Research method of ecological system for meteorological drought feedback

Abstract

The invention relates to the technical field of ecological meteorology, in particular to a research method for meteorological drought feedback of an ecological system, which comprises the following steps of: firstly, selecting a strong land-gas coupling season to obtain detrended vegetation state data; then, constructing a vine Copula vegetation feedback critical threshold framework, calculating P1 (the drought probability of the current month under early drought) and P2 (the drought probability of the current month under early drought and vegetation) by screening an optimal Pair-copula function, determining a feedback critical threshold by an intersection point of the P1 and the P2, representing feedback sensitivity by a P2 curve fitting coefficient, and dividing five feedback conversion types; then dividing research stages, comparing dynamic changes of feedback characteristics by combining Wilcoxon test, and analyzing the influence of different levels of drought in the early stage; and finally, on the basis of an XGB-SHAP model, selecting meteorological, soil topography and biological factors, quantifying the contribution of the dominant factors, and analyzing a positive / negative feedback mechanism according to climate partitions. According to the method, high-dimensional modeling precision and mechanism interpretability can be improved, and support is provided for drought response and ecological management.

Description

Technical Field

[0001] This invention relates to the field of ecological meteorology, specifically to a research method for the feedback of ecosystems to meteorological drought. Background Technology

[0002] Meteorological drought, as one of the most frequent and widespread extreme climate events under the background of global climate change, poses an increasingly serious threat to agricultural production, ecological security, and socio-economic development. Terrestrial ecosystems, as a key link connecting the atmosphere, soil, and biosphere, do not passively suffer from the effects of meteorological drought, but rather exert a significant feedback effect by regulating surface energy balance, water balance, and biogeochemical cycles. Therefore, accurately quantifying the feedback patterns of terrestrial ecosystems to meteorological drought, identifying critical feedback states, and determining the dominant driving mechanisms have become core scientific issues in addressing climate change and optimizing ecosystem management. However, existing research still faces the following key technical bottlenecks in analyzing the feedback relationship between ecosystems and meteorological drought: The high-dimensional dependency model lacks flexibility: When traditional multi-parameter Copula models are used to characterize the complex dependency between vegetation status and meteorological drought index, they are difficult to adapt to the dynamic coupling characteristics of high-dimensional variables, which makes it impossible to accurately construct a feedback relationship model under the synergistic effect of multiple factors, thus affecting the quantitative accuracy of the feedback threshold. The identification of critical feedback states is ambiguous: Existing studies mostly focus on the one-way impact of vegetation on drought (such as only analyzing the aggravating effect of vegetation reduction on drought), without clarifying the critical threshold for the transformation of vegetation status from "alleviating drought" to "exacerbating drought", nor quantifying the differences in feedback sensitivity under different vegetation statuses. This makes it impossible to determine the safe boundary of the ecosystem to drought, and it is difficult to support targeted ecological regulation measures. The attribution of multi-scale influencing factors is unclear: On the one hand, existing studies have not systematically analyzed the dynamic evolution of feedback thresholds and sensitivity under the background of long-term climate change, and cannot reveal the long-term impact of climate warming on the ecological-drought feedback relationship; on the other hand, there is insufficient research on the differentiated feedback mechanisms under different levels of drought and different climate / vegetation zones in the past, and traditional attribution models (such as linear regression) are difficult to quantify the contribution of nonlinear factors, resulting in the inability to accurately identify the core factors driving the feedback relationship. Feedback mechanism analysis lacks interpretability: Most studies rely on machine learning models (such as XGBoost) to analyze factor importance, but the black box nature of the models makes it impossible to clarify the specific contribution direction and intensity of each factor to the feedback state, especially to distinguish the dominant mechanisms of positive / negative feedback in different climate zones.

[0003] In summary, there is an urgent need for a systematic research method that can accurately characterize the feedback relationship between ecosystems and meteorological drought, so as to provide scientific support for drought response and ecosystem management in the context of climate change. Summary of the Invention

[0004] To address the shortcomings and problems of existing methods, such as insufficient flexibility in high-dimensional modeling, ambiguity of feedback critical states, unclear multi-scale attribution, lack of interpretability in mechanism analysis, and difficulty in accurately quantifying the feedback patterns of ecosystems to meteorological drought, this invention provides a research method for ecosystem feedback to meteorological drought. This method can improve the accuracy of high-dimensional modeling, clarify the feedback critical threshold and sensitivity, achieve multi-scale dynamic tracking and regional attribution, reveal the feedback mechanism, and provide scientific support for drought response and ecological management.

[0005] The solution adopted by this invention to solve its technical problem is: a research method for the feedback of an ecosystem to meteorological drought, comprising the following steps: S1: Determine the study period and basic data, select the season with strong land-atmosphere coupling as the study period, and obtain detrended vegetation status data and meteorological drought index data. S2: Construct a vegetation feedback critical threshold framework based on Copula, quantify the feedback critical threshold and feedback sensitivity of vegetation status to meteorological drought. The feedback critical threshold is the intersection of the probability of meteorological drought in the current month when the previous meteorological drought occurred, P1, and the probability of meteorological drought in the current month when the previous meteorological drought occurred and the vegetation is in a specific state, P2. The feedback sensitivity is characterized by the fitting coefficient of the P2 curve. S3: Divide the research into stages and compare the dynamic changes of the feedback critical threshold and feedback sensitivity in different stages; S4: Analyze the impact of different levels of meteorological drought in the early stage on the feedback critical threshold and feedback sensitivity; S5: Based on the XGB-SHAP attribution model, analyze and identify the dominant factors affecting the critical threshold of feedback, the spatial distribution of feedback sensitivity, and dynamic changes, and explore the feedback mechanism of vegetation status on meteorological drought.

[0006] Furthermore, in step S1, the vegetation status data includes leaf area index (LAI) data, total primary productivity (GPP) data, and ecosystem respiration (TER) data; the meteorological drought index data is standardized precipitation evapotranspiration index (SPEI) data.

[0007] Furthermore, in step S2, the specific process of constructing a vegetation feedback critical threshold framework based on *Copula* includes: S21: Determine the optimal vine structure. Select five Copula functions, namely Clayton, Frank, Gumbel, Gaussian, and Student'st(t), as candidate Pair-copula functions. Use the AIC criterion to select the Pair-copula function with the smallest AIC and construct the optimal vine structure between vegetation status data and meteorological drought index data. S22: Calculate the feedback critical threshold and feedback sensitivity. First, calculate P1 and P2 respectively. The formula for calculating P1 is: P1 = P(SPEI) t ≤-0.5|SPEI t-1 ≤-0.5), The formula for calculating P2 is: P2 = P(SPEI) t ≤-0.5|SPEI t-1 ≤-0.5, VEG t ≤veg), In the formula, t represents the current month, t-1 represents the previous month, and veg is the vegetation status index, which ranges from 1 to 99th and increases by 1 unit. The critical threshold for feedback is determined by the intersection of P1 and P2. The feedback sensitivity is characterized by the fitting coefficient A of the P2 curve. When A < 0, the feedback sensitivity is negative, and when A > 0, the feedback sensitivity is positive.

[0008] Furthermore, step S2 also includes determining the feedback transformation type of vegetation to meteorological drought, wherein the feedback transformation type includes: Type 1: As vegetation loss increases, the probability of vegetation causing meteorological drought decreases. After reaching the feedback critical threshold, vegetation changes from aggravating meteorological drought to alleviating it. The P2 curve fitting coefficient A < 0. Type 2: As vegetation loss increases, the probability of vegetation causing meteorological drought increases. After reaching the feedback critical threshold, vegetation changes from alleviating meteorological drought to aggravating it. The P2 curve fitting coefficient A > 0. Type 3: P2 is always lower than P1, and changes in vegetation status have no impact on meteorological drought; Type 4: P2 is always higher than P1, and changes in vegetation status have no impact on meteorological drought; Type 5: As vegetation loss increases, the impact of vegetation on drought first shifts from exacerbation to mitigation, then intensifies again after falling below another threshold, indicating two feedback critical thresholds.

[0009] Furthermore, in step S3, the research phase is divided into phase 1 and phase 2 for different years. When comparing the dynamic changes in different phases, the Wilcoxon group test is also used to verify the significance of the difference in the critical threshold of feedback in different phases. The test criteria are p < 0.001 marked as "***", p < 0.01 marked as "**", p < 0.05 marked as "*", and p > 0.05 marked as "-".

[0010] Furthermore, in step S4, the different levels of meteorological drought in the early stage include moderate drought, severe drought, and extreme drought; when analyzing the impact, the proportion of the decrease in the feedback critical threshold and the proportion of the decrease in feedback sensitivity under different drought levels are statistically analyzed, and the influence of the previous drought level on the feedback critical threshold and feedback sensitivity is judged in combination with the spatial distribution characteristics.

[0011] Furthermore, in step S5, when constructing the XGB-SHAP attribution model, the selected influencing factors include: Meteorological factors: surface albedo ALB, evapotranspiration ET, shortwave radiation flux RAD_S, longwave radiation flux RAD_L, latent heat flux LH, sensible heat flux SH, wind speed WS, saturated vapor pressure difference VPD, air temperature TMP, precipitation PRE, and aridity index AI. Soil and topographic factors: Soil moisture content (SM), silt content (SILT), sand content (SAND), clay content (CLAY), elevation (DEM), soil organic carbon content (ORGC), and total nitrogen content (TOTN); Biological factors: Species diversity (ASR).

[0012] Furthermore, in step S5, the construction process of the XGB-SHAP attribution model includes: S51: Divide the attribution data into a test set and a training set in a 2:8 ratio, and use five-fold cross-validation to determine the optimal parameters of the model. The optimal parameters include colsamplebytree, learningrate, maxdepth, n_estimators, and subsample. S52: Construct an attribution model based on optimal parameters, output the SHAP value of each influencing factor through the model, determine the direction of the factor's contribution to the target variable by the positive or negative sign of the SHAP value, and use the average absolute SHAP value as the characteristic importance of the factor to identify the dominant factor. S53: When exploring the feedback mechanism, the study area was divided into humid zone (AI≥0.65), semi-humid zone (0.65≥AI>0.5), semi-arid zone (0.5≥AI>0.2), arid zone (0.2≥AI>0.05), and extremely arid zone (AI≤0.05) based on the aridity index AI. The feedback mechanism was analyzed by combining the role patterns of dominant factors under positive and negative feedback states in different climate zones.

[0013] The beneficial effects of this invention are: Enhancing the flexibility and accuracy of high-dimensional feedback relationship modeling: This invention uses the vine Copula to construct a vegetation feedback critical threshold framework, decomposing the complex dependencies of high-dimensional variables into multiple binary pair-copulas. The optimal function is selected through the AIC criterion, overcoming the limitation of insufficient flexibility of traditional multi-parameter Copula models in high-dimensional systems. This framework can accurately characterize the dependency structure of different regions, different vegetation indices and SPEI, and the calculation error of the feedback critical threshold is reduced by 15%-20% compared with traditional models, providing a high-precision quantitative basis for subsequent feedback state identification. Clarifying the critical state and sensitivity characteristics of ecological-drought feedback: This invention defines P1 and P2, determines the critical threshold of feedback by their intersection, and quantifies the feedback sensitivity by the curve fitting coefficient A of P2. For the first time, it systematically classifies 5 types of feedback transformation, which not only clarifies the safe boundary of vegetation transforming from drought relief to drought exacerbation, but also judges the feedback intensity by the magnitude of the absolute value of the sensitivity coefficient, providing a precise basis for identifying key areas of drought risk in ecosystems.

[0014] Achieving dynamic tracking and differentiated attribution by region: This invention divides the research period into two stages and combines Wilcoxon grouping tests to reveal the interdecadal changes in feedback thresholds and sensitivity. For the first time, it quantifies the long-term impact of climate warming on the ecological-drought feedback relationship and provides a historical baseline for predicting future feedback trends. For 8 vegetation zones and 5 climate zones, this invention accurately identifies the dominant factors in different regions, improving the regional specificity of factor contribution analysis. Attached Figure Description

[0015] Figure 1 A schematic diagram of a vine structure for a random grid; Figure 2 A schematic diagram of five response patterns of vegetation to meteorological drought; Figure 3 This is a schematic diagram illustrating the impact of influencing factors on the feedback status of different climate zones. Figure 4 P1 distribution map for different vegetation zones; Figure 5 Feedback critical threshold diagrams for different vegetation zones; Figure 6 Feedback sensitivity maps for different vegetation zones; Figure 7 An importance map of the spatial distribution factors affecting the feedback critical threshold; Figure 8 Importance map of factors influencing the spatial distribution of feedback sensitivity. Detailed Implementation

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0017] Please see Figure 1-8 This invention provides a technical solution for studying the feedback of ecosystems to meteorological drought: Example

[0018] The core idea of ​​this invention is to quantify the feedback threshold and sensitivity by constructing the Copula vegetation feedback critical threshold framework, combining it with multi-stage and multi-drought level analysis, and then using the XGB-SHAP attribution model to reveal the feedback mechanism, thereby achieving precise research on the interaction between ecosystems and meteorological drought. The following embodiments use terrestrial ecosystems in China as the research object, and the specific implementation process is as follows: S1. Determining the Research Period and Acquiring Basic Data Considering that the feedback between terrestrial ecosystems and the atmosphere mainly occurs during the season of strong land-atmosphere coupling, this embodiment selects the summer months of June to August as the core research period. During this period, vegetation grows vigorously, and surface energy exchange and water cycle processes are active, which can most directly reflect the feedback effect of vegetation on meteorological drought.

[0019] Basic data undergoes rigorous screening and preprocessing. Vegetation status data utilizes three core indicators after detrending: Leaf Area Index (LAI, representing vegetation cover and growth structure), Gross Primary Productivity (GPP, representing vegetation photosynthetic capacity and carbon fixation), and Ecosystem Respiration (TER, representing ecosystem energy consumption processes). Detrending eliminates interference from long-term vegetation growth trends on short-term drought feedback relationships, ensuring the data focuses on vegetation status changes driven by drought events. The meteorological drought index data uses the Standardized Precipitation Evapotranspiration Index (SPEI), which comprehensively considers the impact of precipitation replenishment and evapotranspiration on drought, accurately quantifying the degree of meteorological drought at different times and providing a unified drought assessment standard for subsequent probability calculations and feedback analysis. All data undergoes preprocessing steps including spatial matching (unifying raster resolution and coordinate system), missing value imputation (using neighborhood interpolation or temporal fitting methods), and outlier removal to ensure data integrity and consistency.

[0020] S2. Construction of a vegetation feedback critical threshold framework based on Copula. S21. Determination of the Optimal Vine Structure: Since multi-parameter copulas lack flexibility in high-dimensional system modeling and cannot effectively construct dependency structures for composite events, this embodiment uses vine copulas for modeling. Vine copulas can decompose complex dependencies into multiple binary pair-copulas through a tree structure, providing greater flexibility for high-dimensional system modeling. Five copula functions—Clayton, Frank, Gumbel, Gaussian, and Student'st(t)—are selected as candidate pair-copula functions. The AIC criterion is used as the selection standard for the optimal pair-copula function; the function with the smallest AIC value is the optimal pair-copula function between corresponding variables. Taking July as an example, three grids—103.25°E, 25.75°N, 112.25°E, 26.75°N, and 113.25°E, 36.75°N—are randomly selected to construct vine structures for LAI, GPP, TER, and SPEI, respectively. The results are as follows: Figure 1 As shown, there are differences between the vegetation index and the optimal pair-copula function of SPEI under different grids, and the vine structure can effectively characterize their dependency relationship.

[0021] S22. Feedback Critical Threshold and Feedback Sensitivity Calculation: First, calculate two key probability indicators, P1 and P2, where P1 is the probability of a meteorological drought occurring in the current month given that a meteorological drought occurred in the previous month. The calculation formula is: P1 = P(SPEI) t ≤-0.5|SPEI t-1 ≤-0.5); P2 represents the probability of a meteorological drought occurring in the current month, assuming a meteorological drought occurred in the previous month and the vegetation is in a certain condition. The formula is: P2 = P(SPEI) t ≤-0.5|SPEI t-1 ≤-0.5, VEG t ≤veg); In the formula, t represents the current month, t-1 represents the previous month, and veg represents the vegetation state index, which ranges from 1 to 99th, increasing by 1 unit in length, representing the complete state gradient of the vegetation from poor to good.

[0022] The intersection of curves P1 and P2 is used as the critical threshold for feedback. Before and after this point, the feedback effect of vegetation status on meteorological drought changes fundamentally. The fitting coefficient A of curve P2 is used to characterize the feedback sensitivity. When A < 0, the feedback sensitivity is negative, and when A > 0, the feedback sensitivity is positive. The larger the absolute value of the sensitivity, the stronger the feedback effect of vegetation on drought.

[0023] S23. Determination of Feedback Transformation Types: Based on the relationship between P1 and P2 and the number of feedback critical thresholds, the feedback transformation types of vegetation to meteorological drought are divided into 5 categories (e.g., Figure 2 (as shown) Type 1 (positive-negative feedback transformation): As vegetation loss increases (veg percentile decreases), P2 gradually decreases. When vegetation loss reaches the feedback critical threshold, P2 changes from being greater than P1 to being less than P1. That is, the effect of vegetation on drought changes from "exacerbating" to "alleviating", corresponding to A < 0. This type is mainly concentrated in southern China (high forest coverage, excessive evapotranspiration of vegetation in the early stage aggravates drought, and reduced evapotranspiration after vegetation loss alleviates drought). Type 2 (Negative-Positive Feedback Transformation): As vegetation loss increases, P2 gradually increases. After reaching the critical feedback threshold, P2 changes from being less than P1 to being greater than P1, meaning that the effect of vegetation on drought changes from "alleviating" to "exacerbating," corresponding to A>0. This type is the dominant type in northern regions. The proportions of LAI, GPP, and TER are 57.3% / 72.3% / 68.7% in June, 63.4% / 78.3% / 71.3% in July, and 45.1% / 63.4% / 63.8% in August, respectively (after vegetation reduction in arid and semi-arid areas of northern China, evapotranspiration decreases, leading to insufficient atmospheric water vapor, and the increased albedo of exposed surfaces exacerbates drought). Type 3 (No feedback - P2 is always lower than P1): The P2 curve is always below the P1 curve, indicating that no matter what state the vegetation is in, it will not change the probability of drought occurrence, and the vegetation has no feedback effect on drought. Type 4 (No feedback - P2 is always higher than P1): The P2 curve is always above the P1 curve. Changes in vegetation status do not affect the probability of drought occurrence, and there is no feedback effect. Type 5 (Dual Threshold Transition): As vegetation loss increases, P2 first decreases and then increases, with two feedback critical thresholds. The effect of vegetation on drought first changes from "exacerbating" to "alleviating" and then back to "exacerbating". This type accounts for less than 3% of all vegetation indices and belongs to a special marginal case.

[0024] Subsequent feedback threshold and sensitivity analysis mainly focused on Type 1 and Type 2, as they are the mainstream forms of terrestrial ecosystem feedback to meteorological drought and have significant climate regulation implications.

[0025] S3. Comparative Analysis of Feedback Characteristics at Different Stages S31. To reveal the evolution of vegetation feedback effects under the background of long-term climate change, the research period is divided into two stages: 1982-1999 (stage 1) and 2000-2022 (stage 2). By comparing the feedback characteristics of the two stages, the dynamic change of vegetation feedback to meteorological drought on a long-term scale is revealed.

[0026] S32. First, based on the two-stage baseline data, calculate P1 and P2 for each stage. Then, obtain the feedback critical threshold and feedback sensitivity coefficient for the corresponding stage using the Copula framework. To verify the statistical significance of the two-stage difference, use the Wilcoxon grouping test (a non-parametric test suitable for non-normally distributed data): test the two-stage threshold / sensitivity coefficient for each grid cell, marking p < 0.001 as "***" (extremely significant difference), p < 0.01 as "**" (significant difference), p < 0.05 as "*" (generally significant difference), and p > 0.05 as "-" (no significant difference).

[0027] From the results of the phased changes: the probability of summer drought occurrence shows obvious phase differences. The average P1 in June and August is higher in phase 1 (0.39, 0.45) than in phase 2 (0.38, 0.44). In July, it rises from 0.38 in phase 1 to 0.47 in phase 2. Spatially, the proportion of areas with increased P1 in July reaches 68.6% (general increase). The increased areas in June are concentrated in the north and the central Qinghai-Tibet Plateau, while the increased areas in August are concentrated in the north and parts of the southwest. In terms of feedback threshold, the negative-positive feedback threshold of LAI increased significantly in the Yellow River Basin in June. In July, the negative-positive feedback thresholds in temperate desert regions (R7) and warm temperate deciduous broad-leaved forest regions (R3) decreased significantly. In August, the negative-positive feedback thresholds of LAI in the Northwest region increased, while those in the Northeast and Southeast regions decreased. In terms of feedback sensitivity, the proportion of regions with positive LAI sensitivity coefficients increased from 50.4% in June to 66.2% in August. The sensitivity of the Northwest inland and northern Yellow River Basin increased significantly in June, while the sensitivity of the Northeast and Southeast regions increased significantly in August. The sensitivity of the southeastern Qinghai-Tibet Plateau (positive-negative feedback type region) showed a downward trend, with some grids changing from positive sensitivity to negative sensitivity.

[0028] From the perspective of vegetation zone differences: In June, the region with the largest increase in P1 was the cold temperate coniferous forest region (R1), and the region with the largest decrease was the alpine vegetation region of the Qinghai-Tibet Plateau (R8); In July, the region with the largest increase in P1 was the temperate desert region (R7, range [-0.65, 0.33]), and the region with the largest decrease was the temperate coniferous-deciduous-broadleaf mixed forest (R2, range [-0.17, 0.40]); In August, the region with the largest increase in P1 was R1 (range [-0.49, 0.17]), and the region with the largest decrease was the warm temperate deciduous-broadleaf forest region (R3, range [-0.24, 0.39]). Significant changes in the two-stage feedback thresholds were mainly concentrated in the temperate desert region (R7) and the warm temperate deciduous broad-leaved forest region (R3). For example, in June, the thresholds of GPP and TER increased significantly in R3 and R7, and decreased significantly in R2. In terms of feedback sensitivity, LAI showed significant changes in all vegetation zones in June, no significant changes in the tropical monsoon rainforest and rainforest region (R5), temperate grassland region (R6), and R8 in July, and increased significantly in all regions except R7 in August.

[0029] S4. Impact Analysis of Different Levels of Meteorological Drought in the Previous Period S41. Drought Level Classification: The level of drought in the early stage will change the initial coupling state between vegetation and drought, thereby affecting the current feedback effect. In this embodiment, the early meteorological drought is divided into three levels: moderate drought (SPEI range [-1.0, -0.5]), severe drought (SPEI range [-1.5, -1.0]), and extreme drought (SPEI ≤ -1.5). Using GPP as the representative vegetation index, the influence of different levels of early drought on the feedback critical threshold and sensitivity is analyzed.

[0030] S42. Impact Pattern Analysis: Based on the definition criteria of different levels of early drought, samples corresponding to the drought levels were selected, and P1 and P2 were calculated for each level. Then, the feedback critical threshold and sensitivity coefficient were obtained through the Copula framework.

[0031] Statistical results show that the higher the drought level in the early stage, the larger P1 is, and the feedback critical threshold shows a "rightward shift" trend. During moderate to severe drought, the proportion of the feedback critical threshold decreasing from June to August is 72.7%, 81.7%, and 77.8%, respectively. During severe to extreme drought, the proportions of the decrease are 72.6%, 80.9%, and 76.4%, respectively. This indicates that the more severe the drought in the early stage, the greater the vegetation loss needs to be before it can shift from "alleviating drought" to "exacerbating drought". The "rightward shift" of the feedback critical threshold is essentially an adaptive adjustment of vegetation to extreme drought.

[0032] Regarding feedback sensitivity, as the drought level in the preceding period increased, the vegetation's feedback sensitivity to meteorological drought mainly decreased: during moderate to severe drought, the reduction rates in sensitivity from June to August were 60.5%, 71.8%, and 63.0%, respectively; during severe to extreme drought, the reduction rates were 61.5%, 71.1%, and 61.0%, respectively. Spatially, the areas with reduced positive sensitivity were mainly concentrated in northern China, especially the temperate grassland region (R6). In this region, the more severe the preceding drought, the more significantly vegetation growth was inhibited, and the impact of vegetation status changes on drought probability weakened, leading to a decrease in sensitivity. In contrast, sensitivity increased slightly in some inland areas of Northwest China, which may be related to the strengthening of the vegetation-soil moisture coupling relationship under extreme drought conditions in this region.

[0033] S5. Exploring the Feedback Mechanism Based on the XGB-SHAP Attribution Model Influence factor selection: To comprehensively identify the dominant factors driving the feedback critical threshold, the spatial distribution of sensitivity, and dynamic changes, a multi-dimensional influence factor system is constructed, including: Meteorological factors: Albedo (ALB, affecting surface energy absorption), Evapotranspiration (ET, affecting water balance), Shortwave Radiation Flux (RAD_S), Longwave Radiation Flux (RAD_L, affecting surface temperature), Latent Heat Flux (LH), Sensible Heat Flux (SH, affecting energy distribution), Wind Speed ​​(WS, affecting water vapor diffusion), Vapor Pressure Difference (VPD, affecting vegetation stomatal conductance), Temperature (TMP), Precipitation (PRE, drought recharge), Aridity Index (AI, characterizing climate dryness / wetness background); Soil and topographic factors: soil moisture content (SM, vegetation water supply), silt content (SILT), sand content (SAND), clay content (CLAY, affecting soil water retention capacity), elevation (DEM, affecting water and heat redistribution), soil organic carbon content (ORGC), and total nitrogen content (TOTN, affecting vegetation productivity). Biological factors: Species diversity (ASR, which affects ecosystem stability).

[0034] XGB-SHAP Attribution Model Construction: The XGB-SHAP model combines the strong regression capability of Extreme Gradient Boosting (XGBoost) with the interpretability of Shapley Additive Explanation (SHAP). The specific construction process is as follows: S51. Data Partitioning and Parameter Optimization: The attribution data is divided into a test set (to verify the model's generalization ability) and a training set (to build the model) in a 2:8 ratio. Five-fold cross-validation is used to determine the optimal parameters of the model. The core parameters include colsamplebytree (feature sampling ratio), learning rate, maxdepth (maximum tree depth), n_estimators (number of trees), and subsample (sample sampling ratio). The optimal parameters differ for different target variables (such as positive-negative feedback threshold and negative-positive feedback sensitivity), as shown in the table below:

[0035]

[0036] S52. Dominant Factor Identification: Construct an attribution model based on optimal parameters and output the SHAP value of each influencing factor. A SHAP value > 0 indicates that the factor has a positive contribution to the target variable (such as increasing the feedback threshold), and a SHAP value < 0 indicates a negative contribution. The "average absolute SHAP value" is used as the characteristic importance of the factor. The higher the value, the greater the influence of the factor on the target variable.

[0037] S53. Analysis of Feedback Mechanisms in Different Climate Zones: Based on the aridity index (AI), the study area was divided into five climate zones: humid zone (AI ≥ 0.65), semi-humid zone (0.65 ≥ AI > 0.5), semi-arid zone (0.5 ≥ AI > 0.2), arid zone (0.2 ≥ AI > 0.05), and extremely arid zone (AI ≤ 0.05). Taking the July GPP as an example, the dominant mechanisms under positive and negative feedback states in different zones were analyzed in conjunction with the SHAP value: Positive feedback mechanism (vegetation exacerbates drought): The dominant factor in extremely arid areas is the previous month's drought (SPEI06). High SPEI06 (drought persistence) exacerbates the current drought through continuous depletion of soil moisture. The dominant factor in arid areas is VPD (SHAP values ​​are concentrated in the negative range). High VPD leads to stomatal closure and reduced evapotranspiration, and insufficient atmospheric water vapor further exacerbates drought. Low PRE and high WS intensify drought by reducing water recharge and accelerating water vapor diffusion. The dominant factors in semi-arid areas are low GPP (decreased vegetation productivity) and high TMP. Low GPP reduces ecosystem water use efficiency, and high TMP increases evapotranspiration demand, both exacerbating drought. The dominant factors in semi-humid areas are high TMP, low GPP, and PRE (with even positive and negative contributions). High TMP and low GPP are the main drivers of drought exacerbation, and the uncertainty of PRE leads to fluctuations in feedback intensity. The dominant factors in humid areas are low PRE and low GPP. Although overall water is sufficient, short-term low PRE and vegetation growth inhibition can still trigger local drought exacerbation.

[0038] Negative feedback mechanism (vegetation alleviates drought): The dominant factors in extremely arid areas are high SPEI06 (drought reduction) and high PRE, which alleviate drought by increasing soil moisture replenishment; the dominant factor in arid, semi-arid, and semi-humid areas is high GPP, which regulates the local microclimate by absorbing CO2 through photosynthesis, while vegetation canopy interception and root water retention increase soil moisture reserves, thus alleviating drought; the dominant factor in humid areas is low RAD_L, which reduces surface energy input, lowers surface temperature and evapotranspiration demand, thereby alleviating drought.

[0039] This embodiment, through the complete process described above, achieves quantitative characterization, dynamic tracking, and mechanism analysis of the feedback relationship between regional terrestrial ecosystems and meteorological drought in China, which can provide scientific support for drought response and ecosystem management under the background of climate change.

[0040] The above description is only a preferred embodiment of the present invention and does not limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A research method for the feedback of ecosystems to meteorological drought, characterized in that, Includes the following steps: S1: Determine the study period and basic data, select the season with strong land-atmosphere coupling as the study period, and obtain detrended vegetation status data and meteorological drought index data. S2: Construct a vegetation feedback critical threshold framework based on Copula, quantify the feedback critical threshold and feedback sensitivity of vegetation status to meteorological drought. The feedback critical threshold is the intersection of the probability of meteorological drought in the current month when the previous meteorological drought occurred, P1, and the probability of meteorological drought in the current month when the previous meteorological drought occurred and the vegetation is in a specific state, P2. The feedback sensitivity is characterized by the fitting coefficient of the P2 curve. S3: Divide the research into stages and compare the dynamic changes of the feedback critical threshold and feedback sensitivity in different stages; S4: Analyze the impact of different levels of meteorological drought in the early stage on the feedback critical threshold and feedback sensitivity; S5: Based on the XGB-SHAP attribution model, analyze and identify the dominant factors affecting the critical threshold of feedback, the spatial distribution of feedback sensitivity, and dynamic changes, and explore the feedback mechanism of vegetation status on meteorological drought.

2. The research method for ecosystem feedback to meteorological drought according to claim 1, characterized in that, In step S1, the vegetation status data includes leaf area index (LAI) data, total primary productivity (GPP) data, and ecosystem respiration (TER) data; the meteorological drought index data is standardized precipitation evapotranspiration index (SPEI) data.

3. The research method for ecosystem feedback to meteorological drought according to claim 1, characterized in that, In step S2, the specific process of constructing a vegetation feedback critical threshold framework based on Copula includes: S21: Determine the optimal vine structure. Select five Copula functions, namely Clayton, Frank, Gumbel, Gaussian, and Student'st(t), as candidate Pair-copula functions. Use the AIC criterion to select the Pair-copula function with the smallest AIC and construct the optimal vine structure between vegetation status data and meteorological drought index data. S22: Calculate the feedback critical threshold and feedback sensitivity. First, calculate P1 and P2 respectively. The formula for calculating P1 is: P1 = P(SPEI) t ≤-0.5|SPEI t-1 ≤-0.5), The formula for calculating P2 is: P2 = P(SPEI) t ≤-0.5|SPEI t-1 ≤-0.5, VEG t ≤veg), In the formula, t represents the current month, t-1 represents the previous month, and veg is the vegetation status index, which ranges from 1 to 99th and increases by 1 unit. The critical threshold for feedback is determined by the intersection of P1 and P2. The feedback sensitivity is characterized by the fitting coefficient A of the P2 curve. When A < 0, the feedback sensitivity is negative, and when A > 0, the feedback sensitivity is positive.

4. The research method for ecosystem feedback to meteorological drought according to claim 3, characterized in that, Step S2 further includes determining the feedback transformation type of vegetation to meteorological drought, wherein the feedback transformation type includes: Type 1: As vegetation loss increases, the probability of vegetation causing meteorological drought decreases. After reaching the feedback critical threshold, vegetation changes from aggravating meteorological drought to alleviating it. The P2 curve fitting coefficient A < 0. Type 2: As vegetation loss increases, the probability of vegetation causing meteorological drought increases. After reaching the feedback critical threshold, vegetation changes from alleviating meteorological drought to aggravating it. The P2 curve fitting coefficient A > 0. Type 3: P2 is always lower than P1, and changes in vegetation status have no impact on meteorological drought; Type 4: P2 is always higher than P1, and changes in vegetation status have no impact on meteorological drought; Type 5: As vegetation loss increases, the impact of vegetation on drought first shifts from exacerbation to mitigation, then intensifies again after falling below another threshold, indicating two feedback critical thresholds.

5. The research method for ecosystem feedback to meteorological drought according to claim 1, characterized in that, In step S3, the research phase is divided into phase 1 and phase 2 for different years. When comparing the dynamic changes in different phases, the Wilcoxon group test is also used to verify the significance of the difference in the critical threshold of feedback in different phases. The test criteria are p < 0.001 marked as "***", p < 0.01 marked as "**", p < 0.05 marked as "*", and p > 0.05 marked as "-".

6. The research method for ecosystem feedback to meteorological drought according to claim 1, characterized in that, In step S4, the different levels of meteorological drought in the early stage include moderate drought, severe drought, and extreme drought. When analyzing the impact, the proportion of the decrease in the feedback critical threshold and the proportion of the decrease in feedback sensitivity under different drought levels are statistically analyzed. The influence of the previous drought level on the feedback critical threshold and feedback sensitivity is judged in combination with the spatial distribution characteristics.

7. The research method for ecosystem feedback to meteorological drought according to claim 1, characterized in that, In step S5, when constructing the XGB-SHAP attribution model, the selected influencing factors include: Meteorological factors: surface albedo ALB, evapotranspiration ET, shortwave radiation flux RAD_S, longwave radiation flux RAD_L, latent heat flux LH, sensible heat flux SH, wind speed WS, saturated vapor pressure difference VPD, air temperature TMP, precipitation PRE, and aridity index AI. Soil and topographic factors: Soil moisture content (SM), silt content (SILT), sand content (SAND), clay content (CLAY), elevation (DEM), soil organic carbon content (ORGC), and total nitrogen content (TOTN); Biological factors: Species diversity (ASR).

8. The research method for ecosystem feedback to meteorological drought according to claim 7, characterized in that, In step S5, the construction process of the XGB-SHAP attribution model includes: S51: Divide the attribution data into a test set and a training set in a 2:8 ratio, and use five-fold cross-validation to determine the optimal parameters of the model. The optimal parameters include colsamplebytree, learningrate, maxdepth, n_estimators, and subsample. S52: Construct an attribution model based on optimal parameters, output the SHAP value of each influencing factor through the model, determine the direction of the factor's contribution to the target variable by the positive or negative sign of the SHAP value, and use the average absolute SHAP value as the characteristic importance of the factor to identify the dominant factor. S53: When exploring the feedback mechanism, the study area was divided into humid zone (AI≥0.65), semi-humid zone (0.65≥AI>0.5), semi-arid zone (0.5≥AI>0.2), arid zone (0.2≥AI>0.05), and extremely arid zone (AI≤0.05) based on the aridity index AI. The feedback mechanism was analyzed by combining the role patterns of dominant factors under positive and negative feedback states in different climate zones.

Images