radar device, which can provide dual-polarization continuous images and is currently one of the important data sources for inverting high-resolution soil moisture products. However, there are still two problems in the inversion of high-resolution soil moisture data in the Qinghai-Tibet Plateau region based on Sentinel-1: First, the soil moisture inverted by Sentinel-1 is easily affected by parameters such as vegetation and surface roughness. How to correct these two types of parameters is a difficult problem in research; second, most existing studies are based on station-scale calibration parameters, and then directly apply them to the entire study area, without considering the regional differences in different soil or vegetation parameters.

[0157] 1.2 Research progress on high-resolution soil moisture inversion

[0158] Most of the published reanalysis and microwave remote sensing soil moisture products have low spatial resolutions, and the evaluation results of existing soil moisture products show that soil moisture products from different sources are affected by their inversion algorithms, input parameters, model structures, etc., and there will be great uncertainty when directly applying them to the Qinghai-Tibet Plateau. In order to obtain high-resolution soil moisture data in the Qinghai-Tibet Plateau, spatial downscaling methods have gradually become one of the hot research directions of passive microwave remote sensing soil moisture. In recent years, some scholars have conducted empirical spatial downscaling studies on mainstream soil moisture products based on machine learning methods such as random forests and combined with optical remote sensing data. A series of soil moisture downscaling products covering about 1 km in the Qinghai-Tibet Plateau have been obtained and published. These products mainly establish empirical downscaling relationship models between passive microwave soil moisture data and high-resolution downscaling factors such as optical or thermal infrared with rich spatial heterogeneity at a coarse spatial resolution scale, and then obtain model parameters. Then, under the assumption that the scale of the downscaling model remains unchanged, the model parameters are applied to the downscaling factors with high spatial resolution to calculate the downscaled soil moisture. However, it is worth noting that no matter which downscaling method is used, the accuracy of their downscaling results depends to a certain extent on the accuracy of the coarse-resolution soil moisture product and the direct or indirect relationship between the downscaling factor and the soil moisture value. Therefore, for the Qinghai-Tibet Plateau region with large spatial heterogeneity, the relationship between soil moisture and the downscaling factor is more complex, which limits the performance of the downscaling algorithm and its applicability at the regional scale is still uncertain.

[0159] There is great potential for inverting high-resolution soil moisture data based on SAR backscatter data. SAR is an important component of microwave remote sensing, which has the advantages of large earth observation coverage, short revisit period and high spatial resolution. The Sentinel-1 satellite was launched in 2014. It carries a C-band microwave sensor with a revisit period of 12 days and a spatial resolution of 5m×20m. It is a good data source for high-resolution soil moisture inversion. Since backscatter data is not only sensitive to soil moisture, but also affected by multiple factors such as soil texture, topography, and surface vegetation conditions, it is very challenging to invert soil moisture based on backscatter data. At present, there have been many site-scale soil moisture inversion works based on Sentinel-1, but there has been no research directly based on the backscatter model to invert high-resolution soil moisture data for the entire study area. This is because most studies calibrate the soil and vegetation parameters of the backscatter model at the site scale, while it is difficult to directly calibrate the soil and vegetation parameters of the entire area. Although existing studies have obtained regional-scale soil moisture data by assuming that soil and vegetation parameters corrected at the site scale are also applicable to similar surface conditions, these parameters do not actually take into account the heterogeneity of the surface in the region, making the inverted regional high-resolution soil moisture data insufficient in spatial applicability. The surface heterogeneity of the Qinghai-Tibet Plateau is strong, but due to the harsh natural environment and inconvenient transportation, there are few and unevenly distributed observation sites, making it more difficult to directly apply site observations and backscatter data to correct soil and vegetation parameters for the entire region. Therefore, there is currently no publicly released high-resolution soil moisture data for the Qinghai-Tibet Plateau region directly inverted based on Sentinel-1 data.

[0160] In view of the low spatial resolution of existing near-surface soil moisture products, based on the accuracy assessment and error source analysis results of soil moisture products on the Qinghai-Tibet Plateau, combined with high-resolution Sentinel-1 backscattering data, a 1km spatial soil moisture inversion algorithm for the Qinghai-Tibet Plateau study area in the melt season was developed, and 1km near-surface soil moisture products for the region from 2017 to 2021 were inverted. Subsequently, based on the melt season observation network data of 5 stations, a temporal and spatial comparison analysis was conducted with the published 1km-scale soil moisture products covering the study area to verify the accuracy of the inversion results.

[0161] Chapter 2 Overview of the Study Area, Data and Research Methods

[0162] 2.2 Data

[0163] This study mainly used the following four types of data: (1) field observation data; (2) soil moisture product dataset; (3) Sentinel-1 backscatter data; and (4) other auxiliary data.

[0164] 2.2.1 Field observation data

[0165] In order to optimize and verify the soil moisture inversion results of this study, this patent application comprehensively collects soil moisture data from five observation networks covering different frozen soil types, vegetation types and different geographical locations in the entire Qinghai-Tibet Plateau. These include the active layer hydrothermal observation network (CRS) established by the National Field Scientific Observation and Research Station for Special Environment and Disasters in the Cryosphere of the Northern Tibetan Plateau of the Chinese Academy of Sciences, the Naqu soil hydrothermal observation network in the central Qinghai-Tibet Plateau (Naqu), the Pali soil hydrothermal observation network in the southern Qinghai-Tibet Plateau (Pali), the Maqu soil moisture observation network in the eastern Qinghai-Tibet Plateau (Maqu), and the Ali soil moisture observation network in the western Qinghai-Tibet Plateau (Ngari) (Table 1). This patent application collects soil hydrothermal data of all available soil depths of different observation networks from January 2016 to December 2021. All data have been strictly quality controlled, and the permafrost area soil moisture observation network (CRS) is independent of the ISMN database.

[0166] Table 1 Observation data of soil moisture on the Qinghai-Tibet Plateau

[0167]

[0168] (1)CRS

[0169] The National Field Scientific Observation and Research Station for Special Environment and Disasters in the Cryosphere of the Tibetan Plateau of the Chinese Academy of Sciences is a field observation and research station for long-term monitoring and research of permafrost in my country (http: / / new.crs.ac.cn / ). In view of the special environmental changes in the plateau and the interrelationships between related spheres (such as atmosphere, water, land, and organisms), many field monitoring and experimental sites have been established in different permafrost areas. Since the 1990s, the CRS active layer soil water and heat observation network has been successively established by researchers from the Tibetan Plateau Cryosphere Research Station of the Chinese Academy of Sciences in the permafrost area of ​​the Qinghai-Tibet Plateau with an average altitude of about 4,651 m (Zhao et al., 2021). The observation sites are mainly located in the hinterland of the Qinghai-Tibet Plateau and are concentrated along the Qinghai-Tibet Railway. This patent application collects active layer soil water and heat observation data from 13 stations with complete permafrost data along the Qinghai-Tibet Railway, which runs through the northern and southern boundaries of the permafrost zone of the Qinghai-Tibet Plateau. Each station is equipped with multiple sensors to measure soil temperature and moisture data at different depths (~3–110 cm), and the sampling interval of each station varies from 30 to 120 minutes (Table 2).

[0170] Table 2 Site information of CRS soil water and heat observation data used in this patent application

[0171]

[0172] (3)Maqu

[0173] The Maqu soil temperature and humidity observation network was established in July 2008 in the northeastern part of the Qinghai-Tibet Plateau, with an average altitude of about 3,477m. The observation network is located in the Yellow River Valley and the surrounding hills with a cold and semi-humid climate, and the surface cover type is uniform low grassland. The observation network consists of 20 stations, each equipped with multiple sensors to measure soil temperature and humidity data at different depths (~5–80cm), and the sampling interval of each station is 15 minutes. This patent application collects soil moisture observation data from 11 stations with complete data in the Maqu observation network (Table 2-3).

[0174] (4) Naqu, Pali and Ali

[0175] The Naqu soil temperature and humidity observation network is located in the central part of the Qinghai-Tibet Plateau. This area has low atmospheric water vapor content and small surface biomass, making it an ideal area for microwave satellite inversion of soil moisture. The network is a multi-scale observation network established by Yang Kun, a researcher at the Institute of Tibetan Plateau Research, Chinese Academy of Sciences, in Naqu over a three-year period. It has an average altitude of about 4,650m and covers an area of ​​about 100×100km. The Pali soil temperature and humidity observation network is located in the southern part of the Qinghai-Tibet Plateau, with an average station altitude of about 4,486m. The Ali soil temperature and humidity observation network is located near the city of Ali in the western part of the Qinghai-Tibet Plateau, with an average station altitude of about 4,269m. Naqu and Pali consist of 56 and 25 stations, respectively. Each station is equipped with 4 sensors to measure soil temperature and humidity data at 0–5, 10, 20 and 40cm, respectively. The sampling interval for each station is 30 minutes. Ali consists of 4 stations located near Ali, each station is equipped with multiple sensors to measure soil temperature and humidity data at different depths (~5–80cm), respectively. The sampling interval for each station is 15 minutes. According to the research period of this patent application and the availability of observation data, soil water and heat observation data with complete data were collected from 36, 19 and 4 stations in the Naqu, Pali and Ali observation networks, respectively, for research (Table 3).

[0176] Table 3 Site information of soil hydrothermal observation data of Maqu, Naqu, Pali and Ali used in this patent application

[0177]

[0178] 2.2.2 Soil moisture product dataset

[0179] Six sets of publicly available 1 km soil moisture products covering the Qinghai-Tibet Plateau were collected. These products were mainly generated by downscaling (Table 4).

[0180] (1) 1km resolution soil moisture product

[0181] SM Chai :This dataset consists of daily 0.01°×0.01° surface soil moisture products produced in the Qinghai-Tibet Plateau in 2005, 2010, 2015, 2017 and 2018. This product is based on a multivariate statistical regression model, and is obtained by downscaling the "SMAP time extension 0.25°×0.25° surface soil moisture data in the Qinghai-Tibet Plateau". The data involved in the multivariate statistical regression include GLASS albedo, leaf area index and vegetation cover (Albedo / LAI / FVC), Zhouji-Western China 1km all-weather surface temperature data, and longitude / latitude information.

[0182] SM Li :This dataset consists of soil moisture products (10-100cm) at different depths of 1km×1km and every 10cm interval in China from 2000 to 2020. This product is based on machine learning, with 10 layers of soil moisture data from 1648 observation stations provided by the China Meteorological Administration as the benchmark, and ERA5-Land meteorological forcing data, leaf area index (LAI), land cover types (Landtypes), topography (DEM) and soil texture as covariates.

[0183] SM Song :This dataset is a daily 1km×1km surface soil moisture product produced in China from 2003 to 2022. The product is generated by downscaling the 36km AMSR-E and AMSR-2 surface soil moisture data, using MODIS optical reflectance data and seamless daily thermal infrared surface temperature as the main input data.

[0184] SM Sg :This dataset consists of daily 1km×1km surface soil moisture products produced in the Qinghai-Tibet Plateau from 2001 to 2020. The product first uses auxiliary factors such as optical vegetation index (NDVI, EVI), land surface temperature (LST), evapotranspiration (ET), precipitation, topography (DEM, slope, aspect, TWI), soil properties and soil moisture related indexes (SWCI, SIWSI, VSDI), and uses five machine / deep learning methods such as artificial neural network (ANN), convolutional neural network (CNN), residual neural network (ResNet), long short-term memory network (LSTM) and XGBoost to downscale the ESA CCI soil moisture data. Then, the soil moisture data obtained by different downscaling methods are fused using the Bayesian three-cornered hat method.

[0185] SM Smap:This dataset is a global 1km×1km daily surface soil moisture product produced from 2015 to 2020. This product is obtained by downscaling the global 9km Level 2 enhanced SMAP soil moisture product based on thermal inertia theory. The downscaling algorithm first simulates the surface temperature difference and the relationship between the corresponding soil moisture and NDVI during the SMAP morning and afternoon transit times based on the 0.25° GLDAS-Noah surface temperature and soil moisture simulation results and LTDR NDVI data. Then the downscaling algorithm is applied to the 1kmMODIS LST and NDVI products.

[0186] SM Han :This dataset is a global daily 1km×1km surface soil moisture product produced from 2000 to 2020. This product is based on the physical random forest method, with the surface soil moisture data of 1725 observation stations provided by the Global Soil Moisture Observation Network as the benchmark, and ERA5-Land meteorological forcing data, vegetation index (NDVI / EVI), soil temperature (LST), topography (DEM) and soil texture data as covariates.

[0187] Table 4 1km soil moisture product information

[0188]

[0189] 2.2.3 Sentinel-1 backscatter data

[0190] Sentinel-1 is a satellite equipped with a C-band (frequency: 5.405GHz, wavelength: 5.6cm) synthetic aperture radar (SAR) launched under the framework of the European Space Agency's Copernicus program. It consists of two polar-orbiting satellites (Sentinel-1A and Sentinel-1B), and is mainly used to monitor global marine environmental changes, surface deformation, and dynamic changes in terrestrial forests, water bodies, soil, etc. Sentinel-1A and Sentinel-1B were launched on April 3, 2014 and April 25, 2016, respectively. The revisit period of a single satellite is 12 days, and the revisit period of a dual satellite can be shortened to 6 days. Sentinel-1 acquires data in four modes: Strip Map (SM), Interferometric Wide Swath (IW), Extra Wide Swath (EW), and Wave (WV). IW is the default acquisition mode on land, with a single-well image width of 250 km and a ground spatial resolution of 5×20 m. Data is acquired using VV and VH dual polarization. At the same time, Sentinel-1 data products include Raw Level-0 data, SLC (Single Look Complex), GRD (Ground Range Detected), and OCN (Ocean). GRD data can obtain the backscatter coefficient of the land surface and can be used to invert the surface soil moisture (Table 5). Since the VH polarization mode is more sensitive to vegetation changes, the VV polarization mode is more sensitive to soil moisture and has more application potential than the VH polarization mode in soil moisture inversion research. At the same time, the IW imaging mode of Sentinel-1GRD can observe two 250km wide orbits, the ascending and descending orbits. Considering that there are more descending orbit data available in the Qinghai-Tibet Plateau, this paper collects descending orbit image data collected in the VV and VH polarization modes of the IW imaging mode of Sentinel-1GRD from 2017 to 2021 to carry out soil moisture inversion research.

[0191] Table 5 Sentinel-1 acquisition mode

[0192]

[0193] 2.2.4 Other auxiliary data

[0194] The auxiliary data used in this paper include MODIS datasets, terrain, soil properties and other data products, which are mainly used for the optimization of soil moisture products and the analysis of inversion results (Table 6). The details are as follows:

[0195] (1) MODIS dataset

[0196] Land cover type: The land cover type data comes from the MCD12Q1 level 3 data product (Friedl and Sulla-Menashe 2019) based on the International Geosphere-Biosphere Programme (IGBP) global vegetation classification scheme at 500m. This product is mainly obtained by information extraction technology based on supervised decision tree classification. The dataset contains 17 major land cover types, including 11 natural vegetation types, 3 land development and mosaic land types, and 3 non-vegetated land type definition classes.

[0197] Optical vegetation index: including Normalized Difference Vegetation Index (NDVI) and Leaf Area Index (LAI). NDVI is calculated by the infrared and near-infrared bands of satellite sensors that are sensitive to vegetation. It can monitor the growth status of vegetation and reflect the vegetation coverage and growth status at a certain temporal and spatial scale. The NDVI data used in this patent application is the MODIS 16-day synthetic product MOD13A1 (Didan 2021) with a resolution of 500m from 2017 to 2021. LAI refers to the total area of ​​plant leaves per unit land area as a multiple of the land area. It is an important parameter for describing the exchange of matter and energy between soil, vegetation and atmosphere. The LAI data used in this patent application is the 500 m resolution MODIS 4-day composite product MCD15A3H for 2017–2018, which is composited by selecting the best available pixels from all acquisitions of the MODIS sensors on NASA’s Terra and Aqua satellites within 4 days (Myneni et al., 2021).

[0198] Optical Water Index: The Normalized Difference Water Index (NDWI) is derived from the 500m MOD09GA data product. This product provides reflectance data calculated from the surface reflectance of the MODIS sensor bands 1 to 7 on the Terra satellite after bidirectional atmospheric correction for water, clouds, heavy aerosols, and cloud shadow masks, and is used as the basic data to obtain a variety of MODIS surface products (Vermote and Vermeulen 2015).

[0199] Surface temperature: The surface temperature product uses the MOD11 series. MOD11 uses a physically based day / night surface temperature inversion method. Its inversion accuracy on the Qinghai-Tibet Plateau has been verified by many scholars (Zhengming and Zhao-Liang 1997). This paper selects the MOD11A2 and MYD11A2 1km spatial resolution products obtained by Terra and Aqua sensors, respectively, synthesized every 8 days in 2017-2018, which contain daily daytime surface temperature and nighttime surface temperature. The full-day surface temperature is obtained based on the calculation formula proposed by Zou et al. (2015). In addition, this paper also collects the near-surface soil temperature data (0-10cm) of GLDAS-Noah from 2017-2021 to divide the freezing period and melting period, and to estimate the frozen soil moisture.

[0200] (2) Grassland type

[0201] In addition to land cover types, this paper also collected grassland classification maps in the Qinghai-Tibet Plateau and used them to analyze the spatial distribution of 1km soil moisture data in the region. This dataset comes from the 500m "China Vegetation Atlas" (ACAS2011), which mainly covers three types of grasslands in the plateau: alpine meadow, alpine steppe and alpine desert.

[0202] (3) Terrain

[0203] The terrain data comes from the global elevation dataset with a spatial resolution of 1 (The Global Multi-resolution Terrain Elevation Data 2010, GMTED2010 (Danielson and Gesch 2011)). The digital elevation data (DEM) used in this paper is mainly used as auxiliary data to analyze the impact of spatial heterogeneity on the verification results of soil moisture products, and is used to generate other 7 basic terrain data related to terrain parameters on the Automated Geoscientific Analyses Geographic Information System (SAGAGIS) software platform (Conrad et al. 2015). The 8 terrain parameters include: DEM, ASPECT, SLOPE, Terrain Surface Convexity (CONVEXITY), Terrain Surface Texture (TEXTURE), Terrain Ruggedness Index (TRI), Plan Curvature (PLANCUR) and Profile Curvature (PROCUR).

[0204] (4) Soil properties

[0205] The soil property data comes from the SoilGridsV2 dataset with a spatial resolution of 250m (Hengl et al. 2017). SoilGridsV2 is trained using machine learning methods based on 150,000 soil profile data and 158 covariates obtained from remote sensing data worldwide. The dataset provides a variety of global digital soil property data at 7 different soil depths (0, 5, 15, 30, 60, 100 and 200cm). This paper collects 15 soil property data at the first and second soil depths in the dataset, and averages the soil property data of the two soil depths.

[0206] (5) Microwave vegetation index data

[0207] In addition to the optical vegetation index data, this paper also uses passive microwave vegetation optical depth (VOD) and active microwave vegetation index (RVI) data. The passive microwave vegetation optical depth data comes from the 36km third-level official SMAP DCA VOD product. The active microwave vegetation index is based on Sentinel-1 VV and VH polarimetric data. Calculated.

[0208]

[0209] Among them, RVI stands for Microwave Radar Vegetation Index;

[0210] represents VH polarized backscattering;

[0211] represents VV polarized backscatter;

[0212] Table 6 Auxiliary data product information

[0213]

[0214] 2.3 Research Methods

[0215] 2.3.1 Soil moisture assessment index

[0216] This patent application mainly uses four evaluation indicators, including Pearson correlation coefficient (R), unbiased Root Mean Square Error (ubRMSE), root mean square error (RMSE), and bias. Among them, the correlation coefficient R and the slope Slope represent the correlation and consistency between the satellite product data and the measured data in the time series at the pixel scale; Bias represents the average deviation of the product sequence relative to the measured sequence. A positive deviation represents an overall overestimation of the product data, and a negative deviation represents an overall underestimation of the product data; RMSE represents the average deviation between the product sequence and the measured sequence; ubRMSE is the average deviation value after removing the sequence average deviation Bias from the original RMSE. The formula is as follows:

[0217]

[0218] Among them, Bias represents the average deviation;

[0219] N represents the total number of data;

[0220] Represents product soil moisture data;

[0221] represents observed soil moisture data;

[0222]

[0223] Where R represents the Pearson correlation coefficient;

[0224] N represents the total number of data;

[0225] Represents product soil moisture data;

[0226] Represents the average value of product soil moisture data;

[0227] represents observed soil moisture data;

[0228] Represents the average value of observed soil moisture data;

[0229]

[0230] Among them, RMSE means root mean square error;

[0231] N represents the total number of data;

[0232] Represents product soil moisture data;

[0233] represents observed soil moisture data;

[0234]

[0235] Among them, ubRMSE represents the unbiased root mean square error;

[0236] RMSE stands for root mean square error;

[0237] Bias represents the average deviation;

[0238] Among them, θ EST represents product data, θ REF represents the observed data, and Represent the average values ​​of product data and observation data respectively. Since it is difficult to strictly match the site and pixel or grid data in space and there are errors to a certain extent, R, ubRMSE and Slope are used as priority evaluation indicators to reduce the impact of scale effect on the evaluation results.

[0239] 2.3.2 Soil moisture backscatter inversion model

[0240] (1) Sentinel-1 data preprocessing

[0241] This patent application uses remote sensing images collected in the IW imaging mode of Sentinel-1GRD. Based on Sentinel-1Toolbox, the image data have been preprocessed, including: 1) orbit correction; 2) thermal noise removal; 3) strip noise removal; 4) radiation correction; 5) speckle filtering (Refined Lee filtering) and 6) terrain correction. Since the Sentinel-1 data provided by the Google Earth Engine (GEE) platform only provides the incident angle, the radiation distortion of the backscattered signal of complex terrain is not considered. Therefore, based on the incident angle and STRM DEM data, our research further calculated the local incident angle of the Sentinel-1 data on the GEE platform. The preprocessed Sentinel-1 data are converted into backscatter coefficients and reprojected to 1km data in the WGS84 coordinate system.

[0242] In addition, in order to obtain high-quality Sentinel-1 backscatter data, we also took the following measures to further control its quality: 1) In order to remove the error caused by the local incident angle offset, all local incident angles were normalized to the center angle of the image (40°). The normalization formula is as follows:

[0243]

[0244] in, represents the reference incident angle θ ref Normalized backscatter;

[0245] represents the backscattering observed at the local incident angle θ;

[0246] θ ref represents the reference incident angle;

[0247] θ represents the local angle of incidence;

[0248] In the formula, and Represent the local incident angle θ and the reference incident angle θ respectively ref Observed backscatter coefficient; 2) In order to ensure reliable soil moisture inversion values, the slope data calculated by DEM (slope>6°) were used for areas affected by steep terrain and land cover types such as forest, snow, water body, and mask processing; 3) Since the Sentinel-1 backscatter signal is affected by soil freezing and has uncertainty, in order to avoid this influence, this study only retained the backscatter data during the melting period; 4) Due to the influence of angular reflection effects, shadows or noise, higher or lower backscatter data contain almost no soil moisture signals. Based on the mask rules proposed in existing studies, this study eliminated these backscatter signals, so only the VV polarization backscatter signals between -20 and -5dB and the VH polarization signals between -26 and -11dB were retained.

[0249] (2) Construction of soil moisture inversion model

[0250] The backscattering models used in this study to invert soil moisture are mainly the Water Cloud Model (WCM) and the Ulaby model. The Water Cloud Model is an empirical model proposed by Attema and Uaby for studying crops. The model assumes that vegetation is composed of many particles of the same size and shape, all of which are regarded as a layer of "cloud" and multiple scattering between vegetation and soil is ignored. The model structure is as follows:

[0251]

[0252] in, represents the total backscatter received by the radar;

[0253] represents vegetation backscatter;

[0254] γ 2 is the vegetation penetration coefficient;

[0255] represents soil backscatter;

[0256] In the formula, Represents the total backscatter coefficient received by the radar (unit: linear); and Represent the vegetation and soil scattering parts respectively. The vegetation scattering formula is as follows:

[0257]

[0258] in, represents vegetation backscatter;

[0259] A represents the backscatter of the vegetation canopy under full vegetation cover;

[0260] V1 represents the vegetation parameter, which is usually set to 1;

[0261] θ ref is the reference incident angle;

[0262] γ 2 is the vegetation penetration coefficient;

[0263] In the formula, θ ref is the reference incident angle (40°); A represents the backscatter of the vegetation canopy under full vegetation cover; V1 is set to 1; γ 2 is the vegetation penetration coefficient, which can be expressed as:

[0264]

[0265] Among them, γ 2 is the vegetation penetration coefficient;

[0266] e is the natural base;

[0267] τ is the vegetation optical thickness;

[0268] θ ref is the reference incident angle;

[0269] τ=B×VWC (2.13)

[0270] Where, τ is the vegetation optical thickness;

[0271] B is an empirical parameter that depends on the vegetation canopy type;

[0272] VWC is vegetation water content;

[0273] Where B is an empirical parameter that depends on the type of vegetation canopy and is set to 0.084 in this study; τ is the vegetation optical thickness; VWC (kg / m2) is the vegetation water content. The soil scattering part can be calculated by the Ulaby soil backscattering model, as follows:

[0274]

[0275] in, represents soil backscatter;

[0276] C is the soil parameter, representing the backscattering under dry soil conditions;

[0277] D is a soil parameter representing the sensitivity of backscattering to soil moisture;

[0278] SM represents soil moisture parameter;

[0279] C and D are soil parameters and vegetation parameter A respectively. The calibration process is detailed in Chapter 3. After obtaining the calibrated soil and vegetation parameters, integrating equations (2.10)–(2.14), soil moisture can be inverted based on the following formula:

[0280]

[0281] Among them, SM represents soil moisture parameter;

[0282] represents the total radar backscatter;

[0283] represents vegetation backscatter;

[0284] γ 2 is the vegetation penetration coefficient;

[0285] C and D are soil parameters;

[0286] 2.3.3 Random Forest Regression Model

[0287] In order to obtain the soil backscattering model parameters of the entire study area, this study also used the Random Forest (RF) regression method to extrapolate the corrected soil parameters to the entire study area (see Section 4.2.1 of Chapter 4 for the specific process). The random forest model is a combination model based on the decision tree (Classification and Regression Tree, CART) proposed by Breiman, which mainly has two algorithms: classification and regression. Its basic idea is based on statistical theory. Through the self-help resampling technology, K samples are randomly selected from the original training sample set N with replacement to generate a new training sample set, and then K decision trees are generated based on the self-help sample set to form a random forest. The model has the characteristics of strong applicability, effective avoidance of overfitting, insensitivity to missing data and multivariate collinearity. For the regression model, the prediction results of multiple decision trees are combined, and the final prediction results are obtained through a voting process. This study uses a repeated cross-validation (CV) scheme to evaluate the model performance. 70% of the correction data are randomly selected for 10 fittings, and the remaining 30% of the correction data are used for verification. After each CV run of all models, the correlation coefficient (R) and root mean square error (RMSD) between the predicted and corrected values ​​of the soil model parameters were compared, and the model with better prediction accuracy was selected to predict the soil backscattering model parameters in the entire study area.

[0288] Chapter 3 Quantitative inversion of soil moisture near the surface at 1 km during the melt season on the Qinghai-Tibet Plateau

[0289] In general, the accuracy and spatial resolution of the mainstream near-surface and root zone soil moisture products in the permafrost region of the Qinghai-Tibet Plateau cannot meet the needs of ecohydrological applications in the region. Although the fusion product ESA CCI soil moisture product has relatively high accuracy, its spatial resolution is relatively coarse (~25km), which is more suitable for global-scale application research. Since soil moisture has strong temporal and spatial heterogeneity, high-temporal and spatial resolution and high-precision soil moisture products are an urgent need for regional-scale applications in many fields such as climate change, hydrological simulation in cold and arid regions, crop yield estimation, and food security in the Earth system. So far, the 1km soil moisture data products publicly available in the Qinghai-Tibet Plateau study area are all generated based on downscaling methods. In addition, some studies have obtained soil moisture data for some small areas on the Qinghai-Tibet Plateau by assuming that the soil and vegetation parameters corrected at the station scale are also applicable to similar surface conditions. However, these parameters do not actually take into account the heterogeneity of the surface in the region, making the spatial applicability of the obtained regional high-resolution soil moisture data insufficient.

[0290] This chapter optimizes soil and vegetation parameters from a spatial scale based on a semi-empirical model (combining the water cloud model (WCM) and the Ulaby model), and obtains soil parameters for the entire study area for the first time. At the same time, different vegetation proxy indicators were selected from the optical, active and passive microwave scales to screen and obtain the optimal vegetation optical thickness. In addition, based on the soil moisture data of the stations in the vegetation coverage area, the calculation formula for vegetation moisture content applicable to the Qinghai-Tibet Plateau study area was calibrated and obtained. Finally, the soil moisture data with a spatial resolution of 1km in the melt season of 2017-2021 in the Qinghai-Tibet Plateau study area were inverted, and their performance was evaluated and compared based on the station observation data of 5 observation networks in the study area and 6 sets of published 1km soil moisture product data covering the Qinghai-Tibet Plateau.

[0291] The first part of this chapter outlines the high-resolution soil moisture backscatter inversion algorithm, process and inversion result evaluation method; the second part shows the correction results of soil and backscatter model parameters; the third part shows the accuracy verification and comparative analysis results of 1km soil moisture inversion data in the melting season; the fourth part analyzes and explains the uncertainties in parameter optimization and evaluation methods in the inversion algorithm; the fifth part is a summary of the contents of this chapter.

[0292] 3.1 Inversion algorithm and process

[0293] The Sentinel-1 soil moisture inversion process mainly includes the introduction of bare land and vegetation backscattering models, parameter correction in the backscattering model, and verification of soil moisture inversion results. Figure 1 As shown in the figure. The bare ground and vegetation backscatter models used in Sentinel-1 soil moisture inversion have been introduced in Chapter 2 and will not be expanded here. The evaluation method of the inverted soil moisture has been introduced in detail in Section 2.3.1. Here we mainly introduce the parameter correction process in the bare ground and vegetation backscatter models in detail.

[0294] 3.1.1 Optimization of soil backscattering model parameters

[0295] For the bare soil backscatter model, soil parameters C and D need to be corrected to make it applicable to the Qinghai-Tibet Plateau study area. Based on the soil moisture and backscatter time series of bare soil and sparse vegetation area pixels in 2017-2018, soil parameters C and D were first corrected; then, based on the random forest regression model, the corrected C and D values ​​were extrapolated to the entire study area. The specific correction steps are as follows:

[0296] (1) Screening pixels in bare land and sparse vegetation areas: Based on the MODIS IGBP land cover type and NDVI data, pixels with a land cover type of “bare land” and a NDVI time series value of less than 0.1 were screened as pixels in bare land and sparse vegetation areas.

[0297] (2) Calculation of soil parameters of the above-mentioned selected pixels: First, the time series data of soil moisture and backscatter are required as initial data to correct the soil parameters C and D of the Ulaby model. Since the verification results in Chapter 3 have shown that the overall performance of the ESA CCI soil moisture product published by ESA in the Qinghai-Tibet Plateau study area is better than other coarse-resolution soil moisture data. Therefore, this study used ESA CCI soil moisture data to correct the soil parameters of bare land and sparse vegetation pixels. Subsequently, in the selected pixels, the time series correlation (R) between the Sentinel-1 backscatter coefficient and the ESA CCI soil moisture data was calculated, and the soil parameters C and D of each pixel were estimated by linear fitting. Considering that the Sentinel-1 backscatter coefficient and ESA CCI soil moisture data may also have certain uncertainties under complex terrain conditions, in order to ensure the robustness of the fitted soil parameters, the soil parameters calculated for the pixel can be retained only when the following conditions are met: (1) there is a significant positive correlation between the Sentinel-1 backscatter coefficient and the ESA CCI soil moisture data (R>0&P-value<0.05); (2) the standard deviations of the Sentinel-1 backscatter coefficient and the ESA CCI soil moisture data must be higher than 0.25dB and 0.03m3 / m3 respectively; (3) the data of the Sentinel-1 backscatter coefficient and the ESA CCI soil moisture data that can be used for fitting must account for at least 30% of the total data volume of the entire time series. R is the Pearson correlation coefficient calculated between the Sentinel-1 backscatter coefficient and the ESA CCI soil moisture data; P-value indicates the significance of the correlation coefficient; P-value less than 0.05 is considered significant.

[0298] (3) Correction of soil parameters in the entire study area: Based on the estimated soil parameters C and D of bare land and sparse vegetation pixels, we used them as prediction targets and used 15 soil texture indices, 3 surface temperature indices calculated based on surface temperature data, and 8 terrain indices, a total of 26 indices, as covariates (Table 7). Based on the random forest method, we established two sets of estimation models for C and D parameters (C RF and D RF ), and predicted the soil parameters of the entire study area. In particular, based on the variable importance index, this paper selected the top importance indexes for the training and prediction of the C and D parameter models.

[0299] Table 7 Soil texture, surface temperature and terrain data information

[0300]

[0301] 3.1.2 Optimization of vegetation backscatter model parameters

[0302] The correction of vegetation backscatter model parameters includes two aspects: (1) correction of vegetation parameter A; and (2) selection of optimal vegetation optical thickness (τ) data to invert higher-precision soil moisture data (Chaudhary et al., 2021).

[0303] Vegetation parameter A represents the backscatter of the vegetation canopy under full vegetation coverage. When in the dense vegetation period (defined here as NDVI>0.4 and NDVI>80th (the top 80 percentile of the NDVI time series of a single pixel)), backscatter can hardly penetrate the vegetation, and the vegetation transmittance is almost 0 (γ2=0) (Konings et al., 2017). The selection of the NDVI threshold for the dense vegetation period defined above is the result of multiple tests and is also a compromise to ensure that the vegetation parameters for the entire melting period can be obtained. In this way, combined with formulas 2.12 and 2.13, the vegetation parameter A for the dense vegetation period is dense(t,i,j) It can be calculated by the following formula:

[0304]

[0305] in, represents the total backscatter received by the radar;

[0306] represents vegetation backscatter;

[0307] A dense(t,i,j) Vegetation parameters indicating dense vegetation period;

[0308] θ ref is the reference incident angle;

[0309] The value of A is equal to the value of all A on the dense vegetation pixel on the tth day. dense(t,i,j) The spatial average of the values ​​was calculated and the vegetation parameter A was assumed to be constant throughout the study area.

[0310]

[0311] Among them, A (t) represents all A on the dense vegetation pixel on the tth day dense(t,i,j) The spatial average of the values;

[0312] A dense(t,i,j) Vegetation parameters indicating dense vegetation period;

[0313] ∑A dense(t,i,j) represents all A on the dense vegetation pixel on the tth day dense(t,i,j) The sum of values;

[0314] For the selection of the best vegetation optical thickness data, this study considered seven proxy indicators, including optical (NDVI, LAI, NDWI), active microwave (RVI) and passive microwave (VOD) vegetation indices (Table 8). SMAP-L3 DCA VOD represents the vegetation optical thickness product data calculated by the DCA algorithm of SMAPL3 level products. Specifically, the vegetation optical thickness data is calculated according to formula 2.13, where the vegetation water content (VWC) can be obtained by formulas 4.2–4.7. Based on this, six types of soil moisture data (hereinafter referred to as SM) were inverted at the site scale. NDVI1 , SM NDVI2 , SM LAI , SM NDWI1 , SM NDWI2 , SM RVI ), and verified it based on soil moisture station observation data to select the best vegetation calculation formula. In addition, in order to make the selected VWC formula more suitable for the Qinghai-Tibet Plateau study area, this study also used the soil moisture data of the Maqu observation network and the above-mentioned calibrated soil and vegetation parameters to further calibrate the coefficients in the VWC formula.

[0315] Table 8: Formulas for calculating vegetation moisture content based on different vegetation proxy indicators

[0316]

[0317] 3.1.3 Evaluation Method

[0318] When correcting the parameters of the soil and vegetation backward models and verifying the Sentinel-1 soil moisture inversion results, four commonly used statistical indicators are mainly used, including R, ubRMSD, RMSD and Bias. When selecting the best vegetation proxy indicator, R and ubRMSD calculated between the soil moisture inverted based on different vegetation indicators and the station data are used as selection criteria. Taking into account the different amounts of data available for different indicator proxies, in order to make a fair comparison, this patent application only calculates statistical indicators for soil moisture data calculated by 7 vegetation indicators when there is available data at a common station. Therefore, among the 81 observation stations, 45 stations were selected to participate in the evaluation of the best vegetation proxy indicators.

[0319] To verify the Sentinel-1 soil moisture results (SM S-1 ), which is directly verified by this patent application based on the observation data of the melting season site, and is also verified by ESA CCI and other 6 published 1km soil moisture (SM Chai ,SM Li ,SM Song ,SM Sg,SM Han and SM SMAP ) were compared with each other (Table 2-5). Considering the different amounts of available data for different soil moisture products, the following two schemes were used to ensure the fairness and independence of the verification: (1) In addition to SM SMAP and ESA CCI products. This study evaluates the other six products based on observational data from 2017–2018. This evaluation period was chosen because SM Chai The available data of most observation stations are concentrated in this period. In order to obtain reliable statistical results, only these six sets of data (SM S-1 ,SM Chai ,SM Li ,SM Song ,SM Sg , and SM Han ) and data with common sites and dates between observation data are retained for calculating performance indicators; (2) Since the calibration of soil and vegetation backscatter model parameters was carried out in 2017-2018, in order to ensure the independence of verification, this study also used the observation data of the Naqu observation network from 2019 to 2021 to calibrate the parameters of the SM Chai The other 8 sets of product data were independently verified to supplement the above verification results. SMAP The limited amount of available data after quality control for ESA CCI products is not used to mask the data of different products, but the performance indicators between all available data and observation data of different stations of different products are directly calculated. For the above two evaluation schemes, if the R value calculated for all soil moisture product data is lower than 0.4, these stations will not be considered for the final product performance evaluation, because such a low R value may indicate that there is a certain uncertainty in the observation data.

[0320] 3.2 Backscattering model parameter correction results

[0321] 3.2.1 Soil backscattering model parameter correction results

[0322] In the bare land and sparse vegetation areas, a total of 101,523 pixels were selected for estimating soil parameters C and D. Figure 2 The figure shows the spatial distribution of the estimated C and D values ​​and the corresponding histograms. It can be seen that the selected pixels are mainly distributed in the northern, central and western regions of the Qinghai-Tibet Plateau study area, where the C value distribution range is -30–-5dB and the D value distribution range is 0–60dB.

[0323] Based on the estimated C and D values, two training models (C RF and D RF) to predict soil parameters in the entire study area. Figure 3 The figure shows the importance ranking of variables for training two regression models. Overall, 13 variables out of 26 were selected as C RF training variables, including 5 soil property indices, 3 soil temperature indices, and 5 terrain indices; 12 of the 26 variables were selected as D RF The training variables include 7 soil property indicators, 1 surface temperature indicator and 4 terrain indicators. Among them, Convexity, CEC and CFVO are the three most important variables for predicting C parameters, accounting for about 30% of the overall importance. For D parameters, terrain factors including DEM and Texture are the most important variables, which can be used to explain 25% of the target variable. As expected, due to the complex terrain of the study area, both soil parameters are highly sensitive to terrain-related variables.

[0324] The scatter plots between the corrected and predicted values ​​of soil parameters C and D show good consistency, indicating that RF and D RF The two regression models have high accuracy, with R values ​​and RMSD values ​​of 0.82 and 1.91 dB, and 0.81 and 6.30 dB, respectively. Figure 4 As shown. Figure 5 The cd in the figure shows the spatial distribution of the predicted values ​​of C and D in the entire study area. Overall, the predicted C values ​​are between -30 and -5 dB, and the C values ​​in high altitude areas are generally higher than those in low altitude areas. For the D parameter, its value is between 5 and 60 dB, and the D values ​​in the surrounding areas are generally higher than those in the hinterland of the Qinghai-Tibet Plateau.

[0325] 3.2.2 Vegetation backscatter model parameter correction results

[0326] On the Qinghai-Tibet Plateau, the value of the vegetation parameter A in the water cloud model is set to The average value of represents the total backscatter of dense vegetation pixels, θ ref represents the reference incident angle, such as Figure 5 The figure shows the temporal variation of the annual mean of vegetation index A from 2017 to 2021, with a temporal mean of 0.204±0.048. Compared with autumn and spring, the A(t) value in summer is more stable.

[0327] In order to determine the best vegetation proxy index formula in the water cloud model in the study area, this study obtained seven soil moisture inversion results at the site scale according to different vegetation proxy index formulas, namely SM NDVI1 , SM NDVI2 , SM LAI , SM NDWI1 , SM NDWI2, SM RVI and SM VOD .like Figure 6 The figure shows the performance indicators of the six soil moisture inversion results calculated between the observed data. It can be seen that SM NDVI1, SM NDVI2 , and SM LAI (median R>0.45, ubRMSD<0.07m 3 / m 3 ) outperformed the other three soil moisture inversion results (median R<0.27, ubRMSD<0.08m 3 / m 3 ), which indicates that NDVI and LAI have great potential for obtaining accurate vegetation backscatter components in the study area. Considering that the spatial coverage of NDVI in the study area is higher than that of LAI products, and SM can be seen at most observation stations NDVI2 The R and ubRMSD are slightly higher than SM NDVI1 Therefore, this study determined the calculation formula based on NDVI2 as the optimal vegetation equation for plateau soil moisture inversion. Figure 7 shown.

[0328] In order to further make the selected VWC calculation formula applicable to the Qinghai-Tibet Plateau study area, this study corrected the coefficients of the selected optimal vegetation calculation formula (i.e., Formula 4.3 VWC = a × NDVI) based on the site data of the Maqu observation network and the soil and vegetation parameters in the calibrated model through iterative calculation. 2–b×NDVI+c×(((NDVImax-NDVImin) / (1-NDVImin))), where VWC is the vegetation water content; a, b, and c represent the fitting coefficients of the vegetation water content formula; NDVI represents the normalized vegetation index; NDVImax represents the maximum value of the normalized vegetation index of the pixel; and NDVImin represents the minimum value of the normalized vegetation index of the pixel. Referring to the default parameters of the calculation formula (a=1.1934; b=0.3215; c=0.3), we first set different step sizes for each parameter and use them as initial values ​​for soil moisture inversion, where a=(0.5:0.4:2.5); b=(0.2:0.1:0.5); c=(0:0.05:0.3). Since a large number of tests have been carried out in the early stage, in order to focus on showing the main correction results, the step size range of different coefficients is relatively narrow when showing. The present invention lists the statistical indicators between the soil moisture values ​​inverted using different coefficient combinations and all the observation data of the Maqu observation network. Comparison shows that when the coefficients are set to a = 0.5, b = 0.5 and c = 0.15, the statistical indicators are the best, and the accuracy of the soil moisture inverted at this time is also better than the soil water inverted using the default coefficients. Therefore, this study uses the optimized VWC calculation formula to invert the soil moisture of the entire study area of ​​the Qinghai-Tibet Plateau.

[0329] 3.3 Analysis on the applicability of 1km soil moisture inversion algorithm in the Qinghai-Tibet Plateau during the melting season

[0330] 3.3.1 Comparative analysis of spatial characteristics of soil moisture inversion results during the melting season

[0331] In terms of spatial performance, the soil moisture (SM S-1 ) and 7 other soil moisture data sets (ESACCI, SM Chai ,SM Li .SM Song ,SM Sg ,SM Han , and SM SMAP ) were compared with each other. In addition, since the soil moisture values ​​of different grasslands in the study area are: alpine meadow (AM) > alpine grassland (AS) > alpine desert (AD). Therefore, this study also conducted regional statistics on soil moisture products according to different grassland types. Figure 8 The figure shows the SM in the Qinghai-Tibet Plateau study area from 2017 to 2021. S-1 And the spatial distribution of the time series means of the other 7 sets of soil moisture data. SM S-1 The spatial distribution of SM is similar to that of ESACCI. Its soil moisture value decreases from the southeast to the northwest of the study area. However, compared with the coarse-resolution ESACCI product, SM S-1Can reflect more spatial details. At the same time, SM S-1 The spatial pattern also reflects the distribution characteristics of grassland in the study area. Fig. 9 As shown, it can further explain SM S-1 It can reflect more spatial details. Specifically, the areas with high soil moisture values ​​are distributed in the Three Rivers Source area and the Yarlung Zangbo River Basin, which are mainly alpine meadows; while the areas with low soil moisture values ​​are mainly distributed in the Qaidam Basin, which are mainly bare land and alpine deserts. SM Sg Basically, it shows a similar spatial pattern, but the soil moisture values ​​are lower, especially in the densely vegetated areas.

[0332] In order to better analyze SM S-1 The spatial pattern of soil moisture products was also analyzed according to different grassland types. Fig.10 As shown in Figure 2, as the grassland type changes from alpine meadow to alpine steppe to alpine desert, the mean moisture value of all soil moisture products generally shows a downward trend, but the mean moisture value of different products in different grassland types varies. S-1 The average moisture values ​​in alpine meadows, alpine steppes and alpine deserts are 0.24m 3 / m 3 、0.22m 3 / m 3 and 0.19m 3 / m 3 , and ESA CCI (0.25m 3 / m 3 , 0.21m 3 / m 3 and SM=0.18m 3 / m 3 ) and SM Sg (0.21m 3 / m 3 , 0.20m 3 / m 3 and 0.17m 3 / m 3 ) is closer to the mean moisture value. Chai , SM Song and SM SMAP The average moisture content of the soil in different vegetation types is lower than 0.21m 3 / m 3 (AM), 0.14m 3 / m 3 (AS) and 0.11m 3 / m 3 (AD), while SM Li and SM HanThe average moisture content of 3 / m 3 (AM), 0.23m 3 / m 3 (AS) and 0.22m 3 / m 3 (AD).

[0333] 3.3.2 Comparison and verification of the time series characteristics of soil moisture inversion results during the melting season

[0334] In evaluating SM S-1 In terms of time series reliability, we first used site observation data to directly verify it and compared it with the other seven sets of soil moisture products. Fig.11 and Table 9 show the SM S-1 and 5 other soil moisture products (SM Chai , SM Li , SM Song , SM Sg and SM Han ) and the performance statistical indicators of the observation data of the five observation networks, as well as the median values ​​of all indicators of different observation networks and all observation sites. In terms of R value, except for SMSg (the median R value is 0.69), SM S-1 The median R value (0.56) is better than that of several other soil moisture products, followed by SM Han , SM Chai , SM Song and SM Li , and the median R values ​​were 0.51, 0.45, 0.44, and 0.42, respectively. S-1 The median RMSD of ub(0.064m 3 / m 3 ) is slightly higher than other soil moisture products (between 0.040m 3 / m 3 –0.058m 3 / m 3 SM S-1 , SM Chai , SM Song and SM Sg The median RMSD values ​​of 3 / m 3 Between. SM Han The median RMSD of 3 / m 3 , while SM Li The median RMSD of 3 / m 3. Relative to the observed soil moisture, SM Li It shows a low dry deviation (median deviation is -0.044m 3 / m 3 ). SM Li The station observation data was overestimated, and the median wet deviation was higher at 0.128m 3 / m 3 , while SM Chai The observed data are underestimated, with a higher median dry deviation of -0.096m 3 / m 3 For the other three sets of soil moisture products, their median deviations are all low, ranging from -0.022–0.019m 3 / m 3 .

[0335] For different observation networks, SM S-1 and SM Sg The performance of most observation networks is good (R median value > 0.4), among which the Naqu observation network (which contains the largest number of stations) has the best performance index, with R median values ​​of 0.62 and 0.81 respectively. It is followed by the Maqu, CRS, Ali and Pali observation networks, and SM S-1 and SM Sg The median R values ​​of SM were 0.60 / 0.45 / 0.51 and 0.38 and 0.45, 0.67, 0.68 and 0.42, respectively. S-1 (ubRMSD median is 0.048–0.068m 3 / m 3 In most soil moisture observation networks, the median RMSD of ub is slightly higher than that of other products (0.019–0.082 m 3 / m 3 Between), SM Li The median RMSD values ​​were the highest in CRS and Pali, which were 0.149m 3 / m 3 and 0.222m 3 / m 3 Except for the Maqu observation network (the median deviation is -0.129m 3 / m 3 ) In addition, SM S-1 The bias is low in most observation networks, ranging from -0.076–0.051 m 3 / m 3 In all observation networks, SM Chai All showed dry deviation (-0.137–-0.022m 3 / m 3 ), SM LiAll showed wet deviation (0.038–0.218m 3 / m 3 ). In addition, in the CRS observation network, SM Li , SM Sg and SM Han The wet deviation is also large, ranging from 0.106–0.157m 3 / m 3 .like Fig.12 –14 shows the SM S-1 And 7 other sets of soil moisture product data, Sentinel-1 backscatter and NDVI value time series changes. S-1 The soil moisture value of SM can well capture the dry-wet changes of the observed data, while Chai The moisture estimate is drier than the observed value, SM Li The estimated value of is wetter than the observed value.

[0336] Table 9 2017–2018 SM S-1 ,SM Chai ,SMLi,SM Song ,SM Sg ,and SM Han Statistical indicators between soil moisture products and observation data

[0337]

[0338] The last column is the median of the statistical indicators of all stations. The best performance among the six soil moisture products of different observation networks is bolded.

[0339] In addition, considering that the calibration of the backscatter model in this study was carried out in 2017-2018, the independence of the above verification based on this period may be insufficient. Therefore, this study further verified the calibration of the backscatter model based on the site data of the Naqu observation network in 2019-2021 except for SM Chai The product data of SM S-1 Supplementary to data performance evaluation. Fig.15 The shown shows the SM S-1 and 6 other soil moisture products (SM Li , SM Song , SM Sg , SM Han , SM SMAP and ESA CCI) assessment results, you can see SM S-1 , SM Sg , SM Hanand ESA CCI outperformed other soil moisture products with higher R median (R>0.75), lower ubRMSD (ubRMSD<0.060m 3 / m 3 )、RMSD(RMSD<0.100m 3 / m 3 ) and deviation (Bias<0.074m 3 / m 3 ). SM SMAP The R value is the highest, but the ubRMSD value is also the largest (R = 0.89, ubRMSD = 0.102m 3 / m 3 ). SM Li and SM Song The overall performance is poor, with a low R median value of less than 0.69 and a high ubRMSD value (ubRMSD>0.064m 3 / m 3 ). In addition, SM Li RMSD (RMSD = 0.178 m 3 / m 3 ) and Bias(Bias=0.153m 3 / m 3 )Highest.

[0340] 3.4 Uncertainty analysis

[0341] The verification results show that the soil moisture inversion algorithm proposed in this study can be well used to invert the soil moisture data of 1 km in the Qinghai-Tibet Plateau study area. Compared with the observed data, its overall R median value is 0.56 and the ubRMSD is 0.066m 3 / m 3 And the deviation value is -0.042m 3 / m 3 SM S-1 The data show a similar spatial pattern to the ESA CCI, which is a natural consequence of using the ESA CCI soil moisture data in the backscatter model to correct its soil parameters and the use of high-resolution Sentinel-1 backscatter data. At the same time, compared with other available 1km soil moisture products, SM S-1 It also presents a reasonable spatial distribution. Compared with other products, it is more consistent with the spatial distribution of grassland type maps in the Qinghai-Tibet Plateau study area.

[0342] The above verification results also show that SM Sg It has a similar spatial pattern and better time series variation to ESA CCI, which is due to the fact that SM SgThe downscaling study was conducted using the ESA CCI soil moisture product as the prediction target data. In addition, the evaluation found that SM Sg In high vegetation cover areas, the SM S-1 This may be partly due to the lower value of SM Sg It is generated by machine / deep learning methods using optical / thermal infrared index as predictor, which lacks consideration of the physical relationship between soil moisture and predictor in complex terrain areas. The study also found that at most stations, SM Chai , SM Li and SM Song It performs poorly in capturing the temporal variation of in situ SM, with low R values ​​compared with field observations and data, and SM Chai Than SM S-1 Lower values ​​are drier, while SM Li Than SM S-1 The possible reasons are as follows: (1) We noticed that SM Chai The temporal changes of the NDVI values ​​are basically consistent with those of the NDVI values ​​( Figure 4-12 ), this is because when using multivariate statistical regression methods, SM Chai The temporal changes in moisture values ​​are mainly determined by the temporal changes in vegetation or temperature variables (such as NDVI, LST), which limits the capture of fluctuations in soil moisture data over a short period of time; (2) Li et al. (2022) used soil moisture observation data from regional stations in China as the target variable and obtained soil moisture data for the entire Chinese region based on machine learning methods. However, we found that the number of stations used for machine learning in the Qinghai-Tibet Plateau study area is limited and unevenly distributed, which will cause uncertainty in soil moisture values ​​in the region to a certain extent. In addition, SM Li Compared with the observed data, the large wet bias is likely due to the fact that the errors in the covariates used in machine learning are transferred to the soil moisture downscaling results. Li When downscaling, ERA5-Land soil moisture data was used as one of the covariates, but in fact, this product has been shown to have a large wet bias in the Qinghai-Tibet Plateau study area. This also shows that it is very important to choose a suitable covariate to downscaling soil moisture data.

[0343] Nevertheless, on some sites, SM S-1 There is also some uncertainty in the inversion, which is related to the correction of soil and vegetation backscattering models, SM S-1 This is related to the use of some empirical parameters in algorithm retrieval and factors such as uncertainty in the evaluation method.

[0344] 3.4.1 Soil parameters

[0345] When correcting the soil backscatter model, the ESA CCI soil moisture product is used to estimate the soil parameters C and D of the pixels in bare land and sparse vegetation areas. The implicit assumption is that within the 0.25°×0.25° pixel, the soil moisture value of each 1km pixel is equal to the ESA CCI soil moisture value of the corresponding pixel. Although previous studies have shown that the ESA CCI soil moisture product is more accurate than other coarse resolution products in the Qinghai-Tibet Plateau study area, it may still be inaccurate in the inversion of SM. S-1 Some uncertainties are introduced when calibrating soil parameters. Due to the complex surface and large variations in soil properties in the study area, soil moisture data of a coarse-scale pixel will have large spatial heterogeneity. Therefore, in the future, when calibrating soil parameters, it is necessary to further consider using area weighting methods related to altitude, topography, and land cover type to improve SM S-1 The inversion accuracy is high. At the same time, the soil parameter sample data used to train the random forest regression model are limited and have limitations in spatial distribution, mainly affected by the following factors: (1) After quality control, the spatial distribution of the ESA CCI data in the Qinghai-Tibet Plateau study area is difficult to cover the entire study area; (2) Only the Sentinel-1 backscatter and ESA CCI soil moisture data during the melting period can be used. Therefore, when screening pixels for soil moisture parameter correction, this paper sets a relatively loose threshold to avoid eliminating too many sample data, which will introduce certain uncertainties. However, the selection of this threshold is a compromise, because although using a stricter threshold can obtain more accurate soil parameter samples, it will also greatly reduce the data samples used for random forest regression model training.

[0346] 3.4.2 Vegetation parameters

[0347] For the correction of the soil backscatter model, this study is based on the work of Bai et al. (2022). By assuming that the vegetation transmittance is close to 0, the soil parameter A can be calculated during the dense vegetation period. However, this assumption is difficult to fully meet in the Qinghai-Tibet Plateau where the vegetation cover is relatively low. In addition, this study sets the A parameter as a constant in the study area every day, lacking consideration of the spatial heterogeneity of different regions of the plateau and the possible inconsistency of the A parameter values ​​of different grassland types. For the above reasons, in future studies, obtaining the soil parameter A of different grassland types may improve the accuracy and spatial pattern of soil moisture inversion. At the same time, it is necessary to further optimize the parameters of the selected VWC equation to make it more suitable for the study area. In order to illustrate the accuracy of the VWC equation correction coefficient, we compared the soil moisture data calculated by the default coefficient and the correction coefficient in the optimal VWC equation with the validation data of Maqu and other four observation networks. The results show that in the Maqu observation network, the soil moisture data obtained by correcting the coefficients is more accurate than the moisture value obtained based on the default coefficients, with higher R and lower ubRMSD. However, there is no significant difference in the accuracy between the two in the evaluation results of other observation networks. Fig.16 This result is reasonable because the surface vegetation coverage of other observation networks is low and their NDVI values ​​are also low, so the calculated vegetation optical thickness values ​​are also low, which leads to a low influence of vegetation optical thickness on the soil moisture values ​​inverted by these observation networks. In addition, when inverting soil moisture in this study, the empirical value of the B parameter was directly selected with reference to relevant studies. In the future, the soil moisture inversion accuracy of the study area can be improved by simultaneously correcting and optimizing the vegetation parameters A and B.

[0348] 3.4.3 Uncertainty of verification methods

[0349] In SM S-1 In terms of performance evaluation, due to the limited time series of observation points in the Qinghai-Tibet Plateau study area, the calibration of soil and vegetation parameters in the study was mostly based on observation data from 2017 to 2018. S-1 The evaluation of SM also used observation data from the same period, which may bring some uncertainty to the research results. In addition, this study directly compared the data at the 1km pixel scale with the site data of the sparse network, and the results may be affected by the representative error of the site. However, despite these limitations, the evaluation method in this study is still reasonable. First, as an evaluation of SM S-1To supplement the evaluation results, this paper further collected observation data from the Naqu study area from 2019 to 2021 for independent evaluation and observed similar evaluation results; secondly, when the correlation of all soil moisture products at a single observation station is lower than 0.4, the performance indicators of the observation station are abandoned, because this may indicate that there is a certain uncertainty in the observation data of the station. Accordingly, this also shows that a dense observation network should be deployed in the Qinghai-Tibet Plateau study area in the future to achieve authenticity verification, which plays an important role in further developing high-resolution soil moisture inversion algorithms in this region.

[0350] 3.5 Data Security

[0351] Data security is the lifeline of the information age. In this era of information explosion, data has become an intangible asset, and its value is self-evident. However, data leakage, tampering or abuse may bring immeasurable losses to individuals, enterprises and even society. Therefore, data security is not only a basic requirement for enterprise operations, but also an important guarantee for enterprise development.

[0352] In order to ensure the security of the kilometer-level soil moisture inversion working method, prevent the leakage of the kilometer-level soil moisture inversion working method during transmission, and avoid losses, this patent application encrypts the studied kilometer-level soil moisture inversion working method, and transmits the data through a U disk, a mobile hard disk or a network after encryption; the specific implementation is as follows:

[0353] The first step is to convert the kilometer-level soil moisture inversion working method studied in this patent application into binary, that is, only including the values ​​0 and 1, wherein the length of the kilometer-level soil moisture inversion working method converted into binary is L;

[0354] ζ2=Binaryway <inputofdata>4.8)

[0355] Among them, ζ2 represents the binary kilometer-level soil moisture inversion working method data;

[0356] Binaryway<> indicates a binary conversion method. If the kilometer-level soil moisture inversion working method data developed in this patent application is binary, no conversion is required; if the kilometer-level soil moisture inversion working method data developed in this patent application is other bases, it needs to be converted into binary;

[0357] Inputofdata represents the kilometer-level soil moisture inversion working method data developed in this patent application;

[0358] The second step is to convert the kilometer-level soil moisture inversion method developed in this patent application into a key:

[0359] SecurityKey=SecurityKeyway <inputofdata>4.9)

[0360] Among them, SecurityKey represents the key that has been studied;

[0361] SecurityKeyway<> key conversion method; sha512 is used in this patent application;

[0362] Inputofdata represents the kilometer-level soil moisture inversion working method data developed in this patent application;

[0363] SecurityKey2=Binaryway <securitykey>4.10)

[0364] SecurityKey2 represents a binary key;

[0365] Binaryway<> indicates a binary conversion method. If the key developed by the present patent application is binary, no conversion is required. If the key developed by the present patent application is other bases, it needs to be converted into binary.

[0366] SecurityKey represents the key that has been researched;

[0367] The third step is to perform the following calculations on the key and the binary kilometer-level soil moisture inversion working method data ζ2 to obtain secure kilometer-level soil moisture inversion working method data:

[0368]

[0369] in, Indicates that safe kilometer-level soil moisture inversion working method data is obtained;

[0370] ζ2 represents the binary kilometer-level soil moisture inversion working method data;

[0371] Represents binary exclusive or exclusive operation;

[0372] SecurityKey2 represents a binary key;

[0373] This means to remove the 0 or 1 after SecurityKey2 to make them equal in length.

[0374] If indicates a logical conditional assumption;

[0375] |ζ2| represents the length of ζ2, and its value is L;

[0376] |SecurityKey2| indicates the length of SecurityKey2;

[0377] ‖SecurityKey2 means adding 1 to the front of SecurityKey2 to make their lengths equal;

[0378] Step 4: Up to the third step above, the safe kilometer-level soil moisture inversion working method data is obtained. The obtained safe kilometer-level soil moisture inversion working method data can be stored in a USB flash drive, a mobile hard disk, or the data can be transmitted over the network to ensure security; at the same time, the key is saved in the hands of the user to decrypt the data; the decryption process is the reverse process of encryption, and the kilometer-level soil moisture inversion working method is used after decryption. The decryption process is:

[0379] Get the key from the user and use the key SecurityKey to decrypt the encrypted data in the USB flash drive and mobile hard disk:

[0380]

[0381] Among them, ξ2 represents the decrypted data;

[0382] SecurityKey2 represents a binary key;

[0383] This means to remove the 0 or 1 after SecurityKey2 to make them equal in length.

[0384] Represents binary exclusive or exclusive operation;

[0385] Indicates that safe kilometer-level soil moisture inversion working method data is obtained;

[0386] If indicates a logical conditional assumption;

[0387] |ζ2| represents the length of ζ2, and its value is L;

[0388] |SecurityKey2| indicates the length of SecurityKey2;

[0389] ‖SecurityKey2 means adding 1 to the front of SecurityKey2 to make their lengths equal;

[0390] The decrypted data ξ2 is converted into the same base as the kilometer-level soil moisture inversion working method data developed in this patent application, namely ξ2′; the key corresponding to ξ2′ is determined:

[0391] Key=SecurityKeyway<ξ2′> 4.13)

[0392] Among them, Key represents the calculated key;

[0393] SecurityKeyway<> key conversion method; sha512 is also used;

[0394] ξ2′ represents the decrypted data;

[0395] Finally, the two keys are consistent, that is, Key = SecurityKey, then the decrypted data ξ2′ can be used to invert the kilometer-level soil moisture and identify the high-resolution soil moisture content.

[0396] 3.6 Chapter Summary

[0397] Based on the Sentinel-1 backscatter data, this chapter proposes a soil moisture inversion method that combines the water cloud model and the Ulaby model and corrects the soil and vegetation parameters respectively. Based on this method, the soil moisture data of 1 km in the Qinghai-Tibet Plateau study area from 2017 to 2021 was inverted. Among them, the soil parameters were first corrected in bare land and sparse vegetation areas, and then extrapolated to the entire study area based on the random forest regression model; at the same time, it was assumed that the vegetation parameter B was fixed and unchanged, and the vegetation parameter A was corrected based on the dense vegetation period. The implicit assumption is that in the dense vegetation period, the vegetation transmittance is almost 0. In addition, different soil moisture values ​​were inverted based on 6 vegetation proxy formulas to select the optimal vegetation proxy calculation formula, and on this basis, the coefficients of the optimal vegetation proxy formula were corrected using the observed data and the calibrated soil and vegetation parameters to make it applicable to the study area. Finally, the SM was corrected based on the observation data of the stations in the 5 observation networks in the study area. S-1 The inversion results were evaluated and compared with six other publicly available 1km soil moisture products. The main conclusions are as follows:

[0398] (1) This patent application is the first to obtain 1km-scale soil moisture data of the Qinghai-Tibet Plateau study area based on a semi-empirical backscattering model that takes into account the spatial heterogeneity of soil parameters;

[0399] (2) Compared with the observed data, SM S-1 The inversion results are better than most 1km soil moisture products, with a median R value of 0.57 and a ubRMSD of 0.064m 3 / m 3 , RMSD is -0.107m 3 / m 3 , the deviation is -0.042m 3 / m 3 Although SM S-1 The performance indicators are slightly inferior to SM Sg , but SM S-1 is the result of inversion based on the semi-empirical backscattering model, while SM Sg It is the result of the fusion of different machine learning methods. The former can explore the relationship between microwave backscattering and soil and vegetation parameters based on the constructed mathematical model;

[0400] (3)SM S-1 The inversion results show a reasonable spatial distribution pattern, which is consistent with the spatial distribution of the grassland type map. The areas with high soil moisture values ​​are distributed in the Three Rivers Source Area and the Yarlung Zangbo River Basin, which are mainly alpine meadows; while the areas with low soil moisture values ​​are mainly distributed in the Qaidam Basin, which are mainly bare land and alpine deserts.

[0401] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.< / securitykey> < / inputofdata> < / inputofdata>

Claims

1. A kilometer-level soil moisture inversion method for the Qinghai-Tibet Plateau based on massive big data, characterized in that: The following steps are involved: S1, obtains the collected multi-source remote sensing observations and soil moisture big data; S2, optimizing soil and vegetation parameters from a spatial scale according to the soil moisture big data collected in step S1 to obtain spatial soil parameters; S3, selecting different vegetation proxy indicators from one or any combination of optical, active and passive microwave data to obtain vegetation optical thickness; S4, based on the soil moisture data of the vegetation coverage area, calibrate and obtain the vegetation moisture content suitable for the alpine grassland area. The moisture content is calculated as follows: in, represents the reference incident angle θ ref Normalized backscatter; represents the backscattering observed at the local incident angle θ; θ ref represents the reference incident angle; θ represents the local angle of incidence; in, represents the total backscatter received by the radar; represents vegetation backscatter; γ 2 is the vegetation penetration coefficient; represents soil backscatter; in, represents vegetation backscatter; A represents the backscatter of the vegetation canopy under full vegetation cover; V1 represents vegetation parameters; θ ref is the reference incident angle; γ 2 is the vegetation penetration coefficient; Among them, γ 2 is the vegetation penetration coefficient; e is the natural base; τ is the vegetation optical thickness; θ ref is the reference incident angle; τ=B×VWC (2.13) Where, τ is the vegetation optical thickness; B is an empirical parameter that depends on the vegetation canopy type; VWC is vegetation water content; in, represents soil backscatter; C is the soil parameter, representing the backscattering under dry soil conditions; D is a soil parameter representing the sensitivity of backscattering to soil moisture; SM represents soil moisture parameter; Among them, SM represents soil moisture parameter; represents the total backscatter received by the radar; represents vegetation backscatter; γ 2 is the vegetation penetration coefficient; C and D are soil parameters.

2. The kilometer-level soil moisture inversion method of the Qinghai-Tibet Plateau based on massive big data according to claim 1 is characterized in that: The method for collecting soil moisture big data in step S1 includes field observation data, soil moisture product data set, Sentinel-1 backscatter data, other auxiliary data, or any combination thereof.

3. The kilometer-level soil moisture inversion method of the Qinghai-Tibet Plateau based on massive big data according to claim 1 is characterized in that: In step S2, spatial downscaling and active microwave remote sensing are used to optimize soil and vegetation parameters.

4. The kilometer-level soil moisture inversion method of the Qinghai-Tibet Plateau based on massive big data according to claim 1 is characterized in that: The correction method for soil parameters C and D is: Screening of pixels of bare land and sparse vegetation areas; Calculate soil parameters for filtered cells.

5. The method for inverting kilometer-level soil moisture in the Qinghai-Tibet Plateau based on massive big data according to claim 1 is characterized in that: Calibration of soil parameters for the entire study area was also included.

6. The method for inverting kilometer-level soil moisture in the Qinghai-Tibet Plateau based on massive big data according to claim 1 is characterized in that: The correction method of vegetation parameter A is: in, represents the total backscatter received by the radar; represents vegetation backscatter; A dense(t,i,j) Vegetation parameters indicating dense vegetation period; θ ref is the reference incident angle; Among them, A (t) represents all A on the dense vegetation pixel on the tth day dense(t,i,j) The spatial average of the values; A dense(t,i,j) Vegetation parameters indicating dense vegetation period; ∑A dense(t,i,j) represents all A on the dense vegetation pixel on the tth day dense(t,i,j) The sum of values; A represents the backscatter of the vegetation canopy under full vegetation cover.

7. The method for inverting kilometer-level soil moisture in the Qinghai-Tibet Plateau based on massive big data according to claim 1 is characterized in that: Also included are evaluation indicators, the evaluation indicators including one or any combination of mean deviation, Pearson correlation coefficient, root mean square error, and unbiased root mean square error; The average deviation is calculated as: Among them, Bias represents the average deviation; N represents the total number of data; Represents product soil moisture data; represents observed soil moisture data; The Pearson correlation coefficient calculation formula is: Where R represents the Pearson correlation coefficient; N represents the total number of data; Represents product soil moisture data; Represents the average value of product soil moisture data; represents observed soil moisture data; Represents the average value of observed soil moisture data; The formula for calculating the root mean square error is: Among them, RMSE means root mean square error; N represents the total number of data; Represents product soil moisture data; represents observed soil moisture data; The unbiased root mean square error calculation formula is: Among them, ubRMSE represents the unbiased root mean square error; RMSE stands for root mean square error; Bias represents the average deviation.

8. The method for inverting kilometer-level soil moisture in the Qinghai-Tibet Plateau based on massive big data according to claim 1 is characterized in that: The correction method for vegetation water content (VWC) is one of the following: Among them, VWC is vegetation water content; a1, a2, b1, and b2 represent the fitting coefficients of the vegetation moisture content formula; NDVI stands for Normalized Difference Vegetation Index; VWC=a×NDVI 2 –b×NDVI+c×(((NDVImax-NDVImin) / (1-NDVImin))), Among them, VWC is vegetation water content; a, b, and c represent the fitting coefficients of the vegetation moisture content formula; NDVI stands for Normalized Difference Vegetation Index; NDVImax represents the maximum value of the normalized difference vegetation index of the pixel; NDVImin represents the minimum value of the normalized difference vegetation index of the pixel; VWC=exp(LAI / a3)-b3 Among them, VWC is vegetation water content; Among them, LAI stands for leaf area index; a3 and b3 represent the fitting coefficients of the vegetation moisture content formula; VWC=a4*NDWI+b4 Among them, VWC is vegetation water content; a4 and b4 represent the fitting coefficients of the vegetation moisture content formula; NDWI stands for Normalized Difference Water Index; <h2 style=";text-align:left;direction:ltr">VWC=a5*NDWI<h2 style=";text-align:left;direction:ltr"> 2 <h2 style=";text-align:left;direction:ltr"> +b5*NDWI+c5 Among them, VWC is vegetation water content; a5, b5, and c5 represent the fitting coefficients of the vegetation water content formula; NDWI stands for Normalized Difference Water Index; VWC=a6*RVI+b6 Among them, VWC is vegetation water content; a6, b6, and c6 represent the fitting coefficients of the vegetation moisture content formula; represents VH polarized backscattering; represents VV polarized backscatter.

9. The method for inverting kilometer-level soil moisture in the Qinghai-Tibet Plateau based on massive big data according to claim 1 is characterized in that: The correction method of vegetation water content VWC is: VWC=a×NDVI 2 –b×NDVI+c×(((NDVImax-NDVImin) / (1-NDVImin))), Among them, VWC is vegetation water content; a, b, and c represent the fitting coefficients of the vegetation moisture content formula; NDVI stands for Normalized Difference Vegetation Index; NDVImax represents the maximum value of the normalized difference vegetation index of the pixel; NDVImin represents the minimum value of the normalized difference vegetation index of the pixel.

Images