Mudslide early identification method and system based on SAR technology combined with three-dimensional model

Abstract

This application provides a method and system for early debris flow identification based on SAR technology combined with a 3D model. Interferometric SAR data is selected as the wide-area coverage data for the target area, and fully polarimetric SAR data is selected as the local refined data for the target area. Phase unwrapping is performed on the wide-area coverage data to generate a wide-area digital elevation model (DEM). Polarization decomposition and phase optimization are performed on the local refined data to generate a local DEM. The wide-area DEM and the local DEM are fused to obtain a hybrid DEM. Based on the hybrid DEM, topographic factors are extracted to generate DEM spatial data to support debris flow susceptibility analysis. Early debris flow identification is performed based on historical debris flow data, historical precipitation data, regional soil moisture content, and DEM spatial data, yielding identification results. A differential monitoring and early warning system based on SAR technology and a 3D model is established. Remote sensing data generated by SAR technology is applied to debris flow disaster identification, improving the early identification capability of debris flow hazards and conducting debris flow hazard deformation prediction analysis.

Description

Technical Field

[0001] This application relates to the field of geological disaster monitoring, and in particular to a method and system for early identification of debris flows based on SAR technology combined with a three-dimensional model. Background Technology

[0002] Debris flows are one of the major types of geological disasters, often triggered by factors such as heavy rainfall, snowmelt, or earthquakes. They are characterized by their insidious formation process, rapid development, and severe destructive power. Due to their weak early signs, which are easily obscured by complex terrain and vegetation cover, traditional ground monitoring methods (such as rain gauges, mud level gauges, and video surveillance) are insufficient for effectively identifying the gestation stage. While existing remote sensing technology can provide wide-area observation capabilities, it still has limitations in terms of temporal resolution, penetration ability, and sensitivity to minor surface deformations.

[0003] Synthetic Aperture Radar (SAR) technology is widely used in the identification of geological hazards such as landslides and ground subsidence due to its all-weather, all-time, and high-precision surface deformation monitoring capabilities. In particular, Differential Interferometric SAR (D-InSAR) and temporal InSAR (such as PS-InSAR and SBAS-InSAR) technologies can effectively invert millimeter-level surface deformation rates, providing crucial information for identifying potentially unstable slopes. However, single SAR data cannot fully reflect the multi-factor coupling mechanism of debris flow occurrence, especially in the absence of supporting background knowledge such as topography, geology, and hydrology, which can easily lead to misjudgments or omissions.

[0004] Existing debris flow monitoring and early warning technologies still have significant shortcomings in detection and identification, especially in the early stages of disaster formation, where there is a lack of sensitive, robust, and low-cost identification methods. Summary of the Invention

[0005] The following is an overview of the topics described in detail in this article.

[0006] The purpose of this application is to at least partially solve one of the technical problems existing in the related technologies. The embodiments of this application provide a method and system for early identification of debris flow based on SAR technology combined with three-dimensional models, thereby improving the early identification capability of debris flow disaster hazards.

[0007] An embodiment of this application provides a method for early identification of debris flows based on SAR technology combined with a three-dimensional model, comprising: Obtain historical debris flow data and historical precipitation data for the target area; Interferometric SAR data was selected as the wide-area coverage data for the target area, and fully polarimetric SAR data was selected as the local fine-grained data for the target area. Phase unwrapping is performed based on the wide-area coverage data to generate a wide-area digital elevation model (DEM). Based on the local refined data, polarization decomposition and phase optimization are performed to generate a local DEM; A hybrid DEM is obtained by fusing wide-area DEM and local DEM; Based on the hybrid DEM, the topographic factors required to support debris flow susceptibility analysis are extracted from the DEM spatial data. Early identification of debris flows is performed based on the historical debris flow data, the historical precipitation data, the regional soil moisture content, and the DEM spatial data, and the identification results are obtained.

[0008] According to certain embodiments of this application, the step of performing early debris flow identification based on the historical debris flow data, the historical precipitation data, and the DEM spatial data to obtain identification results includes: Based on the DEM spatial data, topographic factors such as slope, aspect, confluence path, and longitudinal profile of the channel are extracted as favorable physical conditions for judging the formation and flow of debris flows. Using remote sensing images and image enhancement processing, we can identify linear and ring structures at different scales, focusing on areas with abundant loose deposits, well-developed joints and fissures, fractured rock masses, and ancient landslides or debris flow deposit fans, as material conditions for initiating debris flows under rainfall conditions. Remote sensing imagery is used to interpret key information on land cover changes, vegetation disturbance, gully morphology evolution, and potential source areas, serving as relevant conditions for determining the formation and flow of historical debris flows in the basin.

[0009] According to certain embodiments of this application, the step of performing early debris flow identification based on the historical debris flow data, the historical precipitation data, and the DEM spatial data to obtain identification results includes: A three-dimensional terrain model of the target area is constructed using DEM spatial data. Topographic factors such as slope, aspect, curvature, runoff accumulation, and channel density are extracted. Combined with multi-source geographic information, historical debris flow points, active fault zones, areas with high rainfall, steep terrain, and areas rich in loose sediment sources are spatially overlaid and comprehensively analyzed. A preliminary debris flow susceptibility assessment map is constructed, and areas are delineated as high-risk debris flow zones based on the debris flow susceptibility assessment map to generate a spatially continuous debris flow susceptibility probability map.

[0010] According to certain embodiments of this application, the step of identifying debris flows based on the historical debris flow data, the historical precipitation data, and the DEM spatial data to obtain identification results includes: A three-dimensional basic geographic base map is constructed. Point cloud data of the target area is acquired through LiDAR. Ground points and non-ground points are separated by filtering algorithms to generate bare land DEM data. Key terrain factors are extracted based on the bare land DEM data to identify potential loose debris source areas. Combining point cloud intensity, local elevation anomalies, debris source areas, accumulation fans, and spoil heaps, a three-dimensional spatial model of debris flow susceptibility is output.

[0011] According to certain embodiments of this application, the step of performing early debris flow identification based on the historical debris flow data, the historical precipitation data, and the DEM spatial data to obtain identification results includes: Acquire multi-source time-series SAR data; The multi-source time-series SAR data is subjected to interferometric pair generation, interferogram generation and filtering, coherence assessment and deformation parameter inversion. Sliding window analysis is performed on the deformation time series to calculate local acceleration. Based on the local acceleration and preset threshold, accelerated creep regions are marked, and spatially localized but temporally sudden uplift or subsidence is identified. Combined with the geological background, the inducing factors are determined, and the soil deformation field is obtained. Based on the physical relationship between the time series of SAR backscattering coefficients and soil dielectric constant, an empirical relationship for soil volumetric water content is established. Based on the empirical relationship for soil volumetric water content, the soil saturation state is determined, and the soil moisture field is obtained. By spatially aligning the soil deformation field and soil moisture field and performing physical correlation analysis, potential instability risk levels are classified.

[0012] According to certain embodiments of this application, the step of performing early debris flow identification based on the historical debris flow data, the historical precipitation data, and the DEM spatial data to obtain identification results includes: Alignment based on bare ground DEM data, soil deformation field, and soil moisture field; A spatial distribution map of debris flow probability was constructed based on aligned bare land DEM data, soil deformation field, and soil moisture field.

[0013] According to certain embodiments of this application, the step of performing phase unwrapping based on the wide-area coverage data to generate a wide-area digital elevation model (DEM) includes: Interferometric phase diagrams are generated by differential interferometry based on the wide-area coverage data; A phase filtering algorithm is applied to the interferometric phase map to suppress noise; Phase unwrapping is performed on the interferometric phase map to obtain continuous unwrapped phases; Based on the unwrapped phase and orbital parameters, a wide-area digital elevation model (DEM) is generated by inversion.

[0014] According to certain embodiments of this application, the step of performing polarization decomposition and phase optimization based on the local refined data to generate a local DEM includes: The total backscattering of the localized refined data is decomposed into surface scattering, dihedral scattering, and volume scattering; The localized refined data is decomposed by introducing a helical scattering term; The volume scattering-dominant region and the surface scattering-dominant region are quantitatively identified based on the decomposition results. By utilizing the phase consistency differences between volume scattering and surface scattering in different polarization channels, the interference phase can be optimized by weighting. Alternatively, a coherent polarization joint mask can be used to eliminate regions with low coherence and high volume scattering, while retaining high coherence surface scattering pixels. Estimate vegetation height and correct for ground phase; The local refined data, which has undergone polarization decomposition and phase optimization, is then unwrapped and its elevation is inverted again to generate a local DEM.

[0015] A second aspect of this application provides an early debris flow identification system based on SAR technology combined with a three-dimensional model. The system is constructed using the debris flow identification method based on SAR technology as described in the first aspect of this application. The system performs the following steps: Based on the stability assessment and changing trends of the debris flow area, conduct sensitivity analysis or probability assessment to determine the hazard level and plan emergency response plans and evacuation routes; By integrating new SAR data and weather forecasts, and utilizing GIS platforms, 3D terrain models, and dynamic charts, monitoring data is combined with geographic information to generate heat maps or risk zoning maps. Fuzzy comprehensive evaluation method is then used to classify the probability of debris flow occurrence in different areas. The monitoring data, model prediction results, and risk zoning maps are imported into the GIS platform to construct a 3D model of the debris flow area. The terrain data is overlaid on the GIS platform to simulate the movement path of the debris flow, calculate the threatened area, display the risk level distribution through a heat map, and overlay population density and infrastructure layers to construct a 3D visualization platform.

[0016] According to certain embodiments of the second aspect of this application, the system performs the following steps: Based on multidimensional monitoring data, the static susceptibility and dynamic hazard of debris flows are analyzed. Combined with the probability of disaster in precipitation forecasts, a debris flow early warning model is constructed to dynamically judge the changing trend of debris flow areas. Based on the changing trend of debris flow areas, hazard analysis and risk avoidance decisions are made. The results of debris flow hazard analysis and evacuation decision-making are visualized to obtain interactive information on hazard analysis and evacuation decision-making, and the debris flow early warning model is dynamically corrected based on the interactive information.

[0017] The above-mentioned scheme has at least the following beneficial effects: It establishes a differential monitoring and early warning system based on SAR technology and three-dimensional models; utilizes SAR technology to generate remote sensing data for debris flow disaster identification, greatly improving the early identification capability of debris flow hazards and enabling predictive analysis of debris flow hazard deformation. It solves the problem of low efficiency in applying traditional manual geological disaster survey methods to debris flow identification and analysis, making it difficult to identify the morphology and deformation signs of debris flow hazards, especially in large watersheds where investigators cannot determine the watershed conditions, thus hindering the identification, analysis, and monitoring of debris flow hazards. It solves the problem of inaccurate identification and trend analysis of small debris flows, reducing survey difficulty and expenses. It addresses the issue of identifying the susceptibility of debris flow hazards based on changes in debris flow source and terrain, improving the precision of debris flow trend analysis and monitoring indicators. Attached Figure Description

[0018] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0019] Figure 1 This is a step-by-step diagram of a debris flow early identification method based on SAR technology and a three-dimensional model; Figure 2 This is a flowchart illustrating the steps performed by the debris flow early identification and warning system. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0021] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, or the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0022] The embodiments of this application provide a method and system for early identification of debris flows based on SAR technology combined with a three-dimensional model.

[0023] The embodiments of this application will be further described below with reference to the accompanying drawings.

[0024] Reference Figure 1 An early identification method for debris flows based on SAR technology and a three-dimensional model includes the following steps: Step S110: Obtain historical debris flow data and historical precipitation data for the target area; Step S120: Select interferometric SAR data as the wide-area coverage data for the target area, and select fully polarimetric SAR data as the local fine-grained data for the target area; Step S130: Perform phase unwrapping based on wide-area coverage data to generate a wide-area digital elevation model (DEM); Step S140: Perform polarization decomposition and phase optimization based on local refined data to generate a local DEM; Step S150: The wide-area DEM and the local DEM are fused to obtain a hybrid DEM; Step S160: Extract DEM spatial data based on the topographic factors required to support debris flow susceptibility analysis from the hybrid DEM. Step S170: Based on historical debris flow data, historical precipitation data, regional soil moisture content and DEM spatial data, debris flow identification is performed to obtain the identification results.

[0025] For example, historical debris flow data for the target area can be obtained, including the time, location, scale, triggering factors (such as rainstorm intensity, seismic activity, etc.), losses caused, and subsequent investigation reports. Spatial cluster analysis can then be used to identify hotspots or watersheds where debris flows frequently occur or occur in clusters.

[0026] Based on basic geological data such as regional geological maps, distribution of active fault zones, lithological combinations, weathering degree and stratigraphic structure, the stability and erosion resistance of different rock and soil masses under hydrodynamic action are assessed.

[0027] Obtain historical precipitation data (such as annual average rainfall, number of rainstorm days, maximum 1-hour / 24-hour rainfall intensity, etc.) and climate zoning maps of the target area to delineate areas of concentrated heavy rainfall, areas significantly affected by monsoons, or zones prone to extreme weather. Further refine the spatial distribution of rainfall-induced risks by combining topographic lifting effects (such as windward slopes).

[0028] Choose appropriate bands and data sources. L-band (~23 cm) and P-band (~70 cm) have strong vegetation penetration capabilities, especially in sparse or moderately covered forest areas. They can partially penetrate tree canopies and branches, receiving backscattered signals from the ground or thick branches, thus more closely approximating the actual ground surface. Typical systems include ALOS-2 (L-band), NASA's UAVSAR (L / P-band), and the BIOMASS satellite (P-band). SAR data at L-band and higher wavelengths can be used for surface elevation inversion in vegetated areas, but cannot completely strip away vegetation like LiDAR.

[0029] The following data selection criteria were used for wide-area coverage. L-band repeating orbit or single-flight interferometric SAR data (such as ALOS-2, PALSAR-2, and future TanDEM-L mission data) were preferred. The L-band has a longer wavelength (~23.6 cm), strong penetration through vegetation, and weak temporal decorrelation effect, making it suitable for generating large-area, medium-precision initial DEMs.

[0030] The following data selections were made for localized refinement. Fully polarimetric SAR (PolSAR) data (such as Sentinel-1 dual-polarization enhanced mode, ALOS-2 fully polarimetric mode, or future NISAR data) were introduced in key areas (e.g., debris flow hazard zones, forest cover areas). Fully polarimetric data provides complete scattering matrix information, supporting subsequent scattering mechanism analysis.

[0031] Prioritize the use of L-band repeat orbit or single-pass interferometric data (such as ALOS PALSAR, TanDEM-L future data) for wide-area coverage; for more refined requirements, use polarimetric SAR (PolSAR) or fully polarimetric data for local areas, and use decomposition techniques (such as Freeman-Durden, Yamaguchi) to distinguish between volume scattering (vegetation) and surface scattering (land surface), and assist in phase optimization to form local high-precision data.

[0032] Data preprocessing is performed. All SAR imagery undergoes radiometric calibration, multi-look processing, terrain correction (e.g., using SRTM or ASTER GDEM as initial values), and registration (with sub-pixel accuracy). Interferometric pairs are baseline-selected (vertical baseline < critical baseline) to ensure interferometric phase quality.

[0033] The process of unwrapping phase data based on wide-area coverage data to generate a wide-area digital elevation model (DEM) includes the following steps: Interferometric phase diagrams are generated based on wide-area coverage data through differential interferometry. A phase filtering algorithm is applied to the interferometric phase image to suppress noise; Phase unwrapping is performed on the interferometric phase map to obtain continuous unwrapped phases; Based on the unwrapped phase and orbital parameters, a wide-area digital elevation model (DEM) is generated by inversion.

[0034] Specifically, interferometric phase maps are generated using differential interferometry (DInSAR) technology. Advanced phase filtering algorithms (such as Goldstein filtering and NL-SAR filtering) are employed to suppress noise. Robust phase unwrapping algorithms (such as MinimumCost Flow, SNAPHU) are used to obtain continuous phase. Based on the unwrapped phase and orbital parameters, a wide-area initial DEM is generated (spatial resolution typically 10m to 30m, absolute elevation accuracy approximately 5m to 10m).

[0035] Based on local refined data, polarization decomposition and phase optimization are performed to generate a local DEM, including the following steps: The total backscattering of locally refined data is decomposed into surface scattering, dihedral scattering, and volume scattering. A spiral scattering term is introduced to decompose the locally refined data; The volume scattering-dominant region and the surface scattering-dominant region are quantitatively identified based on the decomposition results. By utilizing the phase consistency differences between volume scattering and surface scattering in different polarization channels, the interference phase can be optimized by weighting. Alternatively, a coherent polarization joint mask can be used to eliminate regions with low coherence and high volume scattering, while retaining high coherence surface scattering pixels. Estimate vegetation height and correct for ground phase; The local refined data, which has undergone polarization decomposition and phase optimization, is then unwrapped and its elevation is inverted again to generate a local DEM.

[0036] Specifically, polarization decomposition is performed. Polarization decomposition methods are applied to the full polarization data of local areas. The Freeman-Durden three-component decomposition decomposes the total backscattering into surface scattering (land surface), dihedral scattering (buildings / steep slopes), and volume scattering (vegetation canopy). The Yamaguchi four-component decomposition further introduces a spiral scattering term to enhance the ability to distinguish complex land features. The decomposition results are used to quantitatively identify areas dominated by volume scattering (e.g., forests) and areas dominated by surface scattering (bare surfaces, water bodies, roads).

[0037] Phase optimization processing guided by scattering mechanisms. In areas with strong volume scattering (such as dense forests), traditional interferometric phases are easily affected by vegetation disturbance, leading to elevation deviations.

[0038] PolInSAR or polarization-assisted phase filtering strategies are introduced. The phase consistency differences between volume scattering and surface scattering in different polarization channels are utilized to weighted optimize the interferometric phase; alternatively, a coherence-polarization joint mask is employed to remove low-coherence, high-volume-scattering regions, retaining high-coherence surface-scattering pixels for elevation inversion. A stochastic volume over-ground (RVoG) model can be combined to estimate vegetation height and correct ground phase (applicable to PolInSAR data).

[0039] Based on the optimized phase, the unwrapping and elevation inversion are performed again to generate a high-precision local DEM (spatial resolution up to 3–5 m, elevation accuracy better than 2 m).

[0040] The local and wide-area DEMs are uniformly projected and resampled to a consistent grid, and then registered at the sub-pixel level. A multi-resolution fusion framework (such as wavelet fusion or Gaussian pyramid fusion) or confidence-weighted fusion is employed: the local region is dominated by the high-precision DEM; the transition zone uses smooth interpolation or a weighted function based on coherence / polarization entropy; the wide-area region retains the initial L-band interferometric DEM. Uncertainty information from the original data is preserved during the fusion process, generating accompanying quality maps (such as elevation error maps and coherence maps).

[0041] In key areas such as sparsely vegetated valleys and source areas, SAR DEMs are generally highly reliable and can be used to extract key factors such as slope and gully morphology. They can be combined with optical remote sensing or field surveys to verify terrain features. By fusing LiDAR DEM (local high precision) and SAR DEM (wide coverage), a hybrid elevation model can be constructed to obtain a hybrid DEM.

[0042] The constructed hybrid elevation model (Hybrid DEM) has the following structural features.

[0043] Bottom layer: Wide-area basic DEM generated by L-band interferometric SAR (full coverage, medium precision); Upper layer: Locally mosaicked PolSAR / PolInSAR optimized DEM (high precision, high resolution); Metadata layer: contains auxiliary information such as data source, processing time, coherence, polarization decomposition results, and elevation uncertainty.

[0044] The hybrid DEM has dynamic update capabilities, supporting seamless embedding of newly added high-precision regional data to achieve continuous model optimization.

[0045] This technology effectively integrates the wide-area coverage advantage of L-band interferometric SAR with the scattering mechanism analysis capability of fully polarimetric SAR, constructing a new generation of hybrid elevation models that combine wide-area coverage, high accuracy, and physical interpretability. The wide-area layer is suitable for macro-topographic analysis and watershed modeling; the local layer is suitable for refined applications such as debris flow disaster investigation, disaster monitoring and early warning, and forestry resource surveys.

[0046] In addition, vegetation error correction is required. For wide-area areas, high-vegetation zones are identified using external data (such as NDVI and land use maps). For highly incoherent areas, interpolation, fusion with other DEMs (such as SRTM and ASTER GDEM), or machine learning methods are used for restoration. For finer details, PolInSAR (polarimetric interferometric SAR) technology is used to separate the vegetation layer from the surface layer through multi-polarity phase center height estimation, achieving elevation inversion of the "subvegetated surface" (requires multi-baseline or fully polarimetric data).

[0047] For mixed DEMs, GIS software (such as ArcGIS, QGIS, GRASS) can be used to fill depressions, calculate slope and aspect, extract water flow direction, generate runoff accumulation, automatically extract channel network and longitudinal profile, and generate DEM spatial data that supports the topographic factors required for debris flow susceptibility analysis.

[0048] Based on DEM spatial data, a method for identifying regional debris flow susceptibility conditions using SAR technology is provided, including the following steps: Based on DEM spatial data, topographic factors such as slope, aspect, confluence path, and longitudinal profile of the channel are extracted as favorable physical conditions for judging the formation and flow of debris flows. Using remote sensing images and image enhancement processing, we can identify linear and ring structures at different scales, focusing on areas with abundant loose deposits, well-developed joints and fissures, fractured rock masses, and ancient landslides or debris flow deposit fans, as material conditions for initiating debris flows under rainfall conditions. Remote sensing imagery is used to interpret key information on land cover changes, vegetation disturbance, gully morphology evolution, and potential source areas, serving as relevant conditions for determining the formation and flow of historical debris flows in the basin.

[0049] For example, under certain conditions, synthetic aperture radar (SAR) technology can penetrate vegetation canopy to obtain surface or near-surface information, which can then be used to generate a digital elevation model (DEM) and further extract topographic factors such as slope, aspect, confluence path, and channel longitudinal profile as favorable physical conditions for judging the formation and flow of debris flows.

[0050] Using multi-resolution SAR remote sensing images and image enhancement processing, we can identify linear structures (such as faults and densely jointed zones) and ring structures at different scales. We pay particular attention to areas with abundant loose deposits, well-developed joints and fissures, fractured rock masses, or ancient landslide / debris flow depositional fans to determine the material conditions that trigger debris flows under rainfall conditions.

[0051] By using multi-temporal and multi-source remote sensing images, key information such as land cover change, vegetation disturbance, gully morphology evolution, and potential sediment source areas can be interpreted to determine the relevant conditions for the formation and flow of historical debris flows in the watershed.

[0052] A method for constructing a 3D model of regional debris flow susceptibility based on DEM spatial data is provided, including the following steps: A three-dimensional terrain model of the target area is constructed using DEM spatial data. Topographic factors such as slope, aspect, curvature, runoff accumulation, and channel density are extracted. Combined with multi-source geographic information, historical debris flow points, active fault zones, areas with high rainfall, steep terrain, and areas rich in loose sediment sources are spatially overlaid and comprehensively analyzed. A preliminary debris flow susceptibility assessment map was constructed, and areas were delineated as high-risk debris flow zones based on the debris flow susceptibility assessment map to generate a spatially continuous debris flow susceptibility probability map.

[0053] For example, a three-dimensional terrain model of the study area is constructed using a high-resolution DEM (such as LiDAR or UAV photogrammetry) to extract terrain factors such as slope, aspect, curvature, runoff accumulation, and channel density.

[0054] This study combines multi-source geographic information such as lithology, soil type, vegetation cover, and land use; it spatially overlays and comprehensively analyzes historical debris flow points, active fault zones, areas with high rainfall, steep terrain (such as slopes greater than 25°), and areas rich in loose sediment sources. Machine learning algorithms (such as random forests and XGBoost) or information volume models are employed in the spatial overlay analysis method within a Geographic Information System (GIS).

[0055] A preliminary debris flow susceptibility assessment map is constructed using expert scoring, information content models, logistic regression, or machine learning algorithms (such as random forest). Based on this map, areas with high material supply potential, strong rainfall background, unfavorable geological conditions, and historical disaster support are delineated as high-risk debris flow areas. A spatially continuous debris flow susceptibility probability map is generated and incorporated into a subsequent refined early identification and dynamic monitoring system based on SAR and 3D models.

[0056] Based on DEM spatial data, a high-precision, multi-dimensional, and dynamic debris flow hazard monitoring system is constructed, including the following steps: A three-dimensional basic geographic base map is constructed. Point cloud data of the target area is acquired through LiDAR. Ground points and non-ground points are separated by filtering algorithms to generate bare land DEM data. Key terrain factors are extracted based on the bare land DEM data to identify potential loose debris source areas. Combining point cloud intensity, local elevation anomalies, debris source areas, accumulation fans, and spoil heaps, a three-dimensional spatial model of debris flow susceptibility is output.

[0057] For example, in terms of data foundation, LiDAR answers the question "Where are debris flows likely to occur?" to reflect static susceptibility. LiDAR provides high-precision 3D point clouds, generating "bare land" DEMs (with vegetation removed), slope, aspect, channel network, source area volume, and boundaries, accurately characterizing topographic control factors and the spatial distribution of loose source material, which is the basis for susceptibility assessment. SAR answers the question "Is it currently evolving into a debris flow?" (dynamic hazard), providing key information such as surface deformation rate (mm level), soil dielectric properties (reflecting moisture content), and changes in land cover, capturing dynamic signals of the disaster-inducing process (such as slope creep and soil saturation) and triggering states; combined with precipitation forecasts, it answers the question "Is an outbreak likely in the next few hours to days?" to reflect the probability of an impending disaster.

[0058] In terms of dynamic monitoring, a dynamic monitoring framework for debris flow fusion based on LiDAR+SAR is constructed.

[0059] A high-precision 3D basic geographic base map (LiDAR-driven) is constructed. Point cloud data of the study area is acquired using airborne or UAV LiDAR. Ground points and non-ground points are separated by filtering algorithms (such as progressive morphological filter) to generate a high-resolution (0.5m to 2m) bare land DEM. Based on the DEM, key topographic factors such as slope, aspect, and curvature are extracted. The runoff accumulation, channel network and longitudinal profile, and watershed boundaries are calculated. Potential loose sediment source areas are identified by combining point cloud intensity, local elevation anomalies, sediment source areas, depositional fans, spoil heaps, etc. The output is a 3D spatial model of debris flow susceptibility, which serves as a static base.

[0060] SAR dynamic monitoring of surface condition and water content (SAR-dominated), acquisition of time-series SAR data (such as Sentinel-1 C-band or ALOS-2 L-band), execution of time-series InSAR processing (such as SBAS or PS-InSAR), inversion of surface deformation rate and time series, identification of accelerated creep zones (sudden increase in deformation rate), location of local uplift / subsidence (which may indicate changes in groundwater pressure), and simultaneous utilization of SAR backscattering coefficient (σ). 0 Inverting surface soil moisture content (requires roughness correction) is crucial. In gullies, slope toes, and alluvial areas without dense vegetation cover, L / C band SAR can effectively reflect changes in soil moisture. This is especially important after heavy rainfall if σ... 0 A significant and persistent increase indicates that the soil is approaching saturation. The outputs are the surface deformation field and the soil moisture field, which are dynamic.

[0061] The process of constructing a joint monitoring model for SAR dynamic land surface condition and soil moisture content includes the following steps: Acquire multi-source time-series SAR data; Interferometric pair generation, interferogram generation and filtering, coherence assessment, and deformation parameter inversion are performed on multi-source time-series SAR data. Sliding window analysis is performed on the deformation time series to calculate local acceleration. Based on the local acceleration and preset threshold, accelerated creep regions are marked, and spatially localized but temporally sudden uplift or subsidence is identified. Combined with the geological background, the inducing factors are determined, and the soil deformation field is obtained. Based on the physical relationship between the time series of SAR backscattering coefficients and soil dielectric constant, an empirical relationship for soil volumetric water content is established. Based on the empirical relationship for soil volumetric water content, the soil saturation state is determined, and the soil moisture field is obtained. By spatially aligning the soil deformation field and soil moisture field and performing physical correlation analysis, potential instability risk levels are classified.

[0062] Specifically, the steps for acquiring and filtering multi-source time-series SAR data are as follows.

[0063] C-band data source: Sentinel-1 A / B (6-day revisit period, VV / VH polarization) is preferred. It is suitable for areas with moderate vegetation cover and is sensitive to surface micro-deformation and humidity changes.

[0064] L-band data sources: ALOS-2 PALSAR-2 (14-day revisit period, HH / HV polarization) or future NISARL band data are selected. They have stronger penetration capabilities and are suitable for sparsely vegetated or moderately covered areas. They are also robust to deep soil moisture and slow deformation.

[0065] The time span requirement is: to cover at least one complete hydrological year in order to capture dynamic processes such as the alternation of dry and wet seasons and responses to rainfall events.

[0066] Orbit consistency means using either ascending or descending orbit data uniformly to avoid geometric distortion; if multi-orbit fusion is required, strict geometric and phase corrections must be performed.

[0067] The steps for preprocessing SAR data are as follows.

[0068] Radiation calibration: converting the original digital signal (DN value) into backscattering coefficients σ 0 (Unit: dB), to eliminate the influence of system gain.

[0069] Multi-view processing: Perform multi-view averaging (e.g., 5×1) in the range and azimuth directions to balance resolution and signal-to-noise ratio.

[0070] Terrain correction: Geocoding and radiometric terrain correction (such as SigmaNaught Terrain Correction) are performed using high-precision DEMs (such as LiDAR DEMs) to eliminate backscattering bias caused by slope.

[0071] Registration: Using the main image as a reference, all slave images are registered at the subpixel level (accuracy ≤ 0.01 pixels) to ensure consistency of interference and scattering.

[0072] Orbit refinement: Improve the accuracy of the interferometric baseline by using precise orbital ephemeris (such as RESTITUTED or PRECISE orbits provided by ESA).

[0073] The steps for time-series InSAR processing (deformation field inversion) are as follows.

[0074] Interference pair generation: An interferometric network is constructed based on spatiotemporal baseline thresholds (e.g., time baseline < 180 days, vertical baseline < 300 m for C-band; L-band can be relaxed).

[0075] Interferogram generation and filtering: Generate differential interferograms (DInSAR), remove terrain phase (using an external DEM); apply adaptive phase filtering (such as Goldstein, NL-InSAR) to suppress phase noise.

[0076] Coherence assessment: Calculate the average coherence of each pixel, remove low coherence regions (e.g., γ < 0.3), and retain stable scatterers.

[0077] Time series inversion method selection: PS-InSAR (Permanent Scattering Object): Applicable to highly coherent targets such as cities and exposed rock masses, extracting millimeter-level deformation; SBAS (Small Baseline Subset): Applicable to natural surfaces (such as slopes and farmland), inverting the average deformation rate through distributed scattering objects.

[0078] Deformation parameter inversion: calculate the line-of-sight (LOS) deformation rate (mm / yr) and complete time series displacement; convert to vertical deformation (requires assumptions of incident angle and azimuth angle, or fusion of rising and falling orbit data).

[0079] The steps for identifying surface deformation anomalies are as follows.

[0080] Accelerated creep zone detection: Perform sliding window analysis (e.g., 30-day window) on the deformation time series, calculate local acceleration (second derivative); set thresholds (e.g., acceleration > 2 mm / month² and lasting ≥ 2 periods) to mark the "accelerated creep" region.

[0081] Local uplift / settlement location: Identify significant uplift (positive displacement) or settlement (negative displacement) that is spatially localized (<1 km²) and temporally sudden (e.g., within 7 days after heavy rainfall); and determine whether it may be caused by changes in groundwater pressure, soil liquefaction, or collapse of underground cavities in conjunction with the geological background.

[0082] The steps for soil moisture content inversion (based on backscattering coefficient) are as follows.

[0083] σ 0 Time series extraction: In areas without dense vegetation cover (NDVI < 0.3, optical remote sensing-assisted masking can be used), extract the σ of VV / VH or HH / HV polarization. 0 Time series.

[0084] Surface roughness correction: using historical dry season (driesest soil) σ 0 As a “reference roughness”; or combined with optical / SAR texture features (such as GLCM entropy, root mean square height) to estimate relative roughness changes; use empirical models (such as Oh model, Dubois model) or machine learning methods (such as random forest) to separate the influence of roughness and dielectric constant.

[0085] Soil moisture content inversion: based on σ 0 Physical relationship with soil dielectric constant (e.g., simplified form of IEM model); establishing σ 0 The empirical relationship between θν and soil volumetric water content (θν) is: θν = a·σ 0 +b·incidence angle +c; where the coefficients a, b, and c are determined by ground-measured data (if no typical regional values ​​are available).

[0086] Saturation state determination: After a heavy rainfall event, if σ 0 Significantly increased (Δσ) 0 >3 dB) and does not drop for more than 5 days, indicating that soil drainage is blocked and tends to be saturated; the response hysteresis is verified by combining rainfall data (such as GPM IMERG).

[0087] The steps for multi-parameter fusion and collaborative analysis are as follows.

[0088] Spatial alignment: unify the deformation field and soil moisture field to the same geographic grid (e.g., 30 m resolution).

[0089] Physical correlation analysis: In key geomorphic units such as slope toe, gully, and alluvial fan, analyze the coupling phenomenon of high water content + accelerated settlement / uplift; identify potential instability precursors: such as soil saturation → increased pore water pressure → reduced effective stress → accelerated soil creep.

[0090] Uncertainty quantification: Deformation field: provides standard deviation and residual RMS; Moisture content: provides inversion error (typically ±0.05 cm³ / cm³).

[0091] The final output product includes the following parts.

[0092] Surface deformation field: Annual average LOS deformation rate map (mm / yr); Deformation time series of key points / regions (CSV / GeoJSON); Vector layer of accelerated creep zone and anomalous uplift / subsidence zone.

[0093] Soil moisture field: Time series raster of surface (0–5 cm) volumetric water content (θ_v, unit cm³ / cm³); Soil saturation warning zone (Boolean mask).

[0094] Comprehensive Risk Indicator Map: Integrating deformation acceleration and soil saturation status, it classifies potential instability risk levels as "low / medium / high".

[0095] The system's update mechanism supports near real-time updates (e.g., Sentinel-1 data latency <3 days), enabling dynamic monitoring closed loop.

[0096] Through the above process, the system constructs a dynamic monitoring model of the land surface condition that is SAR-based and integrates deformation and humidity information, which can provide early warning and scientific decision support for disasters such as debris flows, landslides, and ground subsidence.

[0097] Constructing a debris flow risk assessment system based on DEM spatial data includes the following steps: Alignment based on bare ground DEM data, soil deformation field, and soil moisture field; A spatial distribution map of debris flow probability was constructed based on aligned bare land DEM data, soil deformation field, and soil moisture field.

[0098] Specifically, LiDAR DEM, SAR deformation map, and SAR soil moisture map are unified to the same geographic coordinate system and spatial resolution; at each pixel / sub-basin scale, static attributes (from LiDAR) are associated, including slope >30°, whether it is located in the main channel, and source volume >10. 4 The data includes attributes such as m³, and also correlates dynamic indicators (from SAR), including deformation rate over the past 7 days > 5 mm / yr, soil moisture content > 80% of field capacity, etc. For densely vegetated areas, L-band SAR is preferred; for exposed ditches, C-band SAR can also be used.

[0099] A multi-factor coupled probability model is adopted, such as: Pdebris flow = f(Susceptibility, Deformation, Soil Moisture, R(t)); where Susceptibility is the LiDAR-based susceptibility score (0–1), Deformation is the SAR deformation rate standardized value (e.g., Z-score), Soil Moisture is the relative water content retrieved from SAR (0–1), and R(t) is the forecast rainfall for the next 24–72 hours (from the meteorological model), which is then converted into an effective rainfall index.

[0100] Multiple modeling methods are employed, such as logistic regression, random forest, and physical thresholding, each applicable to regions under different conditions. Logistic regression is suitable for regions with historical disaster cases. Random forest / XGBoost handles non-linear relationships and automatically filters key factors. For physical thresholding, conditions such as "slope > 25° + sudden increase in deformation rate + soil moisture content > 0.35% + predicted rainfall > 50 mm / 24h" are considered high probability, outputting a spatial distribution map of debris flow probability (divided into four levels: low, medium, high, and extremely high).

[0101] Based on the changing trends of debris flow areas, hazard analysis and visualization of risk avoidance decisions are performed, including the following steps: Based on the stability assessment and changing trends of the debris flow area, conduct sensitivity analysis or probability assessment to determine the hazard level and plan emergency response plans and evacuation routes; By integrating new SAR data and weather forecasts, and utilizing GIS platforms, 3D terrain models, and dynamic charts, monitoring data is combined with geographic information to generate heat maps or risk zoning maps. Fuzzy comprehensive evaluation method is then used to classify the probability of debris flow occurrence in different areas. The monitoring data, model prediction results, and risk zoning maps are imported into the GIS platform to construct a 3D model of the debris flow area. The terrain data is overlaid on the GIS platform to simulate the movement path of the debris flow, calculate the threatened area, display the risk level distribution through a heat map, and overlay population density and infrastructure layers to construct a 3D visualization platform.

[0102] Specifically, based on the aforementioned stability assessment and trend of debris flow areas, sensitivity analysis or probability assessment is conducted to determine the hazard level, and emergency plans and evacuation routes are planned. Based on geological disaster prevention and control, yellow, orange, and red warning levels and corresponding response measures are set in the early warning system; a yellow warning indicates a moderate probability, and patrols should be strengthened; an orange warning indicates a high probability, and evacuation preparations should be made; a red warning indicates an extremely high probability, and immediate avoidance of danger is required.

[0103] The system automatically integrates new SAR data (Sentinel-1 revisit period of 6 days, dual-satellite revisit period of 3 days) and the latest weather forecasts. Utilizing a GIS platform, 3D models, and dynamic charts, it combines monitoring data with geographic information to generate heat maps or risk zoning maps. The fuzzy comprehensive evaluation method is used to classify the probability of debris flow occurrence in different areas (medium, high, and extremely high). When the probability of a certain potential hazard area exceeds the threshold, an early warning is triggered. The medium, high, and extremely high probabilities of debris flow occurrence correspond to yellow, orange, and red warnings, respectively.

[0104] The monitoring data, model prediction results, and risk zoning maps are imported into the GIS platform to construct a 3D model of the debris flow area. The terrain data is overlaid on the GIS platform to simulate the movement path of the debris flow, calculate the threatened area, display the risk level distribution on a heat map, and overlay population density and infrastructure layers to construct a 3D visualization platform.

[0105] Based on the threat range of debris flow areas, different levels of early warnings are generated according to the disaster range. Regional evacuation decisions are made based on the connection of early warning information to relevant knowledge graphs. A 3D visualization platform displays: LiDAR terrain, SAR deformation arrows, humidity heatmaps, and warning areas. The visualization interface allows decision-makers to query monitoring data, early warning information, and contingency plan databases in real time, and supports remote assessment by emergency personnel.

[0106] An early identification and warning system for debris flows based on DEM spatial data and SAR technology combined with a three-dimensional model is established.

[0107] Reference Figure 2 The debris flow early identification and warning system performs the following steps: S210: Early identification of debris flows is carried out based on historical debris flow data, historical precipitation data and DEM spatial data. The identification results are obtained and the hazard level is classified according to the regional stability assessment and change trend. S220: Based on multi-dimensional monitoring data and precipitation forecasts, a debris flow early warning model integrating static susceptibility and dynamic hazard is constructed; S230: Integrate SAR data and weather forecasts, and use the fuzzy comprehensive evaluation method to classify the probability of debris flow occurrence in different areas; S240: Import the results of hazard analysis and risk avoidance decisions into the GIS platform, build a three-dimensional visualization early warning system, and dynamically correct the early warning model based on user interaction information.

[0108] It should be noted that in step S210, the static susceptibility index Pstatic is calculated using an information content model or a logistic regression model. Historical debris flow event records, historical precipitation sequences, and high-resolution DEM spatial data are used to conduct early identification of debris flow hazard areas.

[0109] Specifically, topographic factors are extracted as follows.

[0110] Key topographic parameters such as slope S, aspect α, curvature C, and runoff accumulation Ac are calculated based on the DEM: ; Where z is the elevation value.

[0111] Stability assessment uses an Information Value Model (IV) or a logistic regression model to evaluate the regional debris flow susceptibility (Pstatic), and the formula is as follows: ; Among them, X i Let β be the i-th influencing factor (such as slope, lithology, vegetation cover, etc.). i is the regression coefficient.

[0112] The risk level classification is based on regional change trends (such as landslide displacement rate and gully erosion intensity). The weights are determined by the Analytic Hierarchy Process (AHP) or the entropy weight method. Sensitivity zones are divided into four risk levels: low, medium, high, and extremely high. Based on this, emergency plans and optimal evacuation routes are planned (the shortest safe path is solved using Dijkstra's algorithm or A* algorithm).

[0113] It should be noted that in step S220, a dynamic early warning model for debris flow is constructed. Based on the static susceptibility, real-time / forecasted precipitation data and multi-dimensional monitoring data (such as soil moisture content, surface displacement, and gully water level) are integrated to construct a dynamic hazard early warning model.

[0114] Specifically, the pre-disaster precipitation threshold model uses the antecedent precipitation index (API) to characterize the soil moisture state: ; Where Pt is the daily rainfall and k is the attenuation coefficient (usually taken as 0.85~0.95).

[0115] The Dynamic Hazard Index (DHI) is defined as follows: ; Where V SAR Here, is the surface deformation rate (mm / day) retrieved from SAR, and wi is the normalization weight. .

[0116] Calculate the probability of disaster P by combining the precipitation forecast for the next 6–24 hours provided by numerical weather prediction (NWP). trigger An alert is triggered when DHI > θ (threshold).

[0117] It should be noted that in step S230, a fuzzy comprehensive evaluation of the probability of debris flow occurrence is performed based on SAR and meteorological data.

[0118] Specifically, it accesses differential interferometric SAR (D-InSAR) or PS-InSAR data from satellites such as Sentinel-1 in real time to obtain the surface deformation field; it also accesses quantitative precipitation forecasts (QPF) and real-time data from ground rain gauges issued by the meteorological bureau.

[0119] Fuzzy comprehensive evaluation constructs a factor set U={u1,u2,u3,u4}={deformation rate, API, slope, channel blockage}, and a comment set V={v1,v2,v3,v4}={extremely low, low, medium, high}, determining the membership function (such as Gaussian or trapezoidal) μ. ij =μ(v j |u i The weighted average operator is used to synthesize the fuzzy evaluation vector B=A*R, where A is the weight vector and R is the fuzzy relation matrix; the final probability level is determined by the maximum membership principle; in the GIS platform, layers such as DEM, land use, population density, and transportation network are overlaid to generate a debris flow risk zoning heat map.

[0120] It should be noted that in step S240, a three-dimensional visualization early warning system platform and model are constructed. Based on high-precision DEM and oblique photography data, a three-dimensional scene of the debris flow basin is constructed in ArcGIS Pro, SuperMap or CesiumJS.

[0121] Specifically, debris flow dynamic models such as FLO-2D or RAMMS are used, and topographic, source quantity, and flow parameters are input to simulate the debris flow trajectory and deposition range: , ; in, Let S be the flow depth, u and v be the velocity components, S0 be the bed slope, and S be the velocity component. f This refers to the friction slope.

[0122] The simulation results are overlaid with infrastructure (schools, hospitals, roads) and population heat maps to automatically identify high-threat areas.

[0123] Users can confirm warning results, mark false alarms, or adjust parameters through the platform. The system records interaction logs for online learning (such as Bayesian updates or reinforcement learning) to dynamically correct warning model parameters and improve subsequent prediction accuracy.

[0124] It is understandable that a debris flow early identification and warning system based on SAR technology combined with a three-dimensional model achieves the following steps: Based on multidimensional monitoring data, the static susceptibility and dynamic hazard of debris flows are analyzed. Combined with the probability of disaster in precipitation forecasts, a debris flow early warning model is constructed to dynamically judge the regional trend of debris flow and to conduct hazard analysis and risk avoidance decision-making based on the regional trend of debris flow. The results of debris flow hazard analysis and evacuation decision-making are visualized to obtain interactive information on hazard analysis and evacuation decision-making, and the debris flow early warning model is dynamically revised based on the interactive information.

[0125] The system includes: data acquisition, debris flow hazard identification, dynamic monitoring, early warning model, and early warning processing.

[0126] Specifically, multi-source data and historical data of the target area are obtained, information on debris flows that have occurred in the past and related precipitation data in the target area are filtered, and spatial cluster analysis is used to identify "hotspot" gullies or watersheds where debris flows frequently occur or occur in clusters.

[0127] Wide-area coverage is achieved by using L-band repeat orbit or single-flight interferometric data (such as ALOS PALSAR and TanDEM-L future data). For more refined data, polarimetric SAR (PolSAR) or fully polarimetric data is used locally. Decomposition techniques (such as Freeman-Durden and Yamaguchi) are used to distinguish between volume scattering (vegetation) and surface scattering (land surface). Phase optimization is then used to generate high-precision local data, and a hybrid elevation model combining high-precision local data with wide-area coverage is constructed.

[0128] Perform the InSAR processing workflow. Accurately register two SAR images, generate an interferogram, perform phase filtering (to reduce noise) and phase unwrapping, convert the unwrapped phases to elevation, and generate a geographic reference DEM.

[0129] Based on remote sensing data of debris flow areas, a three-dimensional model of the debris flow basin is established to obtain the topographic physical conditions, material conditions and historical information of the region's debris flow susceptibility. Machine learning algorithms (such as random forest, XGBoost) or information volume models are used for spatial overlay analysis in geographic information system (GIS).

[0130] Based on the characteristics and susceptibility of debris flow hazards in the region, regular multidimensional monitoring is conducted to obtain multidimensional and multi-period remote sensing data on crack expansion or displacement changes, as well as dynamic monitoring of surface conditions and soil moisture content. The surface deformation rate and time series are then retrieved to identify accelerated creep zones.

[0131] Based on multidimensional monitoring data, the static susceptibility and dynamic hazard of debris flows are analyzed. Combined with the probability of disaster in precipitation forecasts, a debris flow early warning model is constructed to dynamically judge the changing trend of debris flow areas and to conduct hazard analysis and evacuation decisions based on the changing trend.

[0132] The debris flow hazard analysis and avoidance decision-making are visualized, and interactive information on hazard analysis and avoidance decision-making is obtained. The debris flow hazard early warning model is dynamically revised based on the interactive information.

[0133] Based on the dynamic expression of the individual characteristics of debris flow hazard areas, the analysis of the uncertainty of debris flow disaster time is met, which improves the reliability and accuracy of debris flow disaster early warning and provides a data foundation for geological disaster prevention and mitigation.

[0134] It should be noted that by acquiring remote sensing images, screening physical conditions, and extracting 3D models and feature factors, the system can achieve refined and quantitative identification of potential debris flow hazard areas, thereby improving the early warning capability for geological disasters.

[0135] It should be noted that spatial data of the target watershed should be acquired from various remote sensing platforms (such as satellites, drones, aerial photography, etc.), including but not limited to optical imagery, radar imagery, thermal infrared data, and digital elevation models (DEMs). The data coverage should encompass the entire target watershed and its upstream catchment area, with priority given to the period before the rainy season or before and after a disaster to enhance the timeliness of hazard identification.

[0136] The system utilizes open-source remote sensing platform APIs (such as ESA Copernicus Open Access Hub for Sentinel-2, USGS EarthExplorer for Landsat, and the High Resolution Special Data Center) to automatically retrieve and download images based on the target watershed's geographic boundaries (GeoJSON / WKT format) and time windows. A two-level definition strategy is employed: users input points of interest (e.g., villages, roads) to infer upstream catchment areas; if historical debris flow channels are known, a buffer zone of 500m to 1000m centered on these channels is used as the analysis area. The system supports manual drawing or importing of hydrological boundaries. Based on 30m or higher resolution DEMs (such as ASTER GDEM and AW3D30), the D8 flow direction algorithm is used to calculate the cumulative runoff, with threshold values ​​(e.g., 10) set. 4 The system automatically extracts sub-basin boundaries (pixels); it can also overlay hydrological databases (such as HydroSHEDS) for correction; it employs affine transformations or polynomial corrections based on control points or ground features (such as road intersections and ridgelines) for spatiotemporal registration of multi-source remote sensing data, unifying data from different sensors to the same projection coordinate system, and aligning them temporally through nearest neighbor matching or interpolation (such as linear interpolation NDVI time series); the system has a built-in multi-source adapter pattern that can simultaneously schedule optical (Sentinel-2), radar (Sentinel-1), and thermal infrared (Landsat) data. Data such as TIRS are collected and spatiotemporally registered to generate a fused data cube (DataCube). The spatial resolution of the remote sensing data is ≤10m (preferably using 2m to 5m UAV or WorldView data). The remote sensing data temporal phase includes at least 3 phases (pre-disaster, during-disaster, post-disaster, or the same period for 3 consecutive years). Red and near-infrared bands are used for NDVI, and shortwave infrared bands are used for lithology identification. Data quality assessment is performed using Sentinel-2 cloud mask (SCL band) and Landsat QA band, combined with an adaptive threshold segmentation algorithm to calculate the effective pixel ratio. The signal-to-noise ratio is estimated by the local variance / mean ratio, and low-quality images (such as those with cloud cover >30%) are removed.

[0137] It should be noted that, based on the acquisition of raw remote sensing data, preprocessing such as radiometric correction, atmospheric correction, and geometric fine correction is performed; then, based on prior knowledge of geology, geomorphology, and hydrology, key conditions conducive to debris flow occurrence are screened out: physical conditions, material conditions, and historical factors.

[0138] Physical conditions include: slope > 15°, valley morphology (V-shaped / U-shaped), catchment area, etc. Based on the DEM, ArcGIS Spatial Analyst or GDAL library is used for calculation. The slope is calculated using the Horn algorithm. The topographic humidity index (TWI) = ln(cumulative catchment volume / tan(slope)) is used to identify areas prone to water accumulation. Profile / planar curvature reflects the degree of valley development.

[0139] Material conditions include the distribution of loose deposits, which can be determined through NDVI, bare soil index, lithological interpretation, etc.

[0140] The vegetation / bare soil index is calculated using the following standard index formula: NDVI = (NIR–Red) / (NIR + Red); BSI = [(SWIR+Red)–(NIR + Blue)] / [(SWIR+Red)+ (NIR+Blue)].

[0141] High values ​​indicate exposed loose material. A threshold (e.g., BSI > 0.2) is set to extract potential source areas. The threshold employs a dual mechanism of "regional adaptation + expert calibration," with the initial threshold set based on debris flow statistical patterns. In the target area, K-means clustering is used to perform unsupervised grouping of factors such as slope and BSI, automatically identifying high-risk cluster centers. The threshold is then fine-tuned based on feedback from local geological experts, forming a regional parameter database.

[0142] Historical factors include historical debris flow records, landslide locations, and post-disaster image change detection results. Historical debris flow and landslide records are returned through spatial buffer queries (e.g., a 5km radius), containing structured fields such as occurrence time, scale, and causative factors. A multi-source fusion strategy is employed, connecting to a geological hazard database. NLP techniques (such as the BERT-BiLSTM-CRF model) are used to extract "time-location-hazard" triples, which are then geocoded and stored in the database. An ontology-based debris flow hazard knowledge graph is constructed, containing three types of nodes: "topography-source-trigger" and causal relationships. The weights are determined by AHP (Analytic Hierarchy Process) combined with Delphi expert scoring and encapsulated into a configurable JSON rule file, supporting the switching of weight templates for different geomorphic areas; an improved CVA (Change Vector Analysis) is used to calculate the magnitude of the NDVI difference vector between two time phases, combined with Otsu's adaptive threshold segmentation of change areas; a conditional filtering logic or fuzzy comprehensive evaluation model is constructed: IF Slope > 25° AND BSI > 0.25 AND Landslide records in the past 3 years THEN Hazard level = High, supporting weight adjustment (determined by AHP method).

[0143] It should be noted that by fusing high-precision LiDAR point clouds with dense point clouds or DSMs (Digital Surface Models) generated by UAV photogrammetry, high-resolution DEMs (e.g., 0.5m to 2m) are generated through point cloud registration, denoising, and classification (ground / non-ground). Based on this, micro-topographic features (such as channel longitudinal profiles, cross-sectional morphology, confluence paths, and depressions) are extracted, and a 3D scene model with realistic surface texture and elevation information is constructed for subsequent hazard visualization and dynamic simulation input. An improved ICP (Iterative Closest Point) combined with FPFH (Fast Point Feature Histograms) feature matching is used. First, RANSAC coarse registration is used to eliminate large deviations, and then NDT (Normal Distributions Transform) is used to optimize local alignment, controlling the registration error within ±0.1m. CSF (Cloth Surface Model) is then used. The SimulationFilter algorithm simulates terrain covered by vegetation, efficiently separating ground and non-ground points. For densely vegetated areas, it supplements classification accuracy with multi-echo LiDAR intensity information. It enables high-precision fusion of LiDAR and photogrammetric data by introducing "control point constraint fusion." Ground control points (GCPs) are deployed in overlapping areas, or existing high-precision ground features are used to perform absolute orientation of both LiDAR and photogrammetric point clouds. During fusion, the LiDAR is used as the reference, and an affine transformation is applied to the photogrammetric point cloud to compensate for system bias. Finally, the fused point cloud passes the RMSE (Real-Time Validation) evaluation through cross-validation. The depth requirement is ≤0.3m; a DEM is generated through TIN (Triangulated Irregular Network) interpolation; a DSM is generated from the original point cloud; the two are subtracted to obtain a CHM (Canopy Height Model), used to remove vegetation interference; a web-based 3D scene is constructed based on CesiumJS, with the DEM serving as the elevation base map and the DSM combined with orthophotos (DOM) as the texture map, supporting real-time roaming, profile cutting, and visibility analysis, using a combined thresholding method of "cumulative flow + curvature"; cross-sectional analysis generates cross sections every 10m along the centerline of the gully, calculating the aspect ratio and concavity / convexity; the 3D model includes land cover attributes and supports standardized interfaces; It should be noted that multi-dimensional information is integrated to construct a comprehensive feature vector for debris flow hazards. An interannual NDVI / SWI (soil moisture index) series is constructed, and Theil-Sen slope estimation combined with the Mann-Kendall test is used to identify significant degradation trends (p<0.05), forming dynamic change characteristics of vegetation degradation and surface deformation monitored by InSAR. Continuous features (slope, BSI, etc.) are normalized to the [0,1] interval using Min-Max, while categorical features (lithology, land use) are encoded using One-Hot coding, forming static topographic features and material supply characteristics such as slope, channel density, catchment area, loose deposit area, and lithological vulnerability. The resulting "feature factor conditions" are used for debris flow hazard identification. An AI-assisted mode is used to input the feature vector into a pre-trained random forest / XGBoost model, outputting a probabilistic hazard score. Historical disaster weights are superimposed to ultimately form standardized features for debris flow hazards that can be used for hazard level classification or machine learning training.

[0144] In this embodiment, a differential monitoring and early warning system based on SAR technology and a 3D model is established. SAR technology is used to generate remote sensing data for debris flow hazard identification, significantly improving the early identification capability of debris flow hazards and enabling predictive analysis of debris flow hazard deformation. This solves the problem of low efficiency in traditional manual geological hazard survey methods for debris flow identification and analysis, making it difficult to identify the morphology and deformation signs of debris flow hazards, especially in large watersheds where investigators cannot assess the watershed conditions, hindering the identification, analysis, and monitoring of debris flow hazards. It also solves the problem of inaccurate identification and trend analysis of small debris flows, reducing survey difficulty and expenses. Furthermore, it addresses the issue of identifying the susceptibility of debris flow hazards based on changes in debris flow source and terrain, improving the precision of debris flow trend analysis and monitoring indicators. This system can be widely applied to monitoring and early warning in debris flow hazard areas where fixed monitoring equipment stations cannot be established.

[0145] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A method for early identification of debris flows based on SAR technology combined with a three-dimensional model, characterized in that, include: Obtain historical debris flow data and historical precipitation data for the target area; Interferometric SAR data was selected as the wide-area coverage data for the target area, and fully polarimetric SAR data was selected as the local fine-grained data for the target area. Phase unwrapping is performed based on the wide-area coverage data to generate a wide-area digital elevation model (DEM). Based on the local refined data, polarization decomposition and phase optimization are performed to generate a local DEM; A hybrid DEM is obtained by fusing wide-area DEM and local DEM; Based on the hybrid DEM, the topographic factors required to support debris flow susceptibility analysis are extracted from the DEM spatial data. Early identification of debris flows is performed based on the historical debris flow data, the historical precipitation data, the regional soil moisture content, and the DEM spatial data, and the identification results are obtained.

2. The method for early identification of debris flows based on SAR technology combined with a three-dimensional model according to claim 1, characterized in that, The method of early debris flow identification based on the historical debris flow data, the historical precipitation data, and the DEM spatial data yields identification results, including: Based on the DEM spatial data, topographic factors such as slope, aspect, confluence path, and longitudinal profile of the channel are extracted as favorable physical conditions for judging the formation and flow of debris flows. Using remote sensing images and image enhancement processing, we can identify linear and ring structures at different scales, focusing on areas with abundant loose deposits, well-developed joints and fissures, fractured rock masses, and ancient landslides or debris flow deposit fans, as material conditions for initiating debris flows under rainfall conditions. Remote sensing imagery is used to interpret key information on land cover changes, vegetation disturbance, gully morphology evolution, and potential source areas, serving as relevant conditions for determining the formation and flow of historical debris flows in the basin.

3. The method for early identification of debris flows based on SAR technology combined with a three-dimensional model according to claim 1, characterized in that, The method of early debris flow identification based on the historical debris flow data, the historical precipitation data, and the DEM spatial data yields identification results, including: A three-dimensional terrain model of the target area is constructed using DEM spatial data. Topographic factors such as slope, aspect, curvature, runoff accumulation, and channel density are extracted. Combined with multi-source geographic information, historical debris flow points, active fault zones, areas with high rainfall, steep terrain, and areas rich in loose sediment sources are spatially overlaid and comprehensively analyzed. A preliminary debris flow susceptibility assessment map is constructed, and areas are delineated as high-risk debris flow zones based on the debris flow susceptibility assessment map to generate a spatially continuous debris flow susceptibility probability map.

4. The method for early identification of debris flows based on SAR technology combined with a three-dimensional model according to claim 1, characterized in that, The method of early debris flow identification based on the historical debris flow data, the historical precipitation data, and the DEM spatial data yields identification results, including: A three-dimensional basic geographic base map is constructed. Point cloud data of the target area is acquired through LiDAR. Ground points and non-ground points are separated by filtering algorithms to generate bare land DEM data. Key terrain factors are extracted based on the bare land DEM data to identify potential loose debris source areas. Combining point cloud intensity, local elevation anomalies, debris source areas, accumulation fans, and spoil heaps, a three-dimensional spatial model of debris flow susceptibility is output.

5. The method for early identification of debris flows based on SAR technology combined with a three-dimensional model according to claim 4, characterized in that, The method of early debris flow identification based on the historical debris flow data, the historical precipitation data, and the DEM spatial data yields identification results, including: Acquire multi-source time-series SAR data; The multi-source time-series SAR data is subjected to interferometric pair generation, interferogram generation and filtering, coherence assessment and deformation parameter inversion. Sliding window analysis is performed on the deformation time series to calculate local acceleration. Based on the local acceleration and preset threshold, accelerated creep regions are marked, and spatially localized but temporally sudden uplift or subsidence is identified. Combined with the geological background, the inducing factors are determined, and the soil deformation field is obtained. Based on the physical relationship between the time series of SAR backscattering coefficients and soil dielectric constant, an empirical relationship for soil volumetric water content is established. Based on the empirical relationship for soil volumetric water content, the soil saturation state is determined, and the soil moisture field is obtained. By spatially aligning the soil deformation field and soil moisture field and performing physical correlation analysis, potential instability risk levels are classified.

6. The method for early identification of debris flows based on SAR technology combined with a three-dimensional model according to claim 5, characterized in that, The method of early debris flow identification based on the historical debris flow data, the historical precipitation data, and the DEM spatial data yields identification results, including: Alignment based on bare ground DEM data, soil deformation field, and soil moisture field; A spatial distribution map of debris flow probability was constructed based on aligned bare land DEM data, soil deformation field, and soil moisture field.

7. The method for early identification of debris flows based on SAR technology combined with a three-dimensional model according to claim 1, characterized in that, The step of performing phase unwrapping based on the wide-area coverage data to generate a wide-area digital elevation model (DEM) includes: Interferometric phase diagrams are generated by differential interferometry based on the wide-area coverage data; A phase filtering algorithm is applied to the interferometric phase map to suppress noise; Phase unwrapping is performed on the interferometric phase map to obtain continuous unwrapped phases; Based on the unwrapped phase and orbital parameters, a wide-area digital elevation model (DEM) is generated by inversion.

8. The method for early identification of debris flows based on SAR technology combined with a three-dimensional model according to claim 1, characterized in that, The step of generating a local DEM by performing polarization decomposition and phase optimization based on the local refined data includes: The total backscattering of the localized refined data is decomposed into surface scattering, dihedral scattering, and volume scattering; The localized refined data is decomposed by introducing a helical scattering term; The volume scattering-dominant region and the surface scattering-dominant region are quantitatively identified based on the decomposition results. By utilizing the phase consistency differences between volume scattering and surface scattering in different polarization channels, the interference phase can be optimized by weighting. Alternatively, a coherent polarization joint mask can be used to eliminate regions with low coherence and high volume scattering, while retaining high coherence surface scattering pixels. Estimate vegetation height and correct for ground phase; The local refined data, which has undergone polarization decomposition and phase optimization, is then unwrapped and its elevation is inverted again to generate a local DEM.

9. A debris flow early identification system based on SAR technology combined with a three-dimensional model, characterized in that, The system is constructed using the debris flow identification method based on SAR technology as described in any one of claims 1 to 8, and the system performs the following steps: Based on the stability assessment and changing trends of the debris flow area, conduct sensitivity analysis or probability assessment to determine the hazard level and plan emergency response plans and evacuation routes; By integrating new SAR data and weather forecasts, and utilizing GIS platforms, 3D terrain models, and dynamic charts, monitoring data is combined with geographic information to generate heat maps or risk zoning maps. Fuzzy comprehensive evaluation method is then used to classify the probability of debris flow occurrence in different areas. The monitoring data, model prediction results, and risk zoning maps are imported into the GIS platform to construct a 3D model of the debris flow area. The terrain data is overlaid on the GIS platform to simulate the movement path of the debris flow, calculate the threatened area, display the risk level distribution through a heat map, and overlay population density and infrastructure layers to construct a 3D visualization platform.

10. The debris flow early identification system based on SAR technology combined with a three-dimensional model according to claim 9, characterized in that, The system performs the following steps: Based on multidimensional monitoring data, the static susceptibility and dynamic hazard of debris flows are analyzed. Combined with the probability of disaster in precipitation forecasts, a debris flow early warning model is constructed to dynamically judge the changing trend of debris flow areas. Based on the changing trend of debris flow areas, hazard analysis and risk avoidance decisions are made. The results of debris flow hazard analysis and evacuation decision-making are visualized to obtain interactive information on hazard analysis and evacuation decision-making, and the debris flow early warning model is dynamically corrected based on the interactive information.

Images