A system and method for monitoring grassland desertification by multispectral remote sensing

Abstract

The application discloses a kind of grassland desertification multispectral remote sensing monitoring system and method, it is related to remote sensing image technical field, the system includes following component parts: time series image pre-processing registration module, pixel level phenology track construction module, healthy grassland phenology baseline modeling module, track collaborative comparison module and desertification dynamic quantification module;Corresponding monitoring method is time series image pre-processing registration, pixel level phenology track construction, healthy grassland phenology baseline modeling, track collaborative comparison, desertification dynamic quantification five big steps in turn.This application improves the degree of monitoring refinement, collects multispectral time series remote sensing image and constructs per-pixel phenology track, combined with hierarchical structured phenology baseline model, truly restores the original growth law of different pixels, in-depth to pixel scale monitoring, provides data support for desertification feature analysis;Realize desertification comprehensive determination and dynamic quantification, divide desertification grade, provide technical support for desertification prevention work, improve decision efficiency and implementation effectiveness.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image technology, specifically to a multispectral remote sensing monitoring system and method for grassland desertification. Background Technology

[0002] Grasslands, as a core component of terrestrial ecosystems, possess multiple functions including ecological protection, soil and water conservation, and biodiversity maintenance, serving as a crucial carrier of ecological security barriers. However, influenced by factors such as climate change and overgrazing, global grassland desertification is becoming increasingly severe, triggering a series of chain reactions such as declining land productivity and ecosystem degradation, posing a serious threat to regional ecological security and sustainable development. Traditional grassland desertification monitoring relies on ground surveys, which suffer from drawbacks such as being time-consuming and labor-intensive, having limited coverage, and poor timeliness. Multispectral remote sensing technology, with its advantages of large-scale, rapid, and non-contact monitoring, has become the mainstream technology for desertification monitoring. However, existing remote sensing-based monitoring schemes still struggle to meet the needs for refined and dynamic monitoring.

[0003] Current mainstream remote sensing monitoring technologies for grassland desertification have many shortcomings: they are mostly based on image data from a single or a few time points, lacking continuous tracking of phenological trajectories throughout the entire vegetation growth cycle, making it difficult to reflect the dynamic evolution of the desertification process; the construction of healthy grassland baseline models is relatively simple, often using overall averaging, without considering the influence of differential characteristics such as topography, soil, and grassland type at the pixel scale, resulting in insufficient baseline reference; trajectory similarity calculations mostly rely on a single distance index, failing to achieve coordinated quantification of Euclidean distance and cosine similarity, and the extraction of phenological parameter offsets lacks effective temporal alignment technology, resulting in significant bias; desertification level classification mostly uses a single index for judgment, failing to integrate similarity distance and phenological parameter offsets for comprehensive evaluation, and cannot accurately extract the occurrence time and evolution rate of desertification, resulting in low monitoring accuracy, weak dynamic quantification capabilities, and difficulty in supporting accurate decision-making for desertification control.

[0004] The shortcomings of the existing technologies mentioned above result in problems such as low accuracy, insufficient dynamism, and incomplete information in grassland desertification monitoring results, failing to provide timely and accurate technical support for desertification control and hindering the effectiveness of ecological protection efforts. Therefore, there is an urgent need to develop a multispectral remote sensing monitoring system and method capable of pixel-level phenological trajectory construction, hierarchical baseline model optimization, multi-dimensional collaborative comparison, and dynamic quantification of desertification. By integrating key technologies such as temporal image processing, time series analysis, and dynamic normalization, the system can improve the precision and dynamic tracking capabilities of desertification monitoring, providing a scientific basis for early warning, control, and ecological restoration of grassland desertification. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a multispectral remote sensing monitoring system and method for grassland desertification. It can acquire multispectral time-series remote sensing images, preprocess and register them to ensure data consistency, construct pixel-by-pixel phenological trajectories and screen healthy trajectories, and combine them with a hierarchical phenological baseline model to achieve baseline-pixel feature mapping. Through trajectory collaborative comparison and dynamic quantification of desertification, it can classify desertification levels, extract desertification information and output reports, providing scientific support for grassland desertification prevention and control.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On the one hand, a multispectral remote sensing monitoring system for grassland desertification, the system comprising the following components: a time-series image preprocessing and registration module, a pixel-level phenological trajectory construction module, a healthy grassland phenological baseline modeling module, a trajectory collaborative comparison module, and a desertification dynamic quantification module;

[0007] The time-series image preprocessing and registration module acquires multispectral time-series remote sensing images of the monitoring area, performs preprocessing and spatial registration operations, and outputs a registered time-series image set.

[0008] The pixel-level phenological trajectory construction module: receives the registered time-series image set, calculates the vegetation cover index using the remote sensing image pixel-level spectral analysis algorithm; constructs the phenological trajectory of each pixel using time series analysis technology, and outputs the pixel-level phenological trajectory set and phenological parameters of historical healthy grassland trajectory and target growing season real-time trajectory;

[0009] The healthy grassland phenological baseline modeling module: receives historical healthy grassland trajectories, filters healthy grassland pixel trajectories and performs fitting optimization to construct a pixel-level phenological baseline model;

[0010] The trajectory collaborative comparison module receives the real-time trajectory of the target growing season and the pixel-level phenological baseline model, calculates the collaborative similarity distance between the real-time trajectory of the target growing season and the pixel-level phenological baseline model using a trajectory collaborative similarity algorithm, and extracts the offset of the phenological parameters using dynamic time warping technology, outputting a collaborative similarity distance dataset and a phenological parameter offset dataset.

[0011] The desertification dynamic quantification module receives a collaborative similarity distance dataset and a phenological parameter offset dataset, classifies desertification levels using desertification dynamic quantification rules, extracts the desertification occurrence time and desertification evolution rate, and outputs desertification monitoring maps and statistical reports.

[0012] Furthermore, the process of acquiring multispectral time-series remote sensing images of the monitoring area and performing preprocessing and spatial registration operations in the time-series image preprocessing and registration module to output the registered time-series image set is as follows: Multispectral time-series remote sensing images covering the monitoring area are acquired according to the geographical location of the monitoring area and the time span of the grassland growing season, forming an initial multispectral time-series remote sensing image set; each image in the initial multispectral time-series remote sensing image set is preprocessed to eliminate errors generated during image acquisition, resulting in a preprocessed multispectral time-series remote sensing image set; a geographic reference image is selected, and all images in the preprocessed multispectral time-series remote sensing image set are spatially registered with the geographic reference image, unifying all images to the same geographic coordinate system, resulting in the registered time-series image set.

[0013] Furthermore, the calculation formula for the remote sensing image pixel-level spectral analysis algorithm in the pixel-level phenological trajectory construction module is as follows: ,in, The vegetation cover index; This is the normalization adjustment coefficient for pixel-level spectral analysis; To monitor the near-infrared reflectance of individual pixels within the area; To monitor the red band reflectance of the same single pixel within the monitoring area; To monitor the blue band reflectance of the same single pixel within the monitoring area; This is the atmospheric scattering correction factor for the red band; This is the atmospheric scattering correction factor for the blue band; The soil background adjustment coefficient is used. By performing pixel-by-pixel fusion calculation on the spectral characteristics of different bands of multispectral time-series remote sensing images, the interference of soil background, atmospheric scattering and vegetation canopy shadow on the spectral signal is effectively eliminated. The spectral feature information of grassland vegetation at the single pixel scale in the monitoring area is extracted, and the vegetation coverage index of grassland vegetation coverage and growth vitality is calculated.

[0014] Furthermore, the process of constructing the phenological trajectory for each pixel using time series analysis technology in the pixel-level phenological trajectory construction module, and outputting the pixel-level phenological trajectory set and phenological parameters of historical healthy grassland trajectory and target growing season real-time trajectory, is as follows: The calculated vegetation cover index is integrated into the time-series raw data of the pixel-by-pixel vegetation cover index; the time-series raw data is preprocessed using time series analysis technology, and the preprocessed time-series raw data is fitted to construct the phenological trajectory for each pixel; historical period trajectories and target growing season trajectories are extracted from all constructed phenological trajectories, and historical healthy grassland trajectories are obtained through health screening. The screening criteria are as follows: Normal range of cover index: Based on the measured data of grassland type in the monitoring area, the baseline interval is set to [0.6, 0.95], and for alpine grasslands, it can be adjusted down to [0.6, 0.95]. [0.55, 0.90], desert steppe can be adjusted to [0.65, 0.98]; Quantification of vegetation growth stability: the trajectory variation coefficient CV is used for determination, CV = (time series data standard deviation / time series data mean) × 100%, CV ≤ 15% is determined to be growth stable; Distinguishing abnormal factors: the impact of pests and diseases is manifested by the sudden change of local pixel spectrum in a short period of time, and the near-infrared band reflectance shows an abnormal peak, which is more than 20% higher than the surrounding pixels, while the red band reflectance does not show a continuous downward trend; the early stage of desertification is manifested by the spectrum showing a gradual trend, the coverage index continuously decreasing for more than 3 time series nodes, and the absence of the above-mentioned pest and disease characteristic spectrum; extract the real-time trajectory of the target growing season and integrate it into a pixel-level phenological trajectory set, and at the same time analyze each trajectory to extract phenological parameters that characterize the vegetation growth law, and output the pixel-level phenological trajectory set and phenological parameters.

[0015] Furthermore, the pixel-level phenological baseline model in the healthy grassland phenological baseline modeling module is a hierarchical structured model, specifically including a basic trajectory layer, a fitting optimization layer, and a feature mapping layer. The basic trajectory layer is a set of historical healthy grassland pixel trajectories that have been retained after health screening and conform to the original growth pattern of grassland, and is stored one by one according to the spatial location of the pixels. The fitting optimization layer forms a standardized phenological trajectory after fitting and optimizing the trajectory data of the basic trajectory layer. The feature mapping layer establishes a mapping relationship between the standardized phenological trajectory after fitting and optimizing and the topographic, soil, and grassland type features of the corresponding pixels. The three-layer structure is interconnected and the data is shared, together forming a pixel-level phenological baseline model that can accurately reflect the original phenological characteristics of grassland in the monitoring area.

[0016] Furthermore, the specific steps in the trajectory collaborative comparison module for calculating the collaborative similarity distance between the real-time trajectory of the target growing season and the pixel-level phenological baseline model using the trajectory collaborative similarity algorithm are as follows: The real-time trajectory of the target growing season and the single-pixel reference trajectory in the pixel-level phenological baseline model are synchronized and normalized according to the time dimension to unify the trajectory data dimension; phenological feature sequences of the two sets of trajectories are extracted pixel by pixel from the grassland pixels in the monitoring area to form one-to-one corresponding trajectory feature data; the similarity metric of the two sets of trajectory feature data at the pixel level is quantitatively calculated using the trajectory collaborative similarity algorithm to obtain the deviation value between the real-time trajectory and the reference trajectory at the pixel level; finally, the deviation values ​​of all pixels are integrated according to their geographic spatial location to generate a collaborative similarity distance dataset.

[0017] Furthermore, the calculation formula for the trajectory collaborative similarity algorithm in the trajectory collaborative comparison module is as follows: ,in, For the first Trajectory co-similarity distance of each pixel; The numbering of individual grassland pixels within the monitoring area; For collaborative weighting coefficients; For the first The Euclidean distance between the real-time trajectory of each pixel and the baseline trajectory; To determine the maximum Euclidean distance of all pixels within the monitoring area; For the first The cosine similarity between the real-time trajectory of each pixel and the baseline trajectory is calculated. A pixel-by-pixel collaborative similarity quantification analysis is carried out on the real-time phenological trajectory of the target growing season and the pixel-level phenological baseline model. The analysis takes into account the degree of numerical deviation and the consistency of the trend of change of the trajectory in the time dimension, determines the degree of collaborative matching, and quantifies the deviation difference between trajectories.

[0018] Furthermore, the process of extracting the offset of the phenological parameters and outputting the collaborative similarity distance dataset and the phenological parameter offset dataset in the trajectory collaborative comparison module is as follows: Temporal dimension warping is performed on the real-time trajectory of the target growing season and the pixel-level phenological baseline model trajectory; the temporal nodes of the two trajectories are aligned pixel by pixel using dynamic time warping technology; based on the aligned trajectories, the deviation of each phenological parameter in temporal features is extracted and quantified to form phenological parameter offset data; the calculated collaborative similarity distance is organized according to geospatial association to generate a collaborative similarity distance dataset, and the phenological parameter offset data is classified according to corresponding pixels to form a phenological parameter offset dataset; the collaborative similarity distance dataset and the phenological parameter offset dataset are output synchronously.

[0019] Furthermore, the specific rules for classifying desertification levels using the dynamic quantification rules in the desertification dynamic quantification module are as follows: A co-similarity distance dataset and a phenological parameter offset dataset are received; pixel-by-pixel association matching is performed on the two datasets to obtain a standardized desertification analysis dataset; desertification levels are classified according to the dynamic quantification rules. The thresholds of these rules are derived from the fitting and verification of 5 years of measured data and remote sensing data from 30 typical grassland sample areas nationwide. The baseline thresholds are as follows: No desertification, co-similarity distance ≥ 0.85, phenological parameter offset ≤ 5%; Slight desertification, 0.70 ≤ co-similarity distance < 0.85, 5% < phenological parameter offset ≤ 15%; Moderate desertification, 0.40 For severe desertification, the threshold values ​​are: ≤cooperative similarity distance < 0.70, 15% < phenological parameter offset ≤ 30%; for severe desertification, the threshold values ​​are: cooperative similarity distance < 0.40, phenological parameter offset > 30%; regional adaptation method: alpine grassland: threshold values ​​for each level of cooperative similarity distance are lowered by 0.05-0.10, and threshold values ​​for each level of phenological parameter offset are raised by 5%-8%; temperate grassland: the baseline threshold values ​​are used; desert grassland: threshold values ​​for each level of cooperative similarity distance are raised by 0.03-0.08, and threshold values ​​for each level of phenological parameter offset are lowered by 3%-5%; special areas: the threshold values ​​can be calibrated a second time by introducing a soil salinity correction coefficient, the soil salinity correction coefficient being in the range of 0.9-1.1.

[0020] On the other hand, a multispectral remote sensing monitoring method for grassland desertification is applicable to a multispectral remote sensing monitoring system for grassland desertification in any of the aforementioned schemes. The specific steps of this method are as follows:

[0021] S100, Temporal Image Preprocessing and Registration: Multispectral temporal remote sensing images are collected according to the monitoring area and the span of the growing season. After preprocessing to eliminate errors, they are spatially registered with the geographic reference image to output a temporal image set with a unified coordinate system.

[0022] S200, Pixel-level phenological trajectory construction: The vegetation cover index is calculated based on the pixel-level spectral analysis algorithm of remote sensing images. A pixel-by-pixel phenological trajectory is constructed through time series analysis and fitting. Historical healthy grassland trajectories are obtained by health screening according to the following criteria: Normal range of cover index: baseline interval [0.6, 0.95], alpine grassland [0.55, 0.90], desert grassland [0.65, 0.98]; Growth stability: coefficient of variation CV ≤ 15%; Anomaly differentiation: Pests and diseases are characterized by local spectral mutations + abnormal peaks in the near-infrared band; early desertification is characterized by gradual spectral changes and a continuous decrease in the cover index. Real-time trajectory sets and phenological parameters of the target growing season are extracted.

[0023] S300, Healthy Grassland Phenological Baseline Modeling: Screen healthy grassland pixel trajectories and fit and optimize them to construct a hierarchical pixel-level phenological baseline model containing a basic trajectory layer, a fitting and optimization layer and a feature mapping layer.

[0024] S400, Trajectory Collaborative Comparison: Time-normalize the real-time trajectory of the target growing season with the baseline model trajectory, calculate the similarity distance through the trajectory collaborative similarity algorithm, extract the phenological parameter offset through dynamic normalization, and output two types of datasets;

[0025] S500, Dynamic Quantification of Desertification: This method involves associating and matching two datasets, using benchmark thresholds validated in 30 typical sample areas nationwide. No desertification: S≥0.85 and offset≤5%; Mild desertification: 0.70≤S<0.85 and 5%<offset≤15%; Moderate desertification: 0.40≤S<0.70 and 15%<offset≤30%; Severe desertification: S<0.40 and offset>30%. Adjustments are made according to region type: for alpine grasslands, the S threshold is lowered by 0.05-0.10 and the offset threshold is raised by 5%-8%; for desert grasslands, the S threshold is raised by 0.03-0.08 and the offset threshold is lowered by 3%-5%. Four desertification levels are classified, desertification time and rate are extracted, and desertification monitoring maps and statistical reports are output.

[0026] Compared with existing technologies, this multispectral remote sensing monitoring system and method for grassland desertification has the following advantages:

[0027] I. This application acquires multispectral time-series remote sensing images of the monitoring area, and ensures data consistency through preprocessing and spatial registration. Utilizing pixel-level spectral analysis and time-series analysis techniques, it constructs pixel-by-pixel phenological trajectories and filters historical healthy grassland trajectories. Combined with a hierarchical structured phenological baseline model, it achieves the mapping of baselines to pixel topographic, soil, and grassland type characteristics. This breaks through the limitations of traditional monitoring, such as discontinuous tracking of vegetation growth cycles and lack of differentiated adaptation of baseline models. It can realistically restore the original growth patterns of different pixels, capture subtle changes in vegetation during the target growing season, and allow desertification monitoring to penetrate from the overall regional level to the pixel scale, improving the precision and targeting of monitoring, and providing reliable basic data support for subsequent desertification characteristic analysis.

[0028] II. This application uses time warping to compare the real-time trajectory of the target growing season with the baseline model trajectory, employs a trajectory collaborative similarity algorithm to achieve multi-dimensional similarity quantification, and combines dynamic time warping technology to extract phenological parameter offsets. It classifies desertification levels through association matching of the two datasets, simultaneously extracts the desertification occurrence time and evolution rate, and outputs monitoring maps and statistical reports. This overcomes the problems of single similarity calculation, large deviations in phenological parameter extraction, and one-sided desertification assessment indicators in traditional monitoring. It achieves comprehensive judgment and dynamic quantification of desertification status, accurately distinguishing different degrees of desertification and clearly presenting the desertification evolution process. This provides comprehensive and scientific technical support for early warning of grassland desertification, formulation of governance plans, and evaluation of ecological restoration effects, improving the decision-making efficiency and implementation effectiveness of desertification prevention and control.

[0029] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0031] Figure 1 A structural block diagram and a working logic flowchart of a multispectral remote sensing monitoring system for grassland desertification;

[0032] Figure 2 A schematic diagram of healthy grassland phenological baseline modeling and trajectory collaborative comparison data transmission for a multispectral remote sensing monitoring system for grassland desertification;

[0033] Figure 3 This is a flowchart illustrating the steps of a multispectral remote sensing method for monitoring grassland desertification. Detailed Implementation

[0034] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0035] Example 1: Temporal Image Preprocessing and Registration Module: Acquires multispectral temporal remote sensing images of the monitoring area, performs preprocessing and spatial registration operations, and outputs a registered temporal image set. Specifically:

[0036] The temporal image preprocessing and registration module comprehensively acquires multispectral temporal remote sensing images that completely cover the monitoring area, based on the geographical location and the time span of the complete grassland growing season. This ensures that the images capture the entire growth cycle of the grassland, from germination and vigorous growth to withering, forming an initial multispectral temporal remote sensing image set. Since images are affected by various factors during acquisition, such as atmospheric interference, differences in sensor accuracy, and changes in lighting conditions, direct use can lead to deviations in subsequent analysis results. Therefore, each image in the initial image set needs systematic preprocessing to make the grassland surface information presented in the images closer to the actual situation, resulting in a preprocessed multispectral temporal remote sensing image set. To ensure that all subsequent pixel-level analyses are conducted within a unified spatial framework and to avoid analytical errors caused by mismatches in image spatial locations, a geographic reference image with clear topography, high spatial accuracy, and no significant distortion needs to be selected. All preprocessed images are then rigorously spatially registered with this geographic reference image. Through coordinate transformation, geometric correction, and other methods, all images are unified into the same geographic coordinate system, ensuring that images acquired at different times correspond completely in spatial location. The final output is a registered time-series image set, providing fundamental image data with strong spatial consistency and high data reliability for subsequent pixel-level phenological trajectory construction, baseline modeling, and other stages.

[0037] Pixel-level phenological trajectory construction module: Receives the registered time-series image set, calculates the vegetation cover index using a pixel-level spectral analysis algorithm for remote sensing images; constructs the phenological trajectory for each pixel using time-series analysis techniques, and outputs a pixel-level phenological trajectory set and phenological parameters, including historical healthy grassland trajectories and the target growing season's real-time trajectory. Specifically:

[0038] The pixel-level phenological trajectory construction module receives the registered time-series image set and calculates the vegetation cover index of each pixel using a remote sensing image pixel-level spectral analysis algorithm. The formula is as follows: ,in, The vegetation cover index; This is the normalization adjustment coefficient for pixel-level spectral analysis; To monitor the near-infrared reflectance of individual pixels within the area; To monitor the red band reflectance of the same single pixel within the monitoring area; To monitor the blue band reflectance of the same single pixel within the monitoring area; This is the atmospheric scattering correction factor for the red band; This is the atmospheric scattering correction factor for the blue band; As a soil background adjustment coefficient, this algorithm can accurately capture the differences in spectral reflectance characteristics of grassland vegetation in near-infrared, red, and blue bands corresponding to different pixels. Through scientific calculation logic, it transforms spectral information into a vegetation cover index that directly reflects the degree of vegetation growth, allowing the vegetation growth status of each pixel to be quantitatively presented. The calculated vegetation cover indices of all pixels are integrated in chronological order to form the time-series raw data of the pixel-by-pixel vegetation cover index. This raw data contains basic information on vegetation growth at different time points, but may also contain time-series fluctuations due to short-term weather fluctuations and accidental observation errors. Therefore, time-series preprocessing is required to remove outliers and fill in missing data to make the time-series data more stable. Subsequently, time-series analysis technology is used to fit the preprocessed time-series raw data. This technology analyzes and models the trend and periodicity of the time-series data, eliminates the interference caused by short-term fluctuations, and constructs the phenological trajectory of each pixel. This trajectory can clearly show the growth change pattern of the corresponding pixel's vegetation throughout the growing season. From all constructed phenological trajectories, historical period trajectories and target growing season trajectories were extracted. Historical healthy grassland trajectories were obtained through health screening, with the following screening criteria: Normal range of coverage index: Based on measured data of grassland types in the monitoring area, the baseline interval was set to [0.6, 0.95]. For alpine grasslands, this range could be adjusted down to [0.55, 0.90], and for desert grasslands, it could be adjusted up to [0.65, 0.98]. Quantification of vegetation growth stability: The trajectory coefficient of variation (CV) was used for determination. CV = (standard deviation of time series data / mean of time series data) × 100%. CV ≤ 15% was considered stable growth. Distinguishing abnormal factors: The impact of pests and diseases manifests as sudden changes in the spectrum of local pixels within a short period, with abnormal peaks in near-infrared reflectance exceeding 20% ​​of surrounding pixels, while the red reflectance shows no continuous downward trend. Early desertification is characterized by a gradual change in the spectrum, with the coverage index decreasing for more than three consecutive time nodes, and the absence of the aforementioned pest and disease spectral characteristics. Simultaneously, real-time trajectories of the target growing season are extracted and integrated into a pixel-level phenological trajectory set according to spatial location, ensuring that the growth trajectory of each pixel in the target growing season is completely recorded. Phenological parameters that characterize the vegetation growth pattern, such as vegetation germination time, peak growth period, and withering time, contained in each trajectory are analyzed in a synchronous manner. These parameters can reflect the key node changes in vegetation growth. Finally, a pixel-level phenological trajectory set and phenological parameters are output, providing high-quality historical health data for healthy grassland phenological baseline modeling.

[0039] Healthy grassland phenological baseline modeling module: Receives historical healthy grassland trajectories, filters healthy grassland pixel trajectories, performs fitting optimization, and constructs a pixel-level phenological baseline model. Specifically:

[0040] The healthy grassland phenological baseline modeling module receives historical healthy grassland trajectories after health screening. It further filters out healthy grassland pixel trajectories that fully conform to the natural growth patterns of grasslands and have not been affected by human intervention or abnormal environments. These trajectories are stored one-to-one according to their spatial location, forming a basic trajectory layer. This spatial storage method ensures that every specific location of grassland within the monitoring area has a corresponding healthy growth benchmark, avoiding benchmark failure caused by spatial misalignment. Based on the large amount of trajectory data stored in the basic trajectory layer, the trajectory data is fitted and optimized to eliminate trajectory fluctuations caused by natural growth differences between different years and individual grasslands, forming standardized phenological trajectories that constitute the fitting optimization layer. This layer of trajectories can represent the growth change pattern of a certain type of grassland under ideal health conditions, reducing the impact of individual differences on the accuracy of the benchmark. Subsequently, a feature mapping layer is constructed to establish a close mapping relationship between the fitted and optimized standardized phenological trajectories and the corresponding topographic features, soil type features, and grassland type features of the pixels. This allows each standardized trajectory to be adapted to its corresponding geographical environment and grassland type. For example, grassland trajectories under arid terrain and grassland trajectories under humid terrain differ in terms of growth peak and duration. The feature mapping layer allows the baseline model to fully consider these environmental factors, making the constructed hierarchical structured pixel-level phenological baseline model more targeted and scientific. This avoids the baseline model becoming disconnected from the actual grassland growth pattern due to ignoring environmental differences, and ensures that the benchmark setting can fit the actual growth characteristics of grasslands at different locations in the monitoring area during subsequent trajectory comparison analysis, providing a reliable comparative basis for accurately judging the desertification status.

[0041] The trajectory collaborative comparison module receives the real-time trajectory of the target growing season and a pixel-level phenological baseline model. Using a trajectory collaborative similarity algorithm, it calculates the collaborative similarity distance between the real-time trajectory and the pixel-level phenological baseline model. It also uses dynamic time warping technology to extract the offsets of the phenological parameters and outputs a collaborative similarity distance dataset and a phenological parameter offset dataset. Specifically:

[0042] The trajectory collaborative comparison module receives the real-time trajectory of the target growing season and the constructed pixel-level phenological baseline model. First, it synchronizes and standardizes the real-time trajectory and the single-pixel reference trajectory in the baseline model according to the time dimension, unifying the time interval, data acquisition nodes, and other trajectory data dimensions to ensure comparability between the real-time and reference trajectories on a time scale and avoid comparison bias caused by inconsistencies in the time dimension. For each grassland pixel within the monitoring area, phenological feature sequences of the real-time and reference trajectories are extracted pixel-by-pixel. These feature sequences contain core information such as changes in vegetation cover index and the occurrence time of key growth nodes, forming one-to-one trajectory feature data pairs, enabling refined comparative analysis from the overall to the local level. The similarity measure of the two sets of trajectory feature data for each pixel is calculated using a trajectory collaborative similarity algorithm, with the formula: ,in, For the first Trajectory co-similarity distance of each pixel; The numbering of individual grassland pixels within the monitoring area; For collaborative weighting coefficients; For the first The Euclidean distance between the real-time trajectory of each pixel and the baseline trajectory; To determine the maximum Euclidean distance of all pixels within the monitoring area; For the first The algorithm calculates the cosine similarity between the real-time trajectory and the baseline trajectory for each pixel. It comprehensively considers the distance difference and trend consistency between the two trajectories, accurately capturing both numerical deviations and the compatibility of growth rhythms. This yields the deviation value between the real-time trajectory and the baseline trajectory at the pixel level. The deviation values ​​of all pixels are then integrated according to geospatial location to generate a collaborative similarity distance dataset, intuitively reflecting the deviation of grassland from the health baseline at different spatial locations. Simultaneously, the real-time trajectory and baseline model trajectory for the target growing season are further normalized in terms of temporal dimension to ensure precise correspondence between each temporal node. Dynamic temporal normalization technology is used to align the temporal nodes of the two trajectories pixel by pixel. Even if there are slight differences in the time progress of the two trajectories, this technology can achieve matching of key growth stages. Based on the aligned trajectory, the deviation of each phenological parameter in temporal characteristics is extracted in detail, such as the number of days the germination time is delayed, the reduction in the growth peak, and the duration of the advance in withering. These deviations are quantified to form phenological parameter offset data, which is then categorized by corresponding pixels to form a phenological parameter offset dataset. The system simultaneously outputs a collaborative similarity distance dataset and a phenological parameter offset dataset. These two types of data reflect the differences between real-time trajectories and health benchmarks from different dimensions, providing comprehensive and complementary data support for subsequent dynamic quantification of desertification. This allows desertification assessment to combine overall trajectory deviations and specific parameter offsets, improving the accuracy and reliability of the analysis results.

[0043] The desertification dynamic quantification module receives a collaborative similarity distance dataset and a phenological parameter offset dataset, classifies desertification levels using dynamic quantification rules, extracts the desertification occurrence time and desertification evolution rate, and outputs a desertification monitoring map and statistical report. Specifically:

[0044] The dynamic quantification module for desertification receives a co-similarity distance dataset and a phenological parameter offset dataset. It performs pixel-by-pixel correlation matching on both datasets to ensure a one-to-one correspondence between the co-similarity distance and the corresponding phenological parameter offset for each pixel. This eliminates analytical errors caused by data misalignment, resulting in a standardized desertification analysis dataset that provides an accurate data foundation for subsequent desertification level classification. Based on preset dynamic quantification rules, the thresholds of which are derived from a 5-year fitting and verification of measured data and remote sensing data from 30 typical grassland sample areas nationwide, the module uses both co-similarity distance and phenological parameter offset as dual judgment indicators. By comprehensively considering the magnitude and trend of both values, the module accurately classifies the grassland desertification level within the monitoring area into four categories: no desertification, slight desertification, moderate desertification, and severe desertification. Among them, the no-desertification level corresponds to a co-similarity distance ≥ 0.85 and a phenological parameter deviation ≤ 5%, indicating that the grassland growth status in this area is basically consistent with the healthy baseline; the mild desertification level corresponds to a co-similarity distance < 0.85 and a deviation of 5% < phenological parameter ≤ 15%, indicating that grassland growth is slightly affected; the moderate desertification level corresponds to a co-similarity distance < 0.70 and a deviation of 15% < phenological parameter ≤ 30%, indicating that grassland growth has significantly degraded; the severe desertification level corresponds to a co-similarity distance < 0.40 and a deviation of phenological parameter > 30%, indicating that grassland vegetation cover has been greatly reduced and the degree of desertification is serious. For different regional types, the following adaptation methods are adopted: Alpine grassland: the threshold for each level of co-similarity distance is lowered by 0.05-0.10, and the threshold for each level of phenological parameter offset is raised by 5%-8%; Temperate grassland: the baseline threshold is used; Desert grassland: the threshold for each level of co-similarity distance is raised by 0.03-0.08, and the threshold for each level of phenological parameter offset is lowered by 3%-5%; Special regions: a soil salinity correction coefficient with a value range of 0.9-1.1 can be introduced to perform secondary calibration of the threshold. While classifying desertification levels, the occurrence time of desertification corresponding to each level is accurately extracted through the tracing and analysis of time-series data, clarifying which stage of the growing season desertification began to appear. Simultaneously, the desertification evolution rate is calculated to understand the speed of desertification development from its initial state to the current level, determining whether desertification is in a slow development, rapid spread, or stabilizing state. The final output is a desertification monitoring map that intuitively presents the spatial distribution of desertification. The map clearly shows the spatial location, distribution range, and concentrated areas of different desertification levels within the monitoring area. It also outputs a statistical report containing information such as the area proportion of each desertification level, statistics on the occurrence time of desertification, and analysis of the evolution rate. This provides comprehensive and detailed desertification monitoring data for grassland ecological protection departments and related governance agencies, helping them to accurately grasp the overall situation and specific details of grassland desertification within the monitoring area, and to formulate targeted protection measures and desertification control plans. For example, vegetation restoration projects can be prioritized in severely desertified areas, while preventive measures such as fencing protection and reasonable grazing can be implemented in lightly desertified areas to maximize the protection of grassland ecosystems.

[0045] In summary, this grassland desertification multispectral remote sensing monitoring system, relying on the collaborative operation of five core modules, constructs a refined monitoring system covering the entire process. The system first completes the acquisition, preprocessing, and spatial registration of multispectral temporal remote sensing images; then, through pixel-level spectral analysis and time-series analysis, it constructs pixel-by-pixel phenological trajectories; it screens healthy data and builds a hierarchical phenological baseline model to accurately reconstruct the original growth patterns of grassland; using trajectory collaborative comparison and dynamic time warping technology, it quantifies the deviation between real-time growth trajectories and healthy baselines; combining dual-index judgment rules to classify desertification levels, it simultaneously extracts the desertification occurrence time and evolution rate, and finally outputs desertification monitoring maps and statistical reports, such as... Figure 1 As shown.

[0046] Example 2: For grassland areas affected by extreme weather events, in order to assess the degree and extent of the impact of climate events on grassland desertification, multispectral time-series remote sensing images covering the geographical area and the time span of the first full growing season after the climate event were systematically acquired. The images needed to simultaneously include historical images before the climate event and real-time images after the event, forming a complete time series chain to ensure that the long-term impact of the climate event on grassland growth could be captured through before-and-after comparisons. All images in the initial image set underwent comprehensive preprocessing, focusing on eliminating acquisition errors caused by changes in atmospheric composition due to extreme weather, residual cloud cover, and environmental interference with the sensor. At the same time, a unified preprocessing standard was applied to the images before and after the event to ensure the consistency and comparability of the two types of image data, resulting in a preprocessed multispectral time-series remote sensing image set. A georeferenced image with high spatial accuracy, taken before a climate event and free from significant interference, is selected. All preprocessed historical and real-time images are then rigorously spatially registered with this georeferenced image. Through coordinate correction and geometric alignment, all images are unified into the same geographic coordinate system, ensuring that the same pixel can be accurately corresponded in images before and after the event. This avoids misjudgments in change analysis caused by spatial location deviations. The registered time-series image set is output, providing a spatially accurate and data-reliable foundation for subsequent precise comparison of grassland pixel growth changes before and after climate events. This meets the core requirement of rapidly and accurately capturing desertification dynamics in emergency monitoring.

[0047] After receiving the registered time-series image set, a pixel-level spectral analysis algorithm is used to calculate the vegetation cover index for each pixel. This algorithm can accurately capture abnormal changes in the spectral characteristics of grassland vegetation after extreme climate events, such as decreased near-infrared reflectance and increased red reflectance after vegetation damage. Through scientific spectral analysis logic, these changes are transformed into a quantified vegetation cover index, accurately reflecting the growth status of vegetation in each pixel after the event. The calculated vegetation cover indices are integrated in chronological order into pixel-by-pixel raw time-series data of vegetation cover index. This data includes both healthy growth records before the climate event and recovery or degradation processes after the event. Due to potential interference from short-term extreme weather and observation noise, the raw time-series data needs to undergo time series preprocessing. Through smoothing and outlier removal, the long-term trend changes brought about by the climate event are highlighted, and the interference of short-term fluctuations is reduced. Time series analysis was used to fit the preprocessed time-series raw data to construct a phenological trajectory for each pixel. This trajectory clearly shows the growth changes of vegetation before and after climate events, and intuitively presents the impact of events on the vegetation growth cycle. From the constructed phenological trajectories, historical trajectories before the climate event and trajectories for the target growing season after the event were specifically extracted. Historical healthy grassland trajectories were selected based on the following health screening criteria: normal range of cover index: baseline interval [0.6, 0.95], alpine grassland [0.55, 0.90], desert grassland [0.65, 0.98]; growth stability: coefficient of variation CV ≤ 15%; anomaly differentiation: pests and diseases are characterized by local spectral mutations + abnormal peaks in the near-infrared band, and the early stage of desertification is characterized by gradual spectral changes and a continuous decline in cover index. Anomalies caused by non-climate factors that may exist in the historical trajectories were removed to obtain historical healthy grassland trajectories that can truly represent the grassland in the natural healthy state of the region. At the same time, real-time trajectories for the target growing season were extracted and integrated into a pixel-level phenological trajectory set according to spatial location, which fully records the growth and recovery of grassland after the event. Simultaneously analyze each trajectory to extract phenological parameters characterizing vegetation growth patterns, including germination delay time, peak growth intensity, growth cycle length, and premature withering. These parameters can accurately quantify the degree of interference of climate events on key stages of vegetation growth. For example, extreme drought may lead to delayed germination and reduced peak growth. Finally, a pixel-level phenological trajectory set and phenological parameters are output, providing pure historical health data for healthy grassland phenological baseline modeling. It also provides core data support for subsequent comparative analysis of the differences between grassland and healthy benchmarks after events, helping to quickly assess desertification risk.

[0048] We receive historical healthy grassland tracks after health screening. These tracks, all from before the climate event, are unaffected by the extreme weather and accurately reflect the original growth patterns of the grassland. We further screen healthy grassland pixel tracks that perfectly match the natural growth characteristics of the grassland in the region, free from human intervention and other abnormal disturbances. These tracks are stored one by one according to their spatial location, forming a basic track layer. This ensures that each pixel has its own exclusive, event-independent healthy growth benchmark, providing a pure reference standard for comparing changes after the event. Based on the track data of the basic track layer, a professional fitting optimization algorithm is used to optimize the tracks, eliminating track differences caused by natural growth fluctuations in different years. This forms standardized phenological tracks, constituting the fitting optimization layer. This layer of tracks represents the ideal healthy growth pattern under specific geographical environments and grassland types, reducing the impact of natural fluctuations on the accuracy of the benchmark and allowing subsequent comparisons to focus more on the impact of climate events. A feature mapping layer is constructed to establish a close mapping relationship between the fitted and optimized standardized phenological trajectory and the topographic features, soil type features, and grassland type features of the corresponding pixels. This fully considers the inherent influence of different topographic features, soil fertility, and grassland species on the growth trajectory, enabling the baseline model to adapt to the grassland growth characteristics of different locations within the monitoring area. This avoids misjudging normal growth differences as desertification caused by climate events due to ignoring inherent environmental differences, ensuring the scientific validity and relevance of the baseline model in emergency monitoring scenarios, and providing a reliable comparative benchmark for accurately assessing the degree of desertification caused by climate events.

[0049] The system receives real-time trajectories of the target growing season and pixel-level phenological baseline models. It then synchronizes and standardizes these trajectories with the single-pixel baseline trajectories in the baseline model along the time dimension, unifying their time nodes and data periods. This ensures that the real-time growth trajectory after an event corresponds perfectly with the healthy baseline trajectory before the event on a time scale, facilitating direct comparison of differences at the same growth stage and accurately capturing the impact of climate events. Phenological feature sequences of the real-time and baseline trajectories are extracted pixel-by-pixel from grassland pixels within the monitoring area. These sequences encompass core information such as the changing trend of vegetation cover index and the temporal distribution of key growth nodes, forming one-to-one trajectory feature data pairs. This enables refined pixel-by-pixel comparison, avoiding omissions of details in overall analysis. A trajectory collaborative similarity algorithm is used to quantify the similarity of two sets of trajectory feature data at the pixel level. This algorithm comprehensively considers the dual differences in numerical distance and changing trends, accurately calculating the deviation between the real-time trajectory and the baseline trajectory in terms of coverage index values, and also capturing inconsistencies in their growth rhythms, such as delayed growth peaks and reduced intensity in the real-time trajectory. This yields the deviation value between the real-time trajectory and the baseline trajectory at the pixel level. All pixel deviation values ​​are integrated according to geospatial location to generate a collaborative similarity distance dataset, visually presenting the distribution of the degree to which grasslands in different regions are affected by climate events. Simultaneously, the real-time trajectory and baseline model trajectory for the target growing season are refined and regularized in the temporal dimension. Dynamic temporal regularization technology is used to align the temporal nodes of the two trajectories pixel by pixel. Even if climate events disrupt the growth rhythm of the real-time trajectory, this technology can accurately match the corresponding growth stage. Based on the aligned trajectory, the deviation of each phenological parameter in the temporal characteristics is extracted in detail, quantifying it into phenological parameter offset data, which is then categorized by corresponding pixels to form a phenological parameter offset dataset. The collaborative similarity distance dataset and the phenological parameter offset dataset are output simultaneously. Figure 2 As shown, the two types of data comprehensively reflect the differences in grassland growth and health benchmarks after climate events from different dimensions, providing rich and accurate data support for the dynamic quantification of desertification.

[0050] The system receives a co-similarity distance dataset and a phenological parameter offset dataset, and performs pixel-by-pixel association matching on both datasets to ensure that the similarity distance data of each pixel accurately corresponds to the phenological parameter offset data. This eliminates analytical errors caused by data misalignment, resulting in a standardized desertification analysis dataset, providing a high-quality data foundation for rapid assessment in emergency scenarios. Based on the dynamic quantification rules of desertification and the benchmark threshold verified by 5 years of field measurements in 30 typical sample areas nationwide, the system uses co-similarity distance and phenological parameter offset as dual judgment indicators. The values ​​of both are comprehensively evaluated to classify the grassland desertification level in the monitoring area into four levels: no desertification, slight desertification, moderate desertification, and severe desertification. No desertification indicates that climate events have not significantly affected grassland growth, the real-time trajectory has minimal deviation from the benchmark, the co-similarity distance is ≥0.85, and the phenological parameter offset is ≤5%. Slight desertification indicates that the grassland has been slightly affected by desertification. Minor impacts, with small parameter offsets and distance deviations (0.70 ≤ co-similarity distance < 0.85 and 5% < phenological parameter offset ≤ 15%), still possess strong recovery capabilities. Moderate desertification indicates significant grassland degradation with substantial deviations (0.40 ≤ co-similarity distance < 0.70 and 15% < phenological parameter offset ≤ 30%), requiring timely intervention. Severe desertification indicates severe damage to grassland vegetation with extremely large deviations (co-similarity distance < 0.40 and phenological parameter offset > 30%), indicating severe desertification requiring urgent remediation. Thresholds were adjusted according to the monitoring area type: for alpine grasslands, the co-similarity distance threshold was lowered by 0.05-0.10, and the phenological parameter offset threshold was raised by 5%-8%; for desert grasslands, the co-similarity distance threshold was raised by 0.03-0.08, and the phenological parameter offset threshold was lowered by 3%-5%. The key focus is on extracting the timing of desertification following extreme weather events, clarifying whether desertification occurs immediately after the event or develops gradually during the growing season. Simultaneously, the evolution rate of different desertification levels is calculated to determine whether desertification is in a state of rapid spread, slow development, or stabilization, providing crucial dynamic information for emergency response. The system outputs a clear desertification monitoring map displaying the spatial distribution and classification of desertification levels. This map can quickly locate severely desertified areas and key pathways for desertification spread. It also outputs an emergency monitoring statistical report containing core information such as the scope of desertification impact, area of ​​each level of desertification, evolution rate, and occurrence time. This provides specific evidence for relevant departments to quickly formulate emergency management plans, such as prioritizing emergency measures like vegetation replanting and sand fixation in severely desertified areas, and implementing restoration measures like fencing and water replenishment in lightly desertified areas. This aims to minimize desertification losses caused by extreme weather, curb further desertification spread, and protect the stability of grassland ecosystems.

[0051] In summary, this embodiment addresses the emergency monitoring needs of grassland desertification following extreme weather events, developing a specialized and efficient monitoring scheme. It involves acquiring multispectral temporal remote sensing images of the monitoring area, completing preprocessing and spatial registration, and outputting a registered temporal image set; calculating the vegetation cover index, constructing pixel-by-pixel phenological trajectories, and outputting historical healthy grassland trajectories, real-time trajectories of the target growing season, and phenological parameters; constructing a pixel-level phenological baseline model; calculating trajectory co-similarity distance, extracting phenological parameter offsets, and outputting two types of datasets; classifying desertification levels, extracting desertification occurrence time and evolution rate, and outputting desertification monitoring maps and statistical reports, such as... Figure 3 As shown.

[0052] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A multispectral remote sensing monitoring system for grassland desertification, characterized in that, The system comprises the following components: a temporal image preprocessing and registration module, a pixel-level phenological trajectory construction module, a healthy grassland phenological baseline modeling module, a trajectory collaborative comparison module, and a desertification dynamic quantification module; The time-series image preprocessing and registration module acquires multispectral time-series remote sensing images of the monitoring area, performs preprocessing and spatial registration operations, and outputs a registered time-series image set. The pixel-level phenological trajectory construction module: receives the registered time-series image set, calculates the vegetation cover index using the remote sensing image pixel-level spectral analysis algorithm; constructs the phenological trajectory of each pixel using time-series analysis technology, outputs a pixel-level phenological trajectory set containing historical healthy grassland trajectories and target growing season real-time trajectories, and outputs phenological parameters at the same time; The healthy grassland phenological baseline modeling module: receives historical healthy grassland trajectories, filters healthy grassland pixel trajectories and performs fitting optimization to construct a pixel-level phenological baseline model; The trajectory collaborative comparison module receives the real-time trajectory of the target growing season and the pixel-level phenological baseline model, calculates the collaborative similarity distance between the real-time trajectory of the target growing season and the pixel-level phenological baseline model using a trajectory collaborative similarity algorithm, and extracts the offset of the phenological parameters using dynamic time warping technology, outputting a collaborative similarity distance dataset and a phenological parameter offset dataset. The desertification dynamic quantification module receives a collaborative similarity distance dataset and a phenological parameter offset dataset, classifies desertification levels using desertification dynamic quantification rules, extracts the desertification occurrence time and desertification evolution rate, and outputs desertification monitoring maps and statistical reports.

2. The grassland desertification multispectral remote sensing monitoring system according to claim 1, characterized in that, The process of acquiring multispectral time-series remote sensing images of the monitoring area in the time-series image preprocessing and registration module, performing preprocessing and spatial registration operations, and outputting the registered time-series image set is as follows: according to the geographical location of the monitoring area and the time span of the grassland growing season, multispectral time-series remote sensing images covering the monitoring area are acquired to form an initial multispectral time-series remote sensing image set. Each image in the initial multispectral time-series remote sensing image set is preprocessed to eliminate errors generated during image acquisition, resulting in a preprocessed multispectral time-series remote sensing image set. A geographic reference image is selected, and all images in the preprocessed multispectral time-series remote sensing image set are spatially registered with the geographic reference image to unify all images into the same geographic coordinate system, resulting in a registered time-series image set.

3. The grassland desertification multispectral remote sensing monitoring system according to claim 1, characterized in that, The calculation formula for the remote sensing image pixel-level spectral analysis algorithm in the pixel-level phenological trajectory construction module is as follows: ,in, The vegetation cover index; This is the normalization adjustment coefficient for pixel-level spectral analysis; To monitor the near-infrared reflectance of individual pixels within the area; To monitor the red band reflectance of the same single pixel within the monitoring area; To monitor the blue band reflectance of the same single pixel within the monitoring area; This is the atmospheric scattering correction factor for the red band; This is the atmospheric scattering correction factor for the blue band; This is the soil background adjustment coefficient.

4. The grassland desertification multispectral remote sensing monitoring system according to claim 1, characterized in that, The pixel-level phenological trajectory construction module uses time series analysis technology to construct the phenological trajectory of each pixel, and outputs a set of pixel-level phenological trajectories containing historical healthy grassland trajectories and real-time trajectories of the target growing season. The process of outputting phenological parameters is as follows: the calculated vegetation cover index is integrated into the time series raw data of the pixel-by-pixel vegetation cover index; the time series raw data is preprocessed using time series analysis technology; and the preprocessed time series raw data is fitted to construct the phenological trajectory of each pixel. Historical phenological trajectories and target growing season trajectories are extracted from all constructed phenological trajectories. Historical healthy grassland trajectories are obtained after health screening. Real-time trajectories of the target growing season are extracted and integrated into a pixel-level phenological trajectory set. At the same time, phenological parameters representing vegetation growth patterns are extracted from each trajectory. A pixel-level phenological trajectory set containing historical healthy grassland trajectories and real-time trajectories of the target growing season is output, along with phenological parameters.

5. The grassland desertification multispectral remote sensing monitoring system according to claim 1, characterized in that, The pixel-level phenological baseline modeling module for healthy grassland is a hierarchical structured model, specifically including a basic trajectory layer, a fitting optimization layer, and a feature mapping layer. The basic trajectory layer is a set of historical healthy grassland pixel trajectories that have been retained after health screening and conform to the original growth pattern of grassland, and is stored one by one according to the spatial location of the pixels. The fitting optimization layer forms a standardized phenological trajectory based on the trajectory data of the basic trajectory layer after fitting optimization; the feature mapping layer establishes a mapping relationship between the fitted and optimized standardized phenological trajectory and the terrain, soil and grassland type features of the corresponding pixels.

6. The grassland desertification multispectral remote sensing monitoring system according to claim 1, characterized in that, The specific steps of the trajectory collaborative comparison module in calculating the collaborative similarity distance between the real-time trajectory of the target growing season and the pixel-level phenological baseline model using the trajectory collaborative similarity algorithm are as follows: The real-time trajectory of the target growing season and the single-pixel reference trajectory in the pixel-level phenological baseline model are synchronized and normalized according to the time dimension to unify the trajectory data dimension; Phenological feature sequences of the two sets of trajectories are extracted pixel-by-pixel based on phenological parameters for grassland pixels within the monitoring area. The phenological feature sequences are feature sequences formed by arranging phenological parameters according to the time dimension, forming one-to-one corresponding trajectory feature data; The similarity metric of the two sets of trajectory feature data at the pixel level is quantitatively calculated using the trajectory collaborative similarity algorithm to obtain the deviation value between the real-time trajectory and the reference trajectory at the pixel level; Finally, the deviation values ​​of all pixels are integrated according to their geographic spatial location to generate a collaborative similarity distance dataset.

7. The grassland desertification multispectral remote sensing monitoring system according to claim 6, characterized in that, The calculation formula for the trajectory collaborative similarity algorithm in the trajectory collaborative comparison module is as follows: ,in, For the first Trajectory co-similarity distance of each pixel; The numbering of individual grassland pixels within the monitoring area; For collaborative weighting coefficients; For the first The Euclidean distance between the real-time trajectory of each pixel and the baseline trajectory; To determine the maximum Euclidean distance of all pixels within the monitoring area; For the first Cosine similarity between the real-time trajectory of each pixel and the baseline trajectory.

8. The grassland desertification multispectral remote sensing monitoring system according to claim 1, characterized in that, The process of extracting the offset of phenological parameters and outputting the collaborative similarity distance dataset and phenological parameter offset dataset in the trajectory collaborative comparison module is as follows: Temporal dimension warping is performed on the real-time trajectory of the target growing season and the pixel-level phenological baseline model trajectory; the temporal nodes of the two trajectories are aligned pixel by pixel using dynamic time warping technology; based on the aligned trajectories, the deviation of each phenological parameter in temporal features is extracted and quantified to form phenological parameter offset data; the calculated collaborative similarity distance is organized according to geospatial association to generate a collaborative similarity distance dataset, and the phenological parameter offset data is classified according to corresponding pixels to form a phenological parameter offset dataset; the collaborative similarity distance dataset and the phenological parameter offset dataset are output synchronously.

9. The grassland desertification multispectral remote sensing monitoring system according to claim 1, characterized in that, The specific rules for classifying desertification levels using the dynamic quantification rules in the desertification dynamic quantification module are as follows: receiving a collaborative similarity distance dataset and a phenological parameter offset dataset, performing pixel-by-pixel association matching on the two datasets to obtain a standardized desertification analysis dataset; classifying desertification levels according to the dynamic quantification rules, using collaborative similarity distance and phenological parameter offset as the judgment indicators, the specific desertification levels are: no desertification, light desertification, moderate desertification, and severe desertification.

10. A method for multispectral remote sensing monitoring of grassland desertification, applicable to the multispectral remote sensing monitoring system for grassland desertification as described in any one of claims 1-9, characterized in that, The specific steps of this method are as follows: S100, Temporal Image Preprocessing and Registration: Multispectral temporal remote sensing images are collected according to the monitoring area and the span of the growing season. After preprocessing to eliminate errors, they are spatially registered with the geographic reference image to output a temporal image set with a unified coordinate system. S200, Pixel-level phenological trajectory construction: The vegetation cover index is calculated based on the pixel-level spectral analysis algorithm of remote sensing images. The phenological trajectory of each pixel is constructed through time series analysis technology. The output is a set of pixel-level phenological trajectories containing historical healthy grassland trajectories and real-time trajectories of the target growing season, and the phenological parameters are also output. S300, Healthy Grassland Phenological Baseline Modeling: Screen healthy grassland pixel trajectories and fit and optimize them to construct a hierarchical pixel-level phenological baseline model containing a basic trajectory layer, a fitting and optimization layer and a feature mapping layer. S400, Trajectory Collaborative Comparison: Time-normalize the real-time trajectory of the target growing season with the baseline model trajectory, calculate the collaborative similarity distance through the trajectory collaborative similarity algorithm, extract the phenological parameter offset after dynamic normalization, and output two types of datasets; S500, Desertification Dynamic Quantification: It associates and matches two types of datasets, and classifies them into four levels: no desertification, light desertification, moderate desertification and severe desertification, based on the indicators of collaborative similarity distance and phenological parameter offset. It extracts the desertification occurrence time and desertification evolution rate, and outputs desertification monitoring maps and statistical reports.

Images