Method and system for generating surface soil moisture based on spatio-temporal fusion of remote sensing data

Abstract

The application discloses a method and system for generating surface soil humidity based on spatio-temporal fusion of remote sensing data, and relates to the technical field of soil humidity processing. The method comprises the following steps: obtaining a vegetation parameter image, an active microwave remote sensing image and a passive microwave remote sensing image of a required date, superimposing the vegetation parameter image and the active microwave remote sensing image to obtain a high-resolution parameter image; resampling the parameter image to a low resolution, and performing spatial downscaling on the passive microwave remote sensing image through the low-resolution parameter image; taking a time sequence of each pixel in the passive microwave remote sensing image as a signal, obtaining phase grid data and high-frequency amplitude grid data of different frequencies of the signal, and performing spatial interpolation; and fitting the spatial interpolation result and the spatial downscaling result in a time dimension to obtain a daily-scale surface soil humidity image. The application realizes fitting of spatial resolution and time resolution, obtains a soil humidity image with fine spatio-temporal resolution, and has high application value.

Description

Technical Field

[0001] This invention relates to the field of soil moisture treatment technology, and in particular to a method and system for generating surface soil moisture based on the spatiotemporal fusion of remote sensing data. Background Technology

[0002] Soil moisture is a key factor in the energy and water cycles of the Earth system and an important parameter in many fields, including environmental and agricultural sciences. On a global scale, soil moisture is extremely valuable for weather forecasting, climate change monitoring, and monitoring extreme events such as floods. On a regional scale, soil moisture information is an indispensable part of local agriculture and water resource management. Remote sensing technology overcomes the shortcomings of traditional methods—such as its fast response time, wide monitoring range, long-term dynamic monitoring, and spatial continuity—by providing an effective means of acquiring large-scale soil moisture information.

[0003] Soil moisture retrieval methods based on remote sensing can be broadly categorized into three types: those based on optical, thermal infrared, and microwave radar data. The emission and scattering characteristics of soil in the microwave band are influenced by its dielectric properties, and the soil dielectric constant is directly related to soil moisture content; this is the physical basis for microwave remote sensing inversion of soil moisture. Microwave remote sensing is not limited by weather conditions such as sunlight and fog, has strong cloud penetration capabilities, can detect soil information under vegetation cover, and possesses all-weather simulation capabilities and high sensitivity to changes in soil moisture. In terms of the continuous availability of large-scale data products, microwave-based soil moisture products are far more advanced than those based on optical and thermal infrared data.

[0004] In existing methods, one or two pairs of coarse and fine resolution images are usually required as reference images, along with a large amount of remote sensing and auxiliary data that are difficult to obtain. Then, a coarse resolution image of the target date is used to generate a fine resolution image. This results in a large demand for computing power in existing technologies, low efficiency in generating fine resolution images, and because there is a contradiction between temporal and spatial resolution in microwave remote sensing, it is impossible to directly obtain soil moisture products with fine spatiotemporal resolution, resulting in insufficient spatiotemporal resolution in the obtained soil moisture products. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of the prior art by providing a method and system for generating surface soil moisture based on the spatiotemporal fusion of remote sensing data. This solves the problem in the prior art where the contradiction between temporal and spatial resolution in microwave remote sensing makes it impossible to directly obtain soil moisture products with fine spatiotemporal resolution, resulting in insufficient spatiotemporal resolution in the obtained soil moisture products.

[0006] The present invention specifically provides the following technical solution:

[0007] A method for generating surface soil moisture based on spatiotemporal fusion of remote sensing data includes:

[0008] Acquire vegetation parameter images, active microwave remote sensing images, and passive microwave remote sensing images for the desired dates, and overlay the vegetation parameter images with the active microwave remote sensing images to obtain high-resolution parameter images; wherein the spatial resolution of the high-resolution images is higher than a set threshold, and the spatial resolution of the low-resolution images is lower than a set threshold.

[0009] The spatial resolution of the parametric image is resampled to a low resolution, and the passive microwave remote sensing image is spatially downscaled using the low-resolution parametric image to obtain a temporally discrete initial surface soil moisture image.

[0010] The time series of each pixel in the passive microwave remote sensing image is used as a signal. Phase raster data and high-frequency amplitude raster data of the signal at different frequencies are obtained. The phase raster data and high-frequency amplitude raster data are spatially interpolated. The spatial interpolation result is fitted with the initial surface soil moisture image in the time dimension to obtain a daily-scale surface soil moisture image.

[0011] Preferably, obtaining vegetation parameter images for the desired date includes:

[0012] Obtain different combinations of vegetation parameters;

[0013] Different combinations of vegetation parameters are input into the radiative transfer model to simulate and obtain spectral curves. The spectral curves are then converted into atmospheric apparent reflectance to obtain the band reflectance captured by the model.

[0014] The reflectance of the band is input into a random forest model for inversion to obtain vegetation parameter images for the required dates; wherein, the vegetation parameter images include leaf area index (LAI), canopy water content (CWC), and mean leaf tilt angle (ALA) images.

[0015] Preferably, the step of overlaying the vegetation parameter image with the active microwave remote sensing image to obtain a high-resolution parameter image specifically involves:

[0016] By overlaying active microwave remote sensing images with vegetation parameter images containing leaf area index (LAI), canopy water content (CWC), and mean leaf inclination angle (ALA), a set of images containing seven bands including normalized difference vegetation index (NDVI), leaf area index (LAI), canopy water content (CWC), mean leaf inclination angle (ALA), vertical polarization (VV), cross polarization (VH), and radar wave incident angle (θ) is obtained.

[0017] The images of the seven bands were used as high-resolution parametric images.

[0018] Preferably, the step of spatially downscaling the passive microwave remote sensing image using low-resolution parametric imagery to obtain a temporally discrete initial surface soil moisture image includes:

[0019] The spatial resolution of the high-resolution parametric image is resampled to a low resolution and matched with the pixels of the passive microwave remote sensing image. The pixels are then used as a training set to train the surface soil moisture inversion model.

[0020] High-resolution parametric images are input into the trained surface soil moisture inversion model to obtain high-resolution initial surface soil moisture images under discrete time conditions.

[0021] Preferably, the time series of each pixel in the passive microwave remote sensing image is used as a signal to acquire phase raster data and high-frequency amplitude raster data of the signal at different frequencies, including:

[0022] Perform a Fast Fourier Transform on the signal to obtain the amplitude and phase at different frequencies of the signal;

[0023] The timing SSM at different spatial resolutions is represented by amplitudes and phases at different frequencies, specifically as follows:

[0024]

[0025]

[0026] Where n is the number of signal components, i is the frequency of the signal, and A i The amplitude is given at low resolution, where x is a variable and ψ is a variable. i The phase is at low resolution, t is the total signal duration, and a is the phase. i The amplitude at high resolution. I represents the phase at high resolution, and U represents the phase at low resolution.

[0027] Preferably, the spatial interpolation of the phase grating data and the high-frequency amplitude grating data specifically involves:

[0028] The low-frequency portion of the amplitude raster data is removed, and bilinear interpolation is used to spatially interpolate the high-frequency amplitude raster data to obtain high-frequency amplitude raster data; wherein, the high-frequency and low-frequency are separated by a frequency threshold;

[0029] The nearest neighbor method is used to spatially interpolate the phase grating data for all frequencies to obtain high-resolution phase grating data.

[0030] Preferably, the step of fitting the spatial interpolation result to the initial surface soil moisture image in the time dimension to obtain a daily-scale surface soil moisture image includes:

[0031] Based on the high-resolution phase raster data and high-frequency amplitude raster data, and combined with the soil moisture contribution of the raster data in the low-frequency component of the initial surface soil moisture image under time discrete conditions, downscaling fitting is performed in the time dimension to obtain a high-resolution surface soil moisture image at the daily scale; where low frequency refers to frequencies below a set threshold.

[0032] Based on the measured soil moisture at the site, inverse Fourier transform and random forest were used to correct the high-resolution surface soil moisture images at the daily scale.

[0033] This invention also provides a system for generating surface soil moisture based on spatiotemporal fusion of remote sensing data, comprising:

[0034] The acquisition module is used to acquire vegetation parameter images, active microwave remote sensing images, and passive microwave remote sensing images for the required dates, and to overlay the vegetation parameter images with the active microwave remote sensing images to obtain high-resolution parameter images; wherein the spatial resolution of the high-resolution images is higher than a set threshold, and the spatial resolution of the low-resolution images is lower than a set threshold.

[0035] The downscaling module is used to resample the spatial resolution of the parametric image to a low resolution, and to spatially downscale the passive microwave remote sensing image using the low-resolution parametric image to obtain a temporally discrete initial surface soil moisture image.

[0036] The fusion module is used to take the time series of each pixel in the passive microwave remote sensing image as a signal, acquire phase raster data and high-frequency amplitude raster data of the signal at different frequencies, perform spatial interpolation on the phase raster data and high-frequency amplitude raster data, and fit the spatial interpolation result with the initial surface soil moisture image in the time dimension to obtain a daily-scale surface soil moisture image.

[0037] The present invention provides a computer device, including a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor performs the steps of the above-described method for generating surface soil moisture based on the spatiotemporal fusion of remote sensing data.

[0038] The present invention provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method for generating surface soil moisture based on spatiotemporal fusion of remote sensing data.

[0039] Compared with the prior art, the present invention has the following significant advantages:

[0040] This invention achieves spatial downscaling of passive microwave remote sensing images by resampling the spatial resolution of the acquired parametric images. Only spatial resolution resampling is required, reducing computational demands and further lowering the cost of spatial downscaling. It also improves the downscaling efficiency for objects with significant scale differences. Furthermore, it spatially interpolates the signal phase raster data and high-frequency amplitude raster data from the pixel time series of the passive microwave remote sensing images. The spatial interpolation results are then fitted to the initial surface soil moisture image in the time dimension. This achieves the fitting of spatial and temporal resolution, directly obtaining soil moisture products with fine spatiotemporal resolution, thus possessing high application value. Attached Figure Description

[0041] Figure 1 This invention provides a map showing the distribution of soil moisture observation points and land cover types in the Jiefangzha irrigation area.

[0042] Figure 2 This invention provides a technical roadmap for generating spatiotemporally continuous 20m×20m soil moisture.

[0043] Figure 3 This is a framework diagram of the spatiotemporal fusion algorithm provided by the present invention;

[0044] Figure 4 This is a performance evaluation graph of the random forest alternative to the PROSAIL model provided by this invention; wherein, Figure 4 (a) is a performance evaluation graph of the ALA parameters. Figure 4 (b) is a performance evaluation graph of the LAI parameters. Figure 4 (c) is a performance evaluation graph of CWC parameters;

[0045] Figure 5 This is an accuracy assessment of the surface soil moisture inversion model provided by the present invention; Figure 5 (a) The inversion accuracy of the model for SMAP SSM at a resolution of 9km. Figure 5 (b) The performance of model at 20m resolution;

[0046] Figure 6 This invention provides soil moisture data with a spatial resolution of 20m from July to 18, 2018, obtained through two spatiotemporal fusion algorithms. Figure 6 (a) Surface soil moisture obtained by the ESTARFM algorithm, with reference images dated July-8, 2018 and July-20, 2018. Figure 6 (b) Topsoil moisture obtained by the STFFT algorithm (without ground data correction);

[0047] Figure 7This invention provides the temporal variation of surface soil moisture from in-situ observations and downscaled data at the Hangjinhouqi station in 2018 and 2019. Figure 7 (a) and Figure 7 (b) shows the results for 2018 and 2019, respectively;

[0048] Figure 8 This invention provides a verification of the downscale surface soil moisture at the Hangjinhouqi station. Figure 8 (a) and Figure 8 (b) shows the results for 2018 and 2019, respectively;

[0049] Figure 9 This invention provides a surface soil moisture map of the study area with a daily spatial resolution of 20m; wherein, Figure 9 of (a), Figure 9 (b) Figure 9 (c) Figure 9 (d) and Figure 9 (e) shows the surface soil moisture from July 17, 2018 to July 21, 2018. Figure 9 of (f), Figure 9 of (g), Figure 9 of (h) Figure 9 (i) and Figure 9 (j) represents the surface soil moisture map from May 16, 2019 to May 20, 2019. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0051] The research area of ​​this invention is the Jiefang Gate Irrigation District ( Figure 1 (106°43'~107°27'E,40°34'~41°14'N), with an altitude of 1032m~1050m, covering an area of ​​approximately 2.157*105 hectares, including 1.543*105 hectares of arable land, with soils mainly consisting of silty loam, loam, and clay loam.

[0052] The study area is located in an arid, semi-arid, and semi-desert steppe zone with abundant sunshine, averaging approximately 3180 hours of sunshine annually, and an average frost-free period of 130 to 168 days. The climate is predominantly temperate continental, with cold, snow-scarce winters and hot, dry summers. The average annual temperature is approximately 6.3℃ to 7.7℃, average annual precipitation is 138.2 mm, and average annual evaporation is 2096.4 mm. Agricultural production in the region is highly dependent on irrigation. Irrigation in the irrigation area mainly relies on water from the Yellow River, with an annual water diversion volume reaching 1 billion to 1.2 billion cubic meters. Daily soil moisture products at the 10-meter scale will have significant guiding significance for efficient water resource management in local agriculture.

[0053] Four types of datasets will be used in the process of soil moisture at the time-space scale (Table 1). They are: (1) active microwave remote sensing products (images) of Sentinel-1 GRD IW mode, (2) atmospheric apparent reflectance products (images) of Sentinel-2 L1C, (3) surface soil moisture products of the SMAP (The Soil Moisture Active Passive) Level-4SSM satellite, which are also passive microwave remote sensing image data, and (4) surface soil moisture data observed in situ.

[0054] Table 1. Datasets used for developing methods to reduce soil moisture.

[0055] Dataset Variable Spatial resolution Temporal resolution Sentinel-2L2A Surface reflection 20m×20m 2-3 days Sentinel-1GRD VV / VH / Angle 5m×20m 12 days SMAPLevel-4 SSM 9km×9km 3h MeteorologicalStation SSM Point Daily OrdinaryObservationPoints SSM Point Monthly

[0056] Sentinel-1GRD IW mode active microwave remote sensing products:

[0057] Sentinel-1 consists of a binary constellation, Sentinel-1A and Sentinel-1B. The revisit period for a single satellite is 12 days, which can be reduced to 6 days through the complementary pairing. It can acquire satellite imagery with a resolution of 5-40 meters under all-weather, all-day conditions. The Sentinel-1 satellite carries a C-band SAR sensor operating at 5.4 GHz, featuring multi-polarization imaging capabilities and four imaging modes. The Sentinel-1 imagery used in this invention is down-orbit data in the interferometric wide-swath mode, corresponding to a ground resolution of 5m × 20m, a data level of Level-1, and a ground range detected (GRD) format. Dual polarization includes cross-polarization (VH, Vertical horizontal) and vertical polarization (VV, Vertical vertical). The required data spans from April 1, 2018 to September 30, 2018, and includes data from the same period in 2019, with a temporal resolution of 12 days within the study area. Data was acquired from the Google Earth Engine platform and processed using Refined Lee filtering, then exported at a spatial resolution of 20m.

[0058] Sentinel-2L1C Atmospheric Apparent Reflectance Products:

[0059] Like Sentinel-1, Sentinel-2 is also a binary satellite system consisting of Sentinel-2A and Sentinel-2B. The revisit period for a single satellite is 10 days, which can be reduced to 5 days through binary satellite complementarity, resulting in a revisit period of 2-3 days within the region of this invention. Sentinel-2 carries a multispectral imager with 13 spectral bands and a swath width of 290 km. Ground resolution varies from 10 m to 60 m depending on the band. Since the GEE platform lacks 2018 Sentinel-2L2A (S2L2A) surface reflectance products for the region of this invention, the Sentinel-2L1C (S2L1C) product was selected and atmospherically corrected based on the research results of Yin et al. to obtain the S2L2A product. To match the Sentinel-1 data, the date of S2L2A is chosen to be consistent with the Sentinel-1 overpass date. Due to the high temporal resolution of Sentinel-2, the matching requirements can be met for most of the time. Unmatched dates were filled in using linear interpolation and SG filtering on the S2L2A data. This method was also used to fill in local blank areas in images after cloud removal using the QA60 band. This invention selected 10 bands (B02-B8A, B11, and B12) covering the spectral region from visible light to shortwave infrared at 10m and 20m spatial resolutions, and resampled them all to 20m spatial resolution.

[0060] SMAP Level-4SSM Topsoil Moisture Product:

[0061] The SMAP Level-4 surface soil moisture product uses data assimilation techniques to integrate spaceborne L-band brightness and temperature measurements and precipitation observations into a watershed land surface process model. This generates surface soil moisture (0-5 cm vertical average), root zone soil moisture (0-100 cm vertical average), and other research products (unverified), including surface meteorological forcing variables, soil temperature, evapotranspiration, and net radiation. This invention utilizes the SMAP Level-4 surface soil moisture product acquired from the GEE platform as the background field for SSM downscaling, with a spatial resolution of 9 km and a temporal resolution set to daily output.

[0062] In-situ surface soil moisture observation:

[0063] Two long-term observation stations, Hangjinhouqi Station (40°51′00″N, 107°07′00″E) and Uradhouqi Station (41°04′00″N, 107°03′00″E), provide daily soil volumetric water content at depths of 10, 40, and 100 cm. This invention extracts automatically measured surface soil moisture (SSM) data from these stations from April 30 to September 30, 2018, and during the same period in 2019. Considering the non-negligible errors in SMAP products within the study area, SSM observations from Uradhouqi Station were used for correction of the downscaling product during the spatiotemporal fusion stage, while observations from Hangjinhouqi Station were used for validation of the final surface soil moisture downscaling product. This data was obtained from the China Meteorological Data Service Center (http: / / data.cma.cn).

[0064] In addition to the in-situ data of soil surface moisture obtained from continuous fixed-point observations, soil volumetric water content was measured 14 times at 19 ordinary observation points in the study area from April to September 2018 and 2019. These ground observations will be used to evaluate the performance of the soil moisture inversion model at a resolution of 20m.

[0065] Multiple studies have shown a strong linear relationship between radar backscattering coefficient and soil moisture when there is no vegetation cover. However, during the peak vegetation growth period, the accuracy of directly using the backscattering coefficient to retrieve soil moisture is poor due to the influence of the vegetation canopy. In this case, it is necessary to extract the soil backscattering coefficient from the echo signal collected by SAR and then retrieve soil moisture information from it. This invention, referencing the water-cloud model, considers that the total backscattering from a vegetated surface consists of two parts: one is the volume scattering term directly reflected back from the vegetation canopy, and the other is the ground backscattering term after double attenuation by the vegetation canopy. Both of these are related to vegetation water content (VWC), vegetation type (V), and soil moisture (M).S The total backscattering coefficient (σ) is related to the radar wave wavelength (λ) and the radar wave incident angle (θ). Therefore, the total backscattering coefficient (σ) can be expressed as:

[0066] σ=f(VWC,V,M S ,λ,θ).

[0067] It is assumed that vegetation water content is a function of canopy water content (CWC):

[0068] VWC = f(CWC).

[0069] Canopy water content is defined as the product of leaf area index (LAI) and leaf water content (LWC):

[0070] CWC = LAI × LWC.

[0071] Furthermore, this invention posits that the influence of vegetation type on backscattering primarily stems from differences in canopy structure among different vegetation types. Canopy structure can be simply characterized using leaf area index (LAI) and mean leaf angle (ALA). Additionally, NDVI is used to describe differences in vegetation cover. Therefore, vegetation categories can be represented as:

[0072] V = f(LAI,ALA,NDVI).

[0073] Combining the formulas, we get:

[0074] σ=f(LAI,CWC,ALA,NDVI,M S ,λ,θ).

[0075] Change the format:

[0076] M S =f(LAI,CWC,ALA,NDVI,VV,VH,θ).

[0077] Therefore, the method proposed in this invention firstly calculates NDVI using the S2L2A product and inverts vegetation LAI, CWC, and ALA using the PROSAIL radiative transfer model. Secondly, it fuses NDVI, LAI, CWC, ALA, and Sentinel-1GRD products, and then uses the random forest algorithm to spatially downscale SMAP. Thirdly, it develops a new spatiotemporal fusion algorithm and performs spatiotemporal fusion on soil moisture.

[0078] like Figure 2 As shown, the following describes a method for generating surface soil moisture based on spatiotemporal fusion of remote sensing data, according to an embodiment of the present invention, including the following steps:

[0079] Step S1: Acquire vegetation parameter images, active microwave remote sensing images, and passive microwave remote sensing images for the required dates, and overlay the vegetation parameter images with the active microwave remote sensing images to obtain high-resolution parameter images; wherein the spatial resolution of the high-resolution images is higher than a set threshold, and the spatial resolution of the low-resolution images is lower than a set threshold.

[0080] In step S1, the vegetation parameter image for the desired date is obtained, including:

[0081] Different combinations of vegetation parameters were constructed and input into the radiative transfer model to simulate and generate spectral curves. The spectral curves were then converted into band reflectance that could be captured by the atmospheric apparent reflectance acquisition model (Atmospheric Apparent Reflectance Product - Sentinel-2).

[0082] GEE (Google Earth Engine - a cloud-based geographic information processing platform) has a built-in function for calculating the Normalized Difference Vegetation Index (NDVI). Therefore, this paper focuses on describing how to use the S2L2A (Sentinel-2-L2A) product to retrieve vegetation parameters such as LAI, CWC, and ALA. This invention first uses the PROSAIL radiative transfer model to simulate 5000 different combinations of vegetation parameters (Table 2) and their corresponding spectral curves. Then, the spectral curves are converted into reflectance bands that Sentinel-2 can capture using the spectral response function of Sentinel-2.

[0083] Table 2 Different combinations of vegetation parameters based on the PROSAIL model

[0084] Parameters Abbreviation valuerange Leafchlorophyllcontent <![CDATA[C ab (μg / cm 2 )]]> 0-100 Leaf structure parameter N (unitless) 1-3 Leafcarotenoidcontent <![CDATA[C ar (μg / cm 2 )]]> <![CDATA[C ab / 7]]> Brownpigmentcontent <![CDATA[C b ]]> 0 Drymattercontent <![CDATA[C m (g / cm 2 )]]> 0.002-0.020 Leafwatercontent <![CDATA[LWC(g / cm 2 )]]> 0.005-0.050 Averageleafangle ALA(°) 40-80 Leafareaindex LAI 0.1-5 Hotspot parameter hspot 0.1

[0085] The band reflectance is input into the random forest model for inversion to obtain vegetation parameter images for the required dates; the vegetation parameter images include leaf area index (LAI), canopy water content (CWC), and mean leaf tilt angle (ALA) images.

[0086] The simulated dataset obtained by the PROSAIL model was used as the training set for the random forest. Band reflectance was used as the input variable of the random forest, and leaf area index (LAI), canopy water content (CWC), and mean leaf tilt angle (ALA) vegetation parameters were used as output variables. Then, the trained random forest model was applied pixel by pixel to Sentinel-2 images to obtain LAI, CWC, and ALA images for the required dates.

[0087] Besides using the PROSAIL model to simulate the dataset, the training of the random forest model and the inversion of vegetation parameters were both performed on the GEE platform. The advantage of this is that only the final vegetation parameter images need to be downloaded, without needing to download the original Sentinel-2 images. Furthermore, while the inversion process of the PROSAIL model is not easily deployed on the GEE platform, the random forest model trained on the simulated dataset can easily perform the inversion of parameters such as LAI on the GEE platform.

[0088] In step S1, the vegetation parameter image is overlaid with the active microwave remote sensing image to obtain a parameter image with a spatial resolution of Ukm, specifically:

[0089] Acquire active microwave remote sensing images (images from Sentinel-1).

[0090] By overlaying active microwave remote sensing imagery with vegetation parameter images containing leaf area index (LAI), canopy water content (CWC), and mean leaf inclination angle (ALA), a set of images was obtained, encompassing seven bands: normalized difference vegetation index (NDVI), leaf area index (LAI), canopy water content (CWC), mean leaf inclination angle (ALA), vertical polarization (VV), cross-polarization (VH), and radar incident angle (θ). The active microwave remote sensing imagery used here is a Sentinel-1 imagery after refined Lee filtering.

[0091] The images of the seven bands were used as high-resolution parametric images.

[0092] Step S2: Resample the spatial resolution of the parametric image to a low resolution, and use the low-resolution parametric image to spatially downscale the passive microwave remote sensing image to obtain a temporally discrete initial surface soil moisture image.

[0093] Step S2 specifically includes:

[0094] The spatial resolution of the high-resolution parametric image is resampled to a low resolution using mean aggregation. In this embodiment, the low resolution is 9km, which corresponds one-to-one with the pixels of the passive microwave remote sensing image (the image obtained by the soil moisture active-passive remote sensing satellite SMAP). The pixels are then used as the training set to train the surface soil moisture inversion model.

[0095] High-resolution parametric images are input into the trained surface soil moisture inversion model to obtain an initial surface soil moisture image with high spatial resolution under discrete time conditions.

[0096] The following example illustrates this:

[0097] The surface soil moisture inversion model was applied to a parametric image with a spatial resolution of 20m before resampling to obtain surface soil moisture at a spatial resolution of 20m, thus achieving spatial downscaling of SMAP. Here, 20m represents a practical application of high resolution. This method is widely used for spatial downscaling of soil moisture products. It is important to note that only SMAP SSM spatial downscaling products with Sentinel-1 overpass dates are obtained, which are temporally discontinuous. Figure 3 (g)), that is, a set of surface soil moisture images with high spatial resolution and low temporal resolution were obtained. Figure 3 (g)).

[0098] Step S3: Using the time series of each pixel in the passive microwave remote sensing image as a signal, acquire phase raster data and high-frequency amplitude raster data at different frequencies of the signal, where high frequency refers to frequencies above a set threshold. Spatial interpolation is performed on the phase raster data and high-frequency amplitude raster data. The spatial interpolation result is then fitted to the initial surface soil moisture image in the time dimension to obtain a daily-scale surface soil moisture image. Here, frequency represents events affecting soil moisture, amplitude represents the degree of influence of the event on soil moisture, and phase indicates the start time of the event.

[0099] SMAP itself possesses the characteristics of high temporal resolution and low spatial resolution. A spatiotemporal fusion algorithm combines the advantages of both to obtain a soil moisture product with high spatiotemporal resolution. The temporal nature of the reference image and the variability of soil moisture both influence the application of traditional spatiotemporal fusion algorithms in soil moisture products. This invention proposes a soil moisture spatiotemporal fusion algorithm based on Fourier transform (STFFT) to address the aforementioned problems.

[0100] Soil moisture is influenced by various factors, including precipitation, evapotranspiration, infiltration, groundwater recharge, and irrigation. This algorithm categorizes these factors into high-frequency and low-frequency information in the time series. Irrigation, precipitation, and evapotranspiration are considered high-frequency information, while infiltration and groundwater recharge are considered low-frequency information. Interestingly, spatially, high-frequency information in the time series becomes low-frequency information, and vice versa. Looking at atmospheric phenomena controlling precipitation and evapotranspiration, changes in boundary layer eddies, cumulonimbus clouds, fronts, and squall lines occur on a kilometer-scale. These atmospheric phenomena almost all occur on a diurnal scale, meaning they change daily. Another intuitive observation is that soil texture, which affects water infiltration, varies within a single farmland, but remains largely unchanged over time. This phenomenon of opposite spatiotemporal frequencies for the same information provides the basis for our spatiotemporal fusion algorithm.

[0101] In step S3, the time series of each pixel in the passive microwave remote sensing image is used as a signal to acquire phase raster data and high-frequency amplitude raster data of different frequencies of the signal, including:

[0102] Each cell in an SMAP SSM can be viewed as a signal in the time series. Figure 3 (a) represents the change in soil moisture over time within a low-resolution range (9km in practice).

[0103] Perform a Fast Fourier Transform on the signal to obtain its amplitude and phase at different frequencies. Figure 3 (b)).

[0104] The timing SSM at different spatial resolutions is represented by amplitudes and phases at different frequencies, specifically as follows:

[0105]

[0106]

[0107] Where n is the number of signal components, i is the frequency of the signal, and A i The amplitude is given at low resolution, where x is a variable and ψ is a variable. i The phase is at low resolution, t is the total signal duration, and a is the phase. i The amplitude at high resolution. I represents the phase at high resolution, and U represents the phase at low resolution.

[0108] By performing the same operation on all pixels of the SMAP SSM, a set of amplitude raster data and phase raster data at different frequencies can be obtained. Figure 3 (c), Figure 3 (d)).

[0109] In step S3, spatial interpolation is performed on the phase grating data and the high-frequency amplitude grating data, specifically as follows:

[0110] The low-frequency portion of the amplitude raster data is removed, and bilinear interpolation is used to spatially interpolate the high-frequency amplitude raster data to obtain high-frequency amplitude raster data. High and low frequencies are distinguished using a frequency threshold.

[0111] The nearest neighbor method is used to spatially interpolate the phase grating data for all frequencies to obtain high-resolution phase grating data.

[0112] In step S3, the spatial interpolation result is fitted to the initial surface soil moisture image in the time dimension to obtain a daily-scale surface soil moisture image, including:

[0113] Based on high-resolution phase raster data and high-frequency amplitude raster data, and combined with the contribution of the initial surface soil moisture image in the time discrete case to the soil moisture in the low-frequency component raster data, downscaling fitting is performed in the time dimension to obtain a high-resolution surface soil moisture image at the daily scale.

[0114] Based on the measured soil moisture at the site, inverse Fourier transform and random forest were used to correct the high-resolution surface soil moisture images at the daily scale.

[0115] The following is a detailed explanation of step S3:

[0116] It's important to clarify that, according to the sampling theorem, interpolation or fitting of a finite number of sampling points performs better on low-frequency information than on high-frequency information. The algorithm removes the low-frequency components from the amplitude raster data. Figure 3 (The orange portion in (c)) uses bilinear interpolation to interpolate the high-frequency amplitude raster data. Figure 3 Spatial interpolation was performed on the blue portion of (c), and the nearest neighbor method was used to interpolate the phase grating data for all frequencies, resulting in phase grating data with a spatial resolution of 20m. Figure 3 (f) and high-frequency amplitude raster data ( Figure 3 (e)). This approach is based on the premise that the spatial variations of high-frequency temporal information such as irrigation, precipitation, and evapotranspiration are low-frequency. Therefore, the influence of these factors on soil moisture can be downscaled using interpolation in the spatial dimension, assuming that the initial time of all events within a 9km radius is the same. At this point, only {a0,…,a} remain in the equation. n If there are n+1 unknowns that need to be solved, then if there are more than n+1 equations, then all the unknowns can be solved. Figure 3 (h)) This achieves spatiotemporal downscaling of SSM. The establishment of the solution equations depends on the spatial downscaling results of the second step, SMAP SSM. Figure 3 (g) means that an equation can be established for each day's SSM downscaling results. This approach is based on the fact that high-frequency spatial information such as infiltration and groundwater recharge changes in a low-frequency temporal dimension, and the contribution of these factors to soil moisture can be downscaled in the time dimension.

[0117] However, since the second step of constructing the surface soil moisture inversion model using SMAP SSM as the dependent variable, it was assumed that the accuracy of the SMAP SSM product was sufficiently high, which is undoubtedly an ideal state. In fact, in some areas lacking ground data, the accuracy of the SMAP SSM product still needs to be improved. Taking the area of ​​this invention as an example, soil moisture is severely underestimated; therefore, it is necessary to correct the downscaling product. The specific correction method is to use a random forest model instead of the original polynomial summation during the inverse Fourier transform. The SSM observations from the Urat Rear Banner station were used to train the random forest model.

[0118] The above steps can be used to downscale SMAP SSM to obtain surface soil moisture products with a daily spatial resolution of 20m. Figure 3 (j)).

[0119] Based on the above method, the present invention provides a system for generating surface soil moisture based on the spatiotemporal fusion of remote sensing data, comprising: an acquisition module, a downscaling module, and a fusion module.

[0120] The acquisition module is used to acquire vegetation parameter images, active microwave remote sensing images, and passive microwave remote sensing images for the required dates. It then overlays the vegetation parameter images with the active microwave remote sensing images to obtain high-resolution parameter images. The spatial resolution of the high-resolution images is higher than a set threshold, while the spatial resolution of the low-resolution images is lower than the set threshold. The downscaling module is used to resample the spatial resolution of the parameter images to a low resolution. The low-resolution parameter images are then used to spatially downscale the passive microwave remote sensing images to obtain a temporally discrete initial surface soil moisture image. The fusion module uses the time series of each pixel in the passive microwave remote sensing images as a signal to acquire phase raster data and high-frequency amplitude raster data at different frequencies. It then performs spatial interpolation on the phase raster data and high-frequency amplitude raster data, and fits the spatial interpolation results with the initial surface soil moisture image in the time dimension to obtain a daily-scale surface soil moisture image.

[0121] The present invention provides a computer device, including a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor performs the steps of a method for generating surface soil moisture based on the spatiotemporal fusion of remote sensing data.

[0122] According to the disclosed embodiments, the computer device can communicate with one or more external devices (e.g., keyboard, pointing device, Bluetooth communication, etc.) or with any device that enables the computing device to communicate with one or more other computing devices (e.g., router, demodulator, etc.).

[0123] The present invention provides a storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of a method for generating surface soil moisture based on spatiotemporal fusion of remote sensing data.

[0124] According to the disclosed embodiments, the storage medium can be a non-volatile computer-readable storage medium, such as, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, the storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0125] The above description, in conjunction with specific preferred embodiments, provides a more detailed explanation of the present invention. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered to fall within the scope of protection of the present invention.

Claims

1. A method for generating surface soil moisture based on spatiotemporal fusion of remote sensing data, characterized in that, include: Acquire vegetation parameter images, active microwave remote sensing images, and passive microwave remote sensing images for the desired dates, and overlay the vegetation parameter images with the active microwave remote sensing images to obtain high-resolution parameter images; wherein the spatial resolution of the high-resolution images is higher than a set threshold, and the spatial resolution of the low-resolution images is lower than a set threshold. The spatial resolution of the parametric image is resampled to a low resolution, and the passive microwave remote sensing image is spatially downscaled using the low-resolution parametric image to obtain a temporally discrete initial surface soil moisture image. The time series of each pixel in the passive microwave remote sensing image is used as a signal. Phase raster data and high-frequency amplitude raster data of the signal at different frequencies are obtained. The phase raster data and high-frequency amplitude raster data are spatially interpolated. The spatial interpolation result is fitted with the initial surface soil moisture image in the time dimension to obtain a daily-scale surface soil moisture image. The spatial interpolation of phase raster data and high-frequency amplitude raster data specifically involves: The low-frequency portion of the amplitude raster data is removed, and bilinear interpolation is used to spatially interpolate the high-frequency amplitude raster data to obtain high-frequency amplitude raster data; wherein, the high and low frequencies are separated by a frequency threshold. Spatial interpolation of phase grating data for all frequencies is performed using the nearest neighbor method to obtain high-resolution phase grating data. The step of fitting the spatial interpolation result to the initial surface soil moisture image in the time dimension to obtain a daily-scale surface soil moisture image includes: Based on the high-resolution phase raster data and high-frequency amplitude raster data, and combined with the soil moisture contribution of the initial surface soil moisture image in the time discrete case to the raster data in the low-frequency component, downscaling fitting is performed in the time dimension to obtain a high-resolution surface soil moisture image at the daily scale. Based on the measured soil moisture at the site, inverse Fourier transform and random forest were used to correct the high-resolution surface soil moisture images at the daily scale.

2. The method for generating surface soil moisture based on spatiotemporal fusion of remote sensing data as described in claim 1, characterized in that, The process of obtaining vegetation parameter images for the desired date includes: Obtain different combinations of vegetation parameters; Different combinations of vegetation parameters are input into the radiative transfer model to simulate and obtain spectral curves. The spectral curves are then converted into atmospheric apparent reflectance to obtain the band reflectance captured by the model. The reflectance of the band is input into a random forest model for inversion to obtain vegetation parameter images for the required dates; wherein, the vegetation parameter images include leaf area index (LAI), canopy water content (CWC), and mean leaf tilt angle (ALA) images.

3. The method for generating surface soil moisture based on spatiotemporal fusion of remote sensing data as described in claim 1, characterized in that, The process of overlaying vegetation parameter images with active microwave remote sensing images to obtain high-resolution parameter images specifically involves: By overlaying active microwave remote sensing images with vegetation parameter images containing leaf area index (LAI), canopy water content (CWC), and mean leaf inclination angle (ALA), a set of images containing seven bands including normalized difference vegetation index (NDVI), leaf area index (LAI), canopy water content (CWC), mean leaf inclination angle (ALA), vertical polarization (VV), cross polarization (VH), and radar wave incident angle (θ) is obtained. The images of the seven bands were used as high-resolution parametric images.

4. The method for generating surface soil moisture based on spatiotemporal fusion of remote sensing data as described in claim 1, characterized in that, The method of spatially downscaling passive microwave remote sensing images using low-resolution parametric images to obtain temporally discrete initial surface soil moisture images includes: The spatial resolution of the high-resolution parametric image is resampled to a low resolution and matched with the pixels of the passive microwave remote sensing image. The pixels are then used as a training set to train the surface soil moisture inversion model. High-resolution parametric images are input into the trained surface soil moisture inversion model to obtain high-resolution initial surface soil moisture images under discrete time conditions.

5. The method for generating surface soil moisture based on spatiotemporal fusion of remote sensing data as described in claim 1, characterized in that, Using the time series of each pixel in the passive microwave remote sensing image as a signal, phase raster data and high-frequency amplitude raster data of the signal at different frequencies are acquired, including: Perform a Fast Fourier Transform on the signal to obtain the amplitude and phase at different frequencies of the signal; Timing at different spatial resolutions is represented by amplitude and phase at different frequencies. SSM Specifically: ； ； Where n is the number of signal components. i For the frequency of the signal, The amplitude at low resolution. x As variables, Phase at low resolution t The total duration of the signal. The amplitude at high resolution. I represents the phase at high resolution, and U represents the phase at low resolution.

6. A system for generating surface soil moisture based on spatiotemporal fusion of remote sensing data, characterized in that, include: The acquisition module is used to acquire vegetation parameter images, active microwave remote sensing images, and passive microwave remote sensing images for the required dates, and to overlay the vegetation parameter images with the active microwave remote sensing images to obtain high-resolution parameter images; wherein the spatial resolution of the high-resolution images is higher than a set threshold, and the spatial resolution of the low-resolution images is lower than a set threshold. The downscaling module is used to resample the spatial resolution of the parametric image to a low resolution, and to spatially downscale the passive microwave remote sensing image using the low-resolution parametric image to obtain a temporally discrete initial surface soil moisture image. The fusion module is used to take the time series of each pixel in the passive microwave remote sensing image as a signal, acquire phase raster data and high-frequency amplitude raster data of the signal at different frequencies, perform spatial interpolation on the phase raster data and high-frequency amplitude raster data, and fit the spatial interpolation result with the initial surface soil moisture image in the time dimension to obtain a daily-scale surface soil moisture image. The spatial interpolation of phase raster data and high-frequency amplitude raster data specifically involves: The low-frequency portion of the amplitude raster data is removed, and bilinear interpolation is used to spatially interpolate the high-frequency amplitude raster data to obtain high-frequency amplitude raster data; wherein, the high and low frequencies are separated by a frequency threshold. Spatial interpolation of phase grating data for all frequencies is performed using the nearest neighbor method to obtain high-resolution phase grating data. The step of fitting the spatial interpolation result to the initial surface soil moisture image in the time dimension to obtain a daily-scale surface soil moisture image includes: Based on the high-resolution phase raster data and high-frequency amplitude raster data, and combined with the soil moisture contribution of the initial surface soil moisture image in the time discrete case to the raster data in the low-frequency component, downscaling fitting is performed in the time dimension to obtain a high-resolution surface soil moisture image at the daily scale. Based on the measured soil moisture at the site, inverse Fourier transform and random forest were used to correct the high-resolution surface soil moisture images at the daily scale.

7. A computer device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of a method for generating surface soil moisture based on spatiotemporal fusion of remote sensing data as described in any one of claims 1 to 5.

8. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for generating surface soil moisture based on the spatiotemporal fusion of remote sensing data as described in any one of claims 1 to 5.

Images