Method and system for monitoring and health evaluation of alpine wetland based on multi-source remote sensing

Abstract

The application relates to a high-cold wetland dynamic monitoring and health evaluation method and system based on multi-source remote sensing, which comprises the following steps: step 1, data acquisition and preprocessing; step 2, multi-modal data feature extraction; step 3, multi-modal data fusion; step 4, dynamic monitoring; and step 5, wetland health evaluation. The unique seasonal change and surface complexity of the high-cold wetland optimize a SAR polarization decomposition model and a calculation method of an optical wetland index, provide a scientific basis for wetland protection area management, improve ecological restoration efficiency, support high-cold region water resource management and ecological protection policy making, and provide data support for research on wetland carbon sink function and climate change influence.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing monitoring and ecological assessment technology, specifically involving a method for conducting all-time and all-space dynamic monitoring and ecological health assessment of alpine wetlands by combining multimodal SAR data and multispectral remote sensing data and applying deep learning algorithms. Background Technology

[0002] Alpine wetlands play an irreplaceable role in water conservation, carbon sequestration, and climate regulation. However, global climate change and intensified human activities have led to increasingly serious problems such as degradation, shrinkage, and biodiversity loss in alpine wetlands.

[0003] Existing technologies typically utilize single optical or SAR data for wetland monitoring, which has the following main drawbacks:

[0004] (1) Data uniformity

[0005] Optical remote sensing data is easily limited by cloud cover and sunlight conditions, making it difficult to guarantee multi-temporal and full-coverage acquisition; although SAR data can penetrate cloud layers, it has limitations in extracting vegetation details and wetland type differences.

[0006] (2) Insufficient dynamic monitoring

[0007] Most studies focus on static monitoring of short-term wetland conditions, lacking a comprehensive evaluation of the hydrological and ecological evolution of wetlands on multi-year and seasonal scales; the monitoring indicators are singular and cannot fully reflect the dynamic changes of multiple elements such as wetland water, vegetation and soil moisture.

[0008] (3) The health assessment model is too simple.

[0009] Traditional wetland health assessments often rely on a limited number of indicators or simple single indices, making it difficult to fully utilize the advantages of SAR and optical multi-source data in comprehensive wetland ecological analysis; the accuracy and applicability of the models are also insufficient to meet the special environmental needs of alpine wetlands.

[0010] To this end, this invention deeply integrates SAR and optical remote sensing data, combines spatiotemporal convolutional neural networks (ST-CNN) and attention mechanisms to conduct long-term, dynamic monitoring of hydrology, vegetation and soil moisture in alpine wetlands, and adopts a multi-index ecological health assessment system (AHP) to quantify the health status of wetlands, so as to achieve refined management and protection of alpine wetlands. Summary of the Invention

[0011] This invention aims to overcome the shortcomings of single-source remote sensing data in monitoring alpine wetlands, including incomplete dynamic monitoring and health assessment. It provides a method and system for dynamic monitoring and health assessment of alpine wetlands based on multi-source remote sensing. By integrating the complementary advantages of SAR and optical data, and combining deep learning multimodal feature extraction and the AHP health assessment model, it achieves comprehensive monitoring and health classification of soil moisture, hydrological range, and vegetation cover in alpine wetlands at different seasonal and annual scales. This provides a scientific basis for decision-making regarding the ecological protection and restoration of wetlands in alpine regions.

[0012] To solve the aforementioned technical problems, the present invention adopts the following solution:

[0013] (I) Methodology

[0014] A method for dynamic monitoring and health assessment of alpine wetlands based on SAR and optical remote sensing includes the following steps:

[0015] Step 1: Data Acquisition and Preprocessing;

[0016] Acquire SAR image data, optical image data, auxiliary data, etc., and perform preprocessing on each.

[0017] Step 2: Multimodal feature extraction;

[0018] Optical features are extracted using SAR features.

[0019] Step 3: Multimodal data fusion;

[0020] By employing a spatiotemporal convolutional neural network (ST-CNN), inputting SAR and optical multi-temporal data, and combining an attention mechanism, the sensitivity to wetland features is improved.

[0021] Step 4: Dynamic monitoring;

[0022] By using the fused feature data, wetlands are classified into permanent wetlands, seasonal wetlands, peatlands, and degraded wetlands based on random forest (RF) or support vector machine (SVM) classification algorithms. The dynamic change trends of wetlands, wetland area, vegetation cover, and water body expansion and contraction are analyzed through time series data.

[0023] Step 5: Wetland health assessment;

[0024] Constructing health indicators: Water Health Index (WHI): based on changes in water area and water connectivity; Vegetation Health Index (VHI): combining NDVI and surface temperature changes to quantify the health status of vegetation growth; Soil Moisture Index (SMI): estimating wetland soil moisture status based on SAR texture features.

[0025] Comprehensive evaluation model: Construct a health evaluation model based on weight allocation (AHP analytic hierarchy process) to integrate various indicators into a comprehensive score for wetland ecological health.

[0026] Furthermore, in step 1, the input data source is:

[0027] SAR imagery: VV / VH polarization data from the European Space Agency's (ESA) Sentinel-1 satellite or JAXA ALOS PALSAR data. Sentinel-1 has a repeat period of 6 days and a spatial resolution of approximately 5 meters; ALOS PALSAR has a spatial resolution of approximately 10 meters and a repeat period of approximately 14 days. The focus is on acquiring the scattering characteristics of wetland water and soil moisture, and analyzing wetland hydrological dynamics using multi-polarization and multi-temporal characteristics.

[0028] Optical imagery: Using 10-meter resolution multispectral imagery (repeat every 5 days) from the European Space Agency (ESA) Sentinel-2 satellite or 30-meter resolution imagery (repeat every 16 days) from the U.S. Geological Survey (USGS) Landsat series, information on wetland vegetation cover, surface water distribution, and surface temperature is obtained to support the calculation of vegetation and water indices.

[0029] Supporting data: 30-meter resolution DEM data from the Shuttle Radar Topography Mission (SRTM) or ASTGTM V2 (Advanced Spaceborne Thermal Emission and ReflectionRadiometer Global Digital Elevation Model) are used to extract wetland topographic features, such as low-lying area identification and water catchment point analysis. ERA5 reanalysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF) or regional meteorological data from the China Meteorological Administration (ERA5 provides hourly global grid resolution data at approximately 31 km) are used to analyze the impact of meteorological factors such as precipitation and temperature on wetland hydrological dynamics, supporting the calibration and validation of wetland hydrological models. The Global Land Cover by National Mapping Organizations (GLCNMO) dataset from the U.S. Geological Survey (USGS) or LULC data from the National Forestry and Grassland Administration are used for wetland type classification, land use change analysis, and to assist in wetland classification and health assessment. The SSURGO (Soil Survey Geographic) database from the U.S. Department of Agriculture (USDA) or the China Soil Database are used to calibrate wetland soil moisture inversion models and improve the accuracy of soil moisture estimation.

[0030] The data preprocessing process in step 1:

[0031] 1) Steps for SAR data preprocessing:

[0032] S11. Speckle noise filtering: Lee filters or IDAN filters are used to remove speckle noise in SAR images, thereby improving the signal-to-noise ratio of subsequent feature extraction.

[0033] S12. Geometric and Radiometric Correction: Geometric correction is performed based on accurate orbital data to ensure spatial consistency between SAR and optical images; radiometric correction is used to correct the amplitude values ​​of SAR images and unify data acquired at different times or from different sensors.

[0034] S13. Polarization decomposition: The Freeman-Durden decomposition method is used to decompose dual-polarization (VV / VH) SAR data into surface scattering, vegetation scattering and volume scattering components to extract the surface scattering characteristics of wetlands.

[0035] 2) Optical data preprocessing method steps:

[0036] S21. Cloud and cloud shadow removal: The Fmask algorithm is used to identify and remove cloud and cloud shadow areas in optical images to ensure accurate extraction of vegetation and water features.

[0037] S22. Atmospheric Correction: Atmospheric correction is performed using the 6S (Second Simulation of the Satellite Signal in the Solar Spectrum) or LEDAPS (Landsat Ecosystem Disturbance Adaptive Processing System) methods to obtain surface reflectance;

[0038] S23. Time series interpolation completion: Based on spatiotemporal interpolation algorithms such as STARFM (Spatial and Temporal Adaptive Reflectance Fusion Model), missing data in multi-temporal optical images caused by cloud cover or other factors are completed.

[0039] 3) Data alignment method

[0040] Pixel-level registration algorithms, such as registration methods based on cross-correlation or registration methods based on feature point matching (such as SIFT, SURF, etc.), are used to accurately align SAR images and optical images in the same coordinate system, ensuring the spatial consistency of multi-source data and providing a foundation for subsequent data fusion processing.

[0041] Furthermore, the multimodal features in step 2 mainly include the following features:

[0042] 1) SAR Feature Extraction:

[0043] Scattering characteristics: VV (vertical transmission, vertical reception) and VH (vertical transmission, horizontal reception) polarization information from Sentinel-1 or ALOS PALSAR data are utilized. These two polarization methods are chosen because they exhibit high sensitivity and discriminative power for different land cover types in cold, wetland environments. VV polarized waves, due to their vertical propagation characteristics, are particularly sensitive to specular reflection from water surfaces, effectively distinguishing highly scattering water areas. Furthermore, VH polarized waves, due to their horizontal reception characteristics, can better capture the volumetric and multidirectional scattering effects of vegetation, thus distinguishing moderately scattering vegetation areas. By calculating the VV / VH ratio and the Polarization Difference Index (PDI), the contrast between water and vegetation, and between vegetation and bare land, is further enhanced, improving classification accuracy. In addition, by combining additional scattering characteristics such as the dual polarization ratio and the Polarization Difference Index, this invention can more comprehensively describe different land cover types in wetlands, ensuring high-precision classification and monitoring in complex, cold environments.

[0044] Texture Features: The Gray-Level Co-occurrence Matrix (GLCM) method is used to quantify the roughness and moisture state of wetland soil from SAR imagery. GLCM can capture the spatial distribution characteristics of surface microstructures and assess the texture complexity and moisture variations of the soil surface by calculating statistical indicators such as contrast, homogeneity, energy, and correlation. Alpine wetlands often exhibit freeze-thaw cycles and seasonal moisture changes, and texture features can effectively reflect these dynamic processes. For example, an increase in soil roughness is usually associated with reduced vegetation cover or human disturbance, while changes in moisture state directly affect the reflective properties of soil texture. By combining multi-scale texture features (such as Scale-Invariant Feature Transform (SIFT) and Local Binary Pattern (LBP), this invention can capture detailed changes in the wetland surface at different spatial scales, improving the detail and accuracy of wetland type differentiation, especially showing significant effects under complex surface conditions in alpine environments.

[0045] 2) Optical feature extraction:

[0046] NDVI (Normalized Difference Vegetation Index): Calculated by the difference in reflectance between the near-infrared (NIR) and red (Red) wavelengths, the formula is: NDVI = (NIR - Red) / (NIR + Red). It is used to quantify vegetation density and health status. In alpine wetland environments, vegetation cover has a crucial impact on ecological health. NDVI effectively reflects vegetation density and growth status, making it suitable for capturing seasonal growth changes and assessing the health of vegetation. Due to the short growing season of alpine wetland vegetation, NDVI can provide highly sensitive vegetation monitoring data in a short time, supporting timely assessment of vegetation dynamics.

[0047] MNDWI (Modified Normalized Water Index): Calculated using reflectance in the green and shortwave infrared (SWIR) bands, the formula is: MNDWI = (Green - SWIR) / (Green + SWIR). It enhances the identification of water body boundaries. MNDWI is highly sensitive to water body boundary identification and is suitable for complex wetland environments, especially in areas with interwoven vegetation and water bodies. Compared to traditional water indices, MNDWI reduces interference from buildings and shadows, improving the accuracy and stability of water body extraction. Furthermore, when used in conjunction with NDVI, MNDWI provides more comprehensive vegetation and water body information, enhancing the accuracy of wetland classification and monitoring, and supporting a comprehensive assessment of wetland hydrological dynamics.

[0048] LST (Land Surface Temperature): This estimates the temperature distribution of the land surface using remote sensing data in the thermal infrared band. It typically employs single-band or multi-band radiation temperature calculation methods to reflect the thermal state of the land surface. In alpine wetland environments, land surface temperature is closely related to vegetation health and the thermal properties of water bodies. LST can reflect vegetation transpiration and the heat capacity of water bodies, helping to identify areas of vegetation stress and monitor dynamic changes in the land surface thermal state. Alpine wetlands are significantly affected by air temperature changes; LST provides real-time monitoring of the wetland surface thermal environment, supporting dynamic response assessments to environmental changes.

[0049] 3) Dynamic characteristics:

[0050] A time-series dataset of wetlands is constructed using Sentinel-1 (SAR) and Sentinel-2 / Landsat (optical) imagery acquired over multiple time periods (monthly or quarterly). Data for each time point includes key parameters such as scattering characteristics (VV / VH), spectral indices (NDVI, MNDWI), and land surface temperature (LST). By analyzing seasonal variations, peak growth and decline processes of wetland vegetation, as well as seasonal fluctuations in water area, are identified. For example, time-series NDVI data is used to detect seasonal growth peaks and winter degradation trends in vegetation; MNDWI time series data are used to monitor the expansion and contraction of water bodies during the rainy and dry seasons. By combining multi-year (e.g., 5-year or 10-year) time-series data, long-term trends in wetland hydrological dynamics and vegetation cover are assessed. Statistical analysis methods (such as linear regression and moving average) are applied to detect trends of increase or decrease in wetland water area, as well as long-term upward or downward trends in vegetation indices. Furthermore, anomaly detection algorithms (such as Z-score and anomaly detection models) in time series data are used to identify sudden change events in wetland ecosystems, such as abnormal droughts, floods, or vegetation mutations caused by human disturbance.

[0051] Furthermore, step 3 employs a spatio-temporal convolutional neural network (ST-CNN) combined with an attention mechanism to perform deep fusion processing of SAR and optical multi-temporal data.

[0052] 1) Spatiotemporal Convolutional Neural Network (ST-CNN) Architecture Design:

[0053] ST-CNN aims to capture feature variations in both spatial and temporal dimensions, adapting to the complex spatiotemporal dynamics of alpine wetlands. Its main architecture includes the following components:

[0054] Input layer:

[0055] SAR feature inputs include scattering features (VV, VH), texture features (GLCM index), and dynamic features (time series variation curves).

[0056] Optical feature inputs include spectral indices (NDVI, MNDWI, LST) and time-series variation features (such as EVI, SAVI, GNDVI).

[0057] Convolutional Layers:

[0058] Spatial convolutional layer: Two-dimensional convolutional kernels (such as 3×3 or 5×5) are used to extract the spatial features of the input data and capture the spatial distribution and structural information of ground features.

[0059] Temporal convolutional layer: Employs a one-dimensional convolutional kernel (such as 1×3 or 1×5) to process time series features along the time dimension, capturing the temporal dependence of wetland dynamic changes.

[0060] Pooling Layers:

[0061] Spatial pooling: Using max pooling or average pooling operations to reduce the spatial dimension of the feature map and extract high-level spatial features.

[0062] Temporal pooling: Employs time-dimensional pooling operations to compress time-series features while retaining key dynamic information.

[0063] Attention Mechanism Layers:

[0064] Channel Attention: Adaptively adjusts the importance weights of different feature channels to highlight key features and suppress irrelevant or redundant features. It uses global average pooling and global max pooling to generate channel weight vectors, and then calculates the attention weight for each channel using a fully connected layer and an activation function (such as sigmoid).

[0065] Spatial Attention: By adjusting the feature importance of different spatial regions, it highlights the feature responses of the core wetland area and suppresses interference from background or noise regions. It combines spatial information from different feature maps to generate a spatial weight map, and calculates the attention weight for each spatial location through convolution operations.

[0066] Fusion Layers:

[0067] Feature concatenation: The SAR processed by channel and spatial attention mechanisms is concatenated with the optical feature map to form a comprehensive feature representation.

[0068] Weighted fusion: By using learned weights, feature maps from different sources are summed in a weighted manner to fuse information from multiple sources.

[0069] Fully Connected Layers:

[0070] Function: Maps the fused high-dimensional feature vectors to the output space of classification or regression tasks to achieve wetland type classification and prediction of dynamic change trends.

[0071] Activation function: The Softmax activation function is used to output classification probabilities for multi-class classification of wetland types.

[0072] Output Layers:

[0073] Wetland classification results: Output the classification probability map of various wetland types (permanent wetlands, seasonal wetlands, peatlands, degraded wetlands).

[0074] Dynamic trend chart: Outputs a dynamic trend prediction chart of key indicators such as wetland water area, vegetation coverage and soil moisture.

[0075] 2) Integration methods for attention mechanisms

[0076] The attention mechanism plays a crucial role in this invention, enhancing the model's sensitivity to key features of alpine wetlands by adaptively adjusting the importance weights of features. The specific integration method is as follows:

[0077] Channel Attention Implementation Process:

[0078] Global average pooling and global max pooling are performed on the input feature map to generate two different context description vectors.

[0079] These two vectors are input into a fully connected layer, and after passing through the ReLU activation function, the channel weight vectors are generated by passing through the Sigmoid activation function.

[0080] The channel weight vector is multiplied with the original feature map channel by channel to adjust the feature response intensity of each channel.

[0081] The implementation process of spatial attention:

[0082] The input feature map is subjected to average pooling and max pooling along the channel dimension to generate two single-channel feature maps.

[0083] These two single-channel feature maps are concatenated along the channel dimension to form a dual-channel feature map.

[0084] A spatial weight map is generated using a convolutional layer (such as a 7x7 convolutional kernel) and a sigmoid activation function.

[0085] The spatial weight map is multiplied pixel by pixel with the original feature map to adjust the feature response intensity at each spatial location.

[0086] 3) Data fusion process

[0087] Feature extraction: Scattering features, texture features, dynamic features, and spectral indices are extracted from SAR and optical data, respectively.

[0088] Feature preprocessing: The extracted features are standardized to ensure the consistency of different features in numerical range and improve the fusion effect.

[0089] Feature input: The processed SAR features and optical features are input into different branches of the ST-CNN network, respectively.

[0090] Feature fusion: Within ST-CNN, spatial and temporal features are extracted through convolutional and pooling layers. Channel attention and spatial attention mechanisms are integrated to optimize feature weight allocation and highlight key features. SAR and optical features are deeply fused through feature concatenation or weighted fusion layers to form a comprehensive feature representation.

[0091] Classification and Prediction: Through fully connected layers and the Softmax activation function, we achieve multi-class classification of wetland types and prediction of dynamic change trends.

[0092] 4) Training and Validation Process

[0093] Model training and validation are crucial steps to ensure the effectiveness of data fusion and model performance. The specific process is as follows:

[0094] Sample labeling: Wetland type labels are obtained through field surveys or high-resolution image data to form training and validation sets. This ensures that the samples cover different wetland types and different time points, enhancing the model's generalization ability.

[0095] Loss function: Cross-entropy loss is used to optimize the accuracy of classification tasks. Temporal consistency regularization is introduced to ensure that the model's predictions are smooth and consistent over time, reducing abrupt changes and noise interference.

[0096] Algorithm optimization: The Adam optimizer is used with a learning rate of 0.001, combined with momentum and adaptive learning rate adjustment strategies to accelerate model convergence. The number of training epochs and batch size are dynamically adjusted based on the validation set accuracy to avoid overfitting.

[0097] Model Validation: Model performance is evaluated using an independent validation set, with metrics such as accuracy, precision, recall, and F1 score used to measure classification effectiveness. Cross-validation is used to further validate the model's robustness and generalization ability.

[0098] Further, in step 4, after completing the multimodal data fusion, the dynamic monitoring step uses classification algorithms and time series analysis to accurately classify the types of alpine wetlands and monitor their dynamic changing trends. The specific implementation of this step includes wetland type classification, calculation and analysis of key monitoring indicators, and output and application of monitoring results, as detailed below:

[0099] 1) Classification of wetland types

[0100] Alpine wetlands are classified into different types based on their hydrological characteristics and ecological status, including permanent wetlands, seasonal wetlands, peatlands, and degraded wetlands, in order to carry out targeted ecological management and protection.

[0101] Classification algorithm selection:

[0102] Random Forest (RF): It leverages its advantages in handling high-dimensional features and resisting overfitting to perform multi-class classification of wetland types.

[0103] Support Vector Machine (SVM): Suitable for classification of data in high-dimensional space, especially performing well when the sample size is small and the class distribution is imbalanced.

[0104] Classification characteristics:

[0105] Fusion features: Based on the comprehensive features after fusion in step three, including SAR scattering features, optical spectral index, texture features and dynamic features.

[0106] Time series features: By combining information on the changes in wetland features at different time points, the spatiotemporal accuracy of classification can be improved.

[0107] Classification criteria and judgment basis:

[0108] Permanent wetlands are defined as wetlands where water cover time and humidity levels remain stable for most of the three consecutive years (e.g., more than 9 months of the year), with seasonal water level fluctuations of less than 5%. The criteria for determination are a water cover time of more than 75% of the year and a stable trend in the soil moisture index (SMI).

[0109] Seasonal wetlands: defined as wetlands where water levels or water areas fluctuate significantly throughout the year, with seasonal extreme differences in water area exceeding 30%.

[0110] Peatlands: By combining SAR scattering characteristics (low scattering coefficient, usually less than 0.4), optical vegetation index (low NDVI value, usually <0.3) and soil type data, wetland types with high organic matter content are identified, exhibiting special polarization response and texture characteristics.

[0111] Degraded wetlands are identified as such based on multiple indicators (such as an annual average NDVI value below 0.3 and an interannual decline rate exceeding 5%), combined with a continuous downward trend in the soil moisture index (SMI) or a significant increase in anthropogenic disturbance indices (such as land use change and grazing intensity).

[0112] 2) Calculation and analysis of key monitoring indicators

[0113] Monitoring indicators:

[0114] Wetland area: Based on the classification results, the area of ​​different wetland types is calculated using Geographic Information System (GIS) tools based on remote sensing imagery, and an annual wetland area change curve is generated.

[0115] Vegetation cover: Using vegetation indices such as NDVI, EVI, and SAVI, calculate the vegetation cover area and vegetation health change trends, and generate a time series curve of vegetation cover.

[0116] Water body expansion and contraction: Using MNDWI time series data, analyze the annual increase and decrease of water body area to generate dynamic curves of water body expansion and contraction.

[0117] Soil moisture: Using the soil moisture index (SMI) extracted by SAR and combined with meteorological data, the annual average soil moisture and the moisture change trend during key periods were calculated.

[0118] 3) Output and application of monitoring results

[0119] Output content:

[0120] Annual Wetland Change Map: Shows the spatial distribution and area changes of different wetland types across different years. It visually illustrates the expansion or contraction trends of wetland area, helping to identify ecological hotspots and degradation risk zones.

[0121] Wetland change curves: These include time-series curves of indicators such as wetland area, vegetation cover, water body expansion and contraction, and soil moisture. Analyzing the long-term trends and seasonal fluctuations of wetland dynamics supports continuous assessment and trend prediction of wetland health status.

[0122] Health Level Zoning Map: Based on comprehensive health scores, wetlands are divided into four health levels: excellent, good, medium, and poor, and the areas of each level are marked on the map. This provides an intuitive spatial distribution reference for the delineation of regional wetland protection areas and the priority restoration of degraded areas.

[0123] Further, in step 5, after completing dynamic monitoring, the wetland health assessment step comprehensively evaluates the ecological health status of alpine wetlands by constructing and calculating multiple ecological health indicators. This step includes the calculation of health indicators, comprehensive evaluation using the Analytic Hierarchy Process (AHP), and the formulation of health zoning and restoration recommendations, as detailed below:

[0124] 1) Calculation of health indicators

[0125] By quantifying key ecological elements such as water volume, vegetation, and soil moisture, the ecological health status of wetlands can be comprehensively assessed.

[0126] Health indicators:

[0127] The Water Health Index (WHI) reflects the ecological health of wetland water bodies by comprehensively analyzing the rate of change in water area and water connectivity. It calculates the annual growth rate or decline rate of water area using time-series data obtained from dynamic monitoring. Based on a digital elevation model (DEM) and water distribution data, it analyzes the density and connectivity of the water network. The WHI score is obtained by weighted summation of the rate of change in water area and connectivity indicators.

[0128] Vegetation Health Index (VHI): The VHI quantifies the health status of wetland vegetation by combining the rate of change in vegetation cover, peak vegetation growth, and surface temperature. It uses NDVI time-series data to calculate the annual growth or decline rate of vegetation cover. It identifies the NDVI peak during the growing season, reflecting the intensity of vegetation growth. Combined with LST data, it assesses the impact of high temperatures on vegetation health, and calculates the VHI score using a weighted formula.

[0129] Soil Moisture Index (SMI): SMI reflects the health status of wetland soil moisture by retrieving soil moisture conditions. A soil moisture retrieval model is established using SAR texture features (such as roughness indices extracted by GLCM) and meteorological data to calculate the annual average soil moisture or key period values. Multi-temporal SMI data is analyzed to assess long-term trends in soil moisture (such as the annual average rate of change). The current state of soil moisture is compared with historical data, and an SMI score is obtained through standardization.

[0130] 2) Analytic Hierarchy Process (AHP) Comprehensive Evaluation

[0131] The Analytic Hierarchy Process (AHP) is used to assign weights to multiple health indicators and conduct a comprehensive evaluation to generate a comprehensive score for wetland ecological health.

[0132] Implementation method:

[0133] Hierarchical structure construction: The comprehensive score of wetland ecological health is used as the target layer, the water health index (WHI), vegetation health index (VHI), and soil moisture index (SMI) are used as the criteria layers, and the specific sub-indicators under each criteria layer (such as the rate of change of water area, the growth rate of vegetation coverage, and the average soil moisture) are used as the indicator layers.

[0134] Judgment Matrix Construction: Judgment matrices for the criterion and indicator layers are constructed using expert scoring or comparisons based on historical data, reflecting the importance relationships between the indicators. Weights for each layer are calculated to ensure the consistency of the judgment matrices (CI < 0.1).

[0135] Weighting: Weights calculated using the analytic hierarchy process (AHP) are assigned to each health indicator to reflect their relative importance in the overall evaluation. For example: WHI: 0.4; VHI: 0.4; SMI: 0.2.

[0136] Overall score calculation: The scores of each health indicator are weighted and summed according to their respective weights to obtain the overall score of wetland ecological health. For example, the calculation formula is: Overall score = (WHI × 0.4) + (VHI × 0.4) + (SMI × 0.2).

[0137] 3) Health Zone and Repair Suggestions

[0138] Based on the overall score, wetlands are classified into different health levels, and corresponding ecological restoration and management recommendations are proposed for areas of different levels.

[0139] Implementation method:

[0140] Score range division: Excellent: Overall score > 0.8; Good: 0.6 ≤ Overall score ≤ 0.8; Average: 0.4 ≤ Overall score < 0.6; Poor: Overall score < 0.4.

[0141] Health level zoning map generation: Based on the comprehensive score, the wetland space is divided into four health level areas: excellent, good, medium and poor, and a health zoning map is generated.

[0142] (II) System Part

[0143] A dynamic monitoring and health assessment system for alpine wetlands based on SAR and optical remote sensing includes the following modules:

[0144] (1) Data acquisition and preprocessing module

[0145] It is responsible for acquiring SAR images, optical images, and auxiliary data (DEM, meteorological, LULC, soil type, etc.), and performing preprocessing steps such as denoising, geometric correction, radiometric correction, cloud removal, atmospheric correction, time series interpolation and registration to ensure the spatial and temporal consistency of multi-source data.

[0146] (2) Multimodal feature extraction module

[0147] SAR Feature Extraction:

[0148] Scattering features, texture features, and dynamic features are extracted from the preprocessed SAR images, and soil moisture and roughness are quantified using polarization decomposition and GLCM methods.

[0149] Optical feature extraction:

[0150] Spectral indices such as NDVI, MNDWI, and LST, as well as monthly or quarterly vegetation growth dynamics, are extracted from the preprocessed optical images.

[0151] The output is a multimodal feature grid or feature vector that satisfies the requirements of subsequent deep learning fusion.

[0152] (3) Multimodal data fusion module

[0153] A spatiotemporal convolutional neural network (ST-CNN) combined with an attention mechanism is used to perform deep fusion of SAR and optical multi-temporal data;

[0154] Network branch design:

[0155] Branch 1 (SAR subnet): Processes SAR scattering and texture features, and extracts hydrological and soil moisture features of wetlands;

[0156] Branch 2 (Optical Subnet): Processes optical vegetation and water features, extracts vegetation cover and water distribution features of wetlands;

[0157] Attention mechanism:

[0158] Channel Attention: Assigns adaptive weights to features of different polarizations / bands to highlight important features;

[0159] Spatial Attention: Enhances the feature response of the core wetland area and suppresses interference from irrelevant areas;

[0160] Fusion and Output: By using feature splicing or weighted fusion layers, the features of branch one and branch two are fused to output wetland classification results, dynamic change trend maps, and key feature distribution maps.

[0161] (4) Wetland dynamic monitoring module

[0162] Using the fused features, wetlands are classified into permanent wetlands, seasonal wetlands, peatlands, and degraded wetlands through classification algorithms such as Random Forest (RF) or Support Vector Machine (SVM).

[0163] By combining time-series imagery, we can analyze the dynamic changes in wetlands, including key indicators such as wetland area, vegetation cover, and water body expansion and contraction.

[0164] Output monitoring results such as annual wetland change maps and area change curves.

[0165] (5) Wetland health assessment module

[0166] Health indicator calculation:

[0167] Calculate the Water Health Index (WHI), Vegetation Health Index (VHI), and Soil Moisture Index (SMI).

[0168] Overall evaluation:

[0169] The Analytic Hierarchy Process (AHP) was used to assign weights to each indicator (e.g., WHI 0.4, VHI 0.4, SMI 0.2) to calculate the comprehensive score of wetland ecological health.

[0170] Wetland health levels are classified according to score ranges (excellent, good, average, poor).

[0171] (6) Output and Decision Support Module

[0172] Generate wetland health level zoning maps and annual wetland change reports.

[0173] Provides decision-making recommendations for wetland protection and restoration, including priorities for the restoration of degraded areas and strategies for the delineation of protected areas.

[0174] Provide dynamic monitoring data and health assessment reports to the ecological management departments of high-altitude wetlands to support scientific decision-making.

[0175] This invention discloses a method and system for dynamic monitoring and health assessment of alpine wetlands based on multi-source remote sensing, which has the following beneficial effects:

[0176] (1) The technical solution of this invention overcomes the limitation of optical imagery caused by cloud cover by fusing SAR and optical data, and analyzes the dynamic changes of wetland soil moisture and water body through SAR data. This combination of multi-source data improves the ability to jointly analyze wetland hydrology, vegetation and soil moisture.

[0177] (2) The present invention integrates dynamic monitoring and health assessment. By combining a spatiotemporal convolutional neural network (ST-CNN) with an attention mechanism, it can simultaneously process the spatiotemporal characteristics of SAR and optical data, and realize long-term monitoring of wetland dynamic changes. Especially under the dual impact of climate change and human activities, the system can provide timely ecological health assessment.

[0178] (3) The present invention designs a multi-task deep learning network based on spatiotemporal convolution and attention mechanism, which improves the accuracy of wetland classification and change detection, while meeting the output requirements of multi-task results.

[0179] (4) The present invention optimizes the characteristics of alpine wetlands. In view of the unique seasonal changes and surface complexity of alpine wetlands, the calculation method of SAR polarization decomposition model and optical wetland index is optimized, making the model more suitable for actual application scenarios. Attached Figure Description

[0180] Figure 1 The present invention presents a method and system flowchart for dynamic monitoring and health assessment of alpine wetlands based on multi-source remote sensing.

[0181] Figure 2 The present invention provides a time-series curve of the continuous water area in the Ruoergai wetland region.

[0182] Figure 3 The present invention provides a time-series curve of seasonal water area in the Ruoergai wetland region.

[0183] Figure 4 The present invention defines the distribution range of surface water bodies in the typical area of ​​Ruoergai.

[0184] Figure 5 This invention provides a surface water distribution map of a typical lake region in the Yangtze River source area.

[0185] Figure 6 This invention presents a distribution map of the wetland health index in the Yangtze River source area in 2018.

[0186] Figure 7 This invention presents a distribution map of the wetland health index in the Yangtze River source area in 2022.

[0187] Example 1: The technical solution of the present invention is used for hydrological change monitoring in Ruoergai Wetland.

[0188] The Ruoergai Wetland is a typical example of alpine wetlands in China, and its hydrological changes are crucial to the regional ecosystem. This embodiment aims to monitor the dynamic hydrological changes of the Ruoergai Wetland using multi-temporal SAR and optical imagery.

[0189] step:

[0190] (1) Data collection:

[0191] SAR data: Sentinel-1 images (VV and VH polarization) were acquired annually between 2015 and 2023, with sampling times during the wet season (July) and dry season (October).

[0192] Optical data: Sentinel-2 images were acquired simultaneously (three views each year in April, July, and October).

[0193] Auxiliary data: Obtain DEM data for extraction of low-lying wetland areas; obtain meteorological data for analysis of the correlation between precipitation and wetland hydrological changes.

[0194] (2) Data preprocessing:

[0195] 1) Preprocessing SAR data:

[0196] Use Lee filtering to denoise and reduce speckle noise interference;

[0197] Geographic registration ensures consistency with optical imagery;

[0198] Surface scattering types were extracted based on Freeman-Durden decomposition.

[0199] 2) Processing optical data:

[0200] Use the Fmask algorithm to remove clouds and cloud shadows;

[0201] Surface reflectance is obtained by applying the atmospheric correction algorithm (6S);

[0202] Calculate MNDWI (Modified Water Index) to enhance water boundary identification.

[0203] (3) Wetland feature extraction:

[0204] 1) SAR data: High scattering regions of wetland water bodies are extracted using VV polarization signals, and wetland soil moisture is analyzed by combining VH polarization texture.

[0205] 2) Optical data: MNDWI was used to distinguish the boundaries between wetland water bodies and vegetation, and NDVI was used to analyze the relationship between vegetation and water bodies.

[0206] (4) Multimodal fusion and monitoring model:

[0207] 1) Using a Spatiotemporal Convolutional Network (ST-CNN):

[0208] Inputs: SAR scattering characteristics and optical water index;

[0209] Convolutional structure: Convolutional extraction is performed on SAR and optical features respectively, and attention mechanism is combined to enhance the feature representation of wetland areas;

[0210] Time dimension analysis: capturing the dynamic changes of wetlands through time series ST-CNN.

[0211] 2) Results: A wetland area change map was generated from 2015 to 2023, distinguishing the dynamic change range of seasonal wetlands and permanent wetlands.

[0212] (5) Output and Analysis:

[0213] 1) Surface water distribution area change curve: The surface water area of ​​wetlands was calculated year by year and a time-series change curve was plotted. It was found that the surface water fluctuated between 2015 and 2023, with two peaks in 2018 and 2021. The continuous water body has shown an increasing trend in the past 8 years, with a growth rate of about 1.5 km2 / a. The seasonal water body has shown a slight overall downward trend.

[0214] 2) Spatial Distribution Map: A water body distribution map of a typical area of ​​the Ruoergai Wetland was generated, showing that seasonal water bodies are mainly distributed in the areas surrounding lakes and rivers, as well as in low-lying areas of the wetland. The seasonal water body distribution area changes most significantly in low-lying areas of the wetland.

[0215] 3) Report generation: By combining meteorological data to analyze the correlation between water body changes and precipitation, it was found that the reduction of water bodies is closely related to wetland precipitation and surface evapotranspiration.

[0216] It is recommended to implement dynamic water body monitoring in areas with significant water body changes; and to output reports on dynamic changes in wetlands for wetland management departments to refer to.

[0217] Example 2: Health Assessment of Yangtze River Source Wetlands Based on the Technical Solution of the Invention

[0218] The Yangtze River source wetlands are an important ecological barrier for the Qinghai-Tibet Plateau, and their health directly affects the stability of downstream river systems. This example focuses on assessing the health status of the wetlands and providing recommendations for protection and restoration.

[0219] step:

[0220] (1) Data collection:

[0221] SAR data: Acquired Sentinel-1 multi-temporal images from 2018 to 2022, including monthly images for each growing season (April to October).

[0222] Optical data: Acquired concurrent Sentinel-2 multispectral images, with a focus on analyzing wetland vegetation cover.

[0223] Auxiliary data: Obtain meteorological data (precipitation, temperature) and DEM data.

[0224] (2) Data preprocessing:

[0225] The same processing method as in Example 1 is used to complete the denoising, correction and registration of SAR and optical images to ensure the consistency of multimodal images.

[0226] (3) Extraction of wetland health indicators:

[0227] Water Health Index (WHI): The water health score is calculated by extracting water distribution through MNDWI and combining it with the trend of water area change.

[0228] Vegetation Health Index (VHI): Based on NDVI and LST (Land Surface Temperature), it quantifies the status of vegetation growth and cover.

[0229] Soil Moisture Index (SMI): Soil moisture is estimated using the texture features of SAR images, reflecting wetland hydrological conditions.

[0230] (4) Comprehensive evaluation model:

[0231] The overall health score was calculated using the Analytic Hierarchy Process (AHP).

[0232] The weights for WHI, VHI, and SMI are set to 0.4, 0.4, and 0.2, respectively.

[0233] The overall score is divided into four levels: Excellent (>0.8), Good (0.6-0.8), Average (0.4-0.6), and Poor (<0.4).

[0234] (5) Evaluation results output:

[0235] Zoning Map: Generate a healthy zoning map of the Yangtze River source wetlands;

[0236] Trend analysis: Combining time series analysis with wetland health change trends, it was found that the vegetation health index declined year by year, which was significantly correlated with rising temperature and reduced precipitation.

[0237] (6) Repair suggestions:

[0238] Implement degraded wetland restoration projects in "poor" areas and restrict grazing activities;

[0239] The proposal suggests zoned management, with strict protection for "superior" zones and ecological water replenishment for "medium" zones.

[0240] Conclusion: This study provides priority planning for wetland protection and restoration, assists in the formulation of ecological protection policies, and outputs health assessment reports to support the design of wetland ecological compensation schemes.

[0241] This invention discloses a method and system for dynamic monitoring and health assessment of alpine wetlands based on multi-source remote sensing, which has the following beneficial effects:

[0242] (1) The technical solution of this invention overcomes the limitation of optical imagery caused by cloud cover by fusing SAR and optical data, and analyzes the dynamic changes of wetland soil moisture and water body through SAR data. This combination of multi-source data improves the ability to jointly analyze wetland hydrology, vegetation and soil moisture.

[0243] (2) The present invention integrates dynamic monitoring and health assessment. By combining a spatiotemporal convolutional neural network (ST-CNN) with an attention mechanism, it can simultaneously process the spatiotemporal characteristics of SAR and optical data, and realize long-term monitoring of wetland dynamic changes. Especially under the dual impact of climate change and human activities, the system can provide timely ecological health assessment.

[0244] (3) The present invention designs a multi-task deep learning network based on spatiotemporal convolution and attention mechanism, which improves the accuracy of wetland classification and change detection, while meeting the output requirements of multi-task results.

[0245] (4) The present invention optimizes the characteristics of alpine wetlands. In view of the unique seasonal changes and surface complexity of alpine wetlands, the calculation method of SAR polarization decomposition model and optical wetland index is optimized, making the model more suitable for actual application scenarios.

[0246] (5) This invention provides a scientific basis for the management of wetland protected areas and improves the efficiency of ecological restoration; supports the formulation of water resource management and ecological protection policies in high-altitude and cold regions; and provides data support for the study of wetland carbon sink function and the impact of climate change.

[0247] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamic monitoring and health assessment of alpine wetlands based on multi-source remote sensing, comprising the following steps: Step 1: Data Acquisition and Preprocessing; Acquire SAR image data, optical image data, and auxiliary data, and perform preprocessing on each. Step 2: Multimodal feature extraction; This system extracts SAR, optical, and dynamic features. The SAR feature extraction includes: for scattering features, calculating the VV / VH ratio and the polarization difference index (PDI) enhances the contrast between water bodies and vegetation, and between vegetation and bare land, improving classification accuracy; for texture features, extracting wetland soil moisture and roughness based on the gray-level co-occurrence matrix, and calculating statistical indicators including contrast, homogeneity, energy, and correlation to assess the texture complexity and moisture changes of the soil surface; the optical feature extraction includes: the Normalized Difference Vegetation Index (NDVI) which provides highly sensitive vegetation monitoring data in a short time, and the Dynamic Feature Corrected Normalized Water Index (MNDWI) used to enhance water body boundaries. The combination of MNDWI and NDVI can... It can provide more comprehensive vegetation and water information, enhance the accuracy of wetland classification and monitoring, and support a comprehensive assessment of the dynamic changes in wetland hydrology. For surface temperature (LST), in alpine wetland environments, LST provides real-time monitoring of the wetland surface thermal environment and supports dynamic response assessment of environmental changes. The extraction of dynamic features includes: constructing a time series dataset of wetlands, where the data at each time point includes scattering characteristics VV / VH, spectral index NDVI, MNDWI, and surface temperature (LST). By analyzing the characteristic changes in different seasons, the peak growth and decline processes of wetland vegetation and the seasonal fluctuations of water area are identified. Statistical analysis methods are applied to detect the growth or decrease trend of wetland water area and the long-term upward or downward trend of vegetation indices. Step 3: Multimodal data fusion; A spatiotemporal convolutional neural network (ST-CNN) combined with an attention mechanism is employed to enhance the sensitivity to wetland features. SAR and optical multi-temporal data are input and fused together. The ST-CNN aims to simultaneously capture feature changes in both spatial and temporal dimensions, adapting to the complex spatiotemporal dynamic environment of alpine wetlands. It outputs wetland classification, dynamic change trend maps, and key feature distribution maps. Its main architecture includes two branches: Branch 1, which processes SAR scattering and texture features; and Branch 2, which processes optical vegetation and water features. Specifically, it comprises the following components: Input layer: SAR feature input, including scattering features, texture features, and dynamic features; optical feature input, including spectral indices and time-series change features. Convolutional layers: Spatial convolutional layers use 3×3 or 5×5 two-dimensional convolutional kernels to extract spatial features from input data, capturing the spatial distribution and structural information of ground features; The temporal convolutional layer uses a one-dimensional convolutional kernel of 1×3 or 1×5 to process time series features along the time dimension and capture the temporal dependence of wetland dynamic changes. Channel attention adaptively adjusts the importance weights of different feature channels to highlight key features and suppress irrelevant or redundant features. It uses global average pooling and global max pooling operations to generate channel weight vectors, and calculates the attention weight of each channel through fully connected layers and activation functions. Spatial attention, by adjusting the feature importance of different spatial regions, highlights the feature response of the core wetland area and suppresses the interference of background or noise areas. It combines the spatial information of different feature maps to generate a spatial weight map and calculates the attention weight of each spatial location through convolution operation. The fusion layer stitches together the SAR feature map processed by the channel and spatial attention mechanisms with the optical feature map to form a comprehensive feature representation. Through the learned weights, the feature maps from different sources are weighted and summed to fuse multi-source information. The fully connected layer maps the fused high-dimensional feature vector to the output space of the classification or regression task to realize wetland type classification and prediction of dynamic change trends. The output layer outputs probability maps of the classification of various wetlands as permanent wetlands, seasonal wetlands, peatlands, and degraded wetlands. Dynamic trend map, outputting a dynamic trend prediction map of wetland water area, vegetation coverage and soil moisture; Step 4: Dynamic monitoring; Based on the fused feature data and random forest or support vector machine classification algorithms, wetlands are classified into permanent wetlands, seasonal wetlands, peatlands, and degraded wetlands. The dynamic change trends of wetlands, including wetland area, vegetation cover, and water body expansion and contraction, are analyzed through time series data. Classification criteria and judgment basis: Permanent wetlands are defined as wetlands in which the water coverage time and humidity status remain stable for most of the time phases over at least three consecutive years, with seasonal water level fluctuations of less than 5%, and are judged based on the proportion of water coverage time to the whole year being greater than 75% and the stable trend of soil moisture index (SMI). Seasonal wetlands are characterized by significant seasonal fluctuations in water level or water area throughout the year, with seasonal extreme differences in water area exceeding 30%. Peatlands, combined with SAR scattering characteristics, optical vegetation index and soil type data, identify wetland types containing a large amount of organic matter, exhibiting special polarization response and texture characteristics; Degraded wetlands are identified as such based on a combination of multiple indicators and a continuously decreasing trend in the Soil Moisture Index (SMI) or a significant increase in the anthropogenic disturbance index. Step 5: Wetland health assessment; The ecological health status of alpine wetlands is comprehensively assessed by constructing and calculating multiple ecological health indicators. The ecological health indicators include the Water Health Index (WHI), which uses time series data of water area obtained from dynamic monitoring to calculate the annual growth rate or reduction rate of water area. Based on the digital elevation model (DEM) and water distribution data, the density and connectivity of the water network are analyzed. The WHI score is obtained by weighted summation of the water area change rate and connectivity indicators. The Vegetation Health Index (VHI) quantifies the health status of wetland vegetation by combining the rate of change in vegetation cover, peak vegetation growth, and surface temperature. It uses NDVI time series data to calculate the annual growth rate or reduction rate of vegetation cover, identifies the peak NDVI during the vegetation growing season to reflect the growth intensity of vegetation, and combines LST data to assess the impact of high temperature on vegetation health. The VHI score is calculated by a weighted formula. The Soil Moisture Index (SMI) reflects the soil moisture status and the health of wetland soil moisture. Using roughness indexes extracted from GLCM and meteorological data, a soil moisture inversion model is established to calculate the annual average or key period values ​​of soil moisture. Multi-temporal SMI data are analyzed to assess the annual average rate of change of soil moisture. The current state of soil moisture is compared with historical data, and the SMI score is obtained through standardization. The weights of the water health index (WHI), vegetation health index (VHI), and soil moisture index (SMI) were assigned and comprehensively evaluated using the analytic hierarchy process (AHP) to generate a comprehensive score for wetland ecological health. Score range division: Excellent: Overall score > 0.8; Good: 0.6 ≤ Overall score ≤ 0.8; Average: 0.4 ≤ Overall score < 0.6; Poor: Overall score < 0.

4.

2. The method for dynamic monitoring and health assessment of alpine wetlands based on multi-source remote sensing according to claim 1, characterized in that, In step 1, SAR image data is acquired using Sentinel-1 VV / VH polarization data or ALOS PALSAR data, which supports multi-temporal and multi-polarization data to obtain the scattering characteristics of wetland water and soil moisture. Optical image data is acquired using Sentinel-2 dataset 10m resolution multispectral images or Landsat series 30m resolution images to obtain wetland vegetation cover, surface water distribution, and surface temperature. Auxiliary data include digital elevation models, meteorological precipitation, and temperature data.

3. A dynamic monitoring and health assessment system for alpine wetlands based on SAR and optical remote sensing, comprising the following modules: The data acquisition and preprocessing module is used for the acquisition and preprocessing of SAR image data, optical image data, and auxiliary data. A multimodal feature extraction module is used for the extraction of SAR and optical features; The multimodal data fusion module employs a spatiotemporal convolutional neural network STCNN, combined with an attention mechanism, to enhance the sensitivity to wetland features. It inputs SAR and optical multitemporal data and performs deep fusion of the SAR and optical multitemporal data. The network branch design includes: Branch 1 SAR subnetwork: which processes SAR scattering and texture features and extracts hydrological and soil moisture features of wetlands. Branch 2 Optical Subnet: Processes optical vegetation and water features, extracting vegetation cover and water distribution characteristics of wetlands; the network architecture also includes the following components: channel attention, which assigns adaptive weights to features of different polarizations or bands to highlight important features; spatial... Attention enhancement: Enhance the feature response of the core wetland area and suppress interference from irrelevant areas; Fusion and output: Through feature splicing or weighted fusion layer, the features of branch one and branch two are fused to output wetland classification results, dynamic change trend map and key feature distribution map; The wetland dynamic monitoring module utilizes fused feature data and classifies wetlands into permanent wetlands, seasonal wetlands, peatlands, and degraded wetlands based on random forest or support vector machine classification algorithms. It analyzes the dynamic change trends of wetlands through time series data, including wetland area, vegetation cover, and water body expansion and contraction. It outputs annual wetland change maps and wetland dynamic curve area change curves. The wetland health assessment module calculates the water health index (WHI), vegetation health index (VHI), and soil moisture index (SMI), constructs a health assessment model based on weight allocation, and uses the analytic hierarchy process (AHP) to integrate the various indicators into a comprehensive score for wetland ecological health, which is used for wetland health assessment. The output and decision support module generates a wetland health level zoning map, including excellent, good, medium, and poor. Provide recommendations for wetland protection and restoration, and provide information on the delineation of protected areas and the priority of restoration for degraded areas.

Images