Analysis Of Vegetation Cover-environment Change-response Relationship Based On Spatio-temporal Bayesian Models

To explore the spatial and temporal changes of the environment in the Beijing-Tianjin Sandstorm Source Control(BTSSC)program areasince the implementation of the Beijing-Tianjin Sandstorm Source Control Project,this article considers spatio-temporal factors throug the analysis of the paper based on spatio-temporal series data with a time resolution of monthly from 2001 to 2017 in BTSSC.The spatio-temporal kriging was used to interpolate meteorological data,and the nonparametric dynamic trend model was used to analyze trends of the vegetation and environmental changes indicators in BTSSC.And then through the Stochastic Partial Differential Equation model(SPDE),the model combines Laplace approximation(INLA)algorithm to analyze the response relationship between vegetation and selected indicators.Spatio-temporal prediction of NDVI and GPP were conducted in the study area at last.The main conclusions are summarized as follows:(1)For having taken the spatio-temporal correlation into account,the spatio-temporal kriging gets higher interpolation accuracy for stationary data.Among them,the fitting of the spatiotemporal model with the sum-metric covariance model gets better outcome.The nonparametric spatio-temporal trend model with Type II interactive mode model has the ability of performing different time scale due to the inclusion of the temporal structure.The presented trend outcomes can accurately reflect the period,maximum and minimum value of the research object,which is consistent with the actual phenomenon.And the value of the effect value shows that the trends in the study area aremainly of spatio-temporal interaction trend.(2)In terms of response relationship,for NDVI,net photosynthesis(PSN)and temperature have a greater positive impact on it.For each additional standard deviation,NDVI increases by 0.480 and 0.382 standard deviations respectively;nighttime land surface temperature,daytime and nighttime land surface temperature difference and 0-10 cm soil Moisture content has a greater negative impact on it,and for each additional standard deviation,it decreases by 0.077,0.074,and 0.065 standard deviations respectively.For GPP,PSN,temperature,and actual evapotranspiration(ET)have a greater positive impact on it.With each additional standard deviation,GPP added with 0.692,0.165 and 0.143 standard deviations respectively.Land surface temperature difference and potential evapotranspiration(PET)have a greater negative impact on it than other indicators.For each additional standard deviation,GPP decreased by 0.073 and 0.033 standard deviations.(3)For land cover types,compared to needleleaved deciduous forest,NDVI of broadleaved deciduous forest and needleleaved evergreen forest increase by 0.05 and 0.03,respectively,with the remaining land cover types of NDVI decrease while GPP did not show significant response relationship with those types.The NDVI values of humid areas are higher than other areasand subhumid arid area3 climate type is significantly responsed with GPP with a reduce of about 21.4kg C/m2 compared with the humid area Neither NDVI nor GPP show significant response relationship with 10-40 cm soil water content.Both of Luvisols and Greyzems are more suitable for vegetation growth,while Solonchaks,Solonetz and Arenosols are not conducive to vegetation growth.The positive effects of Greyzems,Chernozems and Luvisols on NDVI and GPP exceeds precipitation.The phenomenon of wind speed is more serious than the natural resistance of soil disadvantage for increasing vegetation coverage accelerates vegetation degradation in the area.(4)In terms of spatio-temporal modelling,incorporating the spatio-temporal effect and soil unit's type can greatly improve the accuracy of the modelthat consideringmeteorological indicators.Under this circumstance,the impact of precipitation on NDVI and GPP decrease by about 70%,the wind speed decrease by 62% and 15%,the effect of altitude increase by 15 times and 6 times,and the effect of temperature on NDVI and GPP increase by about 30%.It implies that,to a greater extent,altitude affects NDVI through spatio-temporal effects;the effects of altitude and temperature on NDVI and GPP are partially reflected by the differences in soil units' types,and precipitation and wind speed affect NDVI and GPP through space-time interaction effects mostly.(5)Accuracy test shows that the R-square of spatio–temporal Bayesian modeling NDVI and GPPare 0.82 and 0.91 respectively,which is better than 0.62 and 0.67 of the fixed-effect model,respectively.It has the ability of capturing the space-time interaction effects that are usually overlooked,and also the ability of presenting a probability map of exceeding the threshold which is sensitive to model sets.Only 3 kinds of easy-to-obtain datacontaining meteorological data,elevation data and soil type data is needed for achieving higher accuracy under the space-time Bayesian model,and it provides an effective wayfor predicting future vegetation cover changes.