Mangrove forest protection effect intelligent evaluation and scene prediction system

Abstract

The invention discloses a mangrove forest protection effect intelligent evaluation and scene prediction system, and relates to the field of ecological environment information intelligence, and the system comprises a key driving factor recognition module which is used for determining a core variable influencing the protection effect and a dynamic response interval of the core variable based on a statistical analysis and nonlinear fitting method; the protection effect quantitative evaluation module is used for calculating a net improvement effect of protection intervention by comparing ecological index differences inside and outside the protection area; the causal effect analysis module is used for separating independent contributions of natural factors and artificial protection by adopting a dual machine learning framework; the space-time dynamic modeling module is used for performing uncertainty modeling on the space-time evolution of the ecological indexes by using a Bayesian hierarchical structure; and the multi-scene prediction module is used for realizing ecological response prediction under multi-factor driving and supporting input and simulation of user-defined scenes. According to the scheme, multi-dimensional and high-precision evaluation and future trend reliable prediction of mangrove forest protection effects can be realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent ecological and environmental information, and in particular to an intelligent assessment and scenario prediction system for mangrove protection effectiveness. Background Technology

[0002] Faced with the multiple stresses posed by climate change, marine environmental shifts, and human activities to typical marine ecosystems such as mangroves under the backdrop of global change, their ecological response processes exhibit high complexity and uncertainty. As a key blue carbon ecosystem, mangroves play irreplaceable roles in carbon sequestration, shoreline protection, and biodiversity maintenance; however, in recent years, habitat degradation, spatial shrinkage, and functional weakening have become increasingly prominent problems. Scientifically assessing the effectiveness of conservation strategies and predicting their evolutionary trends under different climate-social scenarios has become an urgent need in the field of ecological conservation and adaptive management. Traditional methods often rely on static indicator comparisons or single-path predictions, which often struggle to integrate multi-source heterogeneous data, identify causal mechanisms, quantify uncertainties, and achieve multi-scenario coupled simulations, thus limiting the explanatory power and decision support capabilities of the assessment results.

[0003] Existing methods for assessing the effectiveness of mangrove conservation typically suffer from the following limitations: First, most studies fail to effectively distinguish the independent contributions of natural changes and conservation strategies to ecological responses, lacking quantitative identification and verification mechanisms for causal relationships; second, traditional models often neglect nonlinear interaction effects and ecological threshold behaviors among multiple factors, leading to oversimplified mechanism explanations; third, in terms of prediction, existing methods rely heavily on deterministic modeling, making it difficult to integrate multiple future scenario paths and lacking systematic quantification and spatial explicit expression of outcome uncertainties; fourth, the assessment process is often fragmented, failing to achieve integrated processing from data processing and mechanism mining to predictive output, thus limiting the practical application of the findings in cross-departmental decision-making and public communication. Therefore, based on these challenges, this invention proposes an intelligent assessment and scenario prediction system for mangrove conservation effectiveness. Summary of the Invention

[0004] Purpose of the invention To address the aforementioned issues, the present invention aims to provide an intelligent assessment and scenario prediction system for mangrove conservation effectiveness. This system is designed to scientifically quantify the net effect of conservation strategies, accurately predict the spatiotemporal evolution trends of ecological indicators, and reliably estimate conservation effectiveness under different climate and social pathways. This will provide data-driven decision support for adaptive management of mangrove ecosystems, optimization of conservation strategies, and precise risk control.

[0005] Technical solution To achieve the above objectives, this invention provides an intelligent assessment and scenario prediction system for mangrove conservation effectiveness. This system automatically integrates and standardizes multi-source ecological data, combines dual machine learning algorithms to isolate the independent causal effects of natural factors and conservation measures, utilizes a Bayesian hierarchical model to quantify the uncertainty of the spatiotemporal evolution of ecological indicators, and integrates multiple machine learning algorithms to achieve probabilistic prediction and sensitive factor identification under multiple scenarios. Finally, it outputs various charts and strategy reports through visualization to support dynamic assessment of conservation effectiveness, simulation of future trends, and risk management decisions.

[0006] In a first aspect, the present invention provides an intelligent assessment and scenario prediction system for mangrove conservation effectiveness, comprising: The data preprocessing module is configured to automatically clean, standardize, and spatially register multi-source heterogeneous ecological data to improve data quality and consistency. The key driver identification module is used to determine the core variables affecting conservation effectiveness and their dynamic response ranges based on statistical analysis and nonlinear fitting methods, so as to identify ecological driving elements with significant explanatory power. The conservation effectiveness quantitative assessment module is used to calculate the net improvement effect of conservation interventions by comparing the differences in ecological indicators inside and outside the protected area, thereby achieving objective quantification of the effects of conservation measures. The causal effect analysis module is used to separate the independent contributions of natural factors and human conservation using a dual machine learning framework, thereby accurately identifying the causal effects of policy treatments. The spatiotemporal dynamic modeling module is used to perform uncertainty modeling of the spatiotemporal evolution of ecological indicators using Bayesian hierarchical structure, so as to provide highly reliable trend estimation and heterogeneity characterization, and integrate spatiotemporal autocorrelation features and environmental gradient differences to construct nonlinear response relationships, thereby enhancing the ability to characterize the spatiotemporal heterogeneity of ecological indicators and the reliability of future trend prediction. The multi-scenario prediction module integrates various machine learning algorithms to achieve ecological response prediction driven by multiple factors, and supports user-defined inputs and simulations of climate, social and management scenarios to enhance the system's predictive adaptability and strategic reference value in a variable environment. The results output module provides the ability to generate and export various visualization charts and recommendation reports, enhancing the interpretability of the results and the ability to support decision-making.

[0007] Furthermore, the system also includes a data input interface that supports the access and real-time updates of remote sensing image data, ground observation data, environmental variable data, and human activity intensity data, ensuring the comprehensiveness and timeliness of the data source.

[0008] Furthermore, the data preprocessing module has the ability to automatically process missing values, time series alignment, spatial interpolation and resampling of multi-source ecological monitoring data, so as to improve data availability and model input quality.

[0009] Furthermore, the key driving factor identification module identifies key influencing factors of conservation effectiveness and their suitable ecological thresholds through variable correlation analysis and ecological response curve fitting, thereby optimizing the scientificity and reliability of factor selection.

[0010] Furthermore, the quantitative assessment module for conservation effectiveness is configured to calculate the differences in biomass, vegetation index, and productivity indicators inside and outside the protected area, and output the average treatment effect and geographical distribution heterogeneity of conservation measures to support differentiated adjustments to regional conservation strategies.

[0011] Furthermore, the causal effect analysis module employs a machine learning-based causal inference method to quantify the net causal effect between protection strategies and ecological responses, effectively overcoming the interference of confounding variables.

[0012] Furthermore, the causal effect analysis module further includes an adaptive covariate balancing unit, which is used to suppress the interference of high-dimensional covariate distribution imbalance on causal effect estimation through a dynamic weight adjustment mechanism, thereby improving the robustness and accuracy of the net effect assessment of the protection strategy.

[0013] Furthermore, the spatiotemporal dynamic modeling module establishes a Bayesian hierarchical model by introducing spatiotemporal autocorrelation structures and environmental covariates, which is used to estimate the posterior distribution and uncertainty of ecological state variables, thereby achieving more accurate spatiotemporal prediction and risk mapping.

[0014] Furthermore, the multi-scenario prediction module integrates various tree-based and ensemble learning algorithms to achieve probabilistic prediction of ecological indicators and identification of sensitive factors under different development scenarios, thereby improving the system's simulation capabilities and robustness in complex scenarios.

[0015] Furthermore, the multi-scenario prediction module allows the embedding of multiple climate evolution paths and socio-economic scenarios, and supports dynamic simulation and effect comparison of protection strategy adjustment scenarios, thereby providing multi-option decision support for coping with the future volatile environment.

[0016] Furthermore, the output module can generate spatiotemporal variation maps of ecological indicators, risk zoning maps, uncertainty distribution maps, and multi-format assessment reports. It has interactive visualization and cross-platform export capabilities, which significantly improves the ease of application and communication efficiency of the results.

[0017] In a second aspect, the present invention also provides a non-transitory computer-readable medium storing computer-executable instructions, which, when executed by a processor, control the operation of the system described in the first aspect; the medium includes, but is not limited to, solid-state memory, optical storage devices, or cloud storage systems, ensuring that the system can be stably implemented on a variety of hardware platforms.

[0018] This invention provides an integrated software system for evaluating the effectiveness of mangrove conservation and predicting future trends. Based on an integrated technical architecture, the system integrates eight core modules: multi-source ecological data acquisition and preprocessing, key driver factor screening and threshold setting, net effect assessment of conservation strategies, causal inference based on dual machine learning, spatiotemporal hierarchical Bayesian modeling, ensemble learning algorithm prediction, multi-scenario simulation, and visualization of results. Through modular collaboration and feedback optimization mechanisms, the system achieves closed-loop management of the entire process from data processing and mechanism mining to trend prediction. It supports precise quantification of mangrove ecosystem conservation effectiveness, analysis of driving factors, simulation of dynamic evolution under multiple future scenarios, and quantification of uncertainties. Ultimately, it outputs graphs, reports, and risk maps with spatial explicit expression and decision support functions.

[0019] This system enables multi-dimensional and high-precision assessment of mangrove conservation effectiveness and reliable prediction of future trends, significantly improving the scientific rigor and mechanistic explanatory power of identifying the net effects of conservation strategies. By integrating causal inference, spatiotemporal modeling, and multi-scenario analysis, it effectively overcomes the limitations of traditional methods in characterizing nonlinear responses, integrating factor interactions, and predicting long-term time series. The system possesses adaptability and forward-looking extrapolation capabilities for ecological responses under various climatic and social pathways, and can output visualized results and customized reports. It provides systematic, quantitative, and visualized technical support for the formulation of ecological protection strategies, the identification of risk areas, and the optimization of management measures, and has broad operational application and promotion value.

[0020] Beneficial effects By implementing the intelligent assessment and scenario prediction system for mangrove conservation effectiveness provided by the present invention, the following technical effects are achieved: (1) It supports the dynamic embedding and combined analysis of multiple climate, social and strategic scenarios, and can systematically assess the response differences and potential risks of protection effectiveness under different development paths. It has achieved a leap from single-point prediction to multi-scenario extrapolation, and significantly enhanced the system's decision support capabilities, strategic adaptability and foresight in dealing with future uncertainties.

[0021] (2) Through modular design, a closed-loop system for data, models, and decisions is achieved, supporting automatic flow and feedback optimization of multi-source information. This effectively solves the pain points of fragmented processes and difficulty in coordinating results in traditional evaluation, improves the overall computational efficiency and output consistency of the system, and provides a stable and reliable technical foundation for cross-scale and multi-terminal business applications.

[0022] (3) The dynamic weight adjustment mechanism effectively suppresses the interference of high-dimensional covariate distribution imbalance on the estimation of causal effects, and significantly improves the robustness and accuracy of the net effect assessment of protection strategies. It effectively overcomes the estimation bias caused by the intertwining of confounding factors in complex ecosystems by traditional causal models, and enhances the scientific explanatory power and statistical reliability of the independent contribution of protection measures.

[0023] (4) By organically integrating spatiotemporal autocorrelation characteristics and environmental gradient differences through dynamic coupling coefficients, the ability to characterize the nonlinear response and spatial heterogeneity of ecological indicators is significantly improved. It enables more accurate simulation and uncertainty quantification of ecosystem evolution trends under multiple future scenarios, providing a reliable model basis for identifying high-risk areas and formulating differentiated protection strategies. Attached Figure Description

[0024] To make the above-mentioned intelligent assessment and scenario prediction system for mangrove protection effectiveness of the present invention more obvious and understandable, the accompanying drawings used in the specific embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0025] Figure 1 This is a schematic diagram of the technical route of this application; Figure 2 A schematic diagram illustrating the technical roadmap of a system for assessing the effectiveness of global change risk in the conservation of important marine species and predicting the evolutionary trends of its impacts. Figure 3 A schematic diagram illustrating representative indicators of mangrove biomass conservation effectiveness; Figure 4 A schematic diagram illustrating the cumulative local interaction effect of ON on biomass conservation. Figure 5 A schematic diagram illustrating the cumulative local interaction effect of OP on biomass conservation effectiveness. Detailed Implementation

[0026] Example 1: To address the uncertainties in the ecological responses of important marine species or communities driven by the combined effects of climate variability, marine environmental changes, and human activities under the background of global change, this embodiment constructs a technical framework for predicting the evolution trend of species protection effectiveness under global change, namely SPET-GC (Species Protection Effect Trend prediction under Global Change).

[0027] This framework follows a complete technical process: "causal mechanism identification → spatiotemporal dynamic modeling → multi-scenario prediction → uncertainty assessment and risk combination analysis." It integrates ecological response indicators, conservation intervention intensity, and multi-source driving variables such as climate, ocean, and human activities to gradually achieve key factor identification, ecological state evolution simulation, and future trend prediction. This framework can not only quantitatively identify the core driving factors of conservation effectiveness but also predict their spatial and temporal changes under various future scenario paths, quantify the sources of uncertainty, and ultimately achieve accurate identification and zoning support for high-risk combinations and potential areas of conservation effectiveness transformation.

[0028] The SPET-GC framework aims to achieve systematic identification, modeling, and prediction of the entire chain of "global change - conservation measures - ecological response - evolution trends," and includes the following four core components, such as... Figure 1 As shown, it is applicable to important marine animal species, including mangroves, seagrass beds and other marine plants, as well as small yellow croakers, Chinese white dolphins and other marine animals.

[0029] The evaluation process for this framework is described in detail below.

[0030] First, a dual machine learning approach is used to input species and community conservation effectiveness indicators (such as biomass, distribution range, community structure, habitat density, etc.), conservation measures (such as nature reserves, etc.), and environmental covariates, including climate factors (such as temperature, precipitation, sea surface temperature, etc.), geographical factors (topography, slope, etc.), marine dynamic factors (such as tides, waves, etc.), and human disturbance pressures (such as shoreline development, pollution input, fishing, etc.). By causal inference, the independent effects of various factors are stripped away, and the key driving factors for the evolution of conservation effectiveness of important marine species and communities under the background of global change are identified.

[0031] Secondly, a spatial-temporal hierarchical Bayesian model is constructed, which integrates historical climate change, human disturbance and time series of effectiveness indicators to simulate the dynamic evolution of biomass, habitat quality and functional traits of important marine species and communities at the spatiotemporal scale, providing dynamic modeling support and uncertainty quantification basis for prediction.

[0032] Furthermore, based on causal identification and spatiotemporal simulation, an ensemble learning algorithm is used to construct a predictive model of the impact of species and community conservation effectiveness on evolutionary trends. Multiple global change scenarios are embedded to output spatial distribution maps of the evolutionary trends of species and community conservation effectiveness under different scenarios.

[0033] Finally, Bayesian posterior distribution analysis and sensitivity indicators are introduced to systematically quantify the sources of uncertainty and sensitive factors in the prediction results, identify key risk combinations and their thresholds affecting species and communities, and accurately locate potential areas for changes in the status of species conservation effectiveness under different global change scenarios, providing scientific basis and spatial support for formulating conservation priority areas and response strategies.

[0034] The system framework is described in detail below.

[0035] This system aims to build a platform integrating remote sensing observation, environmental variables, conservation strategies, and climate scenarios to analyze the effectiveness of mangrove conservation and predict future evolution trends. It is divided into eight modules, such as... Figure 2 As shown, it includes: The data preprocessing module is used for the unified aggregation, format standardization, and spatial alignment of multi-source heterogeneous data. Methods and tools include: automated batch preprocessing scripts on the GEE platform; spatial interpolation, resampling, mask extraction, and time series completion.

[0036] The key driver identification module is used to screen variables that significantly impact conservation effectiveness from candidate environmental and anthropogenic factors and define suitability threshold ranges. Methodological tools include: correlation analysis; principal component analysis; factor sensitivity analysis; optimal threshold identification based on ecological response curves; and allowing users to import prior ecological thresholds.

[0037] The quantitative assessment module for conservation effectiveness is used to quantitatively evaluate the degree to which the implementation of protected area strategies improves mangrove ecological indicators. Core functions include: comparing the trends of normalized difference vegetation index, biomass, and net primary productivity changes inside and outside the protected area; multi-level stratification of conservation effectiveness; and outputting treatment effects, protected area response levels, and geographical difference maps.

[0038] The causal effect analysis module is used to separate the mixed effects of natural and policy variables on ecological responses and identify the net policy effect. Methodological tools include: dual machine learning, propensity score matching, causal forest; optional conditional average treatment effect analysis; and outputs causal maps and causal path contribution values.

[0039] The spatiotemporal dynamic modeling module is used to capture the continuous evolution and regional heterogeneity of ecological indicators across spatiotemporal scales. Modeling content includes: constructing a Bayesian hierarchical model based on grids or raster points; introducing temporal variations, autocorrelation structures, and environmental covariates; supporting two inference methods: Markov chain Monte Carlo method or integrated nested Laplace approximation; and outputting predicted mean, confidence intervals, and uncertainty heatmaps.

[0040] The multi-scenario prediction module is used to predict changes in ecological indicators of protected areas driven by multiple factors using ensemble algorithms and to express them in probability distribution. Technical methods include: Random Forest, XGBoost, LightGBM, etc.; it supports multiple scenario inputs; and it integrates interpretation methods such as Shapley value based on game theory, outputting factor importance ranking and prediction confidence intervals.

[0041] The future scenario simulation module is used to construct ecological prediction frameworks under multiple future development paths and supports strategy sensitivity analysis and decision support. Application functions include: input scenarios such as the standard scenario framework from the IPCC climate change study, land use scenarios, and protection intensity settings; visualization of future mangrove spatial distribution and changes in ecological function indicators under different scenarios; and output of trend charts, scenario comparison charts, and risk maps.

[0042] The results output module is used to summarize the protection effectiveness at different future time points, generate graphical outputs, and produce reports. Output includes: time-series trend graphs and interval prediction maps under different emission scenarios; protection effect evolution curves for typical regions; automatically generated maps, report summaries, and exported tables.

[0043] Example 2: Mangroves, as typical blue carbon ecosystems of tropical and subtropical coasts, are mainly distributed between the high and mid-tidal zones, playing a vital ecological role in maintaining coastal stability, enhancing carbon sequestration, protecting against erosion, and preserving biodiversity. In recent years, under the dual pressures of global climate change and human activities, mangroves are facing risks such as spatial compression, habitat fragmentation, and functional degradation. Against the backdrop of rising sea levels, frequent extreme weather events, and intensified coastline development, scientifically assessing their conservation effectiveness is of practical significance for formulating restoration strategies.

[0044] This embodiment takes a typical tropical mangrove system, A Port Nature Reserve and B Bay Ecological Protection Red Line Area as cases, and constructs an integrated framework of "ecological response-factor-driven-effectiveness evaluation". It integrates data such as remote sensing vegetation index, tide level, salinity, topography and disturbance intensity, and systematically evaluates the restoration trend and functional evolution process in three stages: 2000s (S1), 2010s (S2) and 2020s (S3).

[0045] Based on MODIS remote sensing time series and field surveys, a normalized difference in vegetation index (NDI) and leaf area index (LAI) biomass conversion model was established. The results showed that the biomass per unit area in Port A significantly increased during the S1-S3 phases, with an average annual growth rate of 0.005 for the NDI and an increase in biomass from 76.5 Mg / ha to 89.2 Mg / ha, representing an average annual growth rate of 1.2%. Biomass accumulation accelerated and spatial coherence increased in the mid-tidal zone. Growth in Bay B was slower, with an average annual growth rate of only 0.003 for the NDI and an increase in biomass from 71.4 Mg / ha to 78.6 Mg / ha. Biomass showed a phased decline in extreme climate years, and recovery was lagging in the marginal areas.

[0046] In Port A, habitat quality improved significantly during Phase S3, with mangrove forests showing increased contiguousness and reduced fragmentation, and greatly enhanced connectivity between tidal flats, mangroves, and the water transition zone. However, in Bay B, habitat fragmentation remained high, and due to disturbances from aquaculture ponds and hard shorelines, connectivity in marginal areas was insufficient, resulting in limited restoration effects.

[0047] The A-port community has evolved from a single-dominant Kandelia candel structure to a mixed structure of Kandelia candel, Avicennia marina, and Rhizophora stylosa, with increased functional redundancy and improved resilience. Net primary productivity (NPP) has remained stable at 5.2-5.6 tC / ha. The B-bay community, on the other hand, has a simpler structure and lower redundancy, with NNP consistently maintained at 3.8-4.3 tC / ha, and even below 3.5 tC / ha in the marginal areas, reflecting limited system stability and carbon sequestration capacity.

[0048] Through causal modeling and Bayesian posterior analysis, the key driving factors and response thresholds for mangrove conservation effectiveness were identified, such as... Figure 3 As shown, the recovery trend of Port A is mainly controlled by the duration of tidal inundation, water temperature changes, and soil quality. When the tidal level fluctuation is controlled within ±0.15m and the water temperature rise does not exceed 0.2℃, the community biomass is in its optimal state, with a narrow posterior confidence interval, indicating high predictive stability and good buffering capacity of the system. Functional traits are characterized by increased diversity, enhanced functional redundancy, and improved resilience. Bay B, on the other hand, is more affected by human disturbance and eutrophication, resulting in a wider range of posterior biomass fluctuations. The recovery path is easily interrupted by external drivers, manifesting as degradation of community functional traits in the marginal areas, such as simpler structure, reduced functional differentiation, and decreased redundancy.

[0049] LE interaction effect analysis revealed the synergistic mechanism of key factors, such as Figure 4 and Figure 5 As shown, the interaction between pH and inorganic nitrogen revealed that when pH > 8.2 and inorganic nitrogen concentration > 0.3 mg / L, the ALE value was significantly negative, indicating biomass inhibition, possibly due to increased ammonia nitrogen toxicity; while the combination of pH 7.4-7.8 and inorganic nitrogen < 0.1 mg / L exhibited an ecological buffering effect. Similarly, the interaction between pH and reactive phosphate showed that at pH 7.2-7.6 and reactive phosphate < 0.02 mg / L, the cumulative local effect value was significantly positive, which was beneficial to community biomass improvement; when the reactive phosphate concentration was extremely low and pH > 8.0, the cumulative local effect value turned negative, possibly related to high pH-induced phosphorus inavailability, forming a high pH, ​​low reactive phosphate inhibition zone.

[0050] Example 3: Based on the SPET-GC framework described in the preceding embodiments, this embodiment takes a typical mangrove reserve as the research object. Using three climate-social scenario pathways (SSP126, SSP245, and SSP585), it simulates and predicts the temporal evolution trends of core ecological indicators up to 2100, and systematically assesses the differences in the impact of different emission scenarios on conservation effectiveness. The results show that the future evolution of the mangrove ecosystem, while highly dependent on global change pathways, exhibits significant nonlinear response characteristics. The overall evolution trend can be summarized as a gradient pattern of "continuous optimization of low emissions - slowing growth of medium emissions - intensified degradation of high emissions," as shown in Table 1.

[0051] Under the SSP126 (low emissions) scenario, key indicators such as mangrove biomass, leaf area index, and net primary productivity show a steady upward trend. Habitat connectivity is significantly enhanced, community functional structure becomes more diverse and stable, and the system exhibits high resilience and redundancy, with continuous optimization of ecological functions. Under the SSP245 (medium emissions) scenario, the growth rate of ecological indicators slows significantly, the system remains relatively stable but gradually approaches a plateau, exhibiting diminishing marginal returns in conservation efforts. Under the SSP585 (high emissions) scenario, the mangrove system shows obvious signs of degradation, particularly in marginal areas, where significant system instability, decreased carbon sequestration capacity, and increased spatial fragmentation pose serious challenges to the integrity of ecological functions.

[0052] Taking Port A of the nature reserve as an example, its ecosystem resilience is significantly better than other areas, and it is relatively less affected by climate change. Under low-emission scenarios, its biomass and net primary productivity steadily increase, and the habitat structure in the core area is significantly optimized; under moderate-emission scenarios, the system maintains positive functions, and although growth slows down, no significant degradation occurs; even under high-emission pathways, the core area still maintains a certain ecological restoration capacity, with only local functional decline occurring in the peripheral zones. Driving mechanism analysis shows that the protection effectiveness of Port A is mainly affected by tidal inundation duration and seawater temperature changes, exhibiting typical natural factor-dominated response characteristics.

[0053] In contrast, the Bay B area, located within the ecological protection red line, is significantly sensitive to future climate and human disturbance changes, exhibiting disturbance-driven ecological vulnerability. Under a low-emission scenario, its conservation benefits are limited, and habitat fragmentation improvement is slow; under a medium-emission pathway, ecological indicators tend to stagnate, with alternating periods of "growth-degradation" occurring in some areas; under a high-emission pathway, system degradation intensifies, with mangrove biomass projected to decline by 8-12% and net primary productivity by over 15% by 2100, leaving marginal patches facing the dual risks of habitat loss and ecological function deterioration. The main driving factors are increased human activity intensity and dramatic changes in land use structure, indicating that stronger intervention-based conservation strategies are needed in this area in the future.

[0054] Table 1. Changes in representative indicators of mangrove reserves under different future scenarios

[0055] Example 4: Building upon the aforementioned embodiments, this paper proposes an adaptive covariate balancing dual machine learning framework to address the estimation bias caused by high-dimensional nonlinear interference of covariates in traditional causal inference models. The core of this framework lies in introducing a dynamic weight adjustment mechanism. This mechanism calculates the distributional differences between the treatment group and the control group in the high-dimensional feature space and constructs an exponentially weighted balancing factor to finely adjust the sample weights. This effectively suppresses the influence of confounding variables when estimating the average treatment effect of protection strategies, thereby improving the robustness and accuracy of causal effect estimation. This method, by adaptively learning the balance of covariates, outperforms traditional fixed-parameter models and is particularly suitable for scenarios in the field of ecological protection characterized by strong spatial heterogeneity and complex anthropogenic interference.

[0056] The data preprocessing module receives a standardized high-dimensional dataset, which includes processing variables and a set of high-dimensional covariates.

[0057] For each sample, its Mahalanobis distance to the global samples in the covariate space is calculated to measure the degree of anomaly in its covariate distribution. Then, an adaptive balancing weight is constructed, calculated as follows:

[0058] In the formula, For the first Adaptive balancing weights for each sample; This refers to the decay hyperparameter; For the first The covariate vector of each sample; This is the mean vector of global covariates.

[0059] The calculated weights This is introduced into the estimation process of dual machine learning. The first stage uses a weighted machine learning model to fit the processing variables separately. right Prediction model and outcome variables right Prediction model Weight This is used as a weight for sample importance during model training. The second stage calculates and processes the residuals. and the residual of the ending In the formula, For the first One processing variable; For the first There are several outcome variables. Ultimately, the average treatment effect of the protection strategy is... Estimated using weighted linear regression: In the formula, For the residuals, the regression here also uses the aforementioned balanced weights. .

[0060] The final residual Spatial visualization can be performed, or the absolute value can be used as a new dependent variable to conduct correlation analysis with variables such as spatial location and environmental gradient, in order to identify areas with high uncertainty in the estimation of causal effects.

[0061] The estimated average treatment effect The confidence intervals of the data are passed to the results visualization module, along with spatial distribution information of the high uncertainty region, to generate a causal effect report.

[0062] This method, by adaptively balancing the distribution of covariates inside and outside the protected area, significantly reduces estimation bias. In practical applications, compared to the standard DML model, the confidence interval width of the estimated net biomass enhancement effect of the conservation strategy is narrowed by approximately 25%, from ±15 Mg / ha to ±11.25 Mg / ha, signifying a substantial improvement in estimation accuracy. Simultaneously, the method reduces the mean squared error in simulated data by more than 15%, significantly enhancing the reliability and persuasiveness of the quantitative assessment of the net effect of conservation strategies, providing a more accurate scientific basis for management decisions. The results show that the dual machine learning method for quantifying the net contribution of conservation effects based on adaptive covariate balancing effectively overcomes the estimation bias problem caused by the imbalanced distribution of covariates in high-dimensional nonlinear data in traditional causal inference models. By introducing a dynamic weight adjustment mechanism, this method significantly improves the robustness and accuracy of estimating the average treatment effect of conservation strategies, enhances the model's ability to suppress confounding factors in complex ecosystems, and makes the quantitative results of causal effects more explanatory and reliable, providing key technical support for accurately assessing the independent net contribution of conservation measures.

[0063] Example 5: Building upon the aforementioned embodiments, and addressing the challenge of insufficient characterization of the coupling between the instantaneous response and spatiotemporal autocorrelation effects of environmental covariates in traditional spatiotemporal prediction models, a hierarchical Bayesian model is constructed, explicitly introducing a "dynamic coupling coefficient" term into each layer. This coefficient combines spatial neighborhood structure and environmental gradient changes through a nonlinear link function, dynamically modulating the intensity of the influence of environmental covariates on core ecological indicators. This enables the model not only to capture the spatiotemporal dependence of ecological processes but also to quantify the differentiated impacts of environmental factors in different spatiotemporal contexts, thereby more realistically simulating the dynamic evolution of ecosystems.

[0064] Receive a structured spatiotemporal dataset from the spatiotemporal dynamic modeling module, including each grid point. At the point of time biomass observations and a set of environmental covariates .

[0065] Constructing a hierarchical Bayesian model framework: The first layer assumes biomass. It follows a normal distribution with a mean of Defined by a linear predictor. The second layer defines a linear predictor. It consists of three parts: the global intercept, the effects of environmental covariates, and spatiotemporal random effects. . The third layer We model it as a function of the spatiotemporal autocorrelation intensity and the difference in environmental gradient: In the formula, In spatial location Time point Observed biomass; The mean of the predicted biomass; It follows a normal distribution; is the variance parameter of the normal distribution; This is the global intercept; For dynamic coupling coefficients; A vector of environmental covariates; It is a spatiotemporal random effect; The intercept; It controls the scaling of the output of the tanh function; It is the hyperbolic tangent function; For spatial autoregressive hyperparameters; It is a spatial adjacency matrix; This is a vector of neighborhood biomass observations; For environmental gradient hyperparameters; Due to environmental gradient differences. Fourth layer. An autoregressive spatiotemporal process prior is used to capture unexplained spatiotemporal dependencies.

[0066] The posterior distributions of all parameters in the model are estimated using Bayesian inference methods. The prior distributions are set based on ecological knowledge; for example, no-information priors are set for variance parameters.

[0067] Using the well-fitted model, for new spatiotemporal points and By drawing samples from the posterior distribution of its parameters, the predicted value of future biomass and its complete posterior prediction distribution can be obtained, thus providing both point prediction and interval prediction.

[0068] The prediction results and the spatial distribution of the dynamic coupling coefficients are transmitted to the integrated prediction module and the visualization module.

[0069] Validation results show that, while achieving a similar average error to the aforementioned embodiments, this model, through a dynamic coupling mechanism, reveals the spatiotemporal heterogeneity of environmental factors in greater detail. In the case study of mangrove forests on Hainan Island, the model reduced the leave-one-out validation error for biomass by approximately 20%, significantly outperforming the baseline Bayesian model without this mechanism. Simultaneously, the model's prediction accuracy for extreme climate events improved by over 30% because it dynamically adjusts the weighting of environmental pressures through the coupling coefficient. Ultimately, the model's output prediction interval coverage is closer to the theoretical confidence level, providing a more reliable basis for risk management decisions. The results indicate that the Bayesian dynamic coupling prediction model integrating spatiotemporal autocorrelation and environmental gradient significantly improves the ability to characterize the spatiotemporal evolution of ecological indicators. This model effectively captures nonlinear interaction effects and spatial heterogeneity by flexibly modulating the local impact of environmental variables on ecological responses through dynamic coupling coefficients, thereby improving the accuracy and reliability of ecological trend predictions under multiple scenarios. Meanwhile, the posterior prediction distribution output by the model has a good ability to quantify uncertainty, providing a more comprehensive basis for risk management and priority area identification, and enhancing the system's simulation adaptability and strategy support value in complex environments.

Claims

1. A smart evaluation and scenario prediction system for mangrove conservation effectiveness, characterized in that, include: The data preprocessing module is configured to automatically clean, standardize, and spatially register multi-source heterogeneous ecological data. The key driver identification module is used to determine the core variables affecting the effectiveness of protection and their dynamic response ranges based on statistical analysis and nonlinear fitting methods. The conservation effectiveness quantitative assessment module is used to calculate the net improvement effect of conservation interventions by comparing the differences in ecological indicators inside and outside the protected area; The causal effect analysis module is used to separate the independent contributions of natural factors and human conservation using a dual machine learning framework; The spatiotemporal dynamic modeling module is used to model the uncertainty of the spatiotemporal evolution of ecological indicators using a Bayesian hierarchical structure, and to integrate spatiotemporal autocorrelation features with environmental gradient differences to construct nonlinear response relationships. The multi-scenario prediction module integrates various machine learning algorithms to achieve ecological response prediction driven by multiple factors, and supports user-defined scenario input and simulation. The results output module provides functions for generating and exporting various visualization charts and recommended reports.

2. The system according to claim 1, characterized in that: The data preprocessing module has the ability to automatically process multi-source ecological monitoring data, including missing value imputation, time series alignment, spatial interpolation, and resampling.

3. The system according to claim 1, characterized in that: The key driving factor identification module identifies key influencing factors of conservation effectiveness and their suitable ecological thresholds through variable correlation analysis and ecological response curve fitting.

4. The system according to claim 1, characterized in that: The quantitative assessment module for conservation effectiveness is configured to calculate the differences in biomass, vegetation index, and productivity indicators inside and outside the protected area, and output the average treatment effect and geographical distribution heterogeneity of conservation measures.

5. The system according to claim 1, characterized in that: The causal effect analysis module uses a machine learning-based causal inference method to quantify the net causal effect between protection strategies and ecological responses.

6. The system according to claim 5, characterized in that: The causal effect analysis module further includes an adaptive covariate balancing unit, which is used to suppress the interference of high-dimensional covariate distribution imbalance on causal effect estimation through a dynamic weight adjustment mechanism.

7. The system according to claim 1, characterized in that: The spatiotemporal dynamic modeling module establishes a Bayesian hierarchical model by introducing spatiotemporal autocorrelation structures and environmental covariates, which is used to estimate the posterior distribution and uncertainty of ecological state variables.

8. The system according to claim 1, characterized in that: The multi-scenario prediction module integrates various tree-based and ensemble learning algorithms to achieve probabilistic prediction of ecological indicators and identification of sensitive factors under different development scenarios.

9. The system according to claim 8, characterized in that: The multi-scenario prediction module allows the embedding of multiple climate evolution paths and socio-economic scenarios, and supports dynamic simulation and effect comparison of protection strategy adjustment scenarios.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, it controls the operation of the system according to any one of claims 1-9.

Images