Soil moisture content inversion method and system based on l-band dual-polarization data and medium

Abstract

The application discloses a soil moisture content inversion method and system based on L-band dual-polarization data and a medium, and belongs to the field of remote sensing technology and environmental monitoring. The system comprises an SAR data acquisition module, an optical data processing module, a vegetation scattering separation module, an inversion model construction module, a spatial distribution calculation module and a risk assessment module. The SAR data acquisition module is used for regularly collecting L-band dual-polarization SAR data of a target region. The optical data processing module assists in identifying and separating the vegetation scattering component by calculating a vegetation index. The vegetation scattering separation module is used for extracting a pure soil backscattering signal. The inversion model construction module trains and optimizes a soil moisture content inversion model based on the pure soil backscattering signal and in-situ sampling data. The spatial distribution calculation module is used for calculating a spatial distribution map of the surface soil moisture content of the target region. The risk assessment module is used for landslide risk grade division and early warning management.

Description

Technical Field

[0001] This application relates to the fields of remote sensing technology and environmental monitoring, and more specifically, to a method, system, and medium for soil moisture content retrieval based on L-band dual polarization data. Background Technology

[0002] Soil moisture content is a crucial indicator affecting agricultural production, water resource management, and ecological environment protection. Timely and accurate understanding of soil moisture status is essential for disaster prevention and mitigation, as well as precision agriculture. Traditional methods for measuring soil moisture content generally require on-site sampling or the use of sensors. While these methods are relatively accurate, they have limited coverage, low spatial resolution, and high labor costs, making them unsuitable for large-scale and continuous monitoring.

[0003] Remote sensing technology, due to its ability to rapidly cover large areas, has become an effective means of monitoring soil moisture content. Synthetic Aperture Radar (SAR) is particularly prominent, as it can penetrate clouds and support all-weather observation. In recent years, L-band SAR, due to its good surface penetration and sensitivity to soil moisture, has been increasingly used for soil moisture content retrieval. Dual-polarization SAR data can also provide more polarization information, helping to distinguish between soil and vegetation echo signals and improving the accuracy of retrieval.

[0004] However, soil surfaces are often covered with various vegetation types, and the influence of vegetation on radar signals is complex, making it difficult to extract pure soil signals and affecting the accuracy of inversion results. A single data source also struggles to fully reflect the diverse characteristics of the land surface, limiting the effectiveness of the model. Therefore, combining multi-source remote sensing data, such as optical imagery and digital elevation models, using vegetation indices to assist in separating vegetation signals, and then constructing an inversion model using advanced machine learning methods, has become an effective way to improve the level of soil moisture monitoring.

[0005] Currently, although there are some methods for soil moisture content inversion using SAR data, there are still problems such as poor model adaptability, failure to fully eliminate vegetation impact, difficulty in achieving real-time monitoring, and lack of risk warning, which cannot fully meet the requirements for accuracy and timeliness in practical applications.

[0006] In summary, how to utilize L-band dual-polarization data to achieve accurate and dynamic soil moisture content inversion and risk assessment through multi-source data fusion and intelligent modeling has become an urgent technical problem that needs to be solved. Summary of the Invention

[0007] To overcome a series of shortcomings in existing technologies, this application aims to provide a soil moisture content inversion system based on L-band dual-polarization data. The system includes a SAR data acquisition module, an optical data processing module, a vegetation scattering separation module, an inversion model construction module, a spatial distribution calculation module, and a risk assessment module. The SAR data acquisition module is used to periodically collect L-band dual-polarization SAR data of the target area to achieve multi-temporal data acquisition. The optical data processing module is configured to receive optical remote sensing data and assist in identifying and separating vegetation scattering components by calculating vegetation indices. The vegetation scattering separation module responds to the vegetation indices output by the optical data processing module and removes vegetation scattering components based on quantitative removal rules. In addition to the vegetation scattering contribution from SAR data, purified soil backscattering signals are extracted. The inversion model construction module trains and optimizes the soil moisture content inversion model based on the purified soil backscattering signals and field sampling data. The spatial distribution calculation module is used to calculate the spatial distribution map of surface soil moisture content in the target area and verify its accuracy. The risk assessment module is used for landslide risk level classification and early warning management. Specifically, the SAR data acquisition module, optical data processing module, and vegetation scattering separation module work together to purify the soil scattering signals. The inversion model construction module and spatial distribution calculation module jointly achieve high-precision soil moisture content inversion. The risk assessment module completes landslide risk level classification and early warning based on the inversion results.

[0008] Furthermore, vegetation index calculation includes the comprehensive calculation of Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), and Leaf Area Index (LAI), specifically including...

[0009] The NDVI is calculated using the red and near-infrared bands of optical remote sensing data. The specific calculation formula is: NDVI=(NIR-Red) / (NIR+Red), where NIR represents the surface reflectance value in the near-infrared band and Red represents the surface reflectance value in the red band.

[0010] The Enhanced Vegetation Index (EVI) is calculated using the standard formula: EVI = G × [(NIR - Red) / (NIR + C1 × Red - C2 × Blue + L)], where G is the gain factor, with a value of 2.5; Blue is the reflectivity of the blue light band; C1 is the atmospheric impedance coefficient of the red light band, with a value of 6.0; C2 is the atmospheric impedance coefficient of the blue light band, with a value of 7.5; and L is the canopy background adjustment parameter, with a value of 1.0.

[0011] The leaf area index (LAI) is estimated based on the empirical formula LAI = 3.618 × EVI - 0.118.

[0012] A second aspect of this application provides a method for soil moisture content inversion based on L-band dual polarization data, comprising the following steps.

[0013] Step 1: Acquire L-band dual-polarization SAR data, optical remote sensing data, digital elevation model data, and field soil moisture content sampling data of the target area within the set time window, and register and preprocess the multi-source remote sensing data according to the time series characteristics to construct a multi-source remote sensing dataset with time series characteristics.

[0014] Step 2: Calculate the vegetation index based on the optical remote sensing data, and quantitatively separate and remove the vegetation scattering contribution in the L-band dual-polarization SAR data based on the vegetation index to extract the pure soil backscattering signal; at the same time, divide the slope units based on the digital elevation model data.

[0015] Step 3: Divide the field soil moisture content sampling data into training set, validation set and test set in a 6:2:2 ratio. Use the pure soil backscattering signal and training set data to establish a soil moisture content inversion model. Use the validation set to fine-tune the model during training and optimize and calibrate the model parameters of the inversion model.

[0016] Step 4: Apply the optimized inversion model to the L-band dual-polarization SAR data processing, calculate the spatial distribution map of surface soil moisture content in the target area, and conduct accuracy verification and multi-index evaluation in conjunction with test set data.

[0017] Step 5: Combining the soil moisture content inversion results from Step 4 with historical soil moisture content data, establish a landslide risk level assessment model and determine the safe threshold for soil moisture content.

[0018] Step Six: Combine the spatial distribution map of soil moisture content and the slope unit division results to conduct stability analysis, and apply the landslide risk level assessment model to classify the risk level of different areas; if the soil moisture content of a certain area is detected to exceed the preset safety threshold, the area is marked as a high-risk area and an early warning signal is issued; if the soil moisture content is within the safe range, the routine monitoring process continues.

[0019] Step 7: During the dynamic monitoring process, new L-band dual-polarization SAR data is periodically acquired through the SAR data acquisition module. When the vegetation coverage changes by more than 15%, the vegetation index is recalculated and Step 2 is executed. Otherwise, Step 4 and Step 6 are executed directly using the pre-trained inversion model to achieve continuous monitoring of soil moisture content and real-time assessment of landslide risk. When the inversion accuracy decreases by more than 0.1 or the root mean square error increases by more than 20%, Step 3 is re-executed to retrain the model and optimize the parameters.

[0020] Furthermore, the slope unit division process includes the following steps: First, the main topographic factors, including slope, aspect, and plane curvature, are calculated based on digital elevation model data, with the slope gradient obtained using a third-order finite difference algorithm; then, the D8 flow direction algorithm combined with flow accumulation analysis is used to identify the confluence network, and a flow accumulation threshold of 500 pixels is set to extract the river network system; next, initial slope unit division is performed, and the unit boundaries are optimized and adjusted based on constraints that the slope standard deviation is less than 8 degrees and the elevation variation coefficient is less than 0.3; finally, a quality assessment is conducted based on the hydrological connectivity and topographic consistency of the slope units, eliminating irregular units with an area less than 0.5 square kilometers or an aspect ratio greater than 8:1, ensuring that each slope unit has relatively uniform hydrological response characteristics and topographic features, while the slope variation within the control unit is within ±10 degrees.

[0021] Furthermore, the soil moisture content inversion model is constructed based on an ensemble learning framework of support vector machine regression and random forest. The specific steps are as follows: A radial basis function is selected as the kernel function of the support vector machine, and hyperparameters are optimized within the range of C∈[1,1000] and γ∈[0.001,1] using a grid search method; input features include normalized backscattering coefficients of HH polarization and HV polarization, polarization ratio HV / HH, incident angle, local incident angle, slope, and aspect; a random forest consisting of 100 decision trees is constructed, with a maximum tree depth of 10 layers and a minimum number of splits per node of 5; 10-fold cross-validation is used for parameter tuning during model training, and root mean square error and mean absolute error are used as performance evaluation indicators; a weighted average method is used to fuse the two prediction results, with the weight coefficients determined by the validation set performance; the final inversion model achieves an RMSE of less than 6% on the validation set, and a correlation coefficient R0. 2 It reaches 0.70 or higher.

[0022] Furthermore, the model parameter optimization and calibration process includes the following steps: First, a genetic algorithm is used to globally optimize the model hyperparameters, setting the population size to 30, the maximum number of iterations to 150, the crossover probability to 0.8, and the mutation probability to 0.1. In the parameter search space, the regularization parameter of the support vector machine ranges from [1, 500], the kernel parameter ranges from [0.01, 1], and the number of trees in the random forest ranges from [50, 100]. The effective range of the parameters is initially determined through coarse-grained grid search, and then fine-tuned using random search to obtain the optimal hyperparameter combination. Training is stopped when the validation loss shows no improvement after 15 consecutive iterations to prevent overfitting. Finally, the generalization ability of the optimized model is verified using an independent test set to ensure that the root mean square error of soil moisture content retrieval is controlled within 7% and the mean absolute error is controlled within 5%.

[0023] Furthermore, the recalculation process when vegetation cover changes by more than 25% includes the following steps: First, through multi-temporal normalized vegetation index difference analysis, areas with significant changes in vegetation cover are identified, and their boundaries are determined; then, the vegetation index is recalculated for the changed areas, and its time series data is smoothed to suppress short-term fluctuations and observation noise; based on this, the vegetation water cloud model parameters are reconstructed based on the latest vegetation status, and the scattering separation parameters are fitted and optimized using the nonlinear least squares method; by comparing the backscattering statistical characteristics before and after vegetation scattering removal, the scattering removal accuracy is ensured to be maintained above 80%; the seasonal change trend of vegetation cover is predicted through vegetation phenological curve analysis; a regional processing strategy is adopted, and corresponding processing parameters are set for different vegetation types and growth stages; when the rate of change in vegetation cover is detected to exceed 8% / month, the recalculation process is initiated in advance.

[0024] Furthermore, the early warning signal is issued using a four-level early warning mechanism, including blue, yellow, orange, and red warning levels. Specifically: a blue warning is triggered when the soil moisture content reaches 85% of the safe threshold, alerting relevant departments to pay attention, and the warning is valid for 24 hours; a yellow warning is triggered when the soil moisture content reaches 95% of the safe threshold, requiring monitoring to be increased to once every 3 days, and the warning is valid for 12 hours; an orange warning is triggered when the soil moisture content exceeds the safe threshold but the stability coefficient is greater than 1.1, initiating a level-two emergency response procedure, and adjusting the monitoring frequency to once a day; a red warning is triggered when the soil moisture content exceeds the safe threshold and the stability coefficient is less than 1.1, immediately initiating a level-one emergency response.

[0025] Furthermore, the vegetation cover change detection process specifically includes the following steps: First, a vegetation index time series database is established to support long-term monitoring and trend analysis; then, seasonal decomposition is used to extract the trend, seasonal, and random components of the vegetation index, separating periodic changes from anomalous disturbances; the significance level for mutation detection is set to 0.05 to identify significant changes in vegetation cover; when the vegetation index change exceeds 20% and lasts for more than 30 days, it is determined that the vegetation cover in the area has changed significantly; by analyzing the peaks and troughs in the NDVI time series, the budding period, vigorous growth period, and withering period of vegetation are determined; for areas where abnormal changes in vegetation cover are detected, vegetation type classification is re-conducted; the vegetation parameter database is updated, including leaf area index, biomass, and vegetation height parameters.

[0026] A third aspect of this application provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements a method for retrieving soil moisture content.

[0027] Compared with the prior art, this application has the following beneficial effects: By fusing L-band dual-polarization SAR data with multi-source optical remote sensing and digital elevation model data, and combining quantitative removal of vegetation scattering, multi-model integrated inversion and dynamic time series monitoring, this application accurately extracts soil backscattering signals, realizes high-precision surface soil moisture content inversion and landslide risk classification and early warning based on the inversion results. Attached Figure Description

[0028] Figure 1 This is a module communication timing diagram of the soil moisture content inversion system based on L-band dual polarization data in this application embodiment. It shows the data interaction process and collaborative working relationship between the SAR data acquisition module, optical data processing module, vegetation scattering separation module, inversion model construction module, spatial distribution calculation module and risk assessment module.

[0029] Figure 2 This is a schematic flowchart of the soil moisture content inversion method based on L-band dual polarization data disclosed in this application.

[0030] Figure 3 This is a schematic diagram of the landslide risk level classification results obtained after the actual application of the embodiments of this application in Liupanshui area of ​​Guizhou Province, reflecting the actual effect of the inversion results in geological disaster risk assessment. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be described in more detail below with reference to the accompanying drawings. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are some embodiments of this invention, but not all embodiments.

[0032] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] The embodiments and directional terms described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0034] like Figure 1As shown, a soil moisture content inversion system based on L-band dual-polarization data includes a SAR data acquisition module, an optical data processing module, a vegetation scattering separation module, an inversion model construction module, a spatial distribution calculation module, and a risk assessment module. The SAR data acquisition module is used to periodically collect L-band dual-polarization SAR data of the target area to achieve multi-temporal data acquisition. The optical data processing module is configured to receive optical remote sensing data and assist in identifying and separating vegetation scattering components by calculating vegetation indices. The vegetation scattering separation module responds to the vegetation indices output by the optical data processing module and removes vegetation scattering contributions from the SAR data based on quantitative removal rules, improving the soil moisture content retrieval efficiency. The process involves: acquiring purified soil backscattering signals; training and optimizing a soil moisture content inversion model based on the purified soil backscattering signals and field sampling data; calculating a spatial distribution map of surface soil moisture content in the target area and verifying its accuracy; and classifying and managing landslide risk levels. Specifically, the SAR data acquisition module, optical data processing module, and vegetation scattering separation module work together to purify the soil scattering signals. The inversion model construction module and spatial distribution calculation module jointly achieve high-precision soil moisture content inversion. The risk assessment module completes landslide risk level classification and early warning based on the inversion results.

[0035] A soil moisture content retrieval system based on L-band dual-polarization data, employing the aforementioned structure, enables the fusion processing and collaborative analysis of multi-source remote sensing information. It uses optical data to aid in identifying vegetation-covered areas and, combined with a quantitative vegetation scattering removal strategy, effectively eliminates vegetation interference with SAR signals, thereby extracting purer soil backscattering information. Using an retrieval model constructed from fused measured data, it can accurately calculate the distribution of surface soil moisture content on a spatial scale, further supporting risk assessment and early warning of geological hazards such as landslides. Through the collaborative operation of its modules, this system improves the accuracy of soil moisture content retrieval while enhancing continuous spatiotemporal observation capabilities and the timeliness and reliability of landslide risk monitoring, demonstrating strong practical value and potential for widespread application.

[0036] The second aspect of this application, Figure 2 A method for soil moisture content inversion based on L-band dual polarization data is shown, including the following steps.

[0037] Step 1: Acquire L-band dual-polarization SAR data, optical remote sensing data, digital elevation model data, and field soil moisture content sampling data of the target area within the set time window, and register and preprocess the multi-source remote sensing data according to the time series characteristics to construct a multi-source remote sensing dataset with time series characteristics.

[0038] Step 2: Calculate the vegetation index based on the optical remote sensing data, and quantitatively separate and remove the vegetation scattering contribution in the L-band dual-polarization SAR data based on the vegetation index to extract the pure soil backscattering signal; at the same time, divide the slope units based on the digital elevation model data.

[0039] Step 3: Divide the field soil moisture content sampling data into training set, validation set and test set in a 6:2:2 ratio. Use the pure soil backscattering signal and training set data to establish a soil moisture content inversion model. Use the validation set to fine-tune the model during training and optimize and calibrate the model parameters of the inversion model.

[0040] Step 4: Apply the optimized inversion model to the L-band dual-polarization SAR data processing, calculate the spatial distribution map of surface soil moisture content in the target area, and conduct accuracy verification and multi-index evaluation in conjunction with test set data.

[0041] Step 5: Combining the soil moisture content inversion results from Step 4 with historical soil moisture content data, establish a landslide risk level assessment model and determine the safe threshold for soil moisture content.

[0042] Step Six: Combine the spatial distribution map of soil moisture content and the slope unit division results to conduct stability analysis, and apply the landslide risk level assessment model to classify the risk level of different areas; if the soil moisture content of a certain area is detected to exceed the preset safety threshold, the area is marked as a high-risk area and an early warning signal is issued; if the soil moisture content is within the safe range, the routine monitoring process continues.

[0043] Step 7: During the dynamic monitoring process, new L-band dual-polarization SAR data is periodically acquired through the SAR data acquisition module. When the vegetation coverage changes by more than 15%, the vegetation index is recalculated and Step 2 is executed. Otherwise, Step 4 and Step 6 are executed directly using the pre-trained inversion model to achieve continuous monitoring of soil moisture content and real-time assessment of landslide risk. When the inversion accuracy decreases by more than 0.1 or the root mean square error increases by more than 20%, Step 3 is re-executed to retrain the model and optimize the parameters.

[0044] A soil moisture content retrieval method based on L-band dual-polarization data, employing the aforementioned approach, enables time-series fusion and joint analysis of multi-source remote sensing and measured data. It quantitatively removes vegetation scattering contributions from SAR data using vegetation indices extracted from optical remote sensing data, effectively purifying soil backscattering signals and improving the reliability of the inversion input data from the source. By scientifically dividing the training, validation, and test sets to construct the inversion model and introducing model tuning and parameter optimization mechanisms, the accuracy and stability of the soil moisture content retrieval results are significantly improved. Combining high-resolution DEM with slope unit division further enhances the spatial positioning accuracy of the inversion results. Furthermore, a landslide risk assessment model is constructed, incorporating a moisture content safety threshold determination strategy and a real-time early warning mechanism, achieving refined monitoring and dynamic response for geological disaster prevention and control. Through dynamic vegetation change identification and model accuracy detection mechanisms, this method also possesses adaptive updating and long-term operation capabilities, significantly improving the continuity and automation level of soil moisture content monitoring and the real-time performance and accuracy of landslide risk assessment.

[0045] Furthermore, the preprocessing of the multi-source remote sensing data specifically includes the following steps: First, radiometric calibration is performed on the L-band dual-polarization SAR data using σ... 0 The backscattering coefficient was normalized to ensure that the radiometric calibration accuracy was controlled within 0.5 dB. Secondly, the atmospheric scattering effect was removed using the 6S radiative transfer model, and geometric fine correction was performed using ground control points to ensure a registration accuracy better than 1.0 pixel. Then, the digital elevation model data was resampled and filtered, and the resolution was unified to 30 meters using bilinear interpolation. A 5×5 median filter was used to remove noise points from the elevation data. Finally, all data were spatially registered using the UTM projection coordinate system, and cubic spline interpolation and image matching algorithms were combined to achieve spatial consistency of multi-source data, with the registration error controlled within 15 meters.

[0046] The preprocessing methods described above for multi-source remote sensing data effectively improve the radiometric accuracy and geometric consistency of SAR data, ensuring a reliable basis for quantitative analysis of backscattered signals. Simultaneously, atmospheric correction, ground control point calibration, and unified resolution processing significantly enhance the spatial consistency and fusion compatibility among various types of remote sensing data. In particular, the use of the UTM projection coordinate system and high-precision image registration technology controls the registration error to within 15 meters, providing high-quality, highly consistent input data support for the subsequent construction of soil moisture inversion models and landslide risk assessment, thereby significantly improving the overall system's inversion accuracy and assessment reliability.

[0047] Furthermore, the vegetation index calculation includes a comprehensive calculation of the Normalized Difference Vegetation Index (NDVI), the Enhanced Vegetation Index (EVI), and the Leaf Area Index (LAI), specifically including...

[0048] The NDVI is calculated using the red and near-infrared bands of optical remote sensing data. The specific calculation formula is: NDVI=(NIR-Red) / (NIR+Red), where NIR represents the surface reflectance value in the near-infrared band and Red represents the surface reflectance value in the red band.

[0049] The Enhanced Vegetation Index (EVI) is calculated using the standard formula: EVI = G × [(NIR - Red) / (NIR + C1 × Red - C2 × Blue + L)], where G is the gain factor, with a value of 2.5; Blue is the reflectivity of the blue light band; C1 is the atmospheric impedance coefficient of the red light band, with a value of 6.0; C2 is the atmospheric impedance coefficient of the blue light band, with a value of 7.5; and L is the canopy background adjustment parameter, with a value of 1.0.

[0050] The leaf area index (LAI) is estimated based on the empirical formula LAI = 3.618 × EVI - 0.118.

[0051] In the process of vegetation scattering separation, a vegetation water cloud model is constructed: σ° veg = A×LAI×cosθ×(1-τ²) / (1-τ²×R), where A is the vegetation scattering coefficient; θ is the incident angle; τ is the vegetation attenuation coefficient; and R is the soil-vegetation interaction parameter.

[0052] The separation of vegetation scattering was achieved using an iterative method, with the following steps: First, the vegetation scattering contribution σ°veg was calculated based on the vegetation water cloud model; then, the vegetation scattering component was subtracted from the total backscattering coefficient σ°total of SAR observations to obtain the soil backscattering coefficient σ°soil; during the iteration process, the vegetation-soil interaction parameter R was re-estimated based on the initially extracted σ°total, the vegetation scattering model was updated, and the iteration was repeated until σ°soil converged. The iteration termination condition was that the relative difference between two adjacent iterations was less than 1% or the number of iterations reached 10.

[0053] In summary, by introducing the joint calculation of multiple vegetation indices such as NDVI, EVI, and LAI, and combining a vegetation-water cloud model based on physical mechanisms to quantitatively model and iteratively remove the vegetation scattering contribution, we can effectively remove the interference of vegetation on SAR signals and accurately extract the pure soil backscattering coefficient. At the same time, by fitting and optimizing the model parameters using the least squares method, the residual vegetation scattering influence is controlled within 2dB, which significantly improves the accuracy and stability of the inversion input data and provides a highly reliable physical basis and parameter support for subsequent soil moisture content inversion models.

[0054] Furthermore, the slope unit division process includes the following steps: First, the main topographic factors, including slope, aspect, and plane curvature, are calculated based on digital elevation model data, with the slope gradient obtained using a third-order finite difference algorithm; then, the D8 flow direction algorithm combined with flow accumulation analysis is used to identify the confluence network, and a flow accumulation threshold of 500 pixels is set to extract the river network system; next, initial slope unit division is performed, and the unit boundaries are optimized and adjusted based on constraints that the slope standard deviation is less than 8 degrees and the elevation variation coefficient is less than 0.3; finally, a quality assessment is conducted based on the hydrological connectivity and topographic consistency of the slope units, eliminating irregular units with an area less than 0.5 square kilometers or an aspect ratio greater than 8:1, ensuring that each slope unit has relatively uniform hydrological response characteristics and topographic features, while the slope variation within the control unit is within ±10 degrees.

[0055] In summary, by introducing a slope unit division method based on topographic factors, and combining high-precision topographic parameters such as slope, aspect, and plane curvature with cumulative flow analysis, the structure of geomorphic units can be effectively extracted. Furthermore, by optimizing and adjusting unit boundaries using constraints such as slope standard deviation, elevation variation coefficient, and hydrological connectivity, the accuracy and topographic consistency of slope unit division are improved, and the representativeness and stability of each unit in hydrological response and landslide sensitivity analysis are significantly enhanced. Simultaneously, by eliminating units with excessively small areas or abnormal shapes, a reasonable spatial structure segmentation of complex topographic regions is achieved, providing a high-quality topographic zoning foundation for subsequent spatial matching of soil moisture content inversion results and landslide risk level assessment.

[0056] Furthermore, the soil moisture content inversion model is constructed based on an ensemble learning framework of support vector machine regression and random forest. The specific steps are as follows: A radial basis function is selected as the kernel function of the support vector machine, and hyperparameters are optimized within the range of C∈[1,1000] and γ∈[0.001,1] using a grid search method; input features include the normalized backscattering coefficients of HH polarization and HV polarization, the polarization ratio HV / HH, the incident angle, the local incident angle, the slope, and the aspect; a random forest consisting of 100 decision trees is constructed, with a maximum tree depth of 10 layers and a minimum number of splits per node of 5; 10-fold cross-validation is used for parameter tuning during model training, and root mean square error and mean absolute error are used as performance evaluation indicators; a weighted average method is used to fuse the two prediction results, with the weight coefficients determined by the validation set performance; the final inversion model achieves an RMSE of less than 6% on the validation set, and a correlation coefficient R0. 2 It reaches 0.70 or higher.

[0057] In summary, by constructing an ensemble learning model based on support vector machine regression and random forest, and combining multi-source polarimetric SAR features and terrain factors as joint inputs, this inversion model introduces grid search and 10-fold cross-validation mechanisms during the parameter optimization stage, effectively improving the model's generalization ability and robustness. Simultaneously, by utilizing a weighted fusion strategy to integrate the predictive advantages of support vector machine and random forest, the model's uncertainty and error fluctuations are significantly reduced, ultimately achieving an RMSE control within 6% and an R-value of [missing information - likely a missing value] on the validation set. 2 With a performance level greater than 0.70, it achieves high-precision and highly adaptable inversion of soil moisture content.

[0058] Furthermore, the model parameter optimization and calibration process includes the following steps: First, a genetic algorithm is used to globally optimize the model hyperparameters, setting the population size to 30, the maximum number of iterations to 150, the crossover probability to 0.8, and the mutation probability to 0.1. In the parameter search space, the regularization parameter of the support vector machine ranges from [1, 500], the kernel parameter ranges from [0.01, 1], and the number of trees in the random forest ranges from [50, 100]. The effective range of the parameters is initially determined through coarse-grained grid search, and then fine-tuned using random search to obtain the optimal hyperparameter combination. Training is stopped when the validation loss shows no improvement after 15 consecutive iterations to prevent overfitting. Finally, the generalization ability of the optimized model is verified using an independent test set to ensure that the root mean square error of soil moisture content retrieval is controlled within 7% and the mean absolute error is controlled within 5%.

[0059] In summary, by introducing a genetic algorithm to globally optimize the hyperparameters of the support vector machine and random forest models, and combining population evolution strategies and crossover / mutation operations, the parameter search space was effectively expanded, improving optimization efficiency. A multi-stage tuning method combining coarse-grained grid search and random search was employed to accurately pinpoint the optimal parameter range, enhancing the model's adaptability and robustness. An early stopping mechanism was implemented to avoid overfitting, ensuring the stability and generalization ability of the model training. Finally, after verification on an independent test set, the optimized inversion model achieved high-precision inversion results with a root mean square error of less than 7% and a mean absolute error of less than 5%, significantly improving the accuracy and reliability of soil moisture content prediction.

[0060] Furthermore, the construction process of the landslide risk level assessment model includes the following steps: First, a landslide risk assessment index system is constructed, selecting eight main influencing factors: soil moisture content, slope, aspect, elevation, vegetation cover, soil type, lithology, and distance to fault zones. Second, based on the spatial distribution relationship between historical landslide points and each factor, the information content value of each index is determined. Then, a soil moisture content grading standard is formulated, where: moisture content below 20% is defined as a safe zone, 20%–25% as a warning zone, 25%–30% as a caution zone, and above 30% as a danger zone. A landslide probability calculation formula is then constructed. , where w i Let x be the weight of the i-th factor. i The standardized factor values ​​are given, where P represents the probability of a landslide occurring and n is the total number of indicators involved in the calculation. The performance of the evaluation model is validated to ensure that the area under the receiver operating characteristic curve of the model is greater than 0.75.

[0061] In summary, by constructing a landslide risk assessment system encompassing multi-dimensional influencing indicators such as soil moisture content, topographic factors, vegetation cover, and geological conditions, and combining the spatial distribution characteristics of historical landslide events, the information weights of each indicator are scientifically determined, enabling a quantitative description of landslide hazard. Based on a clear soil moisture content grading standard, regional risks are divided into four levels: safe, caution, warning, and dangerous, facilitating risk classification management and early warning. A logistic regression model is used to calculate the probability of landslide occurrence, ensuring the model has good statistical interpretability and predictive performance. Model validation shows that the area under the receiver operating characteristic curve (AUC) exceeds 0.75, indicating that the constructed landslide risk level assessment model has high discrimination ability and practical value, effectively supporting the scientific assessment and dynamic management of regional landslide risks.

[0062] Furthermore, the process of determining the safety threshold includes the following steps: First, collect historical landslide event data and soil moisture content observation data of the target area over the past 15 years to construct a landslide risk database containing more than 300 samples; based on receiver operating characteristic curve analysis, determine the optimal threshold point to achieve the best balance between precision and recall; establish a classification threshold system for different soil types, specifically setting the safety threshold at 22% for sandy soil, 25% for loam, and 28% for clay soil; subsequently, dynamically adjust the threshold based on previous rainfall and soil saturation; finally, calculate the threshold range at a 95% confidence level: 20%–24% for sandy soil, 23%–27% for loam, and 26%–30% for clay soil.

[0063] In summary, by comprehensively analyzing landslide events and soil moisture content observation data in the target area over the past 15 years, a landslide risk database covering more than 300 samples was constructed. The optimal safety threshold was determined based on the Receiver Operating Characteristic (ROC) method, achieving an effective balance between accuracy and recall in risk assessment. A differentiated classification threshold system was established for different soil types, setting safety thresholds for sandy soil, loam, and clay soil respectively, improving the targeting and precision of the assessment. The thresholds were dynamically adjusted based on rainfall and soil saturation, enhancing the model's responsiveness to environmental changes. Finally, the safety threshold range for each soil type at a 95% confidence level was determined, ensuring the scientific validity and reliability of the landslide risk early warning.

[0064] Furthermore, the recalculation process when vegetation cover changes by more than 25% includes the following steps: First, through multi-temporal normalized vegetation index difference analysis, areas with significant changes in vegetation cover are identified, and their boundaries are determined; then, the vegetation index is recalculated for the changed areas, and its time series data is smoothed to suppress short-term fluctuations and observation noise; based on this, the vegetation water cloud model parameters are reconstructed based on the latest vegetation status, and the scattering separation parameters are fitted and optimized using the nonlinear least squares method; by comparing the backscattering statistical characteristics before and after vegetation scattering removal, the scattering removal accuracy is ensured to be maintained above 80%; the seasonal change trend of vegetation cover is predicted through vegetation phenological curve analysis; a regional processing strategy is adopted, and corresponding processing parameters are set for different vegetation types and growth stages; when the rate of change in vegetation cover is detected to exceed 8% / month, the recalculation process is initiated in advance.

[0065] In summary, multi-temporal normalized vegetation index (NZRI) difference analysis accurately identifies areas of significant vegetation cover change and, combined with boundary range determination, enables efficient monitoring of vegetation dynamics. Recalculating vegetation indices in these areas and employing time-series smoothing effectively suppresses short-term fluctuations and observational noise, improving data stability. Based on the latest vegetation conditions, the vegetation water cloud model parameters are dynamically reconstructed, and the scattering separation parameters are precisely fitted and optimized using nonlinear least squares methods, ensuring that vegetation scattering removal accuracy remains above 80%. Seasonal change trends are predicted through vegetation phenological curve analysis, and processing parameters are rationally set for different vegetation types and growth stages using regional processing strategies. Furthermore, a vegetation cover change rate threshold monitoring mechanism is established; when the change rate exceeds 8% / month, a recalculation process is promptly initiated, achieving dynamic adaptive adjustment of vegetation scattering separation and significantly improving the timeliness and accuracy of soil moisture content retrieval.

[0066] Furthermore, the early warning signal is issued using a four-level early warning mechanism, including blue, yellow, orange, and red warning levels. Specifically: a blue warning is triggered when the soil moisture content reaches 85% of the safe threshold, alerting relevant departments to pay attention, and the warning is valid for 24 hours; a yellow warning is triggered when the soil moisture content reaches 95% of the safe threshold, requiring monitoring to be increased to once every 3 days, and the warning is valid for 12 hours; an orange warning is triggered when the soil moisture content exceeds the safe threshold but the stability coefficient is greater than 1.1, initiating a level-two emergency response procedure, and adjusting the monitoring frequency to once a day; a red warning is triggered when the soil moisture content exceeds the safe threshold and the stability coefficient is less than 1.1, immediately initiating a level-one emergency response.

[0067] In summary, the four-level early warning mechanism, by setting four levels—blue, yellow, orange, and red—achieves graded response management for abnormal changes in soil moisture content: when soil moisture content reaches 85% of the safe threshold, a blue warning is triggered, reminding relevant departments to pay attention and maintain a 24-hour warning validity period; when it reaches 95%, a yellow warning is triggered, requiring an increase in monitoring frequency to once every 3 days, and the warning validity period is shortened to 12 hours; when the moisture content exceeds the safe threshold and the stability coefficient is greater than 1.1, an orange warning is triggered, initiating a level-two emergency response procedure, and adjusting the monitoring frequency to once daily; if the moisture content exceeds the safe threshold and the stability coefficient is less than 1.1, a red warning is triggered, immediately initiating a level-one emergency response. This mechanism realizes dynamic grading and response gradients for risk warnings, improves the accuracy of warnings and the timeliness of emergency management, and effectively enhances the prevention and control capabilities for geological disasters such as landslides.

[0068] Furthermore, the vegetation cover change detection process specifically includes the following steps: First, a vegetation index time series database is established to support long-term monitoring and trend analysis; then, seasonal decomposition is used to extract the trend, seasonal, and random components of the vegetation index, separating periodic changes from anomalous disturbances; the significance level for mutation detection is set to 0.05 to identify significant changes in vegetation cover; when the vegetation index change exceeds 20% and lasts for more than 30 days, it is determined that the vegetation cover in the area has changed significantly; by analyzing the peaks and troughs in the NDVI time series, the budding period, vigorous growth period, and withering period of vegetation are determined; for areas where abnormal changes in vegetation cover are detected, vegetation type classification is re-conducted; the vegetation parameter database is updated, including leaf area index, biomass, and vegetation height parameters.

[0069] In summary, by establishing a vegetation index time series database, long-term dynamic monitoring and trend analysis of vegetation cover in target areas can be achieved. The seasonal decomposition method effectively separates the trend, seasonal, and random disturbance components in vegetation indices, enhancing the ability to identify periodic changes and anomalous events. By setting a significance level of 0.05 for mutation detection, significant changes in vegetation cover are accurately identified; when the change exceeds 20% and lasts for more than 30 days, vegetation cover anomalies are promptly determined. Combined with NDVI time series peak-valley analysis, the characteristics of each stage of vegetation growth are clarified, enabling precise control of the vegetation growth cycle. For anomalous areas, vegetation type classification is re-conducted and the vegetation parameter database is dynamically updated to ensure the timeliness and accuracy of parameters such as leaf area index, biomass, and vegetation height. This process significantly improves the detection accuracy and response speed of vegetation cover changes, providing reliable dynamic parameter support for subsequent vegetation scattering separation and soil moisture content inversion.

[0070] A third aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the soil moisture content inversion method provided in this application.

[0071] In Liupanshui, Guizhou Province, a method, system, and medium for soil moisture retrieval based on L-band dual-polarization data have been successfully applied. First, L-band dual-polarization SAR data from the ALOS-2 satellite, collected on August 16 and September 27, 2022, were used. This data, with a resolution of 3 meters, can penetrate clouds to obtain surface backscattering information. Simultaneously, optical data from the Landsat-8 satellite, acquired on August 22, 2022, was used to calculate vegetation indices and other information. Two simultaneous ground sampling operations were conducted in the Liupanshui area before and after the satellite's transit, yielding a total of 222 valid soil moisture data points.

[0072] In the data preprocessing stage, radiometric calibration and geometric correction are performed on the SAR data. Radiometric calibration uses σ 0 Backscattering coefficient normalization correction was performed with an accuracy controlled within 0.5 dB, and geometric correction using ground control points achieved an accuracy better than 1.0 pixel. Atmospheric and geometric corrections were also performed on the optical data to ensure spatial consistency with the SAR data.

[0073] Next, vegetation scattering separation was performed. NDVI was calculated using the red and near-infrared bands of optical remote sensing data, and vegetation cover was comprehensively assessed by combining EVI and LAI. Based on these vegetation indices, a WCM was constructed to simulate the vegetation scattering contribution. The model parameters were determined by least squares fitting, thereby separating the vegetation scattering contribution from the SAR data and extracting the pure soil backscattering signal.

[0074] Then, a soil moisture content retrieval model was constructed. 222 valid soil moisture content data points were divided into training, validation, and test sets in a 6:2:2 ratio. An ensemble learning framework combining support vector machine regression and random forest was used to build the model. Input features included normalized backscattering coefficients of HH and HV polarizations. The hyperparameters of the support vector machine were optimized using a grid search method, and a random forest consisting of 100 decision trees was constructed with appropriate parameters. During training, 10-fold cross-validation was used for parameter tuning. RMSE and MAE were used as performance evaluation metrics, and a weighted average method was used to fuse the two prediction results. The weight coefficients were determined based on the validation set performance.

[0075] The trained model was applied to L-band dual-polarization SAR data processing in Liupanshui area to obtain spatial distribution maps of surface soil moisture content, including soil moisture content distribution maps at depths of 10cm and 30cm under HH and HV polarization on August 16 and September 27, 2022. Validation by comparison with field sampling data showed high inversion accuracy; for example, at a depth of 10cm, R², MAE, MSE, and MAPE under HH polarization reached 0.902, 2.048, 2.685, and 0.062, respectively.

[0076] In addition, a landslide risk assessment was conducted. Historical landslide event data and soil moisture content observation data from the past 15 years were collected in the Liupanshui area to construct a landslide risk database containing more than 300 samples. Based on ROC analysis, safety thresholds for different soil types were determined, such as 22% for sandy soil, 25% for loam, and 28% for clay soil, and the threshold ranges at a 95% confidence level were calculated. Combining the soil moisture content inversion results and safety thresholds, landslide risk levels were classified and early warning management was implemented in the Liupanshui area, including setting trigger conditions for different early warning levels.

[0077] Finally, we obtained the following: Figure 3 The landslide risk level classification results shown are displayed using a dedicated platform developed based on the open-source tool Cesium, enabling data management and 3D visualization for easy user viewing. This application case provides scientific basis and technical support for geological disaster prevention and control in the Liupanshui area, helping to identify potential landslide areas in advance, take timely preventive measures, and reduce losses caused by geological disasters.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A soil moisture content inversion system based on L-band dual-polarization data, comprising a SAR data acquisition module, an optical data processing module, a vegetation scattering separation module, an inversion model construction module, a spatial distribution calculation module, and a risk assessment module, characterized in that, The SAR data acquisition module is used to periodically collect L-band dual-polarization SAR data of the target area to achieve the acquisition of multi-temporal data; the optical data processing module is configured to receive optical remote sensing data and assist in the identification and separation of vegetation scattering components by calculating vegetation indices. The vegetation scattering separation module responds to the vegetation index output by the optical data processing module, removes the vegetation scattering contribution from the SAR data based on quantitative removal rules, and extracts the pure soil backscattering signal. The inversion model construction module trains and optimizes the soil moisture content inversion model based on purified soil backscattering signals and field sampling data; the spatial distribution calculation module calculates the spatial distribution map of surface soil moisture content in the target area and verifies its accuracy; the risk assessment module is used for landslide risk level classification and early warning management; wherein, the SAR data acquisition module, optical data processing module, and vegetation scattering separation module work together to purify soil scattering signals, the inversion model construction module and the spatial distribution calculation module jointly achieve high-precision soil moisture content inversion, and the risk assessment module completes landslide risk level classification and early warning based on the inversion results; Vegetation index calculation includes a comprehensive calculation of the Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), and Leaf Area Index (LAI), specifically including: The NDVI is calculated using the red and near-infrared bands of optical remote sensing data. The specific calculation formula is: NDVI=(NIR-Red) / (NIR+Red), where NIR represents the surface reflectance value in the near-infrared band and Red represents the surface reflectance value in the red band. The Enhanced Vegetation Index (EVI) is calculated using the standard formula: EVI = G × [(NIR - Red) / (NIR + C1 × Red - C2 × Blue + L)], where G is the gain factor (value 2.5), Blue is the reflectivity of the blue light band, C1 is the atmospheric impedance coefficient of the red light band (value 6.0), C2 is the atmospheric impedance coefficient of the blue light band (value 7.5), and L is the canopy background adjustment parameter (value 1.0). The leaf area index (LAI) is estimated based on the empirical formula LAI = 3.618 × EVI - 0.

118.

2. A method for soil moisture content inversion based on L-band dual-polarization data, implemented based on the soil moisture content inversion system based on L-band dual-polarization data as described in claim 1, characterized in that, Includes the following steps: Step 1: Acquire L-band dual-polarization SAR data, optical remote sensing data, digital elevation model data, and field soil moisture content sampling data of the target area within the set time window, and register and preprocess the multi-source remote sensing data according to the time series characteristics to construct a multi-source remote sensing dataset with time series characteristics. Step 2: Calculate the vegetation index based on optical remote sensing data, and quantitatively separate and remove the vegetation scattering contribution in L-band dual-polarization SAR data based on the vegetation index to extract the pure soil backscattering signal; at the same time, divide the slope units based on digital elevation model data. Step 3: Divide the field soil moisture content sampling data into training set, validation set and test set in a 6:2:2 ratio. Use the pure soil backscatter signal combined with the training set data to build a soil moisture content inversion model. Use the validation set to fine-tune the model during the training process and optimize and calibrate the model parameters of the inversion model. Step 4: Apply the optimized inversion model to L-band dual-polarization SAR data processing, calculate the spatial distribution map of surface soil moisture content in the target area, and conduct accuracy verification and multi-index evaluation in conjunction with test set data; Step 5: Combining the soil moisture content inversion results from Step 4 with historical soil moisture content data, establish a landslide risk level assessment model and determine the safe threshold for soil moisture content. Step 6: Combine the spatial distribution map of soil moisture content and the results of slope unit division to conduct stability analysis, and apply the landslide risk level assessment model to classify the risk level of different areas; if the soil moisture content of a certain area is detected to exceed the preset safety threshold, the area is marked as a high-risk area and an early warning signal is issued. If the soil moisture content is within a safe range, continue with the routine monitoring procedures. Step 7: During dynamic monitoring, new L-band dual-polarization SAR data is periodically acquired through the SAR data acquisition module. When the vegetation coverage changes by more than 15%, the vegetation index is recalculated and Step 2 is executed. Otherwise, Step 4 and Step 6 are executed directly using the pre-trained inversion model to achieve continuous monitoring of soil moisture content and real-time assessment of landslide risk. When the inversion accuracy decreases by more than 0.1 or the root mean square error increases by more than 20%, Step 3 is re-executed to retrain the model and optimize the parameters.

3. The method for soil moisture content inversion based on L-band dual-polarization data according to claim 2, characterized in that, The slope unit division process includes the following steps: First, the main topographic factors, including slope, aspect, and plane curvature, are calculated based on digital elevation model data, with the slope gradient obtained using a third-order finite difference algorithm. Then, the D8 flow direction algorithm combined with flow accumulation analysis is used to identify the confluence network, setting a flow accumulation threshold of 500 pixels to extract the river network. Next, initial slope unit division is performed, and the unit boundaries are optimized and adjusted based on constraints of a slope standard deviation of less than 8 degrees and an elevation variation coefficient of less than 0.

3. Finally, a quality assessment is conducted based on the hydrological connectivity and topographic consistency of the slope units, eliminating irregular units with an area less than 0.5 square kilometers or an aspect ratio greater than 8:1, ensuring that each slope unit has relatively uniform hydrological response characteristics and topographic features, while the slope variation within the control unit is within ±10 degrees.

4. The method for soil moisture content inversion based on L-band dual-polarization data according to claim 2, characterized in that, The soil moisture content inversion model is constructed based on an integrated learning framework of support vector machine regression and random forest. The specific steps are as follows: the radial basis function is selected as the kernel function of the support vector machine, and the hyperparameters are optimized in the range of C∈[1,1000] and γ∈[0.001,1] by grid search method; Input features include normalized backscattering coefficients of HH and HV polarizations, polarization ratio HV / HH, incident angle, local incident angle, slope, and aspect. A random forest of 100 decision trees is constructed, with a maximum tree depth of 10 layers and a minimum number of splits per node of 5. During model training, 10-fold cross-validation is used for parameter tuning, and root mean square error (RMSE) and mean absolute error (MAE) are used as performance evaluation metrics. A weighted average method is used to fuse the two prediction results, with weight coefficients determined based on validation set performance. The final inversion model achieves an RMSE of less than 6% on the validation set, and a correlation coefficient R0. 2 It reaches 0.70 or higher.

5. The method for soil moisture content inversion based on L-band dual-polarization data according to claim 2, characterized in that, The model parameter optimization and calibration process includes the following steps: First, a genetic algorithm is used to globally optimize the model hyperparameters, with a population size of 30, a maximum number of iterations of 150, a crossover probability of 0.8, and a mutation probability of 0.

1. In the parameter search space, the regularization parameter of the support vector machine ranges from [1, 500], the kernel parameter ranges from [0.01, 1], and the number of trees in the random forest ranges from [50, 100]. The effective range of the parameters is initially determined through coarse-grained grid search, and then fine-tuned using random search to obtain the optimal hyperparameter combination. Training is stopped when the validation loss shows no improvement after 15 consecutive iterations to prevent overfitting. Finally, the generalization ability of the optimized model is verified using an independent test set to ensure that the root mean square error of soil moisture content retrieval is controlled within 7% and the mean absolute error is controlled within 5%.

6. The method for soil moisture content inversion based on L-band dual-polarization data according to claim 2, characterized in that, The recalculation process when vegetation cover changes by more than 25% includes the following steps: First, through multi-temporal normalized vegetation index difference analysis, areas with significant changes in vegetation cover are identified, and their boundaries are determined. Then, the vegetation index is recalculated for the changed areas, and its time-series data is smoothed to suppress short-term fluctuations and observation noise. Based on this, the vegetation water cloud model parameters are reconstructed according to the latest vegetation conditions, and the scattering separation parameters are fitted and optimized using the nonlinear least squares method. By comparing the backscattering statistical characteristics before and after vegetation scattering removal, the accuracy of scattering removal is ensured to remain above 80%. The seasonal change trend of vegetation cover is predicted through vegetation phenological curve analysis. A regional processing strategy is adopted, setting corresponding processing parameters for different vegetation types and growth stages. When the rate of change in vegetation cover exceeds 8% / month, the recalculation process is initiated in advance.

7. The method for soil moisture content inversion based on L-band dual-polarization data according to claim 2, characterized in that, The early warning signal is issued using a four-level early warning mechanism, including blue, yellow, orange, and red warning levels. Specifically: a blue warning is triggered when the soil moisture content reaches 85% of the safe threshold, alerting relevant departments to pay attention, and the warning is valid for 24 hours; a yellow warning is triggered when the soil moisture content reaches 95% of the safe threshold, requiring monitoring to be increased to once every 3 days, and the warning is valid for 12 hours; an orange warning is triggered when the soil moisture content exceeds the safe threshold but the stability coefficient is greater than 1.1, initiating a level-two emergency response procedure, and adjusting the monitoring frequency to once a day; a red warning is triggered when the soil moisture content exceeds the safe threshold and the stability coefficient is less than 1.1, immediately initiating a level-one emergency response.

8. A method for soil moisture content inversion based on L-band dual-polarization data according to claim 2, characterized in that, The vegetation cover change detection process includes the following steps: First, a vegetation index time series database is established to support long-term monitoring and trend analysis. Then, seasonal decomposition is used to extract the trend, seasonal, and random components of the vegetation index, separating periodic changes from anomalous disturbances. A significance level of 0.05 is set for mutation detection to identify significant changes in vegetation cover. When the vegetation index change exceeds 20% and lasts for more than 30 days, a significant change in regional vegetation cover is determined. The budding, vigorous growth, and withering stages of vegetation are determined by analyzing the peaks and troughs in the NDVI time series. For areas where abnormal changes in vegetation cover are detected, vegetation type classification is re-implemented. Finally, the vegetation parameter database is updated, including leaf area index, biomass, and vegetation height parameters.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the soil moisture content inversion method as described in any one of claims 2-8.

Images