A Study Of Forest Carbon Stock Distribution Based On Spatio-temporal Geographically Weighted Regression

Forest carbon stock plays a crucial role in the integrity and stability of the earth’s ecosystem,so it is important to grasp the spatial and temporal distribution patterns of forest standing carbon stock to maintain ecological balance and dynamic monitoring of forest natural resources,and to provide technical and data support for promoting ecological construction and formulating different afforestation policies and effectively implementing forest management strategies.With the development of technology,the wide application of remote sensing images provides new opportunities and ideas for the study of the spatial and temporal distribution of forest carbon stocks.In this study,remote sensing image data,digital elevation model data and forest type II survey data were collected,and relevant remote sensing factors,stand factors and topographic factors were extracted,and these factors were filtered with the calculated forest standing carbon stock by stepwise regression analysis to obtain significant factors.Global models including least squares(OLS)and linear mixed models(LMM),and local models including geographically weighted models(GWR),multi-scale geographically weighted regression models(MGWR),time-weighted models(TWR),and spatio-temporal geographically weighted models(GTWR)were used to investigate the relationship between forest stand carbon stock and remote sensing factors,stand factors,and topographic factors in Liangshui Nature Reserve in 1989,1999,2009,and 2019.The relationships between distribution and remote sensing factors,stand factors and topographic factors were investigated.The prediction accuracy of global and local models was compared;the inverse distance interpolation method was used to analyze the visibility of the prediction results of each model;the spatial distribution of forest carbon stock estimated by the GTWR model was synergistically analyzed with the topography of the study area to explore the spatial distribution pattern of forest carbon stock.The results showed that:(1)The traditional statistical model OLS is the worst fit,with the model having the largest residual sum of squares(RSS)and root mean square error(RMSE)among all models and the smallest R~2(0.464).the LMM,also as a global model adding random effects to OLS,has an improved R~2of 14%over OLS.The local models based on geographically weighted regression,namely GWR,MGWR,TWR,and GTWR,all have good performance,with R~2improving by23%,27%,39%,and 58%,respectively,compared to OLS,with GTWR being the best fit.(2)By comparing the Moran index and Z scores of each model residual at different scales,it is found that the OLS model has obvious spatial autocorrelation,which leads to biased estimation tests of the model coefficients.LMM solves the problem of spatial autocorrelation to some extent,but the model residuals at small scales are still spatially correlated,and the local models(GWR,MGWR,TWR,GTWR)can obviously overcome the influence of this spatial autocorrelation on the model,and MGWR solves the influence of spatial autocorrelation to the greatest extent,and performs the best without considering the time dimension.(3)The topography of the study area was mapped using digital elevation model data,and the spatial distribution pattern of forest carbon stock at different topography was analyzed,and it was found that the distribution of forest carbon stock basically followed the rule of:sunny slope>semi-sunny slope>semi-shady slope>shady slope.The main reason is that the differences in sunshine duration and sunshine intensity lead to different amounts of solar radiation absorbed by photosynthesis of forest vegetation.The solar radiation received by sunny slopes is higher than that of shady slopes,so the carbon storage capacity is naturally stronger than that of shady slopes.(4)The overall spatial distribution trends of forest carbon stocks in the study area in longitude and latitude directions were statistically analyzed,and hotspot analysis and error analysis were also done on the predicted data from the GTWR model,and it was found that the forest carbon stocks in the northern part of the study area were significantly higher than those in other areas.