A watershed vegetation spatial distribution optimization method and system based on water-carbon coupling

Abstract

A watershed vegetation spatial distribution optimization method and system based on water-carbon coupling, in view of the problems of existing vegetation configuration water-carbon function fragmentation, low spatial optimization calculation efficiency and lack of climate adaptability, a distributed hydrological model and a carbon cycle model are coupled to realize dynamic interaction of water-carbon process through soil moisture content as a bridge; an artificial neural network proxy model is used to replace the original model for fitness evaluation in optimization iteration, and the original model is called every preset generation to verify, and when the error is out of limit, the proxy model is retrained; the optimization objectives include carbon sink (GPP-Rh) and water consumption intensity (ET / precipitation ratio), and CMIP6 future climate scenario data is introduced to calculate climate robustness index CRI, and the configuration scheme stable under different emission scenarios is screened. The method can complete the spatial optimization of watershed vegetation in a short time, and output grid-level vegetation distribution map and water-carbon benefit evaluation results.

Description

Technical Field

[0001] This application relates to the field of watershed ecological management technology, specifically to a method and system for optimizing the spatial distribution of watershed vegetation based on water-carbon coupling. Background Technology

[0002] Watershed vegetation configuration is a crucial aspect of ecological restoration and soil and water conservation projects. Currently, watershed vegetation configuration relies primarily on human experience or a single ecological model for assessment.

[0003] Commonly used hydrological models (such as SWAT) can simulate watershed water cycle processes well, but lack a detailed characterization of carbon cycle processes. Carbon sink assessment models (such as the InVEST carbon module) are usually based on simplified parameters, making it difficult to reflect the dynamic coupling relationship between vegetation transpiration and photosynthesis. In practical applications, optimizing carbon sinks in isolation often leads to an excessively high proportion of water-consuming vegetation, causing regional water deficits; conversely, considering water conservation in isolation may sacrifice the ecosystem's carbon sequestration potential. This fragmented assessment approach to water and carbon functions makes it difficult to support coordinated vegetation configuration decisions.

[0004] Furthermore, existing vegetation spatial optimization methods mostly adjust vegetation proportions based on sub-basins or statistical units, neglecting the explicit spatial distribution characteristics at the grid scale. Spatial pattern indicators such as the shape and aggregation degree of vegetation patches have a significant impact on surface runoff confluence and microclimate, but existing optimization algorithms struggle to incorporate these spatial pattern indicators as variables into the calculation.

[0005] More importantly, vegetation configuration optimization is usually based on static historical climate data. Current methods lack mechanisms to assess the climate robustness of configuration schemes in the face of future climate change scenarios. This means that some schemes optimal under current climate conditions may fail under future drought or extreme precipitation scenarios, or even lead to the failure of ecological engineering projects.

[0006] Finally, high-precision water-carbon coupling process models require a long simulation time per run, while spatial optimization requires tens of thousands of iterative calculations. Existing technologies lack effective acceleration strategies to address the contradiction between high-precision process models and large-scale optimization calculations, making it difficult to implement optimization solutions in practical engineering.

[0007] Therefore, there is an urgent need for a watershed vegetation spatial distribution optimization method that can take into account the water-carbon coupling mechanism, achieve explicit spatial pattern optimization, and has climate adaptability. Summary of the Invention

[0008] This application provides a watershed vegetation spatial distribution optimization method and system based on water-carbon coupling, which can solve the problems in the prior art that vegetation configuration schemes are difficult to achieve water-carbon synergistic optimization, spatial optimization methods fail to fully consider the spatial combination effect of vegetation, and configuration results lack the technical adaptation to climate change.

[0009] In a first aspect, embodiments of this application provide a method for optimizing the spatial distribution of vegetation in a watershed based on water-carbon coupling, comprising the following steps: Acquire basic geographic data, historical meteorological data, and future climate scenario data for the watershed; An ecosystem water-carbon coupling model is constructed to simulate the hydrological and carbon cycle processes under different vegetation spatial distributions, and outputs the simulation results of water and carbon fluxes. An artificial neural network proxy model is constructed and trained using sample data generated by the water-carbon coupling model to predict the water and carbon flux of vegetation spatial distribution schemes. Establish a multi-objective optimization model, and set optimization objectives that include at least maximizing carbon sink and minimizing water consumption intensity; The multi-objective optimization algorithm combined with the artificial neural network surrogate model is used to iteratively optimize the vegetation spatial distribution scheme based on grid coding. During the iteration process, a dynamic calibration step is performed: every preset iteration number, a scheme with a preset proportion or a preset number in the current population is selected and the water-carbon coupling model is called to simulate it. The simulation results of the water-carbon coupling model are compared with the prediction results of the surrogate model. If the error exceeds the preset error threshold, the artificial neural network surrogate model is retrained. The optimized vegetation spatial distribution scheme is output, which is used to guide the configuration of vegetation planting and ecological restoration projects within the watershed.

[0010] In conjunction with the first aspect, in one implementation, the establishment of the multi-objective optimization model further includes a climate adaptation objective: By introducing the future climate scenario data, the climate robustness index of each optimization scheme under different climate scenarios is calculated. The climate robustness index is used as a constraint or weighting factor for the optimization objective to screen for climate-adaptive vegetation spatial distribution schemes.

[0011] In conjunction with the first aspect, in one implementation method, the formula for calculating the climate robustness index is: ; CRI stands for Climate Robustness Index. and These are the overall fitness values ​​of the scheme under medium and high emission scenarios, respectively. The closer the CRI is to 1, the higher the climate stability of the scheme.

[0012] In conjunction with the first aspect, in one implementation method, the construction of the water-carbon coupling model specifically includes: By integrating a distributed hydrological model and a carbon cycle model, the soil moisture content variable simulated by the hydrological model is used as the input of the water stress factor in the carbon cycle model, thereby achieving dynamic coupling between the water cycle process and the carbon cycle process. The model parameters are calibrated and validated using multi-source data, which includes ground observation data and remote sensing product data.

[0013] In conjunction with the first aspect, in one implementation method, the process of generating the vegetation spatial distribution scheme specifically includes: The watershed is divided into regular grid units, and each grid unit is independently set with vegetation type codes and corresponding vegetation physiological parameters; The vegetation physiological parameters include leaf area index, root depth, and water use efficiency; different vegetation spatial distribution schemes are generated by changing the vegetation type coding of the grid cells.

[0014] In conjunction with the first aspect, in one embodiment, the method further includes a landscape ecological assessment step: Landscape ecological indicators are calculated based on the aforementioned vegetation spatial distribution scheme, and the indicators include at least patch shape index, aggregation index, and edge density; The landscape ecology indicators are used as characteristic variables representing the spatial pattern of vegetation and incorporated into the evaluation function or constraints of the multi-objective optimization model to analyze their comprehensive impact on the watershed's water and carbon balance.

[0015] In conjunction with the first aspect, in one implementation, the training process of the artificial neural network proxy model includes: Multiple sets of vegetation distribution schemes and their corresponding water and carbon flux simulation results were randomly generated using a water-carbon coupling model to form a training dataset. The training dataset is divided into a training set and a validation set. The artificial neural network is trained until the prediction accuracy of the validation set meets the pre-set accuracy threshold.

[0016] In conjunction with the first aspect, in one implementation, the iterative update of the vegetation spatial distribution scheme using a multi-objective optimization algorithm specifically includes: A genetic algorithm is used to perform crossover and mutation operations on the vegetation type encoding matrix; The mutation operation includes: randomly selecting the boundary grid cells of vegetation patches and changing their vegetation type codes to adjust the patch shape index and aggregation index.

[0017] Secondly, embodiments of this application provide a watershed vegetation spatial distribution optimization system based on water-carbon coupling, comprising: The data acquisition module is used to acquire basic geographic data, historical meteorological data, and future climate scenario data for the watershed. The model building module is used to build ecosystem water-carbon coupling models and artificial neural network proxy models. The optimization calculation module is used to establish a multi-objective optimization model and use the multi-objective optimization algorithm combined with the artificial neural network proxy model to iteratively optimize the vegetation spatial distribution scheme and execute the dynamic calibration steps as described in claim 1. The output module is used to output the optimized vegetation spatial distribution scheme and the corresponding water and carbon benefit assessment results. The system is used to execute the method described.

[0018] Thirdly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.

[0019] The beneficial effects of the technical solutions provided in this application include: By constructing an ecosystem water-carbon coupling model and an artificial neural network proxy model, and using a multi-objective optimization algorithm combined with the artificial neural network proxy model to iteratively optimize a grid-coded vegetation spatial distribution scheme, and performing a dynamic calibration step during the iteration process, this technical solution solves the technical problems of fragmented water and carbon functions in vegetation configuration, low computational efficiency of spatial optimization, and contradictions between high-precision process models and large-scale optimization calculations in related technologies. It achieves efficient optimization of water and carbon synergy in the spatial distribution of vegetation in the watershed, significantly reduces computational costs while ensuring simulation accuracy, and the optimized scheme is climate-adaptable and can be used to guide the configuration of vegetation planting and ecological restoration projects in the watershed. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating an embodiment of the watershed vegetation spatial distribution optimization method based on water-carbon coupling in this application. Figure 2 This is a schematic diagram of the functional modules of an embodiment of the watershed vegetation spatial distribution optimization system based on water-carbon coupling of this application. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0023] This application provides a method for optimizing the spatial distribution of vegetation in a watershed based on water-carbon coupling. This method can run on a computer terminal equipped with a processor and is applicable to scenarios such as watershed ecological restoration planning, soil and water conservation scheme preparation, and carbon sink management. The core implementation process of this method is explained below with reference to a simulation application process of a typical watershed.

[0024] In this embodiment, refer to Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the watershed vegetation spatial distribution optimization method based on water-carbon coupling according to this application. Figure 1 As shown, the watershed vegetation spatial distribution optimization method based on water-carbon coupling includes: Step S1: Obtain basic geographic data, historical meteorological data, and future climate scenario data for the watershed.

[0025] The basic geographic data includes digital elevation models, soil type maps, and land use status maps; historical meteorological data uses daily precipitation, temperature, and radiation data from meteorological stations in and around the basin; and future climate scenario data is selected from climate model output data released by the Coupled Model Intercomparison Project (CPIP) and interpolated to the basin grid using downscaling methods.

[0026] Step S2: Construct an ecosystem water-carbon coupling model, which is used to simulate the hydrological cycle and carbon cycle processes under different vegetation spatial distributions, and output the simulation results of water and carbon fluxes.

[0027] In this embodiment, the vegetation scenario setting includes not only the current vegetation scenario based on land use / cover data, but also various potential vegetation configuration schemes. The current scenario distinguishes major vegetation functional types such as coniferous forest, broad-leaved forest, shrubs, and grassland; the potential scenario combines climate suitability analysis to design configuration schemes such as carbon sink priority, water-saving, and integrated schemes. In specific implementation, an explicit parameter assignment method based on grid cells is adopted, dividing the watershed into 1km×1km regular grids. Each grid is independently assigned a vegetation type code and corresponding vegetation physiological parameters, including leaf area index (LAI), root depth, and water use efficiency (WUE), constructing a vegetation parameter matrix. All schemes are input into the water-carbon coupling model in the form of an ASCIIGrid matrix file, and the model directly reads the parameter values ​​of each grid for simulation. This method differs from cellular automata (CA) iterative evolution; instead, it generates continuous patches through continuous assignment, or fragmented patterns through random and alternating assignment, or distribution schemes with different shape indices can be generated by defining the boundary orientation of patches. It should be noted that this embodiment uses a regular grid unit as an example. In other embodiments, the spatial unit can also be an irregular polygon, a sub-basin boundary, or an administrative unit, as long as the vegetation parameters can be set independently.

[0028] Step S3: Construct an artificial neural network surrogate model, train it using sample data generated by the water-carbon coupling model, and use it to predict the water and carbon flux of the vegetation spatial distribution scheme.

[0029] Since a single simulation of the water-carbon coupling model takes approximately tens of minutes, and multi-objective optimization requires evaluating tens of thousands to hundreds of thousands of vegetation distribution schemes, directly calling the original model would take months to years to converge within a reasonable timeframe. Therefore, this embodiment introduces an artificial neural network as a surrogate model.

[0030] The input variable X of the surrogate model includes vegetation spatial distribution characteristics and climate-driven data, specifically represented as follows: ,in The vegetation type of the i-th grid cell is encoded (an integer, such as 1 for coniferous forest, 2 for broadleaf forest, 3 for shrubs, and 4 for grassland), n is the total number of grid cells in the watershed, T is the average annual temperature (°C), P is the annual precipitation (mm), LSI is the patch shape index, and AI is the aggregation index. The input layer dimension is n+4.

[0031] The output variable Y includes total primary productivity (GPP) (unit: gC / m² / yr) and evapotranspiration (ET) (unit: mm / yr), which are two key fluxes characterizing the water-carbon coupling relationship. The output layer has two neurons, corresponding to GPP and ET respectively, and uses a linear activation function.

[0032] The network structure employs a three-layer fully connected feedforward neural network with 128 neurons in the hidden layer, using the ReLU activation function. During the training data generation phase, 2000 sets of simulation results for different vegetation distribution schemes are randomly generated using the water-carbon coupling model. Each set includes an input X and a corresponding output Y. 80% of these are used for training, and 20% for validation. Validation requires R² > 0.95 and relative RMSE < 5%. If the accuracy requirement is not met, the number of training samples is increased (e.g., to 3000 sets) until the accuracy requirement is met.

[0033] Step S4: Establish a multi-objective optimization model, and set optimization objectives that include at least maximizing carbon sink and minimizing water consumption intensity.

[0034] The optimization objectives are to maximize carbon sink and minimize water consumption intensity. Carbon sink is represented by the difference between gross primary productivity (GPP) and soil heterotrophic respiration (Rh), while water consumption intensity is represented by the ratio of evapotranspiration (ET) to concurrent precipitation. This setting is based on the water-carbon trade-off between GPP and ET, namely, high carbon sink vegetation is usually accompanied by high transpiration water consumption. The bi-objective optimization aims to generate a Pareto front, showing the water cost required to increase carbon sink by a unit, so as to automatically reduce the proportion of water-consuming vegetation in dry years.

[0035] Step S5: Use a multi-objective optimization algorithm combined with the artificial neural network surrogate model to iteratively optimize the vegetation spatial distribution scheme, and perform a dynamic calibration step during the iteration process.

[0036] During the iterative process, landscape ecology indicators are used to quantitatively characterize the morphological characteristics of vegetation patches, serving as input variables or optimization constraints for the surrogate model. Specifically, the patch shape index (LSI) is calculated using the formula... The calculations are performed, where E is the patch perimeter and A is the patch area; the aggregation index (AI) is calculated based on the adjacency relationship between pixels of the same type of patch; and the edge density (ED) is calculated based on the total length of the patch boundary per unit area. The influence of vegetation patch size and shape on water and carbon elements is analyzed through simulation to establish the vegetation characteristic-ecological function response relationship.

[0037] Specifically, during the initialization phase, 200 initial vegetation distribution schemes are randomly generated as the population. During the iteration process, each scheme preferentially calls the ANN surrogate model to predict its GPP and ET values ​​(single cycle time <0.1 seconds) for fitness calculation and non-dominated sorting.

[0038] Dynamic calibration employs a hierarchical mechanism: The first layer is accuracy calibration. Every 20 generations of optimization, the top 10% of fitness schemes (approximately 20) are selected from the current population, and high-precision simulations are performed using the original water-carbon coupling model (taking approximately tens of minutes per scheme). The results are then compared with the ANN predictions. If the root mean square error (RMSE) between the original model and the ANN predictions exceeds a threshold (e.g., RMSE > 8%), retraining is triggered. The latest simulation data is added to the training set, and the ANN is retrained until the required accuracy is achieved. The second layer is a fine-tuning mechanism. Every 100 generations of optimization, 50 representative schemes from the current Pareto front are selected, and the original water-carbon coupling model is used to resimulate them. Their input and output data are added to the training dataset, and the ANN is fine-tuned. Training continues for another 10-20 epochs. This mechanism ensures that the surrogate model does not introduce systematic biases when exploring new regions.

[0039] The iteration convergence condition is reaching a preset number of generations (usually 500-1000 generations) or the population diversity is below a threshold. The entire optimization process takes approximately 6-12 hours, which is about two orders of magnitude faster than directly calling the original model.

[0040] Step S6: Output the optimized vegetation spatial distribution scheme, which is used to guide the configuration of vegetation planting and ecological restoration projects within the watershed.

[0041] The output includes an optimized vegetation type distribution map and the corresponding water and carbon benefit assessment results.

[0042] Furthermore, in a preferred embodiment of this example, the specific parameter settings for the above steps are as follows: In step S1, the future climate scenario data specifically selects two sets of data: a medium emission scenario (such as SSP2-4.5) and a high emission scenario (such as SSP5-8.5).

[0043] In step S3, the artificial neural network proxy model employs a three-layer fully connected feedforward neural network. The number of hidden layer neurons is set to 128, and the ReLU activation function is used. During training, the training continues until the prediction accuracy of the validation set meets a preset accuracy threshold (e.g., R² > 0.95 and relative RMSE < 5%).

[0044] In step S5, the multi-objective optimization algorithm employs a genetic algorithm (such as NSGA-III). Crossover and mutation operations are performed on the vegetation type encoding matrix. The mutation operation includes randomly selecting boundary grid cells of vegetation patches and changing their vegetation type encoding to adjust the patch shape index and aggregation index. In the dynamic calibration step, the calibration cycle is set to every 20 generations, and the error threshold is set to 8%. Those skilled in the art will understand that the calibration cycle and error threshold can be adjusted according to the convergence speed of the surrogate model; for example, the calibration cycle can be 10-50 generations, and the error threshold can be 5%-10%.

[0045] In step S6, the vegetation spatial distribution scheme is generated based on grid coding. The watershed is divided into regular grid cells, and each grid cell is independently assigned a vegetation type code and corresponding vegetation physiological parameters, including leaf area index, root depth, and water use efficiency.

[0046] This embodiment achieves efficient optimization of watershed vegetation spatial distribution by constructing a collaborative mechanism between a water-carbon coupling model and a surrogate model, and by introducing dynamic calibration and climate adaptability assessment. Compared with traditional methods that directly call the coupling model for optimization, this method significantly reduces computational costs while ensuring optimization accuracy, and the optimized scheme exhibits good stability under future climate scenarios.

[0047] Secondly, embodiments of this application provide a watershed vegetation spatial distribution optimization system based on water-carbon coupling, which is used to execute the optimization method described in Embodiment 1. Figure 2 As shown, this system is mainly deployed on a server or high-performance computer terminal, and includes a data acquisition module 1, a model building module 2, an optimization calculation module 3, and an output module 4.

[0048] The data acquisition module 1 is configured to acquire basic geographic data, historical meteorological data, and future climate scenario data for the watershed. Specifically, this module can connect to a meteorological database or a geographic information system database via a network interface to automatically download and preprocess digital elevation models, soil type maps, and CMIP6 climate model data, and convert the data into a format required by the model, such as ASCIIGrid or NetCDF format.

[0049] The model building module 2 is configured to construct an ecosystem water-carbon coupling model and an artificial neural network proxy model. This module integrates a distributed hydrological model engine and a carbon cycle model engine, and transmits soil moisture content variables output by the hydrological model to the carbon cycle model via a data interface. Simultaneously, this module also includes a neural network construction unit, used to build an artificial neural network based on a preset network structure (such as input layer, hidden layer, and output layer), and train it using sample data generated by the water-carbon coupling model.

[0050] The optimization calculation module 3 is configured to establish a multi-objective optimization model and iteratively optimize the vegetation spatial distribution scheme using a multi-objective optimization algorithm combined with the artificial neural network surrogate model. This module is the core computing unit of the system and is specifically used to execute step S5 in embodiment 1. It has an internal dynamic calibration subunit, which periodically calls the water-carbon coupling model in model building module 2 for high-precision simulation during the iteration process and triggers retraining of the artificial neural network based on the error results. This module also supports parallel computing and can utilize multi-core CPU or GPU resources to accelerate population fitness assessment.

[0051] Output module 4 is configured to output the optimized vegetation spatial distribution scheme and the corresponding water and carbon benefit assessment results. This module can convert the calculated vegetation type coding matrix into a visualized spatial distribution map and generate an assessment report containing carbon sink, water consumption intensity, and climate robustness index. The output results can be displayed through the system interface or exported as a general geographic data format for subsequent engineering configuration.

[0052] The system provided in this embodiment adopts a modular design, embedding the method steps in Embodiment 1 into software functional modules. In practical applications, the system can be deployed on the server cluster of a watershed management agency. Administrators only need to input the watershed boundary and optimization target parameters through the client, and the system can automatically complete the entire process of data acquisition, model training, optimization calculation, and result output.

[0053] Furthermore, this embodiment also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the watershed vegetation spatial distribution optimization method based on water-carbon coupling as described in Embodiment 1. The storage medium includes, but is not limited to, various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0054] In one specific embodiment, a tributary basin in the Loess Plateau was used as the research object to verify the method described in Example 1. The study area covers approximately 500 km², belongs to a semi-arid climate zone, with an average annual precipitation of about 450 mm, and vegetation cover mainly consisting of shrubs and grasslands, resulting in certain soil erosion problems.

[0055] For the basic data, the DEM used 30m resolution SRTM data, the soil type map was obtained from the 1:1,000,000 China Soil Database, and the land use data used the 2020 GlobeLand30 product. Meteorological data were collected from daily observation records of five meteorological stations within and around the basin from 1980 to 2020. Future climate scenario data were downloaded from the CMIP6 website, and the SSP2-4.5 and SSP5-8.5 scenarios of three climate models (EC-Earth3, MPI-ESM1-2-HR, and CESM2) were selected and downscaled to a 1km grid using the Delta method.

[0056] The water-carbon coupling model was improved based on SWAT2012, and the carbon cycle module was coupled with the CASA model framework. Model calibration used measured runoff data from 2010 to 2015, and the validation period was from 2016 to 2020. After calibration, the Nash efficiency coefficient was 0.52, and the coefficient of determination (R²) was 0.61, meeting the simulation requirements. Vegetation parameters were set according to relevant literature: LAI values ​​were 3.5-4.5 for coniferous forests, 4.0-5.5 for broadleaf forests, 1.5-2.5 for shrubs, and 1.0-2.0 for grasslands.

[0057] The surrogate model was built using Python's TensorFlow library. The network structure is a three-layer fully connected layer. The input layer dimension is determined based on the number of grid cells in the study area (in this example, 500 grid cells plus landscape indicators and climate variables, totaling 504 dimensions). The hidden layer has 128 neurons, and the output layer has two nodes corresponding to GPP and ET, respectively. Training samples were generated using Latin hypercube sampling to create 2000 vegetation distribution schemes, which were then divided into training and validation sets at an 8:2 ratio. During training, it was found that the initial validation set RMSE was approximately 12%, which decreased to around 6% after increasing the sample size to 2500 sets. The number of neurons in the hidden layer can be adjusted according to the number of grid cells in the watershed, typically ranging from 64 to 256. When the number of grid cells n in the watershed is large, to reduce the input dimensionality, block statistical features or convolutional neural networks (CNNs) can be used to extract spatial features before using them as input. This embodiment uses a small to medium-sized watershed as an example, directly using grid coding as the input.

[0058] The optimization algorithm uses NSGA-III, with a population size of 200 and a maximum number of iterations of 500. The dynamic calibration parameters were determined after multiple rounds of debugging: the accuracy calibration period was set to 20 generations, the fine-tuning period to 100 generations, and the error threshold to 8%. During debugging, it was found that if the calibration period was too short (e.g., every 5 generations), the computational cost was too high; if the period was too long (e.g., every 50 generations), the surrogate model was prone to drift. A 5% error threshold resulted in excessive retraining, a 10% threshold caused a slight decrease in accuracy, and 8% was the optimal balance. The calibration period can be between 10 and 50 generations.

[0059] In this embodiment, the multi-objective optimization model aims to find the optimal solution set for the spatial distribution of vegetation. Its construction process includes three parts: defining decision variables, setting optimization objectives, and restricting constraints.

[0060] First, regarding the definition of decision variables: the watershed area to be optimized is divided into several spatial computational units (e.g., regular grids), and the vegetation type configuration within each spatial computational unit is used as an independent decision variable. The optimization algorithm generates different vegetation spatial distribution schemes by changing the vegetation type encoding within each spatial computational unit.

[0061] Secondly, regarding the setting of optimization objectives: the multi-objective optimization model simultaneously considers ecological function benefits and spatial pattern stability, specifically including the following three dimensions of optimization objectives: The goal of maximizing carbon sequestration benefits is to calculate the total primary productivity of the watershed under the current vegetation spatial distribution scheme based on the water-carbon coupling process model or proxy model. The optimization direction is to maximize this total value to reflect the carbon sequestration capacity of vegetation.

[0062] Minimize water consumption objective: Based on the water-carbon coupling process model or surrogate model, calculate the total evapotranspiration of the watershed under the current vegetation spatial distribution scheme, and optimize to minimize this total value to reflect the water resource sustainability of the watershed.

[0063] Spatial pattern optimization objective: Based on the spatial arrangement of the current vegetation spatial distribution scheme, calculate landscape ecology indicators (such as landscape shape index or landscape aggregation index). This objective aims to optimize the spatial morphology of vegetation patches, for example, by minimizing the landscape shape index to reduce patch fragmentation, or by maximizing the landscape aggregation index to promote contiguous vegetation distribution, thereby enhancing the structural stability of the ecosystem.

[0064] Finally, regarding the optimization search logic: Multi-objective optimization algorithms (such as the NSGA-III algorithm) simultaneously evaluate the performance of each vegetation spatial distribution scheme across the three dimensions mentioned above during the iteration process. The algorithm does not seek a single optimal solution, but rather finds a set of Pareto optimal solutions through a non-dominated ranking mechanism. In this set, no single scheme can improve a particular objective (such as increasing carbon sequestration) without compromising other objectives (such as not increasing water consumption). The final optimized scheme output is the recommended scheme after climate robustness screening within this Pareto optimal solution set. 。

[0065] To verify the effectiveness of the method, this embodiment sets up two control schemes for comparison. Control scheme one uses the traditional method, that is, directly calling the SWAT model for genetic algorithm optimization without using a surrogate model; control scheme two uses a static surrogate model, that is, no dynamic calibration is performed after training. The three sets of experiments are run in the same server environment.

[0066] Regarding experimental results, the total optimization time for this embodiment was approximately 8 hours, including approximately 1 hour for surrogate model training, approximately 6 hours for optimization iterations, and approximately 1 hour for dynamic calibration. Control scheme one took approximately 1400-1600 hours (fluctuating due to population convergence), and control scheme two took approximately 7 hours, but the simulation verification error of the final scheme was larger. In terms of ecological benefits, compared to the current vegetation distribution, the optimized scheme in this embodiment improved the simulated GPP value by approximately 12%-18% and reduced the ET / P ratio by approximately 5%-10%. Regarding climate robustness, the fitness ratio (CRI) of this embodiment was approximately 0.82-0.88 under both SSP2-4.5 and SSP5-8.5 scenarios, while control scheme two, which did not consider climate adaptability, had a CRI of only approximately 0.65-0.72.

[0067] The output includes a 1km resolution raster file of vegetation type distribution (GeoTIFF format) and a corresponding benefit assessment report. Spatially, the optimized distribution shows an increase in the proportion of shrubs on steep slopes in the upper reaches of the watershed, a moderate increase in the proportion of trees in gentler areas in the middle reaches, and the retention of grasslands as the main vegetation in the lower reaches of the river valley. The overall pattern is more in line with the principle of water and carbon synergy.

[0068] It should be noted that the above data are model simulation results, and the actual ecological benefits still need to be verified by combining long-term field monitoring data. However, this embodiment at least proves that the method is technically feasible and has certain advantages in terms of computational efficiency and scheme stability compared with traditional methods.

[0069] In another specific embodiment, taking a certain Loess Plateau watershed as the object, the implementation process of the "water-carbon-space" three-dimensional optimization framework is specifically described.

[0070] In terms of acquiring and processing future climate scenario data, this embodiment uses future climate projection data released by CMIP6, specifically selecting the SSP2-4.5 moderate emission scenario (representing 2030-2050) and the SSP5-8.5 high emission scenario (representing the extreme scenario 2070-2090). Since the Global Climate Model (GCM) output resolution is relatively coarse (approximately 100 km), the Delta downscaling method is used to interpolate it to the watershed grid (1 km resolution), and diurnal bias correction is performed. The correction process uses historical measured data as a benchmark, calculating the difference between historical observations and model output values ​​to correct the future model output, obtaining monthly temperature, precipitation, and radiation data. The corrected future climate data is directly input into the water-carbon coupled model, replacing the historical climate driving field, to simulate water and carbon processes under future climate conditions.

[0071] To achieve a balance between current benefits and future stability, the optimization algorithm uses both historical and future climate driving data during the fitness assessment phase. Each vegetation distribution scheme is simulated under historical climate, SSP2-4.5 scenario, and SSP5-8.5 scenario, and the weighted comprehensive fitness is used as the final evaluation index. (Comprehensive Fitness) The calculation formula is: Here, w1 and w2 are weighting coefficients. This is a preferred implementation of climate adaptability assessment. The Climate Robustness Index (CRI) can also be used as a constraint, and both can be used individually or in combination. In this specific configuration, w1 is 0.6 and w2 is 0.4, meaning current climate benefits account for 60% and future climate adaptability accounts for 40%. Fitness_historical is the fitness value based on historical climate data, and Fitness_future is the fitness value based on future climate scenario data, typically the average of two future scenario simulations. This weighting mechanism ensures that the optimization results do not excessively sacrifice current ecological benefits while adapting to future climate change trends. It should be noted that the above-mentioned weighted comprehensive fitness is one way to achieve climate adaptability. In another implementation, the Climate Robustness Index (CRI) can also be directly added as a penalty factor to the fitness function.

[0072] When outputting the Pareto optimal solution set, this embodiment additionally calculates the Climate Robustness Index (CRI) for each scheme for decision-making reference. The CRI calculation formula is as follows: ,in and These represent the overall fitness values ​​of the schemes under medium and high emission scenarios, respectively. The closer the CRI value is to 1, the more stable the scheme's performance is under different climate scenarios. In practical applications, schemes with a CRI greater than 0.8 can be prioritized, meaning that a small amount of current benefits are traded for higher future stability.

[0073] To address the unique challenges of optimizing vegetation spatial distribution, this embodiment incorporates a tailored design for the genetic operators in the NSGA-III algorithm. The crossover operation employs a spatial region exchange method. Specifically, a rectangular window is randomly selected within the grid matrix of the two parent schemes, or a patch region is identified based on connected components. The vegetation codes of the two schemes within this region are then exchanged. This approach helps preserve the spatial aggregation characteristics of each scheme and avoids excessively fragmented patterns. The mutation operation includes two forms: single-point mutation and morphological mutation. Single-point mutation randomly changes the vegetation type code of a single grid cell, introducing a new vegetation type. Morphological mutation randomly selects the boundary grid cells of vegetation patches and changes their codes, thereby adjusting the LSI (Latent Size Index) and AI (Alternating Aggregation Index). Through this operator design, the optimization process not only adjusts the proportion of vegetation types but also simultaneously optimizes the spatial pattern, achieving three-dimensional coordinated optimization of water, carbon, and space.

[0074] After optimization and convergence, the system outputs a Pareto front solution set. Each solution includes a 1km resolution vegetation type raster file (GeoTIFF format), an expected benefit table (including indicators such as GPP, ET, NEP, and ET / P), and climate adaptability indicators (CRI value and weighted fitness value). Managers can select specific schemes based on watershed management objectives. For example, in water-scarce areas, a scheme with a lower ET / P and a CRI greater than 0.8 can be selected, or in carbon sink priority areas, a scheme with the highest NEP and a CRI greater than 0.75 can be selected. Testing shows that, compared to traditional single-climate-scenario optimization methods, the refined technical solution in this embodiment can reduce the fluctuation of benefits under future climate scenarios by approximately 30%.

[0075] In this embodiment, the specific value ranges of vegetation physiological parameters and landscape indicators are shown in Table 1 below. It should be noted that the parameters in the table below represent the preferred configuration for this embodiment, and those skilled in the art can make appropriate adjustments based on the actual watershed characteristics.

[0076] Table 1. Schematic diagram of the specific value ranges of vegetation physiological parameters and landscape indicators.

[0077] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0078] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0079] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0080] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0081] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0082] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0083] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for optimizing the spatial distribution of vegetation in a watershed based on water-carbon coupling, characterized in that, Includes the following steps: Acquire basic geographic data, historical meteorological data, and future climate scenario data for the watershed; An ecosystem water-carbon coupling model is constructed to simulate the hydrological and carbon cycle processes under different vegetation spatial distributions, and outputs the simulation results of water and carbon fluxes. An artificial neural network proxy model is constructed and trained using sample data generated by the water-carbon coupling model to predict the water and carbon flux of vegetation spatial distribution schemes. Establish a multi-objective optimization model, and set optimization objectives that include at least maximizing carbon sink and minimizing water consumption intensity; The multi-objective optimization algorithm combined with the artificial neural network surrogate model is used to iteratively optimize the vegetation spatial distribution scheme based on grid coding. During the iteration process, a dynamic calibration step is performed: every preset iteration number, a scheme with a preset proportion or a preset number in the current population is selected and the water-carbon coupling model is called to simulate it. The simulation results of the water-carbon coupling model are compared with the prediction results of the surrogate model. If the error exceeds the preset error threshold, the artificial neural network surrogate model is retrained. The optimized vegetation spatial distribution scheme is output, which is used to guide the configuration of vegetation planting and ecological restoration projects within the watershed.

2. The method according to claim 1, characterized in that, The establishment of the multi-objective optimization model also includes the climate adaptation objective: By introducing the future climate scenario data, the climate robustness index of each optimization scheme under different climate scenarios is calculated. The climate robustness index is used as a constraint or weighting factor for the optimization objective to screen for climate-adaptive vegetation spatial distribution schemes.

3. The method according to claim 2, characterized in that, The formula for calculating the climate robustness index is as follows: ； CRI stands for Climate Robustness Index. and These are the overall fitness values ​​of the scheme under medium and high emission scenarios, respectively. The closer the CRI is to 1, the higher the climate stability of the scheme.

4. The method according to claim 1, characterized in that, The construction of the water-carbon coupling model specifically includes: By integrating a distributed hydrological model and a carbon cycle model, the soil moisture content variable simulated by the hydrological model is used as the input of the water stress factor in the carbon cycle model, thereby achieving dynamic coupling between the water cycle process and the carbon cycle process. The model parameters are calibrated and validated using multi-source data, which includes ground observation data and remote sensing product data.

5. The method according to claim 1, characterized in that, The process of generating the vegetation spatial distribution scheme specifically includes: The watershed is divided into regular grid units, and each grid unit is independently set with vegetation type codes and corresponding vegetation physiological parameters; The vegetation physiological parameters include leaf area index, root depth, and water use efficiency; different vegetation spatial distribution schemes are generated by changing the vegetation type coding of the grid cells.

6. The method according to claim 5, characterized in that, The method also includes a landscape ecological assessment step: Landscape ecological indicators are calculated based on the aforementioned vegetation spatial distribution scheme, and the indicators include at least patch shape index, aggregation index, and edge density; The landscape ecology indicators are used as characteristic variables representing the spatial pattern of vegetation and incorporated into the evaluation function or constraints of the multi-objective optimization model to analyze their comprehensive impact on the watershed's water and carbon balance.

7. The method according to claim 1, characterized in that, The training process of the artificial neural network proxy model includes: Multiple sets of vegetation distribution schemes and their corresponding water and carbon flux simulation results were randomly generated using a water-carbon coupling model to form a training dataset. The training dataset is divided into a training set and a validation set. The artificial neural network is trained until the prediction accuracy of the validation set meets the pre-set accuracy threshold.

8. The method according to claim 1, characterized in that, The iterative update of the vegetation spatial distribution scheme using a multi-objective optimization algorithm specifically includes: A genetic algorithm is used to perform crossover and mutation operations on the vegetation type encoding matrix; The mutation operation includes: randomly selecting the boundary grid cells of vegetation patches and changing their vegetation type codes to adjust the patch shape index and aggregation index.

9. A watershed vegetation spatial distribution optimization system based on water-carbon coupling, characterized in that, include: The data acquisition module is used to acquire basic geographic data, historical meteorological data, and future climate scenario data for the watershed. The model building module is used to build ecosystem water-carbon coupling models and artificial neural network proxy models. The optimization calculation module is used to establish a multi-objective optimization model and use the multi-objective optimization algorithm combined with the artificial neural network proxy model to iteratively optimize the vegetation spatial distribution scheme and execute the dynamic calibration steps as described in claim 1. The output module is used to output the optimized vegetation spatial distribution scheme and the corresponding water and carbon benefit assessment results. The system is used to perform the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 8.

Images