A method for simulating and predicting the soil organic carbon storage in coastal zones

Abstract

The present invention discloses a method for simulating and predicting the soil organic carbon storage in the coastal zone. Through the enhanced regression tree model for estimating and predicting the soil organic carbon storage in the coastal zone constructed, the simulation and prediction of the soil organic carbon storage in the coastal zone are carried out. The method for simulating and predicting the soil organic carbon storage in the coastal zone provided by the present invention uses data such as high-spatial-resolution, open-source Sentinel optical and radar remote sensing as input data, realizes the simulation and monitoring of the coastal wetland at a 10m spatial resolution on a large regional scale, and the average correlation coefficient of the ten-fold cross-validation of the model reaches 0.8, and the accuracy is significantly better than that of the traditional method for estimating soil carbon storage by statistical methods and spatial interpolation methods based on survey sample points according to ecosystem types. It realizes the high-spatial-resolution, high-precision, low-cost, and rapid simulation and prediction of the soil carbon storage in the coastal zone where the space is fragmented and the field sampling survey is difficult.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-source satellite remote sensing ecosystem monitoring, and particularly to a method for simulating and predicting the soil organic carbon storage in the coastal zone. Background Art

[0002] The carbon stored in the soil far exceeds that in the atmosphere and organisms. Among soil carbon, more than half of the carbon is stored in the form of organic carbon, which is twice that of the atmospheric carbon pool. Compared with soil inorganic carbon, soil organic carbon is more active in the carbon cycle process (Batjes, 1996; Crowther et al., 2016). Therefore, the dynamics of soil organic carbon storage have a significant impact on the global and regional climate.

[0003] Compared with terrestrial ecosystems such as forests, grasslands, and farmlands, the coastal zone ecosystem plays a very important role in the regional carbon cycle due to its higher carbon burial rate and absorption rate. Especially as a significant carbon sink ecosystem, it plays an important role in the realization of the "dual carbon" strategic goal. However, due to reasons such as its fragmented spatial distribution, poor accessibility, and large spatial heterogeneity, the estimation and prediction of the carbon storage in the coastal zone ecosystem, especially the accurate estimation and prediction of soil organic carbon storage, still have great uncertainties.

[0004] Existing methods for estimating soil organic carbon storage include: methods based on survey sampling analysis, establishing empirical statistical models or machine learning models through sample point test analysis data for simulation, and simulation based on process models. The method based on survey sampling is only limited to the assessment and analysis of soil organic carbon storage at a small scale. Establishing empirical models or machine learning models to estimate soil organic carbon storage relies on limited sample point observation data and single, medium- and low-resolution satellite remote sensing data, and there are often large uncertainties in the accuracy of simulating and monitoring soil carbon storage. The simulation of soil organic carbon storage based on process models is limited by the complexity of the model, numerous parameters, and complex optimization, with high uncertainties in the estimation of regional soil organic carbon storage and complex and inconvenient applications.

[0005] Satellite remote sensing earth observation technology is widely used in the monitoring of geographical and ecosystem parameters such as surface land cover, vegetation growth, surface soil moisture, and soil temperature, which can be directly monitored or indirectly inverted. However, it is difficult to directly monitor or indirectly invert soil information involving a certain depth such as soil carbon storage through satellite remote sensing. The traditional method of combining sample point sampling test data with medium- and low-resolution remote sensing data is not suitable for monitoring the soil carbon storage in the coastal zone ecosystem. On the one hand, the cost of human and material resources for large-scale soil carbon storage surveys is high and the time cycle is long; on the other hand, the pixels of medium- and low-resolution satellite remote sensing data are not sufficient to distinguish small fragmented coastal wetlands smaller than the resolution scale, resulting in great uncertainties in the current simulation and monitoring results of the carbon storage in the coastal zone ecosystem.

[0006] Although there are many methods for estimating soil organic carbon storage, there are few effective methods for predicting soil organic carbon storage in the coastal zone under the background of future climate change. The coastal zone ecosystem is affected by human activities, climate change, and sea-level rise caused by them. Predicting the coastal zone ecosystem under future climate change scenarios requires considering the impacts of climate change, human activities, and sea-level rise simultaneously, and obtaining reasonable and reliable information on model input variables under future climate change scenarios. In short, there are still great challenges in predicting soil organic carbon storage in the coastal zone under future climate change scenarios. Summary of the Invention

[0007] The object of the present invention is that the traditional method for estimating soil organic carbon storage by combining sample point sampling test data with medium and low-resolution remote sensing data is not suitable for high spatio-temporal resolution monitoring of soil carbon storage in the coastal zone ecosystem, and cannot predict the soil organic carbon storage in the coastal zone under future climate change scenarios. One object of the present invention is to provide a method for estimating the soil organic carbon storage in the coastal zone with high spatial resolution by using currently open-source high spatio-temporal resolution satellite remote sensing data in combination with information such as meteorology, soil properties, and topography and landforms.

[0008] Another object of the present invention is a prediction method based on the above-mentioned simulation method for soil organic carbon storage in the coastal zone.

[0009] The technical solution adopted to achieve the object of the present invention is as follows:

[0010] A method for simulating soil organic carbon storage in the coastal zone, comprising the following steps:

[0011] Step 1: Select model variables required for model construction

[0012] The model variables include: coastal zone vegetation index, normalized vegetation index, normalized water body index, chlorophyll fluorescence index, enhanced vegetation index, normalized vegetation humidity index, improved soil vegetation index, improved vegetation index, normalized salinity index, radar remote sensing polarization index, elevation, terrain humidity index, annual average temperature, annual precipitation, downward solar short-wave radiation, population, distance from the river, soil salinity, and reference soil carbon storage;

[0013] Step 2: Acquisition and processing of model variables

[0014] First, obtain model variable data through the Google Earth Engine geoscience big data cloud platform and open-source basic geographic data products; then use GIS software to perform format conversion and grid matching processing on the obtained model variable data;

[0015] Step 3: Preparation of model training sample data set

[0016] Use the GIS software to extract raster attributes to the point tool to extract the model input variable values corresponding to the soil carbon storage sample points, and obtain the model training sample data set required for constructing the soil carbon storage model;

[0017] Step 4: Model variable optimization

[0018] Through the correlation coefficient matrix analysis and variance inflation factor test methods, select the optimal model variables from the model variables in Step 1, and obtain the optimal model variable data according to the method in Step 2;

[0019] The optimal model variables include: coastal zone vegetation index, normalized vegetation index, normalized difference vegetation moisture index, modified soil adjusted vegetation index, elevation, distance to river, annual precipitation, annual average temperature, population, soil salinity, and reference soil carbon storage;

[0020] Step 5: Construction of the boosted regression tree model for estimating and predicting the coastal zone soil organic carbon storage

[0021] Use the coastal zone spatial distribution data and soil organic carbon sample point data, combined with the model training sample data set obtained in Step 3, to construct the boosted regression tree model for estimating and predicting the coastal zone soil organic carbon storage;

[0022] Step 6: Simulation of the coastal zone soil organic carbon storage

[0023] Use the optimal model variable data obtained in Step 2 as the model input, and through the boosted regression tree model for estimating and predicting the coastal zone soil organic carbon storage constructed in Step 5, realize the simulation and estimation of the coastal zone soil organic carbon storage with a 10m spatial resolution;

[0024] Step 7: Statistical analysis of the coastal zone soil organic carbon storage

[0025] Use the coastal zone soil organic carbon storage spatial distribution data obtained in Step 6 to statistically analyze the total soil organic carbon storage and average organic carbon density of the regional coastal zone.

[0026] On the other hand, a method for predicting the coastal zone soil organic carbon storage of the present invention includes the following steps:

[0027] Step 1: Select the model variables required for model construction

[0028] The model variables include: coastal zone vegetation index, normalized vegetation index, normalized difference water index, chlorophyll fluorescence index, enhanced vegetation index, normalized difference vegetation moisture index, modified soil adjusted vegetation index, modified vegetation index, normalized salinity index, radar remote sensing polarization index, elevation, terrain moisture index, annual average temperature, annual precipitation, downward solar shortwave radiation, population, distance to river, soil salinity, and reference soil carbon storage;

[0029] Step 2: Acquisition and Processing of Model Variables

[0030] First, obtain model variable data through the Google Earth Engine geoscience big data cloud platform and open-source basic geospatial data products; then use GIS software to perform format conversion and grid matching processing on the obtained model variable data;

[0031] Step 3: Preparation of Model Training Sample Datasets

[0032] Use GIS software to extract raster attributes to point tools to extract the model input variable values corresponding to soil carbon storage sample points, and obtain the model training sample datasets required for constructing the soil carbon storage model;

[0033] Step 4: Optimization of Model Variables

[0034] Through correlation coefficient matrix analysis and variance inflation factor tests, screen the optimal model variables from the model variables in Step 1, and obtain the optimal model variable data according to the method in Step 2;

[0035] The optimal model variables include: coastal vegetation index, normalized difference vegetation index, normalized difference moisture index, modified soil adjusted vegetation index, elevation, distance to river, annual precipitation, annual average temperature, population, soil salinity, and reference soil carbon storage;

[0036] Step 5: Construction of an Enhanced Regression Tree Model for Estimation and Prediction of Coastal Zone Soil Organic Carbon Storage

[0037] Use coastal zone spatial distribution data and soil organic carbon sample point data, combined with the model training sample datasets obtained in Step 3, to construct an enhanced regression tree model for estimation and prediction of coastal zone soil organic carbon storage;

[0038] Step 6: Analysis and Screening of the Relative Importance of Optimal Model Variables

[0039] Through the analysis of variable importance and the analysis of partial dependence plot features of the constructed enhanced regression tree model for coastal zone soil organic carbon storage, determine the relative importance of each optimal model variable screened in Step 4 for the simulation and prediction of coastal zone soil organic carbon, and classify them into important variables, unimportant variables, and stable variables;

[0040] Step 7: Acquisition and Processing of Optimal Model Variable Data

[0041] The existing data is used for the stable variables;

[0042] For unimportant variables other than stable variables, use existing multi-year average satellite remote sensing data;

[0043] The meteorological data are MAT and MAP data for the years 2061 - 2080 and 2081 - 2100 under two scenarios of SSP245 and SSP585 predicted by CMIP6 multi - models;

[0044] The population data are obtained from the China population dataset predicted by the model for the corresponding scenarios and time periods;

[0045] The soil salinity data are obtained by simulating and constructing a multi - source linear regression model using the AIC method with the collected soil sample organic carbon sample point data and multi - source data;

[0046] Step 8: Obtaining the coastal zone spatial distribution data under future climate change scenarios

[0047] Based on the average sea - level rise rate calculated from the actual sea - level rise observation data, considering seawater inundation and landward migration on the basis of the existing coastal zone spatial distribution data, the coastal zone spatial distribution data under future climate change scenarios are obtained.

[0048] Step 9: Predicting the soil organic carbon storage in the coastal zone ecosystem under future climate change scenarios

[0049] Taking the optimal model variable data obtained in Step 7 as the input of the enhanced regression tree model for estimating and predicting the soil organic carbon storage in the coastal zone, and combining with the coastal zone spatial distribution data under future climate change scenarios obtained in Step 8, the soil organic carbon storage data in the coastal zone under different warming scenarios are predicted.

[0050] In the above technical solution, in Step 4, the screening principle is that when the correlation coefficient between two variables in the correlation coefficient matrix is greater than or equal to 0.8, the variable with a lower correlation coefficient with the soil organic carbon storage is excluded; variables with a VIF value greater than 0.4 among all variables are excluded.

[0051] In the above technical solution, in Step 5, the construction and simulation of the enhanced regression tree model for simulating and predicting the soil organic carbon storage in the coastal zone are realized by using the gbm.step function in the dismo package of R language.

[0052] In the above technical solution, model parameter optimization is also included;

[0053] The model parameter optimization method is to adopt a step - by - step simulation test method. When the three key parameters of the learning rate, tree complexity, and bagging ratio are set to make the model reach the optimum, the optimal enhanced regression tree model for estimating and predicting the soil organic carbon storage in the coastal zone is determined.

[0054] In the above technical solution, the constructed and optimized enhanced regression tree model for estimating and predicting the soil organic carbon storage in the coastal zone is verified and evaluated through ten - fold cross - validation (10fold cv).

[0055] In the above technical solution, in step 6, the relative importance of the 11 optimal model variables is as follows:

[0056] Annual average temperature (19.8%), soil salinity (19.7%), distance to the river (13.7), population (13.4%), coastal vegetation index (7.6%), normalized difference vegetation moisture index (6.1%), elevation (5.6%), reference soil carbon storage (4.7%), normalized difference vegetation index (3.6%), annual precipitation (2.9%), improved soil vegetation index (2.8).

[0057] In the above technical solution, the division principles for important variables, unimportant variables, and stable variables are as follows:

[0058] Variables with a relative importance greater than 10% are defined as important variables; variables with a relative importance less than 10% are defined as unimportant variables; variables with little change over time are defined as stable variables.

[0059] In the above technical solution, the important variables include annual average temperature, soil salinity, distance to the river, and population;

[0060] The unimportant variables include coastal vegetation index, normalized difference vegetation moisture index, elevation, reference soil carbon storage, normalized difference vegetation index, annual precipitation, and improved soil vegetation index;

[0061] The stable variables include distance to the river, elevation, and reference soil carbon storage.

[0062] In the above technical solution, in step 7, the multi-source linear regression model is shown as the following formula:

[0063] Salinity=-0.0066×Elevation+8.3523×Distance-0.0173×MAP+0.3989×MAT-0.0017×Population+7.3230.

[0064] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0065] 1. The method for simulating the soil organic carbon storage in the coastal zone provided by the present invention uses high-spatial-resolution and open-source Sentinel optical and radar remote sensing data as input data, realizes the simulation and monitoring of the coastal wetland at a 10m spatial resolution on a large regional scale, and the average correlation coefficient of the ten-fold cross-validation of the model reaches 0.8, and the accuracy is significantly better than the traditional statistical method and spatial interpolation method for estimating soil carbon storage based on survey sample points according to ecosystem types. It realizes the high-spatial-resolution, high-precision, low-cost, and rapid simulation and prediction of the soil carbon storage in the coastal zone where space is fragmented and field sampling is difficult.

[0066] 2. The method for predicting the soil organic carbon storage in the coastal zone provided by the present invention aims at the problems of large computational amount, complexity and high uncertainty in the process model for predicting the soil carbon storage in the coastal ecosystem under different future climate change scenarios. In this invention, the impacts of climate warming, sea - level rise and human activities are considered simultaneously. By first identifying the main environmental impact factors of the soil carbon storage in the coastal ecosystem, a soil salinity model based on environmental and human activity factors variables such as meteorology, topography, population, etc., which are easy to obtain, is constructed to simulate and predict the soil salinity under future climate change scenarios required for the soil organic carbon storage prediction model. Then, combined with the CMIP6 future scenario meteorological data and future population open - source data, a data - driven model is constructed to achieve high - precision, high - spatial - resolution (10m) and efficient prediction and simulation of the soil organic carbon storage in the coastal ecosystem under different future scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 The figure shows the flow chart of the method for simulating the soil organic carbon storage in the coastal zone;

[0068] Figure 2 The figure shows the relationship between the independent variables of the soil organic carbon storage prediction model in the coastal zone and the soil organic carbon storage;

[0069] Figure 3 The figure shows the spatial distribution data of environmental factor variables;

[0070] Figure 4 The figure shows the simulation results of the soil organic carbon storage in the coastal zone of Tianjin;

[0071] Figure 5 The figure shows the partial dependence diagram of the BRT model for simulating the soil organic carbon storage in the coastal zone;

[0072] Figure 6 The figure shows the prediction results of the soil organic carbon storage in the coastal zone of Tianjin;

[0073] Among them, (a) soil organic carbon storage from 2041 to 2060 under the SSP245 scenario, (b) soil organic carbon storage from 2081 to 2100 under the SSP245 scenario, (c) soil organic carbon storage from 2041 to 2060 under the SSP585 scenario, (d) soil organic carbon storage from 2081 to 2100 under the SSP585 scenario. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0074] The following further elaborates on the present invention in conjunction with specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0075] Embodiment 1

[0076] A method for simulating the soil organic carbon storage in the coastal zone, comprising the following steps:

[0077] Step 1: Select model variables required for model construction

[0078] According to literature research and the characteristics of soil organic carbon in the coastal zone, initially select 19 variables that may be used for simulating soil organic carbon storage, including currently available open-source, high-spatial-resolution optical satellite remote sensing (Sentinel-2) and microwave (radar) satellite remote sensing (Sentinel-1), meteorology, soil properties, topography, and population, etc.;

[0079] Specifically, the initially determined model variables include satellite remote sensing observation data, topographic data, soil property data, meteorological data, and population data (Population), a total of 19 environmental factor variables. Among them, the optical satellite remote sensing variables include the coastal zone vegetation index (BNDVI), normalized vegetation index (NDVI), normalized water index (NDWI), chlorophyll fluorescence index (LCI), enhanced vegetation index (EVI), normalized vegetation moisture index (NDMI), modified soil vegetation index (SAVI), and modified vegetation index (kNDVI) based on Sentinel-2 optical satellite remote sensing observations; the radar remote sensing data includes the polarization index (POL) based on Sentinel-1 radar satellite remote sensing; the topographic data includes elevation (DEM), topographic wetness index (TWI), and distance to the river (Distance); the soil property data includes normalized salinity index (NDSI), soil salinity (Salinity), and reference soil carbon storage (SoilGrids); the meteorological data includes mean annual temperature (MAT), annual precipitation (MAP), and downward solar shortwave radiation (SRAD).

[0080] Step 2: Acquisition and processing of model variables

[0081] Among the model variables, the satellite remote sensing data are all obtained through the Google Earth Engine (GEE) geoscience big data cloud platform, and the meteorological data, soil property data, topographic data, and population data are all obtained through open-source basic geographic data products;

[0082] Use GIS software to perform preprocessing on all the obtained model variable data, such as data calibration and grid size matching.

[0083] Step 3: Preparation of the model training sample data set

[0084] Use GIS software to extract the raster attributes to point tool to extract the model input variable values corresponding to the soil carbon storage sample points, and obtain the model training sample data set required for constructing the soil carbon storage model.

[0085] Step 4: Optimization of Model Variables

[0086] Through the correlation coefficient matrix analysis and variance inflation factor test method, select the optimal model variables with the least multicollinearity from the model variables in Step 2.

[0087] Specifically, the screening principle is that when the correlation coefficient between two variables in the correlation coefficient matrix is greater than or equal to 0.8, the variable with a lower correlation coefficient with soil organic carbon storage is excluded; variables with a VIF value greater than 0.4 among all variables are excluded.

[0088] The selected optimal model variables include BNDVI, NDMI, NDVI, SAVI, DEM, Distance, MAP, MAT, Population, Soil salinity, SoilGrids, a total of 11 optimal model variables.

[0089] Step 5: Construction, Optimization and Evaluation of the Boosted Regression Tree Model for Simulating and Predicting the Soil Organic Carbon Storage in the Coastal Zone

[0090] Using the model training sample dataset prepared in Step 3, based on the gbm.step function in the dismo package of R language, construct a boosted regression tree model (BRT model) for estimating and predicting the soil organic carbon storage in the coastal zone;

[0091] Among them, the coastal zone spatial distribution data is selected from existing multiple high-spatial-resolution coastal zone data products, such as Sun et al. (2020), National Earth System Science Data Center (http: / / www.geodata.cn / ); the soil organic carbon sampling point data is obtained by literature retrieval and field sampling analysis.

[0092] The parameter optimization of the boosted regression tree (BRT) model adopts the step-by-step simulation test method. When the three key parameters of the learning rate (lr), tree complexity (tc) and bagging ratio (bg) are optimal, the optimal BRT model is determined. The model error distribution is set as the Poisson distribution, and the constructed and optimized optimal BRT model is verified and evaluated through ten-fold cross-validation (10-fold cv).

[0093] Step 6: Simulation of the Soil Organic Carbon Storage in the Coastal Zone

[0094] Taking the data of the 11 optimal model variables obtained in Step 2 as the input, through the boosted regression tree model for estimating and predicting the soil organic carbon storage in the coastal zone constructed in Step 5, simulate and obtain the spatial distribution data of the soil organic carbon storage in the coastal zone with a 10m spatial resolution.

[0095] Step 7: Statistical Analysis of the Soil Organic Carbon Storage in the Coastal Zone

[0096] Using the spatial distribution data of coastal zone soil organic carbon storage obtained in Step 6, statistically analyze the total soil organic carbon storage and average organic carbon density in the regional coastal zone.

[0097] Example 2

[0098] This example is based on the enhanced regression tree model for estimating and predicting coastal zone soil organic carbon storage constructed in Example 1, and introduces a method for predicting the coastal zone soil organic carbon storage under future climate change scenarios using this model.

[0099] A method for predicting coastal zone soil organic carbon storage with high spatial resolution, comprising the following steps:

[0100] Steps 1-5 are the same as Steps 1-5 in Example 1 and will not be elaborated here.

[0101] Step 6: Analysis and screening of the relative importance of the optimal model variables

[0102] Through the analysis of the importance of BRT model variables and the characteristics analysis of partial dependence plots (PDPs), determine the relative importance of the optimal model variables selected in Step 3 for simulating the coastal zone soil organic carbon storage, and classify them into important variables, unimportant variables, and stable variables;

[0103] Specifically, the relative importance of the 11 optimal model variables is 19.8% (MAT), Salinity (19.7%), Distance (13.7), Population (13.4%), BNDVI (7.6%), NDMI (6.1%), DEM (5.6%), SoilGrids (4.7%), NDVI (3.6%), MAP (2.9%), SAVI (2.8).

[0104] Define variables with a relative importance greater than 10% as important variables; define variables with a relative importance less than 10% as unimportant variables; define variables with little change over time as stable variables;

[0105] According to the above definitions, the important variables include MAT, Salinity, Distance, and Population;

[0106] The unimportant variables include BNDVI, NDMI, DEM, SoilGrids, NDVI, MAP, and SAVI;

[0107] The stable variables include Distance, DEM, and SoilGrids.

[0108] Step 7: Acquisition and processing of the optimal model variable data

[0109] The stable variables adopt existing data; for the non-important variables other than the stable variables, the existing multi-year average satellite remote sensing data is adopted;

[0110] For meteorological data (MAT) and population data (Population), the MAT and MAP data for the periods 2061 - 2080 and 2081 - 2100 under the two scenarios of SSP245 and SSP585 predicted by multiple models in CMIP6 and the corresponding scenario and time period of the predicted Chinese population data (https: / / doi.org / 10.7927 / H4JW8BX5) from (http: / / www.nmic.cn / ) are adopted;

[0111] The soil salinity data (Salinity) is obtained by combining the collected soil sample organic carbon sample point data with the model variable data preliminarily determined in Step 1, and using the AIC method (Akaike, 1981) to construct the following optimal multi-source linear regression model to simulate and obtain the soil salinity data under the SSP245 and SSP585 scenarios:

[0112] Salinity = -0.0066×Elevation + 8.3523×Distance - 0.0173×MAP + 0.3989×MAT - 0.0017×Population + 7.3230 (1)

[0113] Step 8: Obtaining the coastal zone spatial distribution data under future climate change scenarios

[0114] The coastal zone spatial distribution data under future warming scenarios is based on the actual conditions of the simulation and prediction area. The global potential coastal zone spatial distribution data under future climate warming and sea-level rise scenarios developed by Schuerch et al. (2018) is adopted, or the possible coastal zone spatial distribution data under future warming scenarios is calculated by considering seawater inundation and landward migration on the basis of the existing coastal zone spatial distribution data according to the average sea-level rise rate calculated from the actual sea-level rise observation data.

[0115] Step 9: Predicting the soil organic carbon storage in the coastal zone ecosystem under future climate change scenarios

[0116] Using the optimal model variable data obtained in Step 7 as the input to the BRT model, and combining with the coastal zone spatial distribution data under future climate change scenarios obtained in Step 8 to predict the soil organic carbon storage data in the coastal zone under different warming scenarios. Such as the soil organic carbon storage data in the coastal zone for the periods 2061 - 2080 and 2080 - 2100 under the SSP245 and SSP585 scenarios, and statistically analyzing its spatio-temporal changes relative to historical and current soil organic carbon.

[0117] Example 3

[0118] This embodiment takes the simulation of the soil organic carbon storage in the coastal ecosystem (offshore administrative region) of Tianjin as an example, and simulates and predicts the soil organic carbon storage in the coastal ecosystem of Tianjin according to the simulation method or prediction method in Embodiment 1 and Embodiment 2.

[0119] (1) Data acquisition and processing of optimal satellite remote sensing and environmental factor variables

[0120] Through literature search and analysis, as well as the analysis of the biogeophysical and biogeochemical properties of soil organic carbon in the coastal zone, 19 initial environmental factor variables that are easy to obtain and closely related to the soil organic carbon storage in the coastal zone and are used to simulate and predict the soil organic carbon storage in the coastal zone are initially determined, that is, the model variables. Their definitions and acquisition sources are shown in the following table:

[0121]

[0122] Combined with literature data, the sample point data of soil organic carbon storage in the coastal zone obtained from actual investigation sample collection, testing and analysis, a model training data sample point data set is constructed; based on the correlation coefficient matrix and variance inflation factor (VIF) test method, the optimal environmental factor variables for model training are screened. That is, for the initially determined 19 environmental factor variables, the correlation coefficient matrix is first calculated. For two variables with a correlation coefficient greater than 0.8, the one with a smaller correlation coefficient with the soil organic carbon storage is removed. On this basis, further VIF tests are carried out to remove variables with a VIF greater than 4. Finally, a total of 11 variables that are the most important for predicting the soil organic carbon storage in the coastal zone are screened out, that is, the optimal model variables, BNDVI, NDMI, NDVI, SAVI, Elevation (DEM), Distance, MAP, MAT, Population, Soil salinity, SoilGrids.

[0123] Further analysis results show that except for Distance (p = 0.057), there is a significant correlation between the soil organic carbon storage (SCT) in the coastal zone ecosystem and the remaining 10 variables screened out (p < 0.05) ( Figure 2 ). Based on the GEE platform and GIS software, the data of the selected 11 model variables are obtained, and preprocessing such as data calibration and grid size matching is carried out. As Figure 3 shown is the spatial distribution data of variables with a 10m spatial resolution in the coastal zone of Tianjin after processing.

[0124] According to the simulation method of soil organic carbon storage in the coastal zone in Embodiment 1, an enhanced regression tree model for estimating and predicting the soil organic carbon storage in the coastal zone is constructed.

[0125] Multiple groups of model experiments were carried out using step-by-step simulation tests (changing parameters step by step at certain gradient intervals) to determine the optimal values of the learning rate (lr), tree complexity (tc), and bagging ratio (bg) parameters of the BRT model for soil organic carbon in the coastal zone of Tianjin as 0.005, 0.5, and 5, respectively, and to determine the optimal BRT model for SCT simulation prediction. The results of the ten-fold cross-validation evaluation of the model simulation showed that the correlation coefficient between the simulation of the constructed BRT model and the SCT sample point observations reached 0.78, meeting the SCT simulation accuracy requirements.

[0126] (III) Simulation of Soil Organic Carbon Stocks in the Coastal Zone

[0127] Using the spatially distributed data of the 11 optimal model variables obtained in Step 1 as model inputs, the spatially distributed data of soil organic carbon stocks in the coastal zone ecosystem of Tianjin at a 10 m spatial resolution was simulated ( Figure 4 ), where the spatially distributed data of the coastal zone was the 10 m spatial resolution coastal zone data developed by Sun et al. (2020) in the Bohai Rim. Based on the simulation results, the statistical analysis showed that the soil organic carbon stocks and average carbon density in the coastal zone of Tianjin were 8.3 Tg and 68.8 Mg ha -1 .

[0128] (IV) Prediction of Soil Organic Carbon Stocks in the Coastal Zone

[0129] Through the variable importance analysis of the BRT model and the characteristics analysis of the partial dependence plots (PDPs), the relative importance of the independent variables in the model input to the simulation of soil organic carbon stocks was determined. The relative importance of the 11 optimal model variables in the simulation of soil organic carbon stocks in the coastal zone ecosystem of Tianjin is as Figure 5 shown, in order: 19.8% (MAT), Salinity (19.7%), Distance (13.7), Population (13.4%), BNDVI (7.6%), NDMI (6.1%), DEM (5.6%), SoilGrids (4.7%), NDVI (3.6%), MAP (2.9%), SAVI (2.8). The relative importance of the mean annual temperature, soil salinity, distance from the river, and population to the simulation of soil organic carbon stocks was 19.8%, 19.7%, 13.7%, and 13.4%, respectively, which was significantly higher than that of other environmental factor variables. The relative importance of BNDVI, NDMI, Elevation (DEM), SoilGrids, NDVI, MAP, and SAVI was all less than 10%. Since future meteorological data and population data can be obtained through CMIP6 and various existing future population data products, therefore, for the prediction of soil organic carbon stocks, the key is how to obtain soil salinity data under future climate change scenarios.

[0130] Taking the prediction of soil organic carbon storage in the coastal zone of Tianjin under two future climate change scenarios of SSP245 and SSP585 in the years 2041 - 2060 and 2081 - 2100 as an example.

[0131] First, the soil salinity data (Salinity) under future climate change scenarios was simulated using formula (1).

[0132] Meteorological data was obtained from the average meteorological data (MAT and MAP) in the years 2041 - 2060 and the multi - year average meteorological data in the years 2081 - 2100 under the CMIP6 SSP245 and SSP585 scenarios. The population data under future climate change scenarios was extracted from the dataset of China's population prediction under future climate change scenarios developed by Chen (et al., 2020).

[0133] For the distance to the river, DEM, and SoilGrids that do not change much over time, existing data can be used. For the relatively less important variables such as BNDVI, NDMI, NDVI, and SAVI, existing multi - year average satellite remote sensing data can be used.

[0134] Using the above - obtained model input variable data as the input for the constructed BRT model for simulating and predicting soil organic carbon storage in the coastal zone, the soil organic carbon storage under two future climate change scenarios of SSP245 and SSP585 in the years 2041 - 2060 and 2081 - 2100 was simulated and predicted.

[0135] Finally, combining the possible coastal zone spatial distribution data under future warming scenarios obtained from the average sea - level rise rate calculated from the actual sea - level rise observation data, the soil organic carbon storage under future climate change scenarios was estimated ( Figure 6 ). The statistical analysis results show that under the SSP245 scenario, the total soil organic carbon storage and average carbon density in the coastal zone of Tianjin in the years 2041 - 2060 and 2081 - 2100 are 6.9 Tg and 57.2 Mg ha -1 , 5.9 Tg and 48.9 Mg ha -1 ; under the SSP585 scenario, the total soil organic carbon storage and average carbon density in the coastal zone of Tianjin in the years 2041 - 2060 and 2081 - 2100 are 7.1 Tg C and 58.9 Mg ha -1 , 5.6 Tg C and 46.4 Mg ha -1 .

[0136] The above - mentioned is only the preferred implementation manner of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for simulating the soil organic carbon storage in the coastal zone, characterized in that: It includes the following steps: Step 1: Select the model variables required for model construction The model variables include: coastal vegetation index, normalized difference vegetation index, normalized difference water index, chlorophyll fluorescence index, enhanced vegetation index, normalized difference vegetation moisture index, modified soil adjusted vegetation index, modified vegetation index, normalized difference salinity index, radar remote sensing polarization index, elevation, terrain moisture index, annual average temperature, annual precipitation, downward solar shortwave radiation, population, distance to the river, soil salinity, and reference soil carbon storage; Step 2: Acquisition and processing of model variables First, obtain the model variable data through the Google Earth Engine geoscience big data cloud platform and open-source basic geographic data products; then use GIS software to perform format conversion and grid matching processing on the obtained model variable data; Step 3: Preparation of the model training sample dataset Use GIS software to extract the raster attributes to the point tool to extract the model input variable values corresponding to the soil carbon storage sample points, and obtain the model training sample dataset required for constructing the soil carbon storage model; Step 4: Optimization of model variables Through the correlation coefficient matrix analysis and variance inflation factor test method, select the optimal model variables from the model variables in Step 1, and obtain the optimal model variable data according to the method in Step 2; The optimal model variables include: coastal vegetation index, normalized difference vegetation index, normalized difference vegetation moisture index, modified soil adjusted vegetation index, elevation, distance to the river, annual precipitation, annual average temperature, population, soil salinity, and reference soil carbon storage; Step 5: Construction of the boosted regression tree model for estimating and predicting the coastal zone soil organic carbon storage Use the coastal zone spatial distribution data and soil organic carbon sample point data, and combine with the model training sample dataset obtained in Step 3 to construct the boosted regression tree model for estimating and predicting the coastal zone soil organic carbon storage; Step 6: Simulation of the coastal zone soil organic carbon storage Take the optimal model variable data obtained in Step 2 as the model input, and through the boosted regression tree model for estimating and predicting the coastal zone soil organic carbon storage constructed in Step 5, realize the simulation and estimation of the coastal zone soil organic carbon storage with a 10m spatial resolution; Step 7: Statistical analysis of the coastal zone soil organic carbon storage Use the coastal zone soil organic carbon storage spatial distribution data obtained in Step 6 to statistically analyze the total soil organic carbon storage and average organic carbon density of the regional coastal zone.

2. The method for simulating the soil organic carbon storage in the coastal zone according to claim 1, wherein: In Step 4, the screening principle is that when the correlation coefficient between two variables in the correlation coefficient matrix is greater than or equal to 0.8, eliminate the variable with a lower correlation coefficient with the soil organic carbon storage; eliminate the variable with a VIF value greater than 0.4 among all variables.

3. The method for simulating the soil organic carbon storage in the coastal zone according to claim 1, characterized in that: In Step 5, the construction and simulation of the boosted regression tree model for estimating and predicting the coastal zone soil organic carbon storage are realized by using the gbm.step function in the dismo package of R language.

4. The method for simulating the soil organic carbon storage in the coastal zone according to claim 3, wherein: It also includes model parameter optimization; The model parameter optimization method is to adopt the step-by-step simulation test method. When the three key parameters of the learning rate, tree complexity, and bagging ratio are set to make the model reach the optimal, determine the optimal boosted regression tree model for estimating and predicting the coastal zone soil organic carbon storage.

5. The method for simulating the soil organic carbon storage in the coastal zone according to claim 4, wherein: The enhanced regression tree model for estimating and predicting the soil organic carbon storage in the coastal zone after construction and optimization is verified and evaluated through ten-fold cross-validation (10fold cv).

6. A method for predicting the soil organic carbon storage in the coastal zone, characterized in that: It includes the following steps: Step 1: Select model variables required for model construction The model variables include: coastal zone vegetation index, normalized vegetation index, normalized water body index, chlorophyll fluorescence index, enhanced vegetation index, normalized vegetation moisture index, improved soil vegetation index, improved vegetation index, normalized salinity index, radar remote sensing polarization index, elevation, terrain moisture index, annual average temperature, annual precipitation, downward solar shortwave radiation, population, distance to the river, soil salinity, and reference soil carbon storage; Step 2: Acquisition and processing of model variables First, obtain model variable data through the Google Earth Engine geoscience big data cloud platform and open-source basic geographic data products; then use GIS software to perform format conversion and grid matching processing on the obtained model variable data; Step 3: Preparation of the model training sample dataset Use the GIS software to extract raster attributes to point tools to extract the model input variable values corresponding to the soil carbon storage sample points, and obtain the model training sample dataset required for constructing the soil carbon storage model; Step 4: Optimization of model variables Through the correlation coefficient matrix analysis and variance inflation factor test methods, screen the optimal model variables from the model variables in Step 1, and obtain the optimal model variable data according to the method in Step 2; The optimal model variables include: coastal zone vegetation index, normalized vegetation index, normalized vegetation moisture index, improved soil vegetation index, elevation, distance to the river, annual precipitation, annual average temperature, population, soil salinity, and reference soil carbon storage; Step 5: Construction of the enhanced regression tree model for estimating and predicting the soil organic carbon storage in the coastal zone Use the coastal zone spatial distribution data and soil organic carbon sample point data, and combine with the model training sample dataset obtained in Step 3 to construct an enhanced regression tree model for estimating and predicting the soil organic carbon storage in the coastal zone; Step 6: Analysis and screening of the relative importance of the optimal model variables Through the variable importance analysis and partial dependence plot feature analysis of the constructed enhanced regression tree model for the soil organic carbon storage in the coastal zone, determine the relative importance of each optimal model variable screened in Step 4 for the simulation and prediction of the soil organic carbon in the coastal zone, and classify them into important variables, unimportant variables, and stable variables; Step 7: Acquisition and processing of the optimal model variable data The existing data is used for the stable variables; For unimportant variables other than stable variables, the existing satellite remote sensing multi-year average data is used; The meteorological data uses the MAT and MAP data for the years 2061 - 2080 and 2081 - 2100 under the two scenarios of SSP245 and SSP585 predicted by multiple models in CMIP6; The population data is obtained from the Chinese population dataset predicted by the model for the corresponding scenarios and time periods; The soil salinity data is simulated and obtained by constructing a multi-source linear regression model using the collected soil sample organic carbon sample point data and multi-source data through the AIC method; Step 8: Acquisition of the coastal zone spatial distribution data under future climate change scenarios The average sea level rise rate calculated based on the observed data of actual sea level rise, and the spatial distribution data of the coastal zone in the future climate change scenario is calculated by considering seawater inundation and landward migration on the basis of the existing spatial distribution data of the coastal zone. Step 9: Prediction of soil organic carbon storage in the coastal zone ecosystem under future climate change scenarios Using the optimal model variable data obtained in Step 7 as the input of the boosted regression tree model for estimating and predicting soil organic carbon storage in the coastal zone, and combining with the spatial distribution data of the coastal zone under future climate change scenarios obtained in Step 8 to predict the soil organic carbon storage data in the coastal zone under different warming scenarios.

7. The method for predicting the soil organic carbon storage in the coastal zone according to claim 6, characterized in that: In Step 4, the screening principle is that when the correlation coefficient between two variables in the correlation coefficient matrix is greater than or equal to 0.8, the variable with a lower correlation coefficient with soil organic carbon storage is excluded; variables with a VIF value greater than 0.4 among all variables are excluded.

8. The method for predicting the soil organic carbon storage in the coastal zone according to claim 6, wherein: In Step 5, the construction and simulation of the boosted regression tree model for simulating and predicting soil organic carbon storage in the coastal zone are implemented using the gbm.step function in the dismo package of R language.

9. The method for predicting the soil organic carbon storage in the coastal zone according to claim 8, wherein: It also includes model parameter optimization; The model parameter optimization method is to adopt a step-by-step simulation test method. When the three key parameters of learning rate, tree complexity, and bagging ratio are set to make the model reach the optimal state, the optimal boosted regression tree model for estimating and predicting soil organic carbon storage in the coastal zone is determined.

10. The method for predicting the soil organic carbon storage in the coastal zone according to claim 9, wherein: The constructed and optimized boosted regression tree model for estimating and predicting soil organic carbon storage in the coastal zone is verified and evaluated through ten-fold cross-validation (10fold cv).

11. The method for predicting the soil organic carbon storage in the coastal zone according to claim 6, wherein: The division principles for important variables, unimportant variables, and stable variables are as follows: Variables with a relative importance greater than 10% are defined as important variables; variables with a relative importance less than 10% are defined as unimportant variables; variables with little change over time are defined as stable variables.

12. The method for predicting the soil organic carbon storage in the coastal zone according to claim 11, wherein: The important variables include annual average temperature, soil salinity, distance to the river, and population; The unimportant variables include coastal zone vegetation index, normalized vegetation moisture index, elevation, reference soil carbon storage, normalized vegetation index, annual precipitation, and modified soil vegetation index; The stable variables include distance to the river, elevation, and reference soil carbon storage.

13. The method for predicting the soil organic carbon storage in the coastal zone according to claim 6, wherein: In Step 7, the multi-source linear regression model is shown as follows: Salinity = -0.0066 × Elevation + 8.3523 × Distance - 0.0173 × MAP + 0.3989 × MAT - 0.0017 × Population + 7.3230.

Images