XGBoost-based ecological quality dynamic evaluation system and method

Abstract

The invention belongs to the technical field of ecological quality assessment, and belongs to an XGBoost-based ecological quality dynamic assessment system and method, and the system comprises an ecological data analysis module which is used for dividing corresponding ecological types for multi-source ecological data to obtain an ecological type distribution diagram, and obtaining ecological characteristic data according to the influence of different ecological types on an ecological system. And the model construction and optimization module is used for extracting remote sensing ecological indexes and ecological influence factors from the multi-source ecological data and constructing an ecological quality evaluation model. And the index system construction module is used for constructing an ecological quality index system. The dynamic analysis and evaluation module is used for inputting the ecological characteristic data in different periods into an ecological quality evaluation model, outputting and obtaining an ecological quality evaluation result, and performing analysis in combination with an ecological quality index system to obtain a dynamic quality evaluation result; according to the method, refined space-time dynamic monitoring and attribution analysis can be supported, and the dynamic property and decision support capability of evaluation are remarkably improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of ecological quality evaluation, and in particular to an ecological quality dynamic evaluation system and method based on XGBoost. BACKGROUND

[0002] Ecological quality is the core basis of regional sustainable development, and its dynamic change is directly related to ecosystem service function, human survival environment and social and economic coordinated development. Precise grasp of the current situation and evolution trend of ecological quality has key supporting significance for ecological protection red line demarcation, environmental governance decision and regional planning optimization.

[0003] At present, ecological quality evaluation mainly relies on traditional remote sensing indexes (such as remote sensing ecological index RSEI) or simple statistical models, which can realize preliminary characterization of ecological state in a large range, but there are significant technical limitations. On the one hand, traditional evaluation methods mainly focus on static analysis of a single time node, and cannot fully integrate natural geographical, climatic, human activity and other multi-source heterogeneous data. Moreover, they cannot be accurately adapted to the characteristics of different ecological types, so that the complex nonlinear relationship of the ecological system cannot be captured, the evaluation results are biased, and the high-precision evaluation demand cannot be met. On the other hand, the existing technology lacks a systematic dynamic tracking mechanism, and can only output isolated ecological quality values, which is difficult to realize effective monitoring and prediction of the time sequence change trend of ecological quality, and cannot analyze the causes of change with driving factors and dynamic benchmarks, so that the evaluation results are difficult to support targeted ecological protection and management decisions. With the complication of ecological environment problems and the intensification of human activity interference, the traditional evaluation method has been unable to meet the real demand for precise evaluation, dynamic tracking and scientific prediction.

[0004] Therefore, it is urgent to build an ecological quality dynamic evaluation system and method based on XGBoost, break through the static and low-precision bottleneck of traditional technology, realize the precise evaluation, dynamic tracking and trend prediction of ecological quality, and provide comprehensive and scientific decision basis for ecological environment management. SUMMARY

[0005] In order to make up for the shortcomings of the prior art, the ecological quality dynamic evaluation system and method based on XGBoost are provided. The present application is mainly used to solve the problem that the existing evaluation system is not accurate in evaluating the ecological environment state, and lacks dynamic tracking and trend prediction of ecological quality.

[0006] The technical scheme adopted by the application to solve its technical problems is: the ecological quality dynamic evaluation system based on XGBoost provided by the application comprises: The ecological data parsing module is used to collect multi-source ecological data of the target assessment area, divide the multi-source ecological data into corresponding ecological types to obtain an ecological type distribution map, and obtain ecological characteristic data based on the impact of different ecological types on the ecosystem.

[0007] The model building and optimization module is used to standardize multi-source ecological data to obtain an ecological dataset, extract remote sensing ecological indices and ecological impact factors from it, and input them into an ecological quality monitoring model built based on the XGBoost regression algorithm for optimization to obtain an ecological quality assessment model.

[0008] The indicator system construction module is used to build an ecological quality indicator system based on ecological characteristic data and ecological datasets. The ecological quality indicator system includes ecological status indicators, driving factor indicators, and dynamic benchmark indicators.

[0009] The dynamic analysis and evaluation module is used to input ecological characteristic data from different periods into the ecological quality assessment model, output ecological quality assessment results, and combine them with the ecological quality indicator system to conduct trend analysis and obtain dynamic quality assessment results.

[0010] The ecological quality dynamic assessment system based on XGBoost provided by this invention includes the following steps in the ecological data parsing module for classifying corresponding ecological types: Vegetation index threshold range, vegetation coverage and vegetation phenology characteristics were extracted from multi-source ecological data as vegetation-related features, and water body index threshold and hydrological connectivity were extracted as hydrological-related features.

[0011] The nighttime light index and the proportion of construction land are extracted as characteristics of human activities on land, and the elevation and slope are extracted as auxiliary topographic features. A type feature set is constructed by combining vegetation-related features and hydrological-related features.

[0012] Ecological features corresponding to multi-source ecological data are extracted based on type feature sets. Preliminary classification rules are established based on land use data as the basic framework and combined with determined thresholds.

[0013] Using GIS tools, the preliminary classification rules were applied to ecological characteristics to determine the type. Conflict areas were further verified based on GDP density, POP density, and slope. For areas with values ​​below the preset threshold, ecological type distribution maps were obtained by merging them according to the dominant types in the surrounding areas.

[0014] The ecological quality dynamic assessment system based on XGBoost provided by this invention includes the following steps in the ecological data parsing module to obtain ecological characteristic data: The carrying capacity of an ecosystem is defined as structural characteristics, while the contribution of different ecosystem types to ecosystem services and their stability in response to human activities and climate change are defined as functional response characteristics.

[0015] Multiple ecological data units are obtained by typological segmentation of multi-source ecological data based on the ecological type distribution map.

[0016] The characteristic indicators of each data unit are calculated based on structural and functional response characteristics, and the ecological characteristic data are obtained by normalization using the Min-Max scaling method.

[0017] The ecological quality dynamic assessment system based on XGBoost provided by this invention includes the following steps in the model construction and optimization module to obtain the ecological dataset: Multi-source ecological data undergoes unified format conversion, spatiotemporal benchmark alignment, and quality cleaning, and is classified according to continuity, categorization, and counting ratio.

[0018] For continuous data, Min-Max scaling and Z-score standardization are used for standardization. For categorical data, it is converted into label encoding or one-hot encoding.

[0019] The count ratio data is calibrated according to the range [0, 1] and standardized using the Min-Max scaling method.

[0020] The standardized data is integrated according to the spatial pixel-time node-indicator dimension to obtain the ecological dataset.

[0021] The ecological quality dynamic assessment system based on XGBoost provided by this invention includes the following steps in the model construction and optimization module for obtaining remote sensing ecological indices and ecological impact factors: Greenness, humidity, dryness, and heat indices are extracted from the ecological dataset and their consistency is verified with raster pixels to obtain basic indices.

[0022] For each raster cell, the values ​​of the basic indicators are arranged as a row to construct a two-dimensional data matrix. Principal component analysis is used to reduce the dimensionality and correct the orientation to obtain the remote sensing ecological index.

[0023] Natural geographical factors, climate factors, human activity factors, and land use factors were selected from the ecological dataset as preliminary factors.

[0024] Calculate the Pearson correlation coefficient between each preliminary factor and the remote sensing ecological index, and retain the preliminary factors that reach the preset correlation coefficient as ecological impact factors.

[0025] The ecological quality dynamic assessment system based on XGBoost provided by this invention includes the following steps in the model construction and optimization module to obtain the ecological quality assessment model: Annual data are selected from remote sensing ecological indices and ecological impact factors, with remote sensing ecological indices as the dependent variable and ecological impact factors as the independent variable.

[0026] Based on the pre-defined ecological quality monitoring requirements, an ecological quality monitoring model is constructed on an annual scale using the XGBoost regression algorithm.

[0027] The annual-scale data is divided into training and test sets according to a preset ratio. The training set is input into the ecological quality monitoring model, and the hyperparameters are iteratively trained using the Bayesian optimization algorithm.

[0028] After the number of iterations reaches the preset number, the parameter combination that minimizes the test set is extracted as the optimal hyperparameter, and the preset accuracy is verified by numerical accuracy, ecological type adaptability and interannual stability.

[0029] If so, construct an ecological quality assessment model based on the current optimal hyperparameters; otherwise, iterate and train again.

[0030] The ecological quality dynamic assessment system based on XGBoost provided by this invention includes an indicator system construction module comprising: The candidate indicator set construction unit is used to extract indicators that reflect the characteristics of different ecological types from ecological characteristic data, and to construct a candidate indicator set by combining them with the basic indicators in the ecological dataset.

[0031] The State and Driving Force Screening Unit is used to screen indicators that reflect the current state of the ecosystem from the candidate indicator set as ecological state indicators, and to screen indicators that lead to changes in the ecological state, which are divided into driving factor indicators according to natural and human activities.

[0032] The indicator calibration unit is used to screen indicators for measuring the trend of ecological quality change and reference standards, and to obtain dynamic benchmark indicators through time, type and spatial calibration.

[0033] The ecological quality dynamic assessment system based on XGBoost provided by this invention includes the following steps in the dynamic analysis and assessment module to obtain ecological quality assessment results: Based on the assessment time point, ecological characteristic data for the corresponding period are selected and converted into a two-dimensional matrix adapted to the ecological quality assessment model.

[0034] The two-dimensional matrices adapted for each period are input into the ecological quality assessment model in chronological order, and the corresponding ecological quality assessment values ​​are calculated based on the characteristic indicators of each period.

[0035] The ecological quality index of each pixel in each period is extracted from the ecological quality assessment value. The ecological quality levels are divided according to the preset standards, and the results are integrated according to the logic of time-space-assessment results to obtain the ecological quality assessment result.

[0036] The ecological quality dynamic assessment system based on XGBoost provided by this invention includes the following steps in the dynamic analysis and assessment module to obtain dynamic quality assessment results: By integrating the ecological quality assessment results from all periods, extracting the time series data of indicators in the ecological quality indicator system, and constructing a dataset of related result indicators.

[0037] Based on the temporal variation characteristics of the ecological status index dataset, the overall ecological trend is obtained. The area proportion and spatial distribution center changes of different ecological quality levels in each period are statistically analyzed, and a level transition matrix is ​​generated to obtain the structural trend.

[0038] The temporal variation characteristics of different ecological types are calculated, and the trend differences between different types are compared to obtain the type trend. The dynamic trend analysis report is obtained by integrating the trend with the overall ecological trend and structural trend.

[0039] The time-series correlation coefficients between each driving factor and the ecological quality index are calculated based on the driving factor indicators to identify the core driving factors.

[0040] Ecological differences are identified by comparing the ecological quality index of each period with dynamic benchmark indicators, and the causes of trends are located by combining core driving factors.

[0041] The rate of change of the ecological quality index in each period is mapped to space to generate a dynamic spatial distribution map, identify the hotspots to which the target area belongs, and determine the ecological type and core driving factors corresponding to the hotspots by combining spatial overlay analysis.

[0042] Based on the mutation points of core driving factors and the fluctuation characteristics of the ecological quality index, the overall time series is divided into different stages, and the driving differences and trend changes in each stage are analyzed to obtain a driving factor correlation analysis scheme.

[0043] The present invention provides an XGBoost-based method for dynamic assessment of ecological quality, comprising: Collect multi-source ecological data of the target assessment area, divide the multi-source ecological data into corresponding ecological types to obtain an ecological type distribution map, and obtain ecological characteristic data based on the impact of different ecological types on the ecosystem.

[0044] Ecological datasets are obtained by standardizing multi-source ecological data, from which remote sensing ecological indices and ecological impact factors are extracted. These indices are then input into an ecological quality monitoring model constructed based on the XGBoost regression algorithm for optimization, resulting in an ecological quality assessment model.

[0045] An ecological quality indicator system is constructed based on ecological characteristic data and ecological datasets. The ecological quality indicator system includes ecological status indicators, driving factor indicators, and dynamic benchmark indicators.

[0046] Ecological characteristic data from different periods are input into the ecological quality assessment model, and the output is the ecological quality assessment result. Trend analysis is then performed using the ecological quality indicator system to obtain the dynamic quality assessment result.

[0047] The beneficial effects of this invention are as follows: 1. This invention categorizes, extracts, and standardizes multi-source ecological data to form a spatiotemporally consistent and formatted ecological dataset and ecological characteristic data, improving the usability and standardization of assessment data. The XGBoost algorithm adapts to the nonlinear coupling relationship of ecological factors, and the Bayesian-optimized hyperparameters enable the model to maintain high prediction accuracy across different ecological types and years, improving the accuracy and stability of the ecological quality assessment model. A comprehensive and dynamic assessment index system is constructed, providing complete support for trend analysis and causal tracing, upgrading the assessment from a single-state description to dynamic process analysis. Through hotspot area identification, stage division, and causal location, key areas and core driving factors for ecological improvement / degradation can be accurately identified. Simultaneously, combined with dynamic benchmarks, anomaly warnings and trend predictions are achieved, providing targeted and actionable decision-making references for ecological protection, environmental governance, and regional planning. Attached Figure Description

[0048] The invention will now be further described with reference to the accompanying drawings.

[0049] Figure 1 This is a schematic diagram of the modules of the dynamic ecological quality assessment system based on XGBoost provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the dynamic ecological quality assessment method based on XGBoost provided in this embodiment of the invention. Figure 3 This is a comparison chart of the training results of the ecological quality assessment model in the XGBoost-based dynamic ecological quality assessment system provided in this embodiment of the invention. Figure 4 This is a comparison chart of the output results of the ecological quality assessment model in the XGBoost-based dynamic ecological quality assessment system provided in this embodiment of the invention. Detailed Implementation

[0050] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0051] like Figures 1 to 2 As shown in the embodiment of the present invention, the dynamic ecological quality assessment system based on XGBoost includes: The ecological data parsing module is used to collect multi-source ecological data of the target assessment area, divide the multi-source ecological data into corresponding ecological types to obtain an ecological type distribution map, and obtain ecological characteristic data based on the impact of different ecological types on the ecosystem.

[0052] Multi-source ecological data includes remote sensing imagery, land use vector data, physical geographic data such as elevation, slope, and aspect, climate data such as temperature and precipitation, surface ecological parameters such as the National Distance Vibration Index (NDVI) and the National Distance Vibration Index (NDWI), human activity data such as the Nighttime Light Index (NTL), GDP, population size (POP), and carbon emissions (CO2). The physical geographic data, climate data, GDP, population size, and carbon emissions data are derived from publicly released data or datasets from relevant departments such as land resources, meteorology, and statistics. The vegetation index, water index, and nighttime light index are extracted using satellite remote sensing data. All data are processed using bilinear interpolation to achieve a uniform spatial resolution of 1 km, while maintaining a consistent temporal resolution.

[0053] The steps for the ecological data parsing module to classify the corresponding ecological types include: Vegetation index threshold range, vegetation coverage and vegetation phenology characteristics were extracted from multi-source ecological data as vegetation-related features, and water body index threshold and hydrological connectivity were extracted as hydrological-related features.

[0054] The nighttime light index and the proportion of construction land are extracted as characteristics of human activities on land, and the elevation and slope are extracted as auxiliary topographic features. A type feature set is constructed by combining vegetation-related features and hydrological-related features.

[0055] NDVI (Vegetation Index) threshold ranges are as follows: Forest NDVI: 0.6-1.0; Grassland NDVI: 0.3-0.6; Desert NDVI: 0-0.3.

[0056] Vegetation coverage, such as farmland coverage: 0.5-0.8. Urban green space coverage: 0.2-0.5.

[0057] Vegetation phenological characteristics, such as seasonal changes in farmland and stable high coverage of forests throughout the year.

[0058] NDWI (Water Body Index) thresholds are as follows: Water NDWI: 0.5-1.0; Wetland NDWI: 0.2-0.5; Non-water body NDWI: <0.2.

[0059] Hydrological connectivity, such as the flow path of river wetlands and the stability of the water area of ​​lake wetlands.

[0060] Nighttime Light Index (NTL) thresholds are as follows: Urban areas: NTL: 0.8-1.0; Farmland: NTL: 0.2-0.5; Natural areas: NTL: 0-0.2.

[0061] The proportion of land used for construction is as follows: in urban areas, the proportion of land used for construction is >50%, while in farmland areas it is <10%.

[0062] Elevation and slope are important factors. For example, mountain forests are mostly distributed in areas with an elevation above 500m and a slope greater than 15°. Plain farmland is mostly located in areas with an elevation less than 200m and a slope less than 5°.

[0063] Ecological features corresponding to multi-source ecological data are extracted based on type feature sets. Preliminary classification rules are established based on land use data as the basic framework and combined with determined thresholds.

[0064] Preliminary classification rules may include: Rule 1: Land use is marked as forest land + NDVI≥0.6 + elevation>500m + slope>15° → forest ecosystem, further subdivided into coniferous forest / broadleaf forest according to temperature / precipitation.

[0065] Rule 2: Land use is labeled as grassland + NDVI 0.3-0.6 + NTL < 0.2 + annual precipitation 200-800mm → grassland ecosystem, and elevation > 3000m is alpine meadow.

[0066] Rule 3: Land use is marked as water area + NDWI≥0.5 + percentage of land without construction land → aquatic ecosystem.

[0067] Rule 4: Land use is labeled as wetland + NDWI 0.2-0.5 + annual precipitation > 400mm → wetland ecosystem, further subdivided into river / lake / swamp wetland according to NDWI.

[0068] Rule 5: Land use is marked as cultivated land + NDVI 0.4-0.7 + NTL 0.2-0.5 + slope <5° → farmland ecosystem. Annual precipitation >800mm is paddy field, otherwise it is dry field.

[0069] Rule 6: Land use is marked as construction land + NTL≥0.7 + GDP density>5 million yuan / km² + POP density>1000 people / km² → urban ecosystem. Areas with NDVI 0.2-0.5 are urban green spaces, and the rest are built-up areas.

[0070] Rule 7: Land use marked as unused land + NDVI < 0.3 + NDWI < 0.2 + annual precipitation < 400 mm → desert ecosystem.

[0071] Using GIS tools, the preliminary classification rules were applied to ecological characteristics to determine the type. Conflict areas were further verified based on GDP density, POP density, and slope. For areas with values ​​below the preset threshold, ecological type distribution maps were obtained by merging them according to the dominant types in the surrounding areas.

[0072] For example: A pixel is marked as forest land, with NDVI=0.75, elevation=800m, slope=20°, and annual precipitation=1000mm → it is determined to be a broad-leaved forest ecosystem.

[0073] Conflict area verification: Extract auxiliary features such as GDP density, POP density, and slope of conflicting pixels. If POP density > 300 people / km² and slope < 5°, it is identified as farmland. If POP density < 200 people / km² and slope > 10°, it is identified as grassland.

[0074] Merging of transitional areas: For isolated water bodies with an area of ​​less than 10 km² (i.e., 10 1 km pixels) (such as scattered small water bodies in urban built-up areas), they are merged according to the dominant type of the surrounding area. If the small water body belongs to the urban park, it is determined to be an attached water body of urban green space and is included in the urban ecosystem.

[0075] The steps by which the ecological data parsing module obtains ecological characteristic data include: The carrying capacity of an ecosystem is defined as structural characteristics, while the contribution of different ecosystem types to ecosystem services and their stability in response to human activities and climate change are defined as functional response characteristics.

[0076] Multiple ecological data units are obtained by typological segmentation of multi-source ecological data based on the ecological type distribution map.

[0077] In GIS tools (ArcGIS / QGIS), coded by ecological type, such as 1=forest, 2=grassland, raster data such as NDVI, NDWI, NTL, elevation, temperature, precipitation, and GDP are processed to retain only the raster pixels corresponding to the ecological type and remove pixels of other types.

[0078] For each ecological type, check the size of the extracted data sample to ensure it is sufficient to support statistical analysis. For example, the number of pixels for a single ecological type should be ≥50 to avoid characteristic bias caused by insufficient samples. If the number of pixels for a certain type is insufficient, adjacent similar subtypes can be merged, such as merging coniferous forest and broadleaf forest into a single forest for unified extraction.

[0079] If it is necessary to analyze the dynamic changes in ecological characteristics, the multi-source data for each ecological type should be extracted annually at a uniform time granularity to ensure consistency in the time dimension.

[0080] The characteristic indicators of each data unit are calculated based on structural and functional response characteristics, and the ecological characteristic data are obtained by normalization using the Min-Max scaling method.

[0081] Calculation of structural characteristic indicators: Vegetation cover: Calculated based on NDVI using a formula, taking the mean and coefficient of variation of all pixels for this ecological type, reflecting the overall cover level and internal cover uniformity.

[0082] Water body proportion: The percentage of raster pixels with NDWI ≥ 0.2 in this ecological type, such as the water body proportion of wetlands and the water area proportion of urban ecosystems.

[0083] Construction land ratio: This refers to the percentage of raster pixels in the ecological type whose land use is marked as construction land, such as the built-up area ratio of urban ecosystems and the scattered construction land ratio of farmland ecosystems.

[0084] Terrain adaptability: Calculate the average elevation and slope of all pixels in this ecological type, such as the average elevation of alpine meadows and the average slope of farmland, to reflect the adaptability of this type to the terrain.

[0085] Functional response characteristic index calculation: Carbon sequestration potential: Using the NDVI-carbon sequestration correlation model, based on existing research conclusions: carbon sequestration amount ≈ a × vegetation cover + b × NDVI - c × CO2 density, where a, b, and c are regional calibration coefficients, the average carbon sequestration amount of this ecological type is calculated. For example, the carbon sequestration potential of forests is significantly higher than that of deserts.

[0086] Hydrological regulation capacity: The hydrological regulation coefficient is calculated using the formula = (mean NDWI × mean precipitation) / mean slope. The higher the coefficient, the stronger the ability of this type of water to intercept and regulate water resources.

[0087] Habitat quality: The comprehensive score is calculated based on the mean of NDVI + terrain complexity (slope standard deviation) - 0.3 × mean of NTL. The higher the terrain complexity and the lower the human disturbance, the better the habitat quality. For example, forest habitat quality > urban built-up area.

[0088] Ecological sensitivity: Calculate the interannual fluctuation coefficient of NDVI for this ecological type. The fluctuation coefficient = standard deviation / mean. The greater the fluctuation, the more sensitive this type is to changes in the external environment.

[0089] Human disturbance resistance: Calculate the Pearson correlation coefficient between NTL and NDVI within this type. If there is a significant negative correlation (r < -0.5), it indicates that human disturbance is highly destructive to this type and the resistance is weak, such as farmland > grassland. If the negative correlation is not significant (r > -0.3), it indicates strong resistance, such as primary forest.

[0090] Climate adaptability: Calculate the correlation coefficients between temperature and NDVI for this type of climate, and the correlation coefficients between precipitation and NDVI. If the precipitation correlation coefficient is significantly positive, it indicates that this type of climate is dependent on precipitation, such as grasslands. If the temperature correlation coefficient is not significant, it indicates strong climate adaptability, such as forests.

[0091] The model building and optimization module is used to standardize multi-source ecological data to obtain an ecological dataset, extract remote sensing ecological indices and ecological impact factors from it, and input them into an ecological quality monitoring model built based on the XGBoost regression algorithm for optimization to obtain an ecological quality assessment model.

[0092] The steps in obtaining the ecological dataset by the model building and optimization module include: Multi-source ecological data undergoes unified format conversion, spatiotemporal benchmark alignment, and quality cleaning, and is classified according to continuity, categorization, and counting ratio.

[0093] Land use vector data and administrative unit statistical data are converted into raster data, consistent with the original raster data format of remote sensing imagery and surface ecological parameters. All data are standardized to a GIS common format to avoid format conflicts. Categorical data, such as land use type and ecological type, are numerically encoded to replace textual descriptions and adapt to model input.

[0094] Convert all data to the same coordinate system (such as WGS84 geographic coordinate system) to avoid spatial misalignment caused by projection differences. Use bilinear interpolation to unify the spatial resolution of all raster data to 1km (as required above) to ensure consistent spatial correspondence of each cell.

[0095] Adjust data from different time frequencies to a consistent time resolution to ensure that the data are from the same evaluation period.

[0096] Use the 3σ principle or domain common sense threshold to remove invalid data, such as sensor noise with NDVI < -0.2 or > 1.0, negative NTL values, and abnormally high GDP values.

[0097] For continuous data, Min-Max scaling and Z-score standardization are used for standardization. For categorical data, it is converted into label encoding or one-hot encoding.

[0098] Label coding: Operation steps: Assign unique integer codes to land use type, ecological type and other classification data according to the importance of the type or natural order, such as land use type: forest land=1, grassland=2, cultivated land=3, construction land=4, water area=5, wetland=6, desert=7.

[0099] Applicable scenarios: When there is an implicit order between types, such as ecological quality levels: poor = 1, medium = 2, good = 3, excellent = 4, or the number of types is ≤ 10.

[0100] One-hot encoding operation steps: For each classification type, such as slope direction: East, South, West, North, Northeast, Southeast, Northwest, Southwest, generate a binary dummy variable, 0 or 1.

[0101] If a pixel's slope is south, then the dummy variable corresponding to south is 1, and the dummy variable for other slopes is 0.

[0102] Convert categorical data into multiple binary columns to prevent the model from treating unordered categories as ordered relationships.

[0103] The count ratio data is calibrated according to the range [0, 1] and standardized using the Min-Max scaling method.

[0104] The Normalized Difference Vegetation Index (NDVI) is an index that reflects vegetation cover using remote sensing technology. The calculation formula is:

[0105] In the formula, and These represent the reflectance in the near-infrared and red light bands, respectively. The NDVI value ranges from -1.0 to 1.0; generally, a higher NDVI value indicates higher vegetation cover.

[0106] The Normalized Difference Water Index (NDWI) is an index that reflects water body information using remote sensing technology. The calculation formula is as follows:

[0107] In the formula, and These represent the reflectance in the green light band and the near-infrared band, respectively. The NDWI index ranges from -1 to 1, with a value closer to 1 indicating a higher likelihood of a body of water.

[0108] The Nighttime Light Index (NTL) is an objective indicator of the intensity of human activity, derived from nighttime light imagery data of human activities extracted using satellite remote sensing technology.

[0109] The standardized data is integrated according to the spatial pixel-time node-indicator dimension to obtain the ecological dataset. All standardized raster indicators are associated by pixel coordinates to generate a multi-band raster dataset, and a data table is constructed with latitude and longitude coordinates as rows and time and indicators as columns.

[0110] The steps in the model building and optimization module to obtain remote sensing ecological indices and ecological impact factors include: Greenness, humidity, dryness, and heat indices are extracted from the ecological dataset and their consistency is verified with raster pixels to obtain basic indices.

[0111] For each raster cell, the values ​​of the basic indicators are arranged as a row to construct a two-dimensional data matrix. Principal component analysis is used to reduce the dimensionality and correct the orientation to obtain the remote sensing ecological index.

[0112] The covariance matrix, eigenvalues, and eigenvectors were calculated, and the first principal component (PC1) was selected as the core component. Since dryness index (NDBSI) and heat index (LST) are negatively correlated with ecological quality, the corresponding eigenvectors in PC1 need to be inverted to ensure that the larger the RSEI value, the better the ecological quality. The Min-Max scaling method was used to map the corrected PC1 values ​​to the [0,1] interval to obtain the final remote sensing ecological index.

[0113] Natural geographical factors, climate factors, human activity factors, and land use factors were selected from the ecological dataset as preliminary factors.

[0114] Natural geographical factors: Extracted standardized elevation, slope, and aspect.

[0115] Climate factors: Extracted standardized annual average temperature and annual precipitation.

[0116] Human activity factors: Extracted standardized nighttime light index, GDP density, population density, CO2 density, and proportion of construction land.

[0117] Land use factors: extracted coded land use types, dummy variables after label coding or one-hot coding.

[0118] Calculate the Pearson correlation coefficient between each preliminary factor and the remote sensing ecological index, and retain the preliminary factors that reach the preset correlation coefficient as ecological impact factors.

[0119] The steps in the model building and optimization module to obtain the ecological quality assessment model include: like Figures 3 to 4 The image shown is a comparison of the model's performance before and after training. Annual data were selected from remote sensing ecological indices and ecological impact factors, with remote sensing ecological indices as the dependent variable and ecological impact factors as the independent variable.

[0120] Based on the pre-defined ecological quality monitoring requirements, an ecological quality monitoring model is constructed on an annual scale using the XGBoost regression algorithm. The formula is as follows:

[0121] In the formula, It is a remote sensing ecological index. It is the vegetation index. It is Gross Domestic Product. It's the slope. It is the total precipitation. It is a water quality index. It is the nighttime light index. It is the surface temperature. It is population density. It's a slope. It's carbon emissions.

[0122] The annual-scale data is divided into training and test sets according to a preset ratio. The training set is input into the ecological quality monitoring model, and the hyperparameters are iteratively trained using the Bayesian optimization algorithm.

[0123] The learning rate, tree depth, number of decision trees, subsample ratio, column sampling ratio, and regularization coefficient are used as the range of hyperparameters for optimization. The root mean square error (RMSE) of the test set is used as the optimization objective; that is, the algorithm iteratively seeks the parameter combination that minimizes the RMSE. During the Bayesian optimization process, 5-fold cross-validation is performed on the training set—the training set is split into 5 subsets, and 4 subsets are used for training and 1 for validation in turn. The average validation error is calculated to avoid parameter bias caused by a single training / validation split. This ensures that the optimized parameters have stable performance across different data subsets, adapting to the evaluation needs across yearly timescales.

[0124] After the number of iterations reaches the preset number, the parameter combination that minimizes the test set is extracted as the optimal hyperparameter, and the preset accuracy is verified by numerical accuracy, ecological type adaptability and interannual stability.

[0125] Numerical precision: Calculate the coefficient of determination, R², which reflects the explanatory power of the model. For annual scale models, R² ≥ 0.8. Root mean square error, RMSE, which reflects the prediction bias and should be ≤ 0.08. Mean absolute error, MAE, which reflects the average prediction error.

[0126] Ecological type suitability: Accuracy indicators are calculated separately for ecological types such as forest, grassland, and city to ensure that the model performs stably across all types and avoids bias towards a particular ecological type.

[0127] Interannual stability: The model accuracy is verified for each assessment year to ensure that the prediction error does not fluctuate significantly across years, which meets the core requirements of dynamic assessment on an annual scale.

[0128] If so, construct an ecological quality assessment model based on the current optimal hyperparameters; otherwise, iterate and train again.

[0129] The indicator system construction module is used to build an ecological quality indicator system based on ecological characteristic data and ecological datasets. The ecological quality indicator system includes ecological status indicators, driving factor indicators, and dynamic benchmark indicators.

[0130] The indicator system construction module includes: The candidate indicator set construction unit is used to extract indicators that reflect the characteristics of different ecological types from ecological characteristic data, and to construct a candidate indicator set by combining them with the basic indicators in the ecological dataset.

[0131] The candidate indicator set includes mean vegetation cover, carbon sequestration potential, ecological sensitivity, resistance to human disturbance, surface ecological parameters, natural geographic data, climate data, human activity data, and land use coding.

[0132] The State and Driving Force Screening Unit is used to screen indicators that reflect the current state of the ecosystem from the candidate indicator set as ecological state indicators, and to screen indicators that lead to changes in the ecological state, which are divided into driving factor indicators according to natural and human activities.

[0133] Ecological status indicators are categorized as follows: focusing on current status characteristics, such as RSEI (Rounded State of Ecosystem), vegetation cover, water body proportion, habitat quality, carbon sequestration potential, and hydrological regulation capacity. These indicators directly reflect the current level of ecological quality, and the data mainly come from the functional characteristic indicators of ecological characteristic data and the surface ecological parameters of ecological datasets.

[0134] Driving factor indicators are categorized as follows: focusing on causal explanation, they are divided into two categories based on natural and human activities—natural driving factors, such as elevation, slope, temperature, and precipitation, and human activity driving factors, such as NTL, GDP, POP population density, CO2 density, and construction land ratio. The data mainly come from the interference response indicators of natural geography, climate, human activity data, and ecological characteristic data in the ecological dataset.

[0135] The indicator calibration unit is used to screen indicators for measuring the trend of ecological quality change and reference standards, and to obtain dynamic benchmark indicators through time, type and spatial calibration.

[0136] Time-based benchmark calibration: Based on multi-year data from the ecological dataset, the average ecological status indicators of the target assessment area over the past 5-10 years are calculated as a dynamic time benchmark. Simultaneously, the annual fluctuation range of the indicators is calculated to determine the normal fluctuation interval, used to judge whether there are abnormal changes in ecological quality.

[0137] Type benchmark calibration: Based on ecological characteristic data, the mean values ​​of state indicators and driving factor thresholds of various ecosystems are calculated according to ecological types such as forest, grassland and city. For example, the suitable precipitation range of grassland ecosystem is used as a specific benchmark for different types to avoid the adaptation bias of a single benchmark to different ecological types.

[0138] Spatial benchmark calibration: Referencing similar ecological assessment results nationwide or in the region, introduce external spatial benchmarks, such as the average RSEI level of similar ecosystems nationwide, and compare them with the indicators of the target region to enhance the cross-regional comparability of the system.

[0139] The dynamic analysis and evaluation module is used to input ecological characteristic data from different periods into the ecological quality assessment model, output ecological quality assessment results, and combine them with the ecological quality indicator system to conduct trend analysis and obtain dynamic quality assessment results.

[0140] The steps by which the dynamic analysis and assessment module obtains ecological quality assessment results include: Based on the assessment time point, ecological characteristic data for the corresponding period are selected and converted into a two-dimensional matrix adapted to the ecological quality assessment model.

[0141] The two-dimensional matrices adapted for each period are input into the ecological quality assessment model in chronological order, and the corresponding ecological quality assessment values ​​are calculated based on the characteristic indicators of each period.

[0142] The ecological quality index of each pixel in each period is extracted from the ecological quality assessment value. The ecological quality levels are divided according to the preset standards, and the results are integrated according to the logic of time-space-assessment results to obtain the ecological quality assessment result.

[0143] The steps by which the dynamic analysis and evaluation module obtains dynamic quality assessment results include: By integrating the ecological quality assessment results from all periods, extracting the time series data of indicators in the ecological quality indicator system, and constructing a dataset of related result indicators.

[0144] Based on the temporal variation characteristics of the ecological status index dataset, the overall ecological trend is obtained. The area proportion and spatial distribution center changes of different ecological quality levels in each period are statistically analyzed, and a level transition matrix is ​​generated to obtain the structural trend.

[0145] The temporal variation characteristics of different ecological types are calculated, and the trend differences between different types are compared to obtain the type trend. The dynamic trend analysis report is obtained by integrating the trend with the overall ecological trend and structural trend.

[0146] The time-series correlation coefficients between each driving factor and the ecological quality index are calculated based on the driving factor indicators to identify the core driving factors.

[0147] If temperature and the index are negatively correlated, and NTL is also negatively correlated with the index, it suggests that global warming and increased human activity intensity may be the main causes of degradation in some areas. Further partial correlation analysis can eliminate interference between factors and clarify the independent influence of individual driving factors, such as the degree of influence of NTL on the index after controlling for precipitation.

[0148] Ecological differences are identified by comparing the ecological quality index of each period with dynamic benchmark indicators, and the causes of trends are located by combining core driving factors.

[0149] Compared with a time benchmark, it determines whether the current quality deviates from the long-term level. For example, if the index in 2023 is 0.05 higher than the average, it is considered a high-quality year. Compared with a type benchmark, it determines whether the trend of different types conforms to their inherent characteristics. For example, if the grassland index is lower than the type benchmark and continues to decline, it requires close attention. Compared with a spatial benchmark, it clarifies the regional position of the local trend. For example, if the local forest improvement rate is higher than the national average.

[0150] The rate of change of the ecological quality index in each period is mapped to space to generate a dynamic spatial distribution map, identify the hotspots to which the target area belongs, and determine the ecological type and core driving factors corresponding to the hotspots by combining spatial overlay analysis.

[0151] Hotspot areas can be classified as areas of continuous improvement, areas of significant degradation, or areas of fluctuating stability. For example, areas of significant degradation are mostly grassland types and correspond to increased NTL and decreased precipitation.

[0152] Based on the mutation points of core driving factors and the fluctuation characteristics of the ecological quality index, the overall time series is divided into different stages, and the driving differences and trend changes in each stage are analyzed to obtain a driving factor correlation analysis scheme.

[0153] Based on the same general inventive concept, this invention also protects an XGBoost-based method for dynamic assessment of ecological quality, the method comprising: Collect multi-source ecological data of the target assessment area, divide the multi-source ecological data into corresponding ecological types to obtain an ecological type distribution map, and obtain ecological characteristic data based on the impact of different ecological types on the ecosystem.

[0154] Ecological datasets are obtained by standardizing multi-source ecological data, from which remote sensing ecological indices and ecological impact factors are extracted. These indices are then input into an ecological quality monitoring model constructed based on the XGBoost regression algorithm for optimization, resulting in an ecological quality assessment model.

[0155] An ecological quality indicator system is constructed based on ecological characteristic data and ecological datasets. The ecological quality indicator system includes ecological status indicators, driving factor indicators, and dynamic benchmark indicators.

[0156] Ecological characteristic data from different periods are input into the ecological quality assessment model, and the output is the ecological quality assessment result. Trend analysis is then performed using the ecological quality indicator system to obtain the dynamic quality assessment result.

[0157] In summary, the XGBoost-based dynamic ecological quality assessment system and method provided in this embodiment comprehensively classifies ecological characteristics and logically verifies conflict areas, making the classification results more consistent with the ecological essence and laying the foundation for subsequent classification assessment and management. By utilizing the optimized model for continuous assessment and combining it with the indicator system for trend analysis, continuous monitoring of the dynamic evolution of ecological quality is achieved. By constructing an indicator system containing driving factors and performing spatiotemporal correlation and statistical analysis between the model output and the driving factors, key natural and anthropogenic driving factors leading to ecological changes can be effectively identified, providing direct evidence for management decisions. This invention can improve the objectivity and spatial accuracy of ecological type classification, support refined spatiotemporal dynamic monitoring and attribution analysis, and significantly improve the scientific rigor, dynamism, spatial precision, and decision support capabilities of the assessment.

[0158] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An ecological quality dynamic evaluation system based on XGBoost, comprising: characterized in that: an ecological data analysis module, configured to collect multi-source ecological data of a target evaluation area, divide the multi-source ecological data into corresponding ecological types to obtain an ecological type distribution map, and obtain ecological characteristic data according to the influence of different ecological types on an ecological system; a model construction and optimization module, configured to perform standardization processing on the multi-source ecological data to obtain an ecological data set, extract remote sensing ecological indexes and ecological influence factors from the ecological data set, input an ecological quality monitoring model constructed based on an XGBoost regression algorithm for optimization to obtain an ecological quality evaluation model; an index system construction module, configured to construct an ecological quality index system based on the ecological characteristic data and the ecological data set, wherein the ecological quality index system comprises ecological state indexes, driving factor indexes and dynamic benchmark indexes; a dynamic analysis and evaluation module, configured to input ecological characteristic data of different periods into the ecological quality evaluation model, output ecological quality evaluation results, and perform trend analysis on the ecological quality evaluation results to obtain dynamic quality evaluation results.

2. The XGBoost-based dynamic ecological quality assessment system according to claim 1, characterized in that: The step of dividing corresponding ecological types by the ecological data analysis module comprises: extracting a vegetation index threshold interval, vegetation coverage and vegetation phenology characteristics as vegetation-related characteristics, and extracting a water body index threshold and hydrological connectivity as hydrological-related characteristics from the multi-source ecological data; extracting a night light index and a construction land proportion as land human activity characteristics, and extracting elevation and slope as terrain auxiliary characteristics, and constructing a type feature set by combining the vegetation-related characteristics and the hydrological-related characteristics; extracting corresponding ecological characteristics of the multi-source ecological data according to the type feature set, and establishing preliminary division rules based on a land use data framework and in combination with determined thresholds; applying the preliminary division rules to the ecological characteristics by using a GIS tool to perform type determination, performing auxiliary verification on conflict areas according to GDP density, POP density and slope, and merging areas smaller than a preset threshold according to a surrounding dominant type to obtain the ecological type distribution map. 3.The XGBoost-based ecological quality dynamic assessment system according to claim 1, characterized in that: The step of obtaining the ecological characteristic data by the ecological data analysis module comprises: defining the carrying capacity of an ecological system as a structural characteristic, and the contribution of different ecological types to ecological system services and the stability to human activities and climate change as functional response characteristics; performing type segmentation on the multi-source ecological data according to the ecological type distribution map to obtain a plurality of ecological data units; calculating characteristic indexes of each data unit according to the structural characteristics and the functional response characteristics, and performing normalization processing by using a Min-Max scaling method to obtain the ecological characteristic data. 4.The XGBoost-based ecological quality dynamic assessment system according to claim 1, characterized in that: The step of obtaining the ecological data set by the model construction and optimization module comprises: performing format conversion, space-time benchmark alignment and quality cleaning processing on the multi-source ecological data, and classifying the multi-source ecological data according to continuity, type and count proportion; performing standardization processing on continuous data by using a Min-Max scaling method and Z-score standardization, and converting type data into label encoding or one-hot encoding; The count proportion type data is range calibrated in the interval [0, 1], and the Min-Max scaling method is used for standardization processing; All the standardized data are integrated according to the spatial pixel-time node-index dimension to obtain the ecological data set.

5. The XGBoost-based dynamic ecological quality assessment system according to claim 1, characterized in that: The steps of obtaining the remote sensing ecological index and the ecological impact factor by the model construction and optimization module include: Greenness index, humidity index, aridity index and heat index are extracted from the ecological data set, and consistency verification is performed with the grid pixel to obtain the basic index; For each grid pixel, the numerical value of the basic index is taken as a row to construct a two-dimensional data matrix, and the principal component analysis method is used for dimension reduction and direction correction to obtain the remote sensing ecological index; Natural geographical factors, climate factors, human activity factors and land use factors are screened from the ecological data set as preliminary factors; The Pearson correlation coefficient of each preliminary factor and the remote sensing ecological index is calculated, and the preliminary factor reaching the preset correlation coefficient is retained as the ecological impact factor. 6.The XGBoost-based ecological quality dynamic assessment system according to claim 1, characterized in that: The steps of obtaining the ecological quality evaluation model by the model construction and optimization module include: Yearly scale data are screened from the remote sensing ecological index and the ecological impact factor, the remote sensing ecological index is taken as the dependent variable, and the ecological impact factor is taken as the independent variable; Based on the preset ecological quality monitoring demand, the XGBoost regression algorithm is combined to construct the ecological quality monitoring model at the annual scale; The annual scale data are divided into a training set and a test set according to a preset proportion, the training set is input into the ecological quality monitoring model, and the Bayesian optimization algorithm is used to iteratively train the hyperparameters; When the iteration number reaches the preset number, the parameter combination that minimizes the test set is extracted as the optimal hyperparameter, and whether the numerical accuracy, ecological type adaptability and interannual stability meet the preset accuracy is verified; Yes, the ecological quality evaluation model is constructed according to the current optimal hyperparameter, otherwise the iterative training is re-performed.

7. The XGBoost-based dynamic ecological quality assessment system according to claim 1, characterized in that: The index system construction module includes: A candidate index set construction unit is configured to extract indexes reflecting different ecological type characteristics from the ecological characteristic data, and combine the basic indexes in the ecological data set to construct a candidate index set; A state and driving force screening unit is configured to screen indexes reflecting the current state of the ecological system from the candidate index set as the ecological state indexes, and screen impact indexes leading to changes in the ecological state, and divide the indexes into the driving factor indexes according to natural and human activities; An index calibration unit is configured to screen indexes for measuring the trend of ecological quality change and reference standards, and perform time, type and space calibration to obtain the dynamic benchmark index.

8. The XGBoost-based dynamic ecological quality assessment system according to claim 1, characterized in that: The steps of obtaining the ecological quality evaluation result by the dynamic analysis and evaluation module include: According to the evaluation time node, the ecological characteristic data of the corresponding period is screened and converted into a two-dimensional matrix adapted to the ecological quality evaluation model; The two-dimensional matrices adapted to each period are input into the ecological quality evaluation model in chronological order, and the ecological quality evaluation value corresponding to each period is calculated based on the characteristic indexes of each period. The ecological quality index of each pixel in each period is extracted from the ecological quality assessment value, the ecological quality grades divided according to the preset standard are integrated according to the logic of time-space-assessment results to obtain the ecological quality assessment result.

9. The XGBoost-based dynamic ecological quality assessment system according to claim 8, characterized in that: The step of obtaining the dynamic quality assessment result by the dynamic analysis and assessment module comprises: The ecological quality assessment results of all periods are integrated, the index time series data in the ecological quality index system are extracted, and a result index correlation data set is constructed; An ecological overall trend is obtained by analyzing the time series variation characteristics of the result index correlation data set according to the ecological state index, the area proportion of different ecological quality grades in each period and the change of spatial distribution center are counted, a grade transfer matrix is generated to obtain a structure trend; The time series variation characteristics of different ecological types are calculated, the trend differences of different types are compared to obtain a type trend, and the dynamic trend analysis report is obtained by integrating the ecological overall trend and the structure trend; The time series correlation coefficients of each driving factor and the ecological quality index are calculated according to the driving factor index, and the core driving factor is identified; The ecological quality index of each period is compared with the dynamic benchmark index to obtain an ecological difference, and the trend cause is located in combination with the core driving factor; The change rate of the ecological quality index of each period is mapped to space to generate a dynamic change spatial distribution map, a hotspot region to which a target region belongs is identified, and the ecological type and the core driving factor corresponding to the hotspot region are determined in combination with spatial superposition analysis; Based on the mutation point of the core driving factor and the fluctuation characteristics of the ecological quality index, the overall time series is divided into different stages, and the driving difference and trend characteristic change of each stage are analyzed to obtain a driving factor correlation analysis scheme.

10. The XGBoost-based dynamic ecological quality assessment method applied to the XGBoost-based dynamic ecological quality assessment system according to any one of claims 1 to 9, characterized in that, The evaluation method comprises: Multi-source ecological data of a target evaluation region are collected, the multi-source ecological data are divided into corresponding ecological types to obtain an ecological type distribution map, and ecological characteristic data are obtained according to the influence of different ecological types on an ecological system; The multi-source ecological data are standardized to obtain an ecological data set, remote sensing ecological indexes and ecological influence factors are extracted from the ecological data set, and an ecological quality monitoring model constructed based on an XGBoost regression algorithm is inputted for optimization to obtain an ecological quality evaluation model; An ecological quality index system is constructed based on the ecological characteristic data and the ecological data set, and the ecological quality index system comprises ecological state indexes, driving factor indexes and dynamic benchmark indexes; The ecological characteristic data of different periods are inputted into the ecological quality evaluation model, and an ecological quality assessment result is outputted, and a dynamic quality assessment result is obtained by trend analysis in combination with the ecological quality index system.

Images