Regional scale future soil carbon spatial-temporal change prediction method based on geographically weighted regression model

Abstract

The invention relates to the technical field of ecological environment management, and provides a regional scale future soil carbon spatial-temporal change prediction method based on a geographically weighted regression model, comprising the following steps: S1, collecting data; s2, constructing a soil carbon future spatial-temporal change prediction model and performing soil carbon density spatial-temporal change prediction in different periods in the future, including S21, calculating the annual change rate of the soil carbon density; s22, selecting a prediction model auxiliary variable; s23, constructing a soil carbon future temporal and spatial change prediction model; and S24, predicting spatial and temporal changes of the soil carbon density in different periods in the future. According to the method, the empirical relationship and the spatial difference between the annual change rate of the soil organic carbon density and the influence factors are captured by using the geographically weighted regression model, and the influence of the soil initial carbon density on the annual change rate of the soil carbon density is fused; and the prediction of the spatial and temporal distribution trend of the soil carbon density under different temperature change scenes in the future of the region is effectively realized.

Description

Technical Field

[0001] The present invention relates to the technical field of ecological environment governance, and in particular to a method for predicting future spatiotemporal changes in soil carbon at a regional scale based on a geographically weighted regression model. Background Art

[0002] Soils store two to three times more carbon (C) than the atmosphere and vegetation, playing a crucial role in the global C cycle. Even small changes in their C reserves can cause changes in atmospheric CO₂ concentrations, which in turn influence global climate change. Promoting C sequestration in terrestrial ecosystems, particularly through soil sequestration, is considered an important green approach to mitigating global warming. Arable land is the largest land use type globally, covering approximately 37% of the land surface. Because soil C fluxes are relatively low compared to natural soils such as forests and grasslands, arable land is considered to have significant potential for C sequestration. Therefore, predicting the future spatiotemporal variations in soil C in cultivated land areas with high human activity under climate change scenarios (especially global warming) is crucial for developing appropriate long-term management measures for cultivated land to achieve soil C sequestration and carbon neutrality.

[0003] Currently, models for predicting future soil C changes primarily include process models and mechanistic models based on soil formation and evolution, such as the Century model, the Roth C model, and the DNDC model, as well as models based on empirical relationships between soil C and influencing factors. These process and mechanistic models require numerous and complex parameters to calibrate and are susceptible to human intervention. Consequently, cross-regional predictions of soil C responses to climate change exhibit significant discrepancies, with different models predicting inconsistent magnitudes and even directions of soil C changes. Furthermore, existing studies often employ extreme treatment conditions (beyond the range of possible climate change over the next few decades), raising concerns about the validity of extrapolated data and potentially leading to biased predictions of future soil C trends. Empirical relationship prediction models predict future soil C changes by fitting relationships between current soil C and its influencing factors. Although these models reflect only empirical relationships and are difficult to generalize to other regions, their relatively simple modeling process allows for relatively good predictions of future soil C changes within the region where the model is developed. For example, multiple linear regression models can be used to establish empirical relationships between various factors and soil C, thereby predicting future soil C trends under varying environmental conditions. However, the relationships between soil C and influencing factors are often spatially nonstationary. Only one multivariate linear regression model was established for the study area, and its regression coefficients were spatially consistent, ignoring the differences in the effects of various influencing factors on soil C at different spatial locations, which inevitably led to high uncertainty in the prediction results.

[0004] The regression coefficients of the independent variables in the geographically weighted regression (GWR) model vary with spatial location, enabling the detection of the varying degrees of influence of influencing factors on the target variable at different spatial locations, thereby better revealing the spatial nonstationarity of the relationships between soil properties and these influencing factors. However, changes in soil C are also related to the initial soil C density. As soil C density increases, soil C gradually approaches saturation, and the rate of change slows. Therefore, incorporating the influence of initial soil C density into the model is essential for accurately predicting future trends in soil C. Summary of the Invention

[0005] In response to the above technical problems, the present invention provides a method for predicting future spatiotemporal changes in soil carbon at a regional scale based on a geographically weighted regression model, which can more accurately express the spatial non-stationarity of the relationship between the soil C change rate and various influencing factors. By introducing the initial soil C density to integrate the influence of the initial soil C density, the soil C change rate and soil carbon density are predicted year by year, in order to provide a method reference for better predicting the future spatiotemporal change pattern of soil C at a regional scale.

[0006] The present invention provides the following technical solutions:

[0007] The present invention provides a method for predicting future spatiotemporal changes in soil carbon at a regional scale based on a geographically weighted regression model, which comprises the following steps:

[0008] S1: Collecting Data

[0009] Collect soil data and influencing factors data for the study geographical area;

[0010] S2: Construct a prediction model for the future spatiotemporal changes of soil carbon and predict the spatiotemporal changes of soil carbon density in different periods in the future, including the following steps:

[0011] S21: Calculation of annual change rate of soil carbon density

[0012] The annual change rate of soil carbon density was calculated by the soil organic or inorganic carbon density of paired sampling points in two historical periods; the annual change rate of soil organic or inorganic carbon density of all paired sampling points was calculated according to the sampling period interval between the two historical periods, which were the annual change rate of soil organic carbon density SOCD and SOCD. c and annual change rate of soil inorganic carbon density SICD c ;

[0013] S22: Select auxiliary variables for the prediction model

[0014] Data on factors influencing the annual change rate of soil organic or inorganic carbon density were screened as auxiliary variables for modeling. Correlation analysis was used to analyze the correlation between the annual change rate of soil organic or inorganic carbon and the screened influencing factor data and the initial soil organic or inorganic carbon density in the first sampling period. Variance inflation factors were then used to perform collinearity analysis on auxiliary variables to select the auxiliary variables ultimately used for modeling.

[0015] S23: Constructing a model to predict future spatiotemporal changes in soil carbon

[0016] A geographically weighted regression model was used to establish an empirical relationship between the annual change rate of soil organic or inorganic carbon and various auxiliary variables. The regression coefficients of the geographically weighted regression model on each auxiliary variable were obtained. The soil organic or inorganic carbon density at the sampling point in the second period was used as the initial soil carbon density. The spatial distribution pattern of the annual change rate of soil organic or inorganic carbon density under different temperature change scenarios was predicted year by year.

[0017] S24: Prediction of spatiotemporal changes in soil carbon density in different periods in the future

[0018] The future spatiotemporal variation pattern of soil organic or inorganic carbon density under different temperature change scenarios is predicted year by year; the annual change rate of soil organic or inorganic carbon in the predicted year is added to the soil organic or inorganic carbon density of the previous year to obtain the spatiotemporal variation pattern of soil organic or inorganic carbon density in the predicted year.

[0019] According to some embodiments, in step S22, the screened influencing factor data include parent rock, average annual temperature, average annual rainfall, topography, cultivated land use type, fertilization intensity, soil type, soil particle composition and carbon input.

[0020] According to some embodiments, in step S23, the GWR model is used to estimate the local spatial variation of the study area and detect its spatial non-stationarity. The calculation formula is as follows:

[0021]

[0022] Where Y i is the dependent variable of sample point i, X ik is the observed value of the kth auxiliary variable at the i-th point, (u i ,v i ) is the position coordinate of the i-th point, β0(u i ,v i ) is the intercept, β k (u i ,v i ) is the regression coefficient of the i-th point, ε i is the error term;

[0023] The GWR model was used to establish SOCDc and SICD c The empirical relationship model between the auxiliary variables is used to predict SOCD under different climate scenarios in different periods in the future. c and SICD c The spatiotemporal changes of

[0024] Firstly, a GWR model was constructed with SOCDc and SICDc as dependent variables and auxiliary variables with variance inflation coefficient less than 5 as independent variables. An adaptive Gaussian function was used to fit the regression coefficients of each regression point, and the golden section search was used to select the bandwidth. The modified Akaike information criterion was used as the bandwidth selection criterion.

[0025] Secondly, the MAE, RMSE and adjusted determination coefficient between the predicted and measured values of the sample points were calculated and compared with the prediction results of the OLS model to evaluate the prediction accuracy of the GWR model;

[0026] Finally, keeping other auxiliary variables unchanged, the annual mean temperature was increased by 1.5°C and 2°C, respectively. The regression coefficients of the GWR model at each sample point were extracted to predict the spatial distribution patterns of SOCDc and SICDc under the two warming scenarios.

[0027] According to some embodiments, step S24 includes the following steps:

[0028] Calculate year-by-year SOCD cn , the calculation formula is as follows:

[0029]

[0030] In the above formula, n represents the year; SOCDcn is the annual change rate of soil organic carbon in the nth year; (u i ,v i ) is the position coordinate of the i-th sample point; SOCD(u i ,v i ) cn is the soil organic carbon change rate at sampling point i in year n, in kg·m -2 ·yr -1 β0(u i ,v i ) is the intercept at sample point i; SOCD(u i ,v i ) (n-1) is the soil organic carbon density at sampling point i in the previous year, in kg·m -2 ;X ik is the observed value of the kth auxiliary variable determined after collinearity detection at the i-th sampling point; β1 is the regression coefficient of the constructed GWR model on the initial carbon density; β k (ui ,v i ) is the regression coefficient of the kth auxiliary variable at the i-th sampling point of the constructed GWR model; ε i is the error term;

[0031] Calculate year-by-year SICD cn , the calculation formula is as follows:

[0032]

[0033] In the above formula, n represents the year; SICDcn is the annual change rate of soil inorganic carbon in the nth year; (u i ,v i ) is the position coordinate of the i-th sample point; SICD(u i ,v i ) cn is the change rate of soil inorganic carbon at sampling point i in year n, in kg·m -2 ·yr -1 β0(u i ,v i ) is the intercept at sample point i; SICD(u i ,v i ) (n-1) is the soil inorganic carbon density at sampling point i in the previous year, in kg·m -2 ;X ik is the observed value of the kth auxiliary variable determined after collinearity detection at the i-th sampling point; β1 is the regression coefficient of the constructed GWR model on the initial carbon density; β k (u i ,v i ) is the regression coefficient of the kth auxiliary variable at the i-th sampling point of the constructed GWR model; ε i is the error term;

[0034] Calculate year-by-year SOCD n and SICD n , the calculation formula is:

[0035] SOCD(u i , v i ) n =SOCD(u i , v i ) n-1 +SOCD(u i , v i ) cn ×1 (6)

[0036] SICD(u i , v i ) n =SICD(ui , v i ) n-1 +SICD(u i , v i ) cn ×1 (7)

[0037] In the above formula, (u i ,v i ) is the position coordinate of the i-th sample point; SOCD(u i ,v i ) n and SICD(u i ,v i ) n are the soil organic carbon density and soil inorganic carbon density at sampling point i in year n, respectively, in kg·m -2 ; SOCD(u i ,v i ) cn and SICD(u i ,v i ) cn are the annual change rates of soil organic carbon density and soil inorganic carbon density at sampling point i in year n, respectively, in kg·m -2 ·yr -1 ; SOCD(u i ,v i ) (n-1) and SICD(u i ,v i ) (n-1) are the soil organic carbon density and soil inorganic carbon density at sampling point i in the previous year, respectively, in kg·m -2 .

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] (1) The present invention uses a geographically weighted regression model to express the relationship between the soil C change rate and the various influencing factors as the spatial position changes. In the embodiment, the soil organic carbon density change rate (SOCD c ) and soil inorganic carbon density change rate (SICD c ) of the fitted correction coefficient of determination R 2 adj The values were 0.370 and 0.557, respectively, which were 36.53% and 85.05% higher than those of the least squares (OLS) model, and more accurately captured the differences in the effects of different influencing factors on soil C changes at different spatial locations.

[0040] (2) The present invention introduces the initial density of soil C as a model input variable to predict the soil C change rate and soil carbon density year by year, fully considering the influence of the initial density of soil C. This is more in line with the law that the soil C change rate changes with the change of the initial density of soil C, thereby more accurately predicting the future change trend of soil C. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 The Sichuan Basin location and soil sample distribution map at different times provided by the embodiment of the present invention.

[0042] Figure 2 Correlation coefficients between the annual change rate of soil organic carbon density (a) and the annual change rate of soil inorganic carbon density (b) in cultivated land in the Sichuan Basin and the influencing factors provided in the embodiments of the present invention (Note: *, **, and *** indicate significant levels (P < 0.05), significant levels (P < 0.01), and extremely significant levels (P < 0.001), respectively.

[0043] Figure 3 The relationship between the predicted values and the measured values of the annual change rate of soil organic carbon density under the OLS model (a), the annual change rate of soil organic carbon density under the GWR model (b), the annual change rate of soil inorganic carbon density under the OLS model (c), and the annual change rate of soil inorganic carbon density under the GWR model (d) provided in the embodiments of the present invention.

[0044] Figure 4 The spatial distribution of the standardized regression coefficient for predicting the change rate of soil organic carbon density based on the GWR model provided in the embodiment of the present invention.

[0045] Figure 5 The spatial distribution of the standardized regression coefficient for predicting the change rate of soil inorganic carbon density based on the GWR model provided in the embodiment of the present invention.

[0046] Figure 6a The present invention provides statistical characteristics of the annual change rate of soil organic carbon density under different climate scenarios.

[0047] Figure 6b The present invention provides statistical characteristics of the annual change rate of soil inorganic carbon density under different climate scenarios.

[0048] Figure 7 The spatial distribution of the annual change rate of soil organic carbon density under different climate scenarios provided by the embodiment of the present invention.

[0049] Figure 8 The spatial distribution of the annual change rate of soil inorganic carbon density under different climate scenarios provided by the embodiment of the present invention.

[0050] Figure 9aThe embodiment of the present invention provides the changing trend of the annual change rate of soil organic carbon density under different climate scenarios from 2020 to 2060.

[0051] Figure 9b The embodiment of the present invention provides the changing trend of the annual change rate of soil inorganic carbon density under different climate scenarios from 2020 to 2060.

[0052] Figure 10 The spatial distribution of the annual change rate of soil organic carbon density under different climate scenarios from 2020 to 2060 provided by the embodiment of the present invention.

[0053] Figure 11 The spatial distribution of the annual change rate of soil inorganic carbon density under different climate scenarios from 2020 to 2060 provided by the embodiment of the present invention.

[0054] Figure 12 The changing trends of soil organic carbon density (a), soil inorganic carbon density (b), and soil carbon density (c) under different climate scenarios from 2020 to 2060 provided in the embodiments of the present invention.

[0055] Figure 13 This is the spatial distribution of soil organic carbon density under different warming scenarios from 2020 to 2060 provided by the embodiments of the present invention.

[0056] Figure 14 The spatial distribution of soil inorganic carbon density under different climate scenarios from 2020 to 2060 is provided in an embodiment of the present invention.

[0057] Figure 15 This is the spatial distribution of soil carbon density under different climate scenarios from 2020 to 2060 provided by the embodiments of the present invention.

[0058] Figure 16 A flowchart of a method for predicting future spatiotemporal changes in soil carbon at a regional scale based on a geographically weighted regression model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The present invention is described in detail below with reference to the embodiments and accompanying drawings. However, it should be understood that the embodiments and accompanying drawings are merely exemplary descriptions of the present invention and do not constitute any limitation on the scope of protection of the present invention. All reasonable variations and combinations within the scope of the inventive concept of the present invention fall within the scope of protection of the present invention.

[0060] The present invention will be further described below with reference to the accompanying drawings.

[0061] Example 1

[0062] The Sichuan Basin, with its high proportion of cultivated land and dense population, plays a crucial role in my country's grain production. Previous studies have found that the region's average annual temperature showed a highly significant increase from the 1980s to the 2010s, increasing by an average of 0.36°C per decade. Average annual precipitation decreased slightly, but without a significant trend. Of the 111 meteorological stations within the basin, only two—Yunyang and Neijiang—showed no significant change in average annual temperature; the remaining stations showed significant or highly significant increases. Regarding average annual precipitation, only eight stations in the western and southern parts of the basin showed significant changes, while the remaining 103 stations showed no significant changes. The patterns of change in average annual temperature and precipitation in the Sichuan Basin align with those in my country.

[0063] Therefore, the regional-scale future soil carbon spatiotemporal change prediction method based on the geographically weighted regression model provided in this embodiment takes the Sichuan Basin as the research object and predicts the spatiotemporal change pattern of soil C under future climate change scenarios, which is of relatively typical significance.

[0064] like Figure 16 , the prediction method includes the following steps:

[0065] S1: Collecting Data

[0066] Collect soil data and influencing factors data for the study geographical area;

[0067] Soil data were obtained from the compilation of soil species in the basin's districts and counties during the second national soil survey in the 1980s (1980-1985) and field soil sampling points in the 2010s (2017-2019). In the 1980s, a total of 4,219 cultivated land soil sample points representing the basin's main soil species were obtained ( Figure 1 b); These historical soil sample data record in detail the detailed environmental information and soil physical and chemical indicators of each sampling point, including the location, topographic location, parent material, soil type, cultivated land use type, soil organic matter, and soil pH value. The collection of soil samples in the 2010s began in 2016 and was completed from 2017 to 2019. The sampling was roughly based on the locations of the soil sample points in the 1980s at a ratio of approximately 1:1. The sampling plots were as close as possible to the locations of the sampling points in the 1980s (within the same village or nearby villages and towns), and had the same cultivated land use type and soil type as well as similar topographic locations. Each sampling location was centered on the recorded geographic coordinates, and 4 to 5 soil samples were collected within a radius of 5m. The samples were then fully mixed using the quartering method to form a mixed sample, and the latitude and longitude coordinates and altitude were recorded using a handheld GPS. Taking into account the huge changes in land use patterns over the past 40 years, sampling points were added around some sampling points, and a total of 4,409 soil surface (0-20cm) sample points were collected. Figure 1c) Approximately 1 kg of soil samples were collected from each sampling point. After being brought back to the laboratory to remove impurities and placed in a ventilated place to air dry, they were ground and sieved, and then the soil particle composition, pH value, organic matter, carbonate and other physical and chemical properties were determined using indoor analysis methods that were exactly the same as those used in the 1980s. For some missing soil bulk density and carbonate content values, a random forest model and an artificial neural network model were used to establish a prediction method for the relationship between environmental factors and existing soil properties and soil bulk density and carbonate content to fill in the missing values. To detect outliers, the carbonate content of soil samples at different periods in soil types and cultivated land use types was determined by adding or subtracting 3 times the standard deviation as the mean to determine outliers. The samples marked as outliers were replaced with the maximum or minimum normal value obtained under the two classification methods.

[0068] Due to the long sampling interval and dramatic land use changes, the locations of sampling points from the two periods were not completely consistent. Based on soil genetics theory, this example paired soil samples from the two periods based on the principles of spatial proximity, identical soil type and land use type, and similar topographical location. Specifically, using the 1980s soil sample as the center, spatial analysis in Arc GIS software was used to generate buffer distances for these historical sample points. Soil samples from 2017 to 2019 with similar topographic locations, identical soil types and cropping systems, and the smallest distance to the historical samples were selected as paired soil samples, resulting in a total of 3,697 paired sample points.

[0069] The influencing factor data were selected based on the soil formation process. The soil formation factors and soil properties closely related to soil carbon changes were selected as input variables for constructing the soil carbon future change prediction model. These included parent rock, annual average temperature and rainfall, topographic factors, cultivated land use type, fertilization intensity (nitrogen fertilizer), soil type, soil mechanical composition, and initial soil carbon density in the 1980s. Among them, the annual average temperature and annual average rainfall were calculated based on the daily temperature and rainfall data from 1961 to 1980 at 144 county-level meteorological stations in and around the study area ( Figure 1a) First, annual and seasonal mean temperature and precipitation values were calculated for each meteorological station. These values were then interpolated using ordinary kriging in Arc GIS 10.6 to generate spatial distribution raster data for the annual and seasonal mean temperature and precipitation values for the Sichuan Basin. Terrain factors were calculated using the ALOS 12.5m DEM data, which was obtained from the Resources and Environmental Science and Data Center, Chinese Academy of Sciences (https: / / www.resdc.cn / ). Based on this data, terrain factors such as elevation, slope, aspect, plan curvature, profile curvature, and terrain wetness index were calculated in Arc GIS 10.6. The average annual fertilization intensity and carbon input of each plot were calculated using the fertilizer application amounts of 143,552 plots in the study area (the N, P, and K amounts per mu of land have been converted respectively) and crop yields in different planting seasons, as well as the grass-to-grain ratio, grass-root ratio, grain moisture content, carbon content of different crops obtained from existing studies, and the straw return rate in different districts and counties. Based on the principle of complete consistency between the planting system and soil type, the fertilizer application amount information and carbon input of the plot closest to the soil sampling point were assigned to the soil sampling point.

[0070] S2: Construction of a prediction model for future spatiotemporal changes in soil carbon, including the following steps:

[0071] S21: Calculation of annual change rate of soil carbon density

[0072] Soil samples from the two periods were collected from 1980 to 1985 and from 2017 to 2019, respectively. According to the collected data from the second census of various districts and counties in the Sichuan Basin, the collection of soil samples in the 1980s was mainly carried out from 1980 to 1982, so 1982 was used as the initial year of the study in this embodiment. The collection of soil samples in the 2010s was mainly carried out from 2017 to 2018, so 2018 was used as the final year of this embodiment. Therefore, the two sampling periods in this embodiment are approximately 36 years apart. Based on this, the calculation formula for the annual change rate of soil organic carbon (SOCDc) at each sampling point is:

[0073] SOCD c =(SOCD1-SOCD0)÷36 (1)

[0074] In formula (1): SOCD c is the annual change rate of soil organic carbon density, in kg·m -2 ·yr -1 ; SOCD0 is the soil organic carbon density at the beginning of the study period, which is the soil organic carbon density in 1982 in this embodiment; SOCD1 is the soil organic carbon density at the end of the study period, which is the soil organic carbon density in 2018 in this embodiment; 36 is the time interval between soil sampling in the two periods, in yr.

[0075] Annual change rate of soil inorganic carbon (SICD) at each sampling point c ) is calculated as:

[0076] SICD c =(SICD1-SICD0)÷36 (2)

[0077] In formula (2): SICD c is the annual change rate of soil inorganic carbon density, in kg·m -2 ·yr -1 ; SICD0 is the soil inorganic carbon density in the initial year of the study period, which is the soil inorganic carbon density in 1982 in this embodiment; SICD1 is the soil inorganic carbon density in the final year of the study period, which is the soil inorganic carbon density in 2018 in this embodiment; 36 is the time interval between soil sampling in the two periods, in yr.

[0078] S22: Select auxiliary variables for the prediction model

[0079] Influencing factor data obtained from S1, factors related to the change in SOCD and SICD were selected as the SOCD of the study area. c and SICD c Initial auxiliary variables for the prediction model included climate (mean annual temperature and mean annual precipitation), topography (elevation, slope, aspect, slope length, profile curvature, plan curvature, and terrain moisture index), parent rock, soil type, soil physical properties (sand, silt, and clay), cropland use type, fertilization intensity, carbon input, and initial SOCD (SOCD0: soil organic carbon density in the initial year) and SICD (SICD0: soil inorganic carbon density in the initial year) in the 1980s. To facilitate model calculations, the category means of qualitative categorical variables such as parent rock, soil type, and cropland use type were used instead of category names for model construction. Variance inflation factors (VIFs) were used to analyze multicollinearity among the initial variables. Variables with VIFs greater than 5 were removed, while those with VIFs less than 5 were retained as the final auxiliary variables for the prediction model. This value was set smaller than the commonly used 7.5 to minimize collinearity among the auxiliary variables ultimately selected for modeling.

[0080] S23: Construction of a prediction model for spatiotemporal changes in the annual change rate of soil carbon density

[0081] The Geographically Weighted Regression (GWR) model, an extension of the Ordinary Least Square (OLS) model, is a typical local regression model. By embedding the geographic location information of the sample data into the regression parameters through weights, the GWR model can estimate the local spatial variation of the dependent variable and effectively detect its spatial non-stationarity. The calculation formula of GWR is as follows:

[0082]

[0083] In formula (3): Y i is the dependent variable of sample point i, X ik is the observed value of the kth auxiliary variable at the i-th point, (u i ,v i ) is the position coordinate of the i-th point, β0(u i ,v i ) is the intercept, β k (u i ,v i ) is the regression coefficient of the i-th point, ε i is the error term.

[0084] The GWR model was used to establish SOCD c and SICD c The empirical relationship model between the auxiliary variables is used to predict SOCD under different climate scenarios in different periods in the future. c and SICD c The specific modeling process is as follows:

[0085] (a) A GWR model was constructed with SOCDc and SICDc as dependent variables and preselected auxiliary variables with variance inflation coefficients less than 5 as independent variables. An adaptive Gaussian function was used to fit the regression coefficients of each regression point. The golden section search was used to select the bandwidth, and the Corrected Akaike Information Criterion (AICc) was used as the bandwidth selection criterion.

[0086] (b) Calculate the MAE, RMSE and adjusted R-squared (R) between the predicted and measured values of the sample points. 2 adj ) and compared them with the prediction results of the OLS model to evaluate the prediction accuracy of the GWR model.

[0087] (c) Keeping other auxiliary variables unchanged, the annual mean temperature was increased by 1.5°C and 2°C, respectively. The regression coefficients of the GWR model constructed in (a) at each sample point were extracted to predict the spatial distribution patterns of SOCDc and SICDc under the two warming scenarios.

[0088] S24: Prediction of spatiotemporal changes in soil carbon density in different periods in the future

[0089] Based on the regression coefficients of the constructed GWR model at each sample point, keeping other auxiliary variables unchanged and using 2018 as the base year, the spatiotemporal distribution of SOCDc and SICDc in the study area was predicted year by year under three climate scenarios: unchanged mean annual temperature, a 1.5°C increase in mean annual temperature, and a 2°C increase in mean annual temperature. SOCDc and SICDc were added to the previous year's SOCD and SICD, respectively, to obtain the current year's SOCD and SICD for the study area. The sum of SOCD and SICD yielded the current year's STCD.

[0090] SOCD by year from 2019 to 2060 cn The calculation formula is as follows:

[0091]

[0092] In formula (4), n represents the year; SOCDcn is the annual change rate of soil organic carbon in the nth year; (u i ,v i ) is the position coordinate of the i-th sample point; SOCD(u i ,v i ) cn is the soil organic carbon change rate at sampling point i in year n, in kg·m -2 ·yr -1 β0(u i ,v i ) is the intercept at sample point i; SOCD(u i ,v i ) (n-1) is the soil organic carbon density at sample point i in the previous year (i.e., initial carbon density SOCD0, such as the predicted SOCD in 2019 c2019 The initial carbon density value is the observation value from 2017 to 2019, and the unit is kg·m -2 ;X ik is the observed value of the kth auxiliary variable determined after collinearity detection at the i-th sampling point; β1 is the regression coefficient of the constructed GWR model on the initial carbon density; β k (u i ,v i ) is the regression coefficient of the kth auxiliary variable at the i-th sampling point of the constructed GWR model; ε i is the error term.

[0093] SICD by year from 2019 to 2060 cn The calculation formula is as follows:

[0094]

[0095] In formula (5), n represents the year; SICDcn is the annual change rate of soil inorganic carbon in the nth year; (u i ,v i ) is the position coordinate of the i-th sample point; SICD(u i ,v i ) cn is the change rate of soil inorganic carbon at sampling point i in year n, in kg·m -2 ·yr -1 β0(u i ,v i ) is the intercept at sample point i; SICD(u i ,v i ) (n-1) is the soil inorganic carbon density at sample point i in the previous year (i.e., initial carbon density SICD0, such as the predicted SICD in 2019 c2019 The initial carbon density value is the observation value from 2017 to 2019, and the unit is kg·m -2 ;X ik is the observed value of the kth auxiliary variable determined after collinearity detection at the i-th sampling point; β1 is the regression coefficient of the constructed GWR model on the initial carbon density; β k (u i ,v i ) is the regression coefficient of the kth auxiliary variable at the i-th sampling point of the constructed GWR model; ε i is the error term.

[0096] SOCD by year from 2019 to 2060 n and SICD n The calculation formula is:

[0097] SOCD(u i , v i ) n =SOCD(u i , v i ) n-1 +SOCD(u i , v i ) cn ×1 (6)

[0098] SICD(u i , v i ) n =SICD(ui , v i ) n-1 +SICD(u i , v i ) cn ×1 (7)

[0099] In formulas (6) and (7): (u i ,v i ) is the position coordinate of the i-th sample point; SOCD(u i ,v i ) n and SICD(u i ,v i ) n are the soil organic carbon density and soil inorganic carbon density at sampling point i in year n, respectively, in kg·m -2 ; SOCD(u i ,v i ) cn and SICD(u i ,v i ) cn are the annual change rates of soil organic carbon density and soil inorganic carbon density at sampling point i in year n, respectively, in kg·m -2 ·yr -1 ; SOCD(u i ,v i ) (n-1) and SICD(u i ,v i ) (n-1) are the soil organic carbon density and soil inorganic carbon density at sampling point i in the previous year, respectively, in kg·m -2 .

[0100] The following provides an evaluation of the prediction model parameters and prediction accuracy in the above soil carbon density annual change rate prediction results. The evaluation process is as follows:

[0101] The first step is to select auxiliary variables:

[0102] The results of correlation analysis showed that ( Figure 1 ), annual average temperature, annual average rainfall, elevation, slope, aspect, plan curvature, profile curvature, terrain moisture index, parent rock, soil type, sand, silt, clay, cultivated land use type, fertilization intensity, C input, and SOCD0 and SOCD of the study area c It is related to the SICD in the study area and can be used as its primary auxiliary variable. cThe primary auxiliary variables are annual mean temperature, annual mean precipitation, elevation, slope, aspect, terrain moisture index, parent rock, soil type, sand, silt, clay, cultivated land use type, fertilization intensity, C input and SICD0.

[0103] In order to reduce the collinearity among auxiliary variables, the OLS model was used to test the collinearity of the primary auxiliary variables, and the primary variables with variance inflation coefficient greater than 5 were removed based on the test results. As shown in Table 1, 10 factors were finally selected as the SOCD of the study area. c The auxiliary variables of the prediction model are annual average temperature, annual average rainfall, slope, terrain moisture index, sand grains, soil parent rock, soil type, cultivated land use type, fertilization intensity and SOCD0; while SICD c There are eight auxiliary variables in the prediction model, namely, annual average temperature, annual average rainfall, elevation, slope, sand grains, parent rock, fertilization intensity and SICD0.

[0104] Table 1 Collinearity diagnosis of different auxiliary variables based on OLS model

[0105]

[0106]

[0107] Secondly, the prediction accuracy of the soil carbon density annual change rate prediction model was evaluated.

[0108] The OLS model and GWR model were used to analyze the SOCD of cultivated land in Sichuan Basin in the past 40 years. c and SICD c Make predictions ( Figure 3 , Table 2). The results show that the GWR model has a significant impact on SOCD in the study area. c Prediction R 2 adj is 0.370, which is better than the OLS model (R 2 adj =0.271); MAE and RMSE were reduced by 30.89% and 23.65% respectively compared with the OLS model. c Prediction R 2 adj The GWR model is 0.557, which is an increase of 85.05%; MAE and RMSE are reduced by 32.29% and 26.87% respectively. c and SICD c Both have better prediction effects.

[0109] Table 2 Evaluation of prediction model results

[0110]

[0111] Spatial distribution of regression coefficients of model pre-explanatory variables

[0112] The GWR model results show that the SOCD of cultivated land in the Sichuan Basin c The explanatory variables of Figure 4 ). When the regression coefficient of the explanatory variable is > 0, it means that the variable has an impact on SOCD. c When the regression coefficient of the explanatory variable is less than 0, it means that the variable has a negative effect on SOCDc. Specifically, the parent rocks of the Minjiang River Basin and the Qujiang River Basin have a positive effect on SOCDc. c has a significant negative impact on SOCD, while the parent rocks in the Yangtze River mainstream area have a significant negative impact on SOCD. c The positive impact is the greatest ( Figure 4 a) Effect of cultivated land use type on SOCD c The impact of the basin increases gradually from the southwest and east to the north. Figure 4 b), while the soil types are distributed from low in the north to high in the center ( Figure 4 c) Effect of fertilization intensity on SOCD in the Minjiang River Basin and the Fujiang River Basin c The negative impact is the greatest, but this impact gradually decreases towards the surrounding areas ( Figure 4 d) Sand particles and SOCD0 and SOCD c The area with the smallest sand regression coefficient is around the basin ( Figure 4 e), and the area with the smallest SOCD0 regression coefficient also includes the Chengdu Plain ( Figure 4 g). The regression coefficients of slope and terrain moisture index have similar distribution patterns in space, and generally show a "low-medium-low-high-low" zonal distribution pattern from northeast to southwest ( Figure 4 f and g). Annual mean temperature and SOCD c The negative effect is dominant in the basin, with positive effects only in the northern and southeastern parts. Figure 4 h) The annual average rainfall is opposite to the annual average temperature. The regression coefficient of the annual average rainfall in most areas of the Sichuan Basin is greater than 0, which is consistent with the SOCD c There is a positive effect; only in the northwest of the basin, the annual average rainfall regression coefficient is less than 0 in the Tuojiang River Basin and parts of the Jialing River Basin, that is, c Has a negative effect.

[0113] Sichuan Basin Cultivated Land SICD c The regression coefficients of each explanatory variable are as follows: Figure 5The results show that the spatial non-stationarity of each explanatory variable is different. Among them, the regression coefficient of soil-forming parent rock shows a spatial distribution trend of high in the middle and low around. c Has a positive effect ( Figure 5 a) The regression coefficients of fertilization intensity and sand content are > 0 in most areas of the Sichuan Basin, except for the area near the Chengdu Plain. c Has a negative impact ( Figure 5 a) The regression coefficients of elevation and annual average rainfall have similar distribution patterns in space, and in most areas the regression coefficients of the two are <0, which is consistent with SICD. c Has a negative effect ( Figure 5 d and g). The regression coefficients of slope and annual mean temperature are generally distributed in a patchy pattern with high in the center and low around the edges. However, the annual mean temperature has a significant effect on SICD. c The area with negative impact is wider ( Figure 5 e and f). The regression coefficients of SICD0 are all <0, that is, SICD0 and SICD c The effects were negative, especially in the Qujiang River Basin, Minjiang River Basin and the Yangtze River mainstream area ( Figure 5 h).

[0114] Based on the spatiotemporal change prediction model of cultivated land surface soil carbon in the Sichuan Basin constructed in this example, the SOCD of the study area was predicted under the scenarios of unchanged annual average temperature, increase of 1.5℃ and 2℃ respectively. c and SICD c The final prediction results are the prediction results under different temperature change scenarios. c In general, it shows a downward trend with the increase of annual average temperature ( Figure 6a When the annual average temperature rises by 1.5℃ and 2℃, the SOCD c The mean values decreased by 32.67% and 43.82% respectively. c With the increase of annual average temperature, the Figure 6b When the annual average temperature rises by 1.5℃ and 2℃, SICD in the study area c The mean values increased by 2.70 times and 3.63 times respectively. The analysis shows that the prediction results are consistent with the actual conditions in different regions.

[0115] The spatial distribution characteristics of surface SOCDc in cultivated land in Sichuan Basin were predicted under different climate scenarios ( Figure 7 ). The results show that SOCD in the study area c There is obvious spatial heterogeneity. When the annual mean temperature remains constant, the SOCD in the southwest of the basin c The SOCD in the northeastern part of the basin and around the basin is higher. cLower, even negative ( Figure 7 a). Under the climate warming scenario ( Figure 7 b and c), Sichuan Basin SOCD c The overall trend is decreasing, SOCD c The number of negative areas is also increasing, especially in the northeastern part of the basin. This suggests that climate warming has accelerated the decomposition rate of surface SOC in cultivated land in the Sichuan Basin, and some areas may experience SOCD losses due to warming.

[0116] Figure 8 Shows the surface SICD of cultivated land in Sichuan Basin c Spatial distribution characteristics under different climate scenarios. Figure 8 a It can be seen that the SICD of the study area at the current temperature c The SICD in the central and western parts of the basin (near the Chengdu Plain) is relatively low, with significant spatial differences. c Higher, while SICD in the northeast and south of the basin c Lower ( Figure 8 a). Under the climate warming scenario ( Figure 8 b and c), Sichuan Basin SICD c The spatial distribution pattern of SICD in the central basin has further changed with the increase of annual average temperature. c The positive area of ​​the SIC gradually expanded, and the SIC sequestration capacity increased; however, the SICD in the northeastern and southwestern parts of the basin c The risk of SIC loss has further increased.

[0117] Spatiotemporal simulation of the annual change rate of soil carbon density from 2020 to 2060

[0118] Prediction of surface SOCD of cultivated land in the Sichuan Basin from 2020 to 2060 under different climate scenarios c and SICD c The spatial and temporal variation trend of Figure 9a It can be seen that at the current temperature, SOCD in the study area c Slightly decreased but not much changed; with the increase of annual mean temperature, SOCD c Significantly decreased. c It is slowly increasing at current temperatures but still negative; as the annual mean temperature rises, SICD c From negative to positive ( Figure 9b This indicates that climate warming has a significant impact on the changes in SOCD and SICD of cultivated land surface in the Sichuan Basin, and there are differences. As the annual average temperature rises, the SOC sequestration capacity in the study area decreases significantly, while the SIC sequestration capacity further increases.

[0119] Figure 10The surface SOCD of cultivated land in the Sichuan Basin from 2020 to 2060 under different climate scenarios is shown. c When the annual mean temperature remains constant ( Figure 10 ae), SOCD of the study area c The values are basically positive, indicating that SOCD is generally in an accumulative state. As time goes by (2020-2060), SOCD in the study area c This indicates that under the condition of constant annual average temperature, the cultivated land in the Sichuan Basin still has the capacity to sequester SOC.

[0120] When the annual average temperature rises by 1.5℃ ( Figure 10 fj), SOCD of the study area c The average value is significantly lower than that before the warming. Among them, SOCD in the northeast, central and southwestern parts of the basin c The SOCD in the western and southern parts of the basin c This indicates that under an annual mean temperature increase of 1.5°C, the SOC sequestration capacity of cultivated land in the Sichuan Basin will decrease and there will be obvious spatial differences.

[0121] When the annual average temperature rises by 2℃ ( Figure 10 ko), SOCD of the study area c The mean value continues to decrease when the annual mean temperature remains unchanged and the annual mean temperature increases by 1.5℃. Spatially, the northeastern, central and southwestern parts of the basin are SOCD. c The area of the decline region has increased, indicating that the range of SOC loss has expanded; while SOCD in the western and southern parts of the basin c It is still positive and still has a certain SOC storage capacity.

[0122] SICD of cultivated land surface in the Sichuan Basin from 2020 to 2060 under different climate scenarios c The spatial distribution trend of SICD in the study area is shown in Figure 6-11. c Maintaining a low level, SICD c The average value is generally between -0.01 and 0.01 kg·m -2 ·yr -1 and will not change much over time (2020-2060). Spatially, the SICD in the northwest and central parts of the basin c The SICD in the southwest and east of the basin is relatively high. c Relatively low.

[0123] When the annual average temperature rises by 1.5℃ (Figure 6-11f-j), the SICD of the study area cThe mean value increased significantly compared to the case when the annual mean temperature remained unchanged, turning from negative to positive, and the spatial distribution was significantly different. c The SICD in the north and southwest of the basin has increased significantly. c Compared with the period before warming, the SICD in the study area will be significantly reduced from 2020 to 2060. c The average value showed an overall downward trend, which indicates that there are significant differences in the SIC sequestration capacity of cultivated land in the Sichuan Basin under different geographical conditions.

[0124] When the annual average temperature rises by 2℃ (Figure 6-11k-o), the SICD of the study area in each period c The mean is significantly higher than when the annual mean temperature remains unchanged and when the annual mean temperature rises by 1.5℃. c The mean value continues to decline. c The spatial distribution pattern of SIC is similar to that observed when the annual mean temperature rises by 1.5°C, but exhibits more pronounced spatial differences. This suggests that warming only enhances the SIC sequestration capacity of cultivated soils in the central region of the basin, while further decreasing in the northern and southwestern regions.

[0125] Spatiotemporal distribution pattern of soil carbon density from 2020 to 2060

[0126] Figure 6-12 shows the changing trends of surface SOCD, SICD, and STCD in cultivated land in the Sichuan Basin from 2020 to 2060 under different climate scenarios. SOCD in the study area shows an increasing trend under all climate change scenarios (Figure 6-12a). With increasing annual average temperature, SOCD accumulation in the study area decreases. This indicates that current temperatures are more conducive to SOCD accumulation in the study area.

[0127] Figure 6-12b shows that, under a constant annual mean temperature, SICD in the study area decreases slightly from 2020 to 2060. However, as the annual mean temperature rises, SICD in the study area gradually increases. This suggests that rising temperatures can enhance the SIC sequestration capacity of cultivated land in the study area.

[0128] Combining changes in SOC and SIC reveals that STCD in the study area shows an increasing trend under all climate scenarios (Figure 6-12c), indicating that cultivated soils in the Sichuan Basin will continue to function as a carbon sink in the future. Under the scenario of constant annual mean temperature, although SIC decreases annually, the amount lost is far less than the increase in SOC. Under the warming scenario, however, while SIC shifts from loss to sequestration, the increase in SOC decreases significantly, resulting in a lower cumulative amount of STCD under the warming scenario than under the scenario of constant annual mean temperature. This difference widens over time.

[0129] Figure 6-13 shows the spatial distribution trends of surface SOCD in cultivated land in the Sichuan Basin from 2020 to 2060 under different climate scenarios. When the annual mean temperature remains constant (Figures 6-13a-e), SOCD in the study area generally shows a gradual increase, with mean SOCD values increasing by 4.02–13.75%, consistent with statistical results. Spatially, SOCD is relatively high in the northwest and southwest of the study area, while it is relatively low in the central and southeastern parts of the basin.

[0130] When the annual mean temperature rose by 1.5°C (Figures 6-13f-j), the mean SOCD values in the study area decreased compared to the pre-temperature period, decreasing by 0.46–6.77%, primarily in the southwestern, central, and northeastern parts of the basin. The mean SOCD values also showed an overall upward trend from 2020 to 2060 (1.89–6.55%), but the rate of increase was lower than when the annual mean temperature remained constant. This suggests that the SOC sequestration capacity of cultivated land in the Sichuan Basin would decrease slightly under a 1.5°C annual mean temperature increase.

[0131] When the annual mean temperature rises by 2°C (Figure 6-13k-o), SOCD in the study area continues to increase from 2020 to 2060, but the rate of increase decreases significantly (1.17-4.13%). Compared with the scenarios of unchanged annual mean temperature and a 1.5°C annual mean temperature increase, the mean SOCD values in the study area decrease by 0.61-9.02% and 0.15-2.42%, respectively. Spatially, the area of SOCD decline expands significantly in the southwestern, central, and northeastern parts of the basin. This suggests that with a 2°C annual mean temperature increase, the sequestration capacity of parts of the Sichuan Basin's SIC will continue to decline, and some areas may even experience SOC losses.

[0132] Figure 6-13 shows the spatial distribution trends of surface SICD in cultivated land in the Sichuan Basin under different climate scenarios from 2020 to 2060. Under current temperatures (Figures 6-14a-e), SICD in the study area has decreased slightly overall (0.79-2.16%), consistent with statistical results. Spatially, SICD in the study area exhibits a high center and low periphery distribution. This suggests that under constant annual mean temperature, the SIC sequestration capacity of cultivated land in the Sichuan Basin remains unchanged.

[0133] When the temperature rose by 1.5°C (Figures 6-14f-j), the mean SICD values in the study area increased (by 2.12% to 39.46%) compared to pre-warming levels. From 2020 to 2060, the SICD values in the study area showed an overall upward trend, with mean SICD values increasing by 9.11% to 33.60%. Spatially, the SICD distribution in the study area remained high in the center and low in the surrounding areas, but the high-value areas in the center and the low-value areas in the southwest and northeast of the basin expanded significantly. This indicates that under a 1.5°C annual mean temperature increase, the SIC sequestration capacity of cultivated land in the Sichuan Basin increased, with significant spatial variations.

[0134] When the temperature rises by 2°C (Figure 6-14k-o), the spatial distribution pattern of SICD in the study area does not change significantly.

[0135] From 2020 to 2060, SICD in the study area continued to increase, with the rate of increase further accelerating (12.67-47.17%), consistent with statistical results. Compared to the scenarios with a constant annual mean temperature and a 1.5°C increase in annual mean temperature, the mean SICD values in the study area increased by 2.83-54.69% and 0.69-10.92%, respectively. This suggests that under a 2°C increase in annual mean temperature, the overall capacity of cultivated land in the Sichuan Basin to sequester SIC will further increase. However, in areas with high annual mean precipitation (such as the northeastern part of the basin), the risk of SIC loss is further increased due to factors such as leaching.

[0136] Figure 6-15 shows the spatial distribution trends of surface soil carbon decomposition (STCD) in cultivated land in the Sichuan Basin from 2020 to 2060 under different climate scenarios. When the annual mean temperature remains constant (Figures 6-15a-e), STCD in the study area shows an overall steady increase over time (2020-2060), with mean STCD values increasing by 3.28-11.33%, consistent with statistical results. Spatially, STCD is relatively high in the central basin, while relatively low in the northeastern and southwestern regions. This suggests that, under a constant annual mean temperature scenario, cultivated land in the Sichuan Basin has a strong carbon sequestration capacity.

[0137] When the annual mean temperature rose by 1.5°C (Figures 6-15f-j), the mean STCD value in the study area decreased slightly (0.06-0.52%) compared to when the annual mean temperature remained unchanged. From 2020 to 2060, STCD in the study area showed an overall increasing trend (3.02-10.79%). Spatially, the STCD distribution remained similar to that before the warming, but the low-value areas in the central, southwestern, and northeastern parts of the basin expanded significantly. This suggests that under a 1.5°C annual mean temperature increase, the increase in STCD would only partially offset the loss in SOCD, and the carbon sequestration capacity of cultivated soils in the Sichuan Basin would decrease.

[0138] When the annual mean temperature rose by 2°C (Figure 6-15k-o), the mean STCD values in the study area decreased slightly (0.08-0.44%) compared to when the annual mean temperature remained unchanged. Over time (2020-2060), STCD in the study area continued to increase, but the rate of increase slowed (2.99-10.94%), consistent with the statistical results. Spatially, the spatial distribution of STCD remained similar to that observed for a 1.5°C warming, but the range of the STCD extremes expanded further. This suggests that under a 2°C annual mean temperature increase, STCD in the central, northeastern, and southwestern Sichuan Basin would primarily be a loss, potentially leading to a "carbon source" effect.

[0139] The above embodiments are merely preferred embodiments of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions that fall within the scope of protection of the present invention are within the scope of protection of the present invention. It should be noted that improvements and modifications that can be made by a person skilled in the art without departing from the principles of the present invention are also considered to be within the scope of protection of the present invention.

Claims

1. A method for predicting future spatiotemporal changes in soil carbon at a regional scale based on a geographically weighted regression model, characterized by: The specific steps include: S1: Collecting Data Collect soil data and influencing factors data for the study geographical area; S2: Construct a prediction model for the future spatiotemporal changes of soil carbon and predict the spatiotemporal changes of soil carbon density in different periods in the future, including the following steps: S21: Calculation of annual change rate of soil carbon density The annual change rate of soil carbon density was calculated by the soil organic or inorganic carbon density of paired sampling points in two historical periods; the annual change rate of soil organic or inorganic carbon density of all paired sampling points was calculated according to the sampling period interval between the two historical periods, which were the annual change rate of soil organic carbon density SOCD and SOCD. c and annual change rate of soil inorganic carbon density SICD c ; S22: Select auxiliary variables for the prediction model Data on factors influencing the annual change rate of soil organic or inorganic carbon density were screened as auxiliary variables for modeling. Correlation analysis was used to analyze the correlation between the annual change rate of soil organic or inorganic carbon and the screened influencing factor data and the initial soil organic or inorganic carbon density in the first sampling period. Variance inflation factors were then used to perform collinearity analysis on auxiliary variables to select the auxiliary variables ultimately used for modeling. S23: Constructing a model to predict future spatiotemporal changes in soil carbon A geographically weighted regression model was used to establish an empirical relationship between the annual change rate of soil organic or inorganic carbon and various auxiliary variables. The regression coefficients of the geographically weighted regression model on each auxiliary variable were obtained. The soil organic or inorganic carbon density at the sampling point in the second period was used as the initial soil carbon density. The spatial distribution pattern of the annual change rate of soil organic or inorganic carbon density under different temperature change scenarios was predicted year by year. S24: Prediction of spatiotemporal changes in soil carbon density in different periods in the future The future spatiotemporal variation pattern of soil organic or inorganic carbon density under different temperature change scenarios is predicted year by year; the annual change rate of soil organic or inorganic carbon in the predicted year is added to the soil organic or inorganic carbon density of the previous year to obtain the spatiotemporal variation pattern of soil organic or inorganic carbon density in the predicted year.

2. The method for predicting future spatiotemporal changes in soil carbon at a regional scale based on a geographically weighted regression model according to claim 1, characterized in that: In step S22, the screened influencing factor data include soil-forming parent rock, annual average temperature, annual average rainfall, topography, cultivated land use type, fertilization intensity, soil type, soil particle composition and carbon input.

3. The method for predicting future spatiotemporal changes in soil carbon at a regional scale based on a geographically weighted regression model according to claim 1, characterized in that: In step S23, the GWR model is used to estimate the local spatial variation of the study area and detect its spatial non-stationarity. The calculation formula is as follows: Where Y i is the dependent variable of sample point i, X ik is the observed value of the kth auxiliary variable at the i-th point, (u i ,v i ) is the position coordinate of the i-th point, β0(u i ,v i ) is the intercept, β k (u i ,v i ) is the regression coefficient of the i-th point, ε i is the error term; The GWR model was used to establish SOCD c and SICD c The empirical relationship model between the auxiliary variables is used to predict SOCD under different climate scenarios in different periods in the future. c and SICD c The spatiotemporal changes of First, with SOCD c and SICD c The GWR model was constructed with the primary selected auxiliary variables with variance inflation coefficient less than 5 as independent variables. The adaptive Gaussian function was used to fit the regression coefficients of each regression point, the golden section search was used to select the bandwidth, and the modified Akaike information criterion was used as the bandwidth selection criterion. Secondly, the MAE, RMSE and adjusted determination coefficient between the predicted and measured values of the sample points were calculated and compared with the prediction results of the OLS model to evaluate the prediction accuracy of the GWR model; Finally, keeping other auxiliary variables unchanged, the annual mean temperature was increased by 1.5℃ and 2℃ respectively, and the regression coefficients of the GWR model at each sample point were extracted to predict SOCD under the two warming scenarios. c and SICD c spatial distribution pattern.

4. The method for predicting future spatiotemporal changes in soil carbon at a regional scale based on a geographically weighted regression model according to claim 3, characterized in that: Step S24 includes the following steps: Calculate yearly SOCD cn , the calculation formula is as follows: In the above formula, n represents the year; SOCD cn is the annual change rate of soil organic carbon in year n; (u i ,v i ) is the position coordinate of the i-th sample point; SOCD(u i ,v i ) cn is the soil organic carbon change rate at sampling point i in year n, in kg·m -2 ·yr -1 ; β0(u i ,v i ) is the intercept at sample point i; SOCD(u i ,v i ) (n-1) is the soil organic carbon density at sampling point i in the previous year, in kg·m -2 ;X ik is the observed value of the kth auxiliary variable determined after collinearity detection at the i-th sampling point; β1 is the regression coefficient of the constructed GWR model on the initial carbon density; β k (u i ,v i ) is the regression coefficient of the kth auxiliary variable at the i-th sampling point of the constructed GWR model; ε i is the error term; Calculate year-by-year SICD cn , the calculation formula is as follows: In the above formula, n represents the year; SICD cn is the annual change rate of soil inorganic carbon in the nth year; (u i ,v i ) is the position coordinate of the i-th sample point; SICD(u i ,v i ) cn is the change rate of soil inorganic carbon at sampling point i in year n, in kg·m -2 ·yr -1 ; β0(u i ,v i ) is the intercept at sample point i; SICD(u i ,v i ) (n-1) is the soil inorganic carbon density at sampling point i in the previous year, in kg·m -2 ;X ik is the observed value of the kth input auxiliary variable determined after collinearity detection at the i-th sampling point; β1 is the regression coefficient of the constructed GWR model on the initial carbon density; β k (u i ,v i ) is the regression coefficient of the kth auxiliary variable at the i-th sampling point of the constructed GWR model; ε i is the error term; Calculate yearly SOCD n and SICD n , the calculation formula is: soCD(u i ,v i ) n =SOCD(u i ,v i ) n-1 +SOCD(u i ,v i ) cn ×1 (6) SICD(u i ,v i ) n =SICD(u i ,v i ) n-1 +SICD(u i ,v i ) cn ×1 (7) In the above formula, (u i ,v i ) is the position coordinate of the i-th sample point; SOCD(u i ,v i ) n and SICD(u i ,v i ) n are the soil organic carbon density and soil inorganic carbon density at sampling point i in year n, respectively, in kg·m -2 ; SOCD(u i ,v i ) cn and SICD(u i ,v i ) cn are the annual change rates of soil organic carbon density and soil inorganic carbon density at sampling point i in year n, respectively, in kg·m -2 ·yr -1 ; SOCD(u i ,v i ) (n-1) and SICD(u i ,v i ) (n-1) are the soil organic carbon density and soil inorganic carbon density at sampling point i in the previous year, respectively, in kg·m -2 .

Images