Debris flow prediction method based on multi-source remote sensing

Abstract

The invention discloses a debris flow prediction method based on multi-source remote sensing, and the method comprises the following steps: 1, data acquisition and preprocessing: obtaining target region data through satellite remote sensing, and eliminating terrain shadow interference; step 2, feature extraction: extracting vegetation coverage and surface temperature from the optical image, extracting deformation magnitude and time sequence from InSAR data, and extracting gradient, slope direction, curvature and terrain humidity index from DEM terrain data; 3, performing multi-dimensional feature fusion, and constructing a multi-dimensional feature vector; 4, model construction: establishing a debris flow susceptible area identification model, and fusing the vegetation-terrain coupling characteristics of the optical image, the deformation dynamic characteristics of the InSAR and the terrain stability characteristics of the terrain data; 5, carrying out susceptibility partitioning and early warning, dividing risk grades based on probability values output by the model, and triggering graded early warning responses; according to the method, the multi-dimensional feature space is constructed through the collaborative interaction of the multi-source remote sensing data and the fused data.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring technology, and in particular to a debris flow prediction method based on multi-source remote sensing. Background Technology

[0002] Debris flow prediction is a science that analyzes geological, meteorological, and hydrological conditions, combined with monitoring technology and mathematical models, to predict the timing, location, scale, and hazards of debris flows. Meteorological departments assess debris flow risk through precipitation forecasts (such as cumulative rainfall in the next 24 hours and short-term heavy rainfall). For example, a daily rainfall exceeding 100 mm or an hourly rainfall threshold that triggers a debris flow may lead to a disaster. The bulk density of debris flows is a key parameter distinguishing them from flash floods. If the bulk density is between 1.2 and 2.3 tons per cubic meter, it is classified as a debris flow. By monitoring the volume and stability of loose deposits in the valley, combined with soil and water separation technology or retaining structures, the risk of material sliding is controlled. Debris flow prediction requires a comprehensive approach, integrating meteorological monitoring, source analysis, physical models, and engineering measures, and developing targeted solutions based on regional characteristics. Future research needs to break through the bottlenecks in mechanism research and technology, improve real-time early warning capabilities at small and medium scales, and minimize disaster losses. However, due to the complexity of debris flow phenomena, the current understanding of the debris flow formation process is still immature. To achieve the goal of more accurate prediction and forecasting of debris flow disasters and their losses, a series of scientific and technological problems need to be solved. Theoretically, research on the formation mechanism of debris flows should be further deepened, the critical conditions for the formation of different types of debris flows should be determined, and prediction and forecasting models based on the formation mechanism and initiation conditions of debris flows should be established to achieve a qualitative breakthrough in debris flow prediction and forecasting. Technically, modern mathematical methods and technologies should be fully utilized to quickly acquire and analyze information related to debris flow formation, construct a debris flow prediction and forecasting technology platform system, and realize the operationalization and standardization of prediction and forecasting. It is recommended that future work focus on the following aspects. Only prediction and forecasting models and methods based on the formation mechanism and conditions of debris flows can make scientific and relatively accurate predictions and forecasts of debris flows. Currently, research on the formation mechanism of debris flows is very weak, becoming a bottleneck restricting the development of prediction and forecasting. Strengthening theoretical research on formation is essential for breakthroughs. For single-gully debris flow prediction and forecasting, the main focus is on the debris flow initiation process, initiation conditions, and confluence mechanism. For regional debris flow prediction and forecasting, the main focus is on the coupling mechanism between the regional activity patterns of debris flows and the environmental background conditions controlling regional debris flow activity. Further research is needed to determine the meteorological and hydrological conditions for debris flow outbreaks in different regions and of different types of debris flow valleys.

[0003] Traditional debris flow monitoring and early warning methods mainly rely on ground station observations and manual inspections, which have the following key drawbacks: insufficient spatial coverage. The low density of ground monitoring stations makes it difficult to capture the deformation and vegetation change characteristics of the entire debris flow valley, resulting in a missed detection rate of up to 40% in high-risk areas. For example, early landslide creep or minute displacements in debris flow source areas are often not detected in time due to monitoring blind spots. Relying on manual patrols is not timely and is difficult to respond to sudden extreme weather events. Data dimension is singular and lagging: Existing technologies are mostly based on a single data source and lack comprehensive analysis of multi-dimensional parameters such as vegetation cover, surface deformation, and terrain stability. Early warning response is lagging: Threshold-triggered early warning systems are usually 6-12 hours behind the disaster, which is difficult to meet the needs of disaster prevention and mitigation. Technical limitations: Ground monitoring equipment is costly, difficult to deploy, and easily affected by harsh environments. Traditional remote sensing technology is limited by cloud and fog obstruction and cannot penetrate vegetation cover, resulting in insufficient detection capabilities for hidden disaster hazards. Therefore, this invention proposes a debris flow prediction method based on multi-source remote sensing to solve the problems existing in the prior art. Summary of the Invention

[0004] To address the aforementioned problems, the present invention aims to propose a debris flow prediction method based on multi-source remote sensing. This method integrates optical imagery, InSAR deformation sequences, and DEM topographic data through synergistic fusion of multi-source remote sensing data. Optical imagery captures vegetation degradation and thermal anomalies, InSAR millimeter-level deformation sequences dynamically track landslide creep, and DEM topographic parameters quantify instability risks. This constructs a multi-dimensional feature space covering the three key inducing factors of "vegetation-deformation-topography." By replacing ground stations with satellite remote sensing, traditional monitoring blind spots are eliminated, supporting large-scale, full-area deformation and vegetation dynamic monitoring of gullies.

[0005] To achieve the objectives of this invention, the invention is implemented through the following technical solution: a debris flow prediction method based on multi-source remote sensing, comprising the following steps:

[0006] Step 1: Data Acquisition and Preprocessing. Optical images, InSAR deformation sequences, and DEM topographic data of the target area are acquired through satellite remote sensing. Radiometric correction, geometric registration, and time synchronization are performed on the multi-source data to eliminate topographic shadow interference.

[0007] Step 2: Feature extraction. Extract vegetation cover and surface temperature from optical images, extract deformation level and time series from InSAR data, and extract slope, aspect, curvature and topographic humidity index from DEM topographic data.

[0008] Step 3: Multi-dimensional feature fusion to construct a multi-dimensional feature vector that includes vegetation-topography coupling features from optical imagery, deformation dynamic features from InSAR, and topographic stability features from DEM.

[0009] Step 4: Model building. Establish a debris flow-prone area identification model by integrating vegetation-topography coupling features from optical imagery, deformation dynamics features from InSAR, and topographic stability features from topographic data.

[0010] Step 5: Risk zoning and early warning. Based on the probability values ​​output by the model, risk levels are divided, and early warning thresholds for high-risk troughs are set to trigger tiered early warning responses.

[0011] Further improvements are made in the following aspects: the optical imagery in step one is used to extract land cover features with a spatial resolution of 30m; the InSAR deformation sequence is obtained by using a small baseline set algorithm to obtain millimeter-level deformation sequences; and the DEM topographic data is used to provide topographic elevation information.

[0012] A further improvement is made in the following step: the InSAR deformation sequence generation step in step one is as follows:

[0013] S1. Data acquisition: acquiring multiple images of the same area via satellite;

[0014] S2. Image registration: Register multiple images to the same master image to ensure phase consistency of corresponding points.

[0015] S3. Interferogram generation: The registered image is subjected to interferometric processing to generate an interferogram containing terrain and deformation information.

[0016] S4. Phase unwrapping: converting the interference phase into surface deformation.

[0017] S5. Time series inversion: Using time series InSAR technology and combining multiple interferograms, the time series of surface deformation is obtained by inversion.

[0018] A further improvement is that the geometric registration in step one is based on ground control points to achieve sub-pixel level registration of multi-source data.

[0019] The further improvement lies in the fact that the disaster-prone area identification model in step four is constructed based on the feature differences between historical disaster sites and background areas.

[0020] A further improvement is that the warning threshold in step five is determined based on statistical analysis of historical disaster data.

[0021] The beneficial effects of this invention are as follows: This invention integrates optical images, InSAR deformation sequences, and DEM topographic data through the synergistic fusion of multi-source remote sensing data. Optical images capture vegetation degradation and thermal anomalies, InSAR millimeter-level deformation sequences dynamically track landslide creep, and DEM topographic parameters quantify instability risks. This constructs a multi-dimensional feature space that covers the three key inducing factors of "vegetation-deformation-topography". By replacing ground stations with satellite remote sensing, traditional monitoring blind spots are eliminated, and large-scale valley deformation and vegetation dynamic monitoring are supported. Attached Figure Description

[0022] Figure 1 This is a flowchart of the steps of the present invention;

[0023] Figure 2 This is a flowchart illustrating the steps for generating the InSAR deformation sequence according to the present invention. Detailed Implementation

[0024] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0025] Debris flows are special types of floods formed by rainfall on gullies or mountain slopes, carrying large amounts of solid materials such as mud, sand, rocks, and boulders. Their water and sediment collection processes are highly complex, a product of the combined effects of various natural and anthropogenic factors. Debris flows are characterized by their suddenness, high velocity, large flow rate, large material capacity, and strong destructive power. They often destroy transportation facilities such as highways and railways, and even villages and towns, causing enormous losses. Debris flows are formed when torrential rains or floods saturate and dilute loose, sandy soil on mountain slopes. They have a large area, volume, and flow rate, while landslides are smaller areas of diluted soil. A typical debris flow consists of viscous mud containing suspended coarse solid debris and rich in silt and clay. Under suitable topographical conditions, a large amount of water permeates the solid deposits on the slopes or in the gully bed, reducing their stability. The water-saturated solid deposits then move under their own gravity, forming a debris flow. Debris flows are a catastrophic geological phenomenon. Debris flows typically occur suddenly and violently, carrying enormous boulders. Due to their high speed and immense energy, they are extremely destructive. The entire process of a debris flow usually lasts only a few hours, sometimes as short as a few minutes. It is a natural disaster widely distributed in regions with unique terrain and geomorphological conditions worldwide. It is a mixed flow of soil, water, and air, containing large amounts of mud, sand, and rocks, triggered by rainstorms, snowmelt, or other water sources, and falling between a sediment-laden water flow and a landslide. Debris flows often occur alongside mountain floods. The difference between debris flows and ordinary floods is that debris flows contain a sufficient amount of solid debris such as mud, sand, and rocks, with a volume content of at least 15% and up to about 80%, making them more destructive than ordinary floods. The main hazards of debris flows include the destruction of towns, businesses, factories, mines, and villages, causing casualties among people and livestock, damaging houses and other engineering facilities, and destroying crops, forests, and arable land. Furthermore, debris flows can sometimes block river channels, not only disrupting navigation but also potentially causing floods. Many factors influence the intensity of debris flows, such as debris flow capacity, velocity, and flow rate, with flow rate having the most significant impact on the severity of the disaster. Furthermore, various human activities exacerbate these factors in multiple ways, promoting debris flow formation. Debris flow prediction and forecasting are crucial, serving as an important step and measure in disaster prevention and mitigation. Currently, my country's debris flow prediction and forecasting research commonly employs the following methods: conducting fixed-point observation studies in typical debris flow gullies to address the formation and movement parameters of debris flows. Examples include observational experiments on debris flows in Jiangjiagou and Daqiaogou in the Xiaojiang River basin of Dongchuan City, Yunnan Province; and observational studies on the Shahe debris flow in Hanyuan County, Sichuan Province. Investigating relevant parameters and characteristics of potential debris flow gullies is also essential. Strengthening hydrological and meteorological forecasting, especially for localized heavy rainfall, is also vital, as heavy rainfall is a triggering factor for debris flows. For instance, when monthly rainfall exceeds 350 mm or daily rainfall exceeds 150 mm, a debris flow warning should be issued.Establish technical archives for debris flows, especially for large debris flow gullies, where detailed records should be kept of watershed elements, formation conditions, disaster conditions, and control measures. Address issues related to information reception and transmission. Delineate debris flow hazard zones and potential hazard zones, or implement debris flow hazard sensitivity zoning. Conduct research on debris flow disaster prevention alarms and indoor debris flow model testing.

[0026] Based on this, according to Figure 1 As shown in the figure, this embodiment provides a debris flow prediction method based on multi-source remote sensing, including the following steps:

[0027] Step 1: Data Acquisition and Preprocessing. Optical images, InSAR deformation sequences, and DEM topographic data of the target area are acquired through satellite remote sensing. Radiometric correction, geometric registration, and time synchronization are performed on the multi-source data to eliminate terrain shadow interference. Radiometric correction is used to eliminate sensor noise and atmospheric interference. Geometric registration is based on ground control points to achieve sub-pixel level registration of multi-source data. Time synchronization is used to construct a spatiotemporal cube model.

[0028] Optical imagery was used to extract land cover features with a spatial resolution of 30m. InSAR deformation sequences were obtained using a small baseline set algorithm to capture millimeter-level deformation sequences. DEM topographic data was used to provide topographic elevation information. The steps for generating the InSAR deformation sequence are as follows:

[0029] S1. Data acquisition: acquiring multiple images of the same area via satellite;

[0030] S2. Image registration: Register multiple images to the same master image to ensure phase consistency of corresponding points.

[0031] S3. Interferogram generation: The registered image is subjected to interferometric processing to generate an interferogram containing terrain and deformation information.

[0032] S4. Phase unwrapping: converting the interference phase into surface deformation.

[0033] S5. Time series inversion: Using time series InSAR technology and combining multiple interferograms, the time series of surface deformation is obtained by inversion.

[0034] By comparing the NDVI change rate over multiple periods (ΔNDVI / Δt), vegetation degradation hotspots are identified, rather than single-phase values. A vegetation-temperature composite index is constructed by combining LST (land surface temperature) to distinguish between drought stress and normal seasonal variations. To address the underestimation of NDVI caused by topographic shading, a DEM-assisted radiometric correction model is used to improve the accuracy of vegetation extraction in shaded areas.

[0035] InSAR deformation sequences are the core data support for geological disaster monitoring and early warning. Their application value is mainly reflected in the following aspects: early identification of hidden dangers: by analyzing the trend of InSAR deformation sequences, it is possible to identify hidden danger points that have not yet shown obvious signs of disaster; dynamic monitoring and early warning: the temporal continuity of InSAR deformation sequences can reflect the dynamic changes of surface deformation, providing real-time data support for disaster early warning; verification and supplementation of traditional data: InSAR deformation sequences can verify the accuracy of traditional ground monitoring data, supplement the deficiencies of point monitoring, and form a "point-area combined" monitoring system.

[0036] Step 2: Feature extraction. This involves extracting vegetation cover and surface temperature from optical images to identify vegetation degradation areas and thermal anomalies; extracting deformation magnitude and time series from InSAR data to extract deformation magnitude and trends; capturing landslide creep or minute displacements in debris flow source areas; and extracting slope, aspect, curvature, and topographic humidity index from DEM topographic data to quantify the driving effect of topography on debris flow formation. Finally, key features related to debris flows are extracted from multi-source data to construct a multi-dimensional feature space.

[0037] Step 3: Multi-dimensional feature fusion to construct a multi-dimensional feature vector containing vegetation-topography coupling features from optical imagery, deformation dynamics features from InSAR, and topographic stability features from DEM. The construction methods include: feature normalization preprocessing to standardize the extracted raw features and eliminate dimensional differences; spatiotemporal feature alignment based on geographic coordinate systems and a unified time base to align the spatial location and timestamps of multi-source data; feature selection and dimensionality reduction using mutual information to screen key features; multi-modal feature fusion to construct a three-dimensional feature tensor; and spatiotemporal correlation enhancement by introducing spatiotemporal correlation features. This is used to integrate heterogeneous features, increasing the model's ability to represent complex disaster causes. By constructing a multi-dimensional vector fusing vegetation-topography-deformation features, the limitations of a single data source are addressed. Fusion of Normalized Difference Vegetation Index (NDVI) with topographic parameters: NDVI is weighted and combined with topographic factors such as slope and curvature, such as NDVI × slope, reflecting the sensitivity of vegetation cover to topographic instability. Thermal anomaly and topographic correlation: Combining surface temperature and topographic humidity index (TWI) to identify the risk of water accumulation in vegetation-degraded areas. Spatiotemporal correlation modeling utilizes geographically weighted regression (GWR) to analyze the local correlation patterns of vegetation cover variation with slope. Feature interaction enhances the model's ability to express the collaborative disaster-causing mechanism of "vegetation-topography".

[0038] Step 4: Model Construction. A debris flow-prone area identification model is established, integrating vegetation-topography coupling features from optical imagery, deformation dynamics from InSAR, and topographic stability features from topographic data. Based on the feature differences between historical disaster sites and background areas, a machine learning method is employed, primarily a Random Forest (RF) classifier, to construct the debris flow-prone area identification model. Specific steps are as follows: Data Preparation: The extracted multi-dimensional feature vectors are divided into training and test sets in a 7:3 ratio, ensuring the training data contains at least 100 historical debris flow event samples; Model Selection: A Random Forest classifier is used, with 500 decision trees and a feature importance screening threshold of 0.05; Model Training: Hyperparameters are optimized through cross-validation (K=5), using the Gini index as the splitting criterion; Model Validation: The optimal classification threshold is determined using ROC curves, and the model is considered effective when the AUC value ≥ 0.85; Deployment Optimization: The trained model is deployed to a cloud-based inference platform, supporting real-time data access and second-level response. Stratified sampling (stratified by historical disaster density) ensures the training set contains at least 100 disaster samples, while the test set covers all landform types. Random Forest (RF) is suitable for processing high-dimensional remote sensing features due to its strong resistance to overfitting and high interpretability. Comparative experiments show that RF's AUC (0.92) is significantly higher than SVM (0.85) and ANN (0.88). By replacing grid search with Bayesian optimization, the training time was reduced from 24 hours to 3 hours, and the accuracy was improved by 2.3%. The disaster-prone area identification model in step four is constructed based on the feature differences between historical disaster points and background areas. The warning threshold is determined based on statistical analysis of historical disaster data.

[0039] Step 5: Risk Zoning and Early Warning. Risk levels are categorized based on the probability values ​​output by the model. Early warning thresholds are set for high-risk valleys, triggering tiered early warning responses. Risk grading is based on the statistical characteristics of historical disaster data and expert experience. Low risk (0.0-0.3): Extremely low probability of debris flow, relatively high regional stability; Medium risk (0.3-0.5): Some risk exists, requiring attention to minor anomalies in terrain or vegetation; Higher risk (0.5-0.7): Significantly increased risk, possibly indicating accelerated deformation or vegetation degradation; High risk (0.7-1.0): Extremely high probability of debris flow, requiring immediate emergency response. InSAR captures millimeter-level deformation trends, and NDVI identifies vegetation degradation areas. The combination of these two methods can distinguish between "slow-moving" and "sudden erosion" debris flows. Sub-pixel-level registration achieves precise alignment of multi-source data in the spatiotemporal dimension, eliminating data heterogeneity. The ratio of temporal InSAR deformation rate to NDVI change rate is used to quantify the inhibitory effect of vegetation cover on surface stability. Data acquisition frequency: Sentinel-1 (12 days / time), Sentinel-2 (5 days / time), InSAR (bi-weekly update). Cloud inference latency from data reception to alert issuance is less than 15 minutes.

[0040] This debris flow prediction method based on multi-source remote sensing first acquires optical images, InSAR deformation sequences, and DEM topographic data of the target area simultaneously through satellite remote sensing. Data errors are eliminated through radiometric correction, geometric registration, and time synchronization. Then, vegetation cover and surface temperature are extracted from the optical images, deformation magnitude and time series are extracted from the InSAR data, and slope, aspect, curvature, and topographic humidity index are extracted from the DEM. Next, a multi-dimensional feature vector is constructed that integrates vegetation-topography coupling characteristics, deformation dynamic characteristics, and topographic stability characteristics. Based on the feature differences between historical disaster points and background areas, a debris flow-prone area identification model is trained, and a high-risk warning threshold is set through statistical analysis. This achieves kilometer-level gridded early warning for high-risk gullies, significantly improving the spatial coverage and dynamic response efficiency of debris flow prediction.

[0041] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A debris flow prediction method based on multi-source remote sensing, characterized in that, Includes the following steps: Step 1: Data Acquisition and Preprocessing. Optical images, InSAR deformation sequences, and DEM topographic data of the target area are acquired through satellite remote sensing. Radiometric correction, geometric registration, and time synchronization are performed on the multi-source data to eliminate topographic shadow interference. Step 2: Feature extraction. Extract vegetation cover and surface temperature from optical images, extract deformation level and time series from InSAR data, and extract slope, aspect, curvature and topographic humidity index from DEM topographic data. Step 3: Multi-dimensional feature fusion to construct a multi-dimensional feature vector that includes vegetation-topography coupling features from optical imagery, deformation dynamic features from InSAR, and topographic stability features from DEM. Step 4: Model building. Establish a debris flow-prone area identification model by integrating vegetation-topography coupling features from optical imagery, deformation dynamics features from InSAR, and topographic stability features from topographic data. Step 5: Risk zoning and early warning. Based on the probability values ​​output by the model, risk levels are divided, and early warning thresholds for high-risk troughs are set to trigger tiered early warning responses.

2. The debris flow prediction method based on multi-source remote sensing according to claim 1, characterized in that: The optical imagery in step one is used to extract land cover features with a spatial resolution of 30m. The InSAR deformation sequence is obtained using the small baseline set algorithm to obtain millimeter-level deformation sequences. The DEM topographic data is used to provide topographic elevation information.

3. The debris flow prediction method based on multi-source remote sensing according to claim 2, characterized in that: The steps for generating the InSAR deformation sequence in step one are as follows: S1. Data acquisition: acquiring multiple images of the same area via satellite; S2. Image registration: Register multiple images to the same master image to ensure phase consistency of corresponding points. S3. Interferogram generation: The registered image is subjected to interferometric processing to generate an interferogram containing terrain and deformation information. S4. Phase unwrapping: converting the interference phase into surface deformation. S5. Time series inversion: Using time series InSAR technology and combining multiple interferograms, the time series of surface deformation is obtained by inversion.

4. The debris flow prediction method based on multi-source remote sensing according to claim 1, characterized in that: The geometric registration in step one is based on ground control points to achieve sub-pixel level registration of multi-source data.

5. The debris flow prediction method based on multi-source remote sensing according to claim 1, characterized in that: The disaster-prone area identification model in step four is constructed based on the characteristic differences between historical disaster sites and background areas.

6. The debris flow prediction method based on multi-source remote sensing according to claim 1, characterized in that: The warning threshold in step five is determined based on statistical analysis of historical disaster data.

Images