Ecological quality monitoring method based on multi-source data and random forest
Patent Information
- Application Number
- CN202611486298.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-23
- Publication Date
- 2026-10-27
AI Technical Summary
现有研究虽已涉及动态变化、驱动机制及修复利用等方面，但长时序、高精度动态监测体系仍有待深化
本发明构建的“GEE+Colab+GIS”多源遥感与人工智能融合方法，在实践中生态质量呈“先退化→稳步恢复→加速改善”三阶段演变，与气候波动及保护工程时间节点吻合。实现了盐碱湿地生态质量的长时序、定量化与空间显式监测，具有良好的可重复性和可验证性。如今残存的重度盐碱化区域仍是未来修复治理的重点，本发明可为松嫩平原西部盐碱湿地保护与修复成效评估提供时空数据支撑和方法参考。
Smart Images
Figure CN122990523A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of saline-alkali wetland ecological quality monitoring technology, specifically, it relates to an ecological quality monitoring method based on multi-source data and random forest. Background Technology
[0002] The western part of the Songnen Plain is one of the world's three major concentrated distribution areas of soda saline-alkali land. Saline-alkali wetlands play important ecological functions such as water conservation, climate regulation, and biodiversity maintenance. Under the dual pressure of climate change and human activities, wetlands in this region have long faced problems such as vegetation degradation, salt accumulation, and shrinkage. Conducting dynamic monitoring of ecological quality is of great significance for assessing the degree of degradation and diagnosing the effectiveness of restoration.
[0003] Research on saline-alkali wetlands began earlier abroad. Sharma and Naik systematically described the structure and function of inland saline-alkali wetlands and summarized their global distribution and evolution characteristics. Sorensen investigated the vegetation distribution of saline-alkali wetlands in eastern Nebraska, USA; Partnership SWC summarized the conservation practices in this region. Herrero and Castaneda established a monitoring index system for small saline-alkali wetlands. Sim et al. constructed a model of salinity-driven ecological state transition; Shokri et al. analyzed the combined impact of human activities and natural factors. Khan et al. used remote sensing indicators to assess the hydrological salinization degradation process. While international research has extensive experience in monitoring techniques and ecological patterns, significant regional differences make it difficult to directly apply the results, and the monitoring system that deeply integrates multi-source remote sensing and AI still needs improvement.
[0004] Domestic scholars have conducted extensive research on saline-alkali wetlands in the western Songnen Plain, including but not limited to: identifying wetland disturbance factors and their feedback mechanisms; exploring the desalination and improvement mechanism of rice cultivation; and proposing salinization control strategies. A "rice-reed-fish" composite ecological model has been constructed; the impact of saline-alkali water environment on aquaculture ecological engineering has been analyzed using the Niuxintao wetland as an example. Heavy metal analysis of saline-alkali wetland soils has been carried out; wetland resource protection measures have been summarized; wetland dynamic evolution and its ecological effects have been analyzed; and soil salinization has been monitored using remote sensing. While existing research has covered dynamic changes, driving mechanisms, and remediation and utilization, a long-term, high-precision dynamic monitoring system still needs further development. Summary of the Invention
[0005] This invention integrates multi-temporal remote sensing data, multi-dimensional ecological indices, and random forest algorithms to construct a multi-dimensional feature space. It proposes an ecological quality monitoring method based on multi-source data and random forest, which is applicable to the dynamic monitoring of ecological quality in saline-alkali wetlands in the western Songnen Plain. This aims to provide new ideas and solutions for regional wetland protection and restoration assessment.
[0006] This invention is achieved through the following technical solution: An ecological quality monitoring method based on multi-source data and random forest, the method specifically includes the following steps: S1. Data Acquisition and Preprocessing: Acquire long-term satellite remote sensing data and related auxiliary data for the target area, preprocess the remote sensing images, and obtain a standardized multi-temporal remote sensing dataset covering the target area. S2. Construct a multidimensional feature dataset: Extract multiple types of remote sensing ecological indices from the preprocessed remote sensing images and construct a multidimensional feature dataset for ecological quality assessment; S3. Training the random forest model: Based on the multidimensional feature dataset, using the remote sensing ecological indices corresponding to the samples as input variables and the ecological quality level as the target variable, train a random forest classification model. S4. Spatiotemporal variation analysis of ecological quality: The trained random forest model is applied to feature images of different periods to generate spatial distribution maps of ecological quality levels for each period. Based on the spatial distribution results, the spatiotemporal evolution characteristics of ecological quality in the target area are analyzed.
[0007] Further, in S1, the preprocessing includes: S1.1. Image Filtering: Filter available remote sensing images according to the set time range and cloud cover threshold; S1.2. Image Compositing: Median composite is performed on the selected multiple images to reduce the impact of thin clouds, cloud shadows and atmospheric aerosols on image quality. S1.3. Target Area Cropping: Cropping the composite image based on the vector boundaries of the target area; S1.4. Unified Spatial Reference: Unify the spatial coordinate system and spatial resolution of all images to complete the spatial registration of multi-temporal images.
[0008] Furthermore, in S2, the various remote sensing ecological indices include: vegetation cover, soil salinity, water body index, and surface environment. The vegetation cover category includes the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EDI); the soil salinity category includes the Salinity Index; the water body index category includes the Normalized Difference Water Body Index (NDDI) and the Improved Normalized Difference Water Body Index (EDI); and the surface environment category includes surface temperature and the Bare Soil Index.
[0009] Furthermore, in S3, a confusion matrix is used to verify the model classification accuracy, and the overall accuracy and Kappa coefficient are used as evaluation indicators of the overall reliability of the model. At the same time, the user accuracy and producer accuracy corresponding to each ecological quality level are calculated. After the random forest model is trained, the feature importance of each input remote sensing ecological index is calculated and ranked to identify the core features that contribute the most to the classification of ecological quality.
[0010] Furthermore, in S4, the spatiotemporal evolution characteristic analysis includes: statistically analyzing the area and proportion of different ecological quality levels in each period, and analyzing the area change trend and evolution trajectory of each level within the complete monitoring period.
[0011] Furthermore, in S4, the spatiotemporal evolution characteristic analysis also includes: spatially superimposing the spatial distribution results of the ecological quality level in the first and last periods of the monitoring cycle to obtain the transition matrix of the ecological quality level, and statistically analyzing the total area and proportion of ecological quality improvement and degradation, as well as the main evolution paths.
[0012] An ecological quality monitoring system based on multi-source data and random forest; The system includes a data acquisition and preprocessing module, a multi-dimensional feature construction module, a random forest model training module, and an ecological quality spatiotemporal change analysis module. The data acquisition and preprocessing module acquires long-term satellite remote sensing data and related auxiliary data of the target area, preprocesses the remote sensing images, and obtains a standardized multi-temporal remote sensing dataset covering the target area. The multidimensional feature construction module is used to extract multiple types of remote sensing ecological indices from preprocessed remote sensing images and construct a multidimensional feature dataset for ecological quality assessment. The random forest model training module trains a random forest classification model based on the multidimensional feature dataset, using each remote sensing ecological index corresponding to the sample as input variables and the ecological quality level as the target variable. The ecological quality spatiotemporal change analysis module is used to apply the trained random forest model to feature images of different periods, generate spatial distribution maps of ecological quality levels for each period, and analyze the spatiotemporal evolution characteristics of ecological quality in the target area based on the spatial distribution results.
[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: The "GEE+Colab+GIS" multi-source remote sensing and artificial intelligence fusion method constructed in this invention demonstrates a three-stage evolution of ecological quality in practice: "initial degradation → steady recovery → accelerated improvement," which aligns with climate fluctuations and the timeline of conservation projects. It achieves long-term, quantitative, and spatially explicit monitoring of the ecological quality of saline-alkali wetlands, exhibiting good repeatability and verifiability. The remaining severely salinized areas remain a key focus for future restoration and management. This invention can provide spatiotemporal data support and methodological reference for evaluating the effectiveness of saline-alkali wetland protection and restoration in the western Songnen Plain. Attached Figure Description
[0014] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2A ranking graph of feature importance; Figure 3 This is a map showing the area change of ecological quality level in the study area from 2005 to 2025 in the example. Detailed Implementation
[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0016] Unless otherwise specified, the experimental methods used in the following examples are conventional methods. Unless otherwise specified, the materials, reagents, methods, and instruments used are all conventional materials, reagents, methods, and instruments in the art, and can be obtained commercially by those skilled in the art.
[0017] The technical approach of this invention can be summarized as "data acquisition - feature extraction - model training - spatiotemporal analysis". First, Landsat long-term satellite remote sensing data is acquired, and image preprocessing, remote sensing feature index extraction, and preliminary analysis are completed based on the Google Earth Engine cloud platform. Then, a multidimensional feature dataset is constructed, and model training and accuracy verification are conducted using the Colab platform (using the Scikit-learn library). The confusion matrix is calculated, and the model reliability is evaluated using overall accuracy and Kappa coefficient. Finally, thematic mapping is performed using ArcMap, and the area and proportion of different levels at each period are statistically analyzed to determine spatiotemporal trends. The results show that the "GEE+Colab+GIS" technical approach developed in this invention can support the assessment of protection and restoration of saline-alkali wetlands in the Songnen Plain.
[0018] The specific technology roadmap is as follows: Figure 1 As shown. In this embodiment, the saline-alkali wetland in the western part of the Songnen Plain is taken as an example: 1. Remote Sensing Index Calculation The core principle of this invention is to construct a multi-dimensional feature space using multi-source and multi-temporal remote sensing data, and select three categories of seven remote sensing ecological indices as feature variables for ecological quality evaluation (Table 1).
[0019] Table 1 Summary of Remote Sensing Feature Indices
[0020] It is in the near-infrared band; Red band; Green band; Blue band; It is the shortwave infrared band 1; The brightness temperature in the thermal infrared band; Random Forest (RF) is an ensemble learning algorithm based on decision trees. It improves the accuracy and robustness of a model by constructing multiple decision trees and combining their predictions. RF randomly samples with replacement from the original data to form multiple subsets for building decision trees. During tree construction, a certain number of features are randomly selected from the sample features each time to split the nodes. Finally, the average of the results from each tree is used to obtain the final result. Its core advantage lies in its ability to effectively handle high-dimensional data, prevent overfitting, and be insensitive to missing and outlier values. RF increases the diversity between trees through "random sampling" and "random feature selection," thereby improving generalization ability.
[0021]
[0022] In the formula, Represents each decision tree, Indicates the number of decision trees. This indicates the final result.
[0023] 2. Overview of the study area and data sources 2.1 Overview of the Study Area This invention comprehensively considers the impact range and geomorphological integrity of wetland-salinization in the Holingol River Basin, taking the entire area of Da'an City as the study area, with geographical coordinates between 123°08′-124°35′ east longitude and 44°50′-45°45′ north latitude, covering a total area of approximately 4879 km². The study area is located on the southwestern edge of the Songnen Plain, with a temperate continental monsoon climate. The average annual precipitation is 400-450 mm, the average annual temperature is 4-6℃, and the average annual evaporation is as high as 1500-1800 mm, far exceeding precipitation, resulting in severe water deficit and exacerbating soil salinization. The terrain is low-lying and flat, with an elevation of 120-160 m, and the landforms are mainly alluvial plains and lacustrine depressions. The Nenjiang River and its tributaries, the Tao'er River and the Holingol River, traverse the entire area, forming large areas of floodplain wetlands and lacustrine wetlands. Niuxintaobao National Wetland Park is located in the south-central part of Da'an City. It is the best-preserved floodplain reed marsh wetland in the Huolin River Basin, covering an area of approximately 4,200 hectares, of which about 3,200 hectares are wetlands. The park and its surrounding areas are severely affected by soil salinization, exhibiting a typical "water-salt-vegetation" gradient pattern from healthy reed marshes to saline-alkali bare land, making it an ideal area for conducting dynamic monitoring and research on the ecological quality of saline-alkali wetlands.
[0024] 2.2 Data Source 2.2.1 Satellite remote sensing data The study used Landsat satellite remote sensing imagery as the primary data source (Table 2), selecting growing season (June-September) images from five periods: 2005, 2010, 2015, 2020, and 2025. Landsat 5 TM data was used for 2005 and 2010, Landsat 8 OLI data for 2015 and 2020, and Landsat 9 OLI-2 data for 2025.
[0025] Table 2 Overview of satellite remote sensing data used in this invention
[0026] 2.2.2 Auxiliary Data (1) High-resolution historical images of Google Earth: used for visual interpretation and grade labeling of ecological quality sample points in various periods.
[0027] (2) Vector boundary of county-level administrative divisions in the study area: used as an auxiliary reference for the production of location maps and spatial analysis of the study area.
[0028] (3) Annual precipitation data of meteorological stations in western Jilin Province: used for analysis of the driving forces of ecological quality changes.
[0029] 2.3 Data Preprocessing All remote sensing data preprocessing was performed on the Google Earth Engine cloud platform, mainly including the following steps: (1) Image screening: For each target year, all available images covering the study area from June 1 to September 30 are screened, and the cloud cover threshold is set to be less than 10%.
[0030] (2) Image synthesis: Median synthesis is carried out on the selected multiple images to effectively reduce the impact of residual thin clouds, cloud shadows and atmospheric aerosols. The median synthesis method has good stability performance for outliers and can preserve the actual spectral characteristics of ground features while ensuring data quality.
[0031] (3) Study area cropping: All composite images are precisely cropped according to the vector boundary of the study area, leaving only the data covered by the study area.
[0032] (4) Unified spatial reference: All images adopt the WGS84 geographic coordinate system and 30 m spatial resolution to ensure that images from different periods are correctly registered in spatial location.
[0033] 3. Results and Analysis 3.1 Model Construction and Parameter Setting This invention uses seven remote sensing feature values (NDVI, EVI, SI, NDWI, MNDWI, LST, BI) of sample points as input variables X, and manually labeled ecological quality levels (0 / 1 / 2 / 3) as target variables y, and uses Colab for model training.
[0034] The core parameters of the model are set as follows: number of decision trees n_estimators=100, depth of a single tree is unlimited, number of features randomly selected when splitting a node is sqrt(n_features), and the Gini coefficient is used as the splitting criterion. The random seed random_state is fixed at 42 to ensure the reproducibility of experimental results.
[0035] 3.2 Model Accuracy Evaluation 3.2.1 Classification accuracy of the full-feature model The confusion matrix of the full-feature set random forest model on the test set (30% of the total samples) is shown in Table 3. The overall accuracy of the full-feature model is 96.43%, and the Kappa coefficient is 0.938. Among them, the classification accuracy of "Excellent" and "Medium" grades is the highest, with user accuracy of 97.56% and 98.63% respectively, and producer accuracy above 95%. There are a few misclassifications in the "Good" grade, mainly because the NDVI and SI gradients between slightly degraded meadows, healthy wetland vegetation, and moderately degraded grasslands are continuous, and the boundary areas are prone to ambiguity. Due to the small sample size of the test set (only 3 samples), the user accuracy of the "Poor" grade is low (66.67%), but its producer accuracy is 100%, indicating that all true "Poor" grade samples were successfully identified without any omissions.
[0036] Table 3. Confusion Matrix of Full Feature Model
[0037] 3.2.2 Experimental Results of Feature Optimization The accuracy comparison of the two feature sets is shown in Table 4. Compared with the simplified model that only uses NDVI and NDWI, the full feature set model improves the overall accuracy by 11.51% and the Kappa coefficient by 0.194.
[0038] In the simplified model, the "good" rating exhibited significant confusion (7 samples were misclassified as "excellent" and 11 as "moderate"), indicating that vegetation greenness and humidity information alone cannot effectively distinguish between slightly saline meadows and healthy wetland vegetation. While their NDVI and NDWI values are similar, there are significant differences in soil salinity and bare soil exposure. By introducing salinity index, bare soil index, and surface temperature, the model can capture these key environmental gradients, significantly improving its ability to identify intermediate transitional ratings.
[0039] Table 4. Accuracy Comparison of Models with Different Feature Combinations
[0040] 3.2.3 Feature Importance Analysis Feature importance ranking (e.g.) Figure 2 The data shows that the Bare Soil Index (BI) contributes the most at 26.63%, indicating that soil exposure is more discriminative than vegetation greenness in the assessment of saline-alkali wetland ecological quality. This confirms the important ecological indicative value of the "Land Stress Index" in the HJ 192-2015 assessment framework. The combined contribution of NDVI and EVI exceeds 42%, indicating that vegetation cover remains a core dimension for ecological quality assessment. The Salinity Index (SI) contributes 8.24%, lower than the top three, but still significantly higher than zero, indicating that salinity information is a key supplementary feature for distinguishing between "good" and "medium" levels. Land Surface Temperature (LST) contributes only 1.88%, mainly because it is highly correlated with bare soil exposure; most thermal environment information has been captured by BI, weakening the independent contribution of LST.
[0041] 3.3 Spatiotemporal variation analysis of ecological quality The trained random forest model was applied to feature images from five periods: 2005, 2010, 2015, 2020, and 2025, to obtain spatial distribution maps of ecological quality levels in the study area during each period.
[0042] Table 5 shows the area and proportion of different ecological quality levels at different periods. Figure 3 As shown, the ecological quality of the study area underwent significant phased changes between 2005 and 2025.
[0043] The area rated "excellent" followed a trajectory of "sharp decline - recovery - leap": it was 298.31 km² (6.20%) in 2005, then plummeted to 41.14 km² (0.84%) in 2010, a drop of 86.2%; it then gradually recovered, rising to 275.44 km² in 2015, leaping to 1071.27 km² in 2020, and reaching a peak of 2046.97 km² (41.98%) in 2025, a net increase of 1748.66 km² over 20 years, an increase of 586%, with particularly significant growth between 2020 and 2025.
[0044] Correspondingly, the area of "medium" vegetation cover first increased and then decreased: it was 2915.06 km² (60.60%) in 2005, rose to a peak of 4121.00 km² (84.52%) in 2010, and then continued to decline to 1295.99 km² (26.58%) in 2025, a cumulative decrease of 2825.01 km², a decrease of 68.5%, reflecting a systematic improvement in vegetation cover.
[0045] The area rated "good" tends to stabilize amid fluctuations: it was 1260.41 km² (26.20%) in 2005, then plummeted to 444.81 km² (9.12%) in 2010, a decrease of 64.7%, before gradually recovering to 1318.79 km² (27.05%) in 2025, basically the same as in 2005.
[0046] The area classified as "poor" decreased overall, from 336.93 km² (7.00%) in 2005 to a low of 156.74 km² (3.22%) in 2020, a decrease of 53.5%, and then slightly rebounded to 214.87 km² (4.41%) in 2025.
[0047] Table 5. Area and Proportion of Each Ecological Quality Level in the Study Area, 2005-2025
[0048] Based on the spatial overlay analysis of the ecological quality classification maps of 2005 and 2025, the transition matrix of each level over the 20 years was calculated (Table 6).
[0049] The ecological quality of the study area has been predominantly positive over the past 20 years. The total area of improvement is approximately 2918.49 km², accounting for 60.7% of the total effective area, of which 945.87 km² has improved from "moderate to good". 2 “Central → Excellent” (863.46 km) 2 ) and "Good → Excellent" (850.18km) 2 The three routes, totaling 2659.51 km², are the main contributors to the improvement.
[0050] Furthermore, the total area classified as "poor → average," "poor → good," and "poor → excellent" was 259.98 km², indicating that some severely salinized areas have also improved. The degraded area was only about 286.49 km², accounting for 6.0%, mainly concentrated along the "average → poor," "good → average," and "good → poor" pathways. The degraded area at the "excellent" level was only 32.51 km². 2 This indicates that healthy wetland vegetation has strong stability. Overall, the area improved over 20 years is approximately 10.2 times the area degraded, showing an evolutionary pattern of "large-scale restoration, localized degradation, and core stability".
[0051] Table 6. Ecological Quality Level Transition Matrix (km²) from 2005 to 2025
[0052] 4. Conclusion The research results show that: (1) The overall classification accuracy of the full-feature random forest model reached 96.43%, and the Kappa coefficient was 0.938. Compared with the simplified model that only uses NDVI and NDWI, the accuracy was improved by 11.51% and 19.4%, respectively.
[0053] (2) Feature importance analysis showed that the bare soil index (BI) contributed the most, accounting for 26.63%, followed by NDVI and EVI, accounting for 23.84% and 18.74% respectively. This indicates that the introduction of multidimensional features such as salinity index, bare soil index and surface temperature can significantly improve the classification accuracy of saline-alkali wetland ecological quality.
[0054] (3) From 2005 to 2025, the overall ecological quality of the study area showed a significant improvement trend. Among them, the area with "excellent" quality increased from 298.31 km² to 2046.97 km², an increase of 586%; the area with "medium" quality decreased from 2915.06 km² to 1295.99 km². 2 The decline was 55.5%; the area of positive succession reached 2918.49 km², while the area of degradation was only 286.49 km². (4) Over the past 20 years, the area of positive succession was approximately 2918.49 km², accounting for more than 60% of the total effective area; the area of degradation was only 286.49 km². The evolution of ecological quality exhibits a three-stage characteristic of "degradation first, steady recovery, and accelerated improvement", which corresponds well with the regional climate dryness and wetness fluctuations and the key implementation nodes of the Niuxintao Moisture Retaining Area Protection Project.
[0055] (5) The "GEE+Colab+GIS" multi-source remote sensing and artificial intelligence fusion method constructed in this invention realizes long-term, quantitative and spatial explicit monitoring of the ecological quality of saline-alkali wetlands, and has good repeatability and verifiability. This study can provide spatiotemporal data support and methodological reference for the assessment of the effectiveness of protection and restoration of saline-alkali wetlands in western Songnen Plain. The remaining severely salinized areas are still the focus of future restoration and management.
[0056] The western part of the Songnen Plain is one of the three major concentrated areas of soda saline-alkali land in the world. This embodiment focuses on Da'an City, using five periods of Landsat satellite remote sensing images from 2005, 2010, 2015, 2020, and 2025 on the GEE platform. A multi-dimensional feature set containing seven ecological indices, including vegetation, salinity, water quality, and surface temperature, is constructed. A random forest algorithm is then used to dynamically monitor the ecological quality (excellent, good, moderate, and poor).
[0057] A computer device system includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-described method. A computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.
[0058] A computer program product includes a computer program / instructions that, when executed by a processor, implement the steps of the method described above.
[0059] The above provides a detailed description of the ecological quality monitoring method based on multi-source data and random forest proposed in this invention, and elucidates the principles and implementation methods of this invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.
Claims
1. An ecological quality monitoring method based on multi-source data and random forest, characterized in that: The method specifically includes the following steps: S1. Data Acquisition and Preprocessing: Acquire long-term satellite remote sensing data and related auxiliary data for the target area, preprocess the remote sensing images, and obtain a standardized multi-temporal remote sensing dataset covering the target area. S2. Construct a multidimensional feature dataset: Extract multiple types of remote sensing ecological indices from the preprocessed remote sensing images to construct a multidimensional feature dataset for ecological quality assessment; the multiple types of remote sensing ecological indices include: vegetation cover, soil salinity, water body index, and surface environment. S3. Training the random forest model: Based on the multidimensional feature dataset, using the remote sensing ecological indices corresponding to the samples as input variables and the ecological quality level as the target variable, train a random forest classification model. S4. Spatiotemporal variation analysis of ecological quality: The trained random forest model is applied to feature images of different periods to generate spatial distribution maps of ecological quality levels for each period. Based on the spatial distribution results, the spatiotemporal evolution characteristics of ecological quality in the target area are analyzed.
2. The method according to claim 1, characterized in that: In S1, the preprocessing includes: S1.
1. Image Filtering: Filter available remote sensing images according to the set time range and cloud cover threshold; S1.
2. Image Compositing: Median composite is performed on the selected multiple images to reduce the impact of thin clouds, cloud shadows and atmospheric aerosols on image quality. S1.
3. Target Area Cropping: Cropping the composite image based on the vector boundaries of the target area; S1.
4. Unified Spatial Reference: Unify the spatial coordinate system and spatial resolution of all images to complete the spatial registration of multi-temporal images.
3. The method according to claim 2, characterized in that: In S2, the vegetation cover category includes the normalized vegetation index and the enhanced vegetation index; the soil salinity category includes the salinity index; the water body index category includes the normalized differential water body index and the improved normalized differential water body index; and the surface environment category includes surface temperature and the bare soil index.
4. The method according to claim 3, characterized in that: In S3, a confusion matrix is used to verify the classification accuracy of the model. The overall accuracy and Kappa coefficient are used as evaluation indicators of the overall reliability of the model. At the same time, the user accuracy and producer accuracy corresponding to each ecological quality level are calculated. After the random forest model is trained, the feature importance of each input remote sensing ecological index is calculated and ranked to identify the core features that contribute the most to the classification of ecological quality.
5. The method according to claim 4, characterized in that: In S4, the spatiotemporal evolution characteristic analysis includes: statistically analyzing the area and proportion of different ecological quality levels in each period, and analyzing the area change trend and evolution trajectory of each level within the complete monitoring period.
6. The method according to claim 5, characterized in that: In S4, the spatiotemporal evolution characteristic analysis also includes: spatially superimposing the spatial distribution results of ecological quality levels at the beginning and end of the monitoring period to obtain the transition matrix of ecological quality levels, and statistically analyzing the total area and proportion of ecological quality improvement and degradation, as well as the main evolution paths.
7. An ecological quality monitoring system based on multi-source data and random forest, characterized in that: The system is used to execute the ecological quality monitoring method based on multi-source data and random forest as described in any one of claims 1 to 6; The system includes a data acquisition and preprocessing module, a multi-dimensional feature construction module, a random forest model training module, and an ecological quality spatiotemporal change analysis module. The data acquisition and preprocessing module acquires long-term satellite remote sensing data and related auxiliary data of the target area, preprocesses the remote sensing images, and obtains a standardized multi-temporal remote sensing dataset covering the target area. The multidimensional feature construction module is used to extract multiple types of remote sensing ecological indices from preprocessed remote sensing images and construct a multidimensional feature dataset for ecological quality assessment. The various remote sensing ecological indices include: vegetation cover, soil salinity, water body index, and surface environment; The random forest model training module trains a random forest classification model based on the multidimensional feature dataset, using each remote sensing ecological index corresponding to the sample as input variables and the ecological quality level as the target variable. The ecological quality spatiotemporal change analysis module is used to apply the trained random forest model to feature images of different periods, generate spatial distribution maps of ecological quality levels for each period, and analyze the spatiotemporal evolution characteristics of ecological quality in the target area based on the spatial distribution results.
8. A computer device system, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.
9. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 6.
10. A computer program product comprising a computer program / instructions, characterized in that, When executed by a processor, the computer program instructions implement the steps of the method according to any one of claims 1 to 6.