| The goal and policy implementation of "carbon peaking and carbon neutralization" reflect China’s responsibility to actively respond to climate change and promote the construction of a community with a shared future for mankind.It is also an inevitable requirement for China to promote high-quality development.The "double carbon" goal is closely related to the construction and improvement of terrestrial ecosystem.As an important component of terrestrial ecosystem,forest ecosystem plays a key role in the carbon cycle,water cycle and radiant energy exchange of regional and global terrestrial ecosystem.It also plays an important role in absorbing carbon dioxide in the atmosphere and mitigating global warming.Forest biomass is one of the most basic characteristics of forest ecosystem.It is not only an important factor in the research of environment and climate model,but also an important parameter to evaluate forest productivity and carbon sink.It can reflect the relationship between material circulation and energy flow between forest and environment,also reflect the management level and utilization value of forest.Accurate estimation of forest biomass is an important method to study the earth’s carbon cycle and global climate change.It is of great significance to further study the uncertainty of carbon cycle in terrestrial ecosystem,get the change pattern of forest biomass and productivity,reduce greenhouse effect,promote sustainable development and achieve the "double carbon" goal.Guizhou is a province with the most widely developed karst landform in China,and large area and wide distribution of carbonate rocks.The vegetation growth environment there is relatively poor.Rocky desertification in karst areas is one of the main environmental factors affecting the functions of forest ecosystem,including biomass.Studying the remote sensing inversion modeling optimization method of forest biomass in karst area and analyzing the temporal and spatial changes and driving factors of karst forest biomass can provide basic data for forest carbon sink evaluation.It can also provide scientific reference for regional ecological planning and the formulation of environmental governance measures,which is of great theoretical and practical significance.Taking Guizhou Province as the study area,the National Forest Resources Continuous Inventory(NFCI)data of Guizhou Province in 1985,1990,1995,2000,2005,2010 and 2015 was taken as the ground measured data in this paper.Using Landsat,MODIS and Sentinel-1 remote sensing image of the corresponding years,the remote sensing inversion of forest aboveground biomass(AGB)and the remote sensing classification of rocky desertification were carried out,and the temporal and spatial changes and driving factors of forest biomass were quantitatively analyzed.The research results are as follows:(1)Based on the NFCI data and the MODIS remote sensing image,the bedrock exposure rate,vegetation coverage,surface temperature difference and other characteristic factors were extracted.Logistic regression(Lo R)model,random forest(RF)model and supported vector machine(SVM)model were constructed respectively.An optimization algorithm based on SVM was proposed by integrating the characteristics of vegetation types and seasonal phase differences,and five remote sensing classification maps of rocky desertification in 2001,2005,2010,2015 and 2020 were produced.The overall accuracy of Lo R,RF,SVM were 73.7 %,78.2 % and 80.6 % respectively and the kappa coefficients were 0.608,0.672 and 0.707 respectively.The vegetation types and seasonal differences of vegetation was related to rocky desertification strongly.The overall accuracy of the optimized remote sensing classification model of rocky desertification was 91.1 %and the kappa coefficient was 0.861.From 2001 to 2020,the spatial distribution of rocky desertification in Guizhou showed a pattern of "heavy in the west,light in the east,heavy in the south and light in the north".During this period,rocky desertification has been continuously ameliorated.(2)Based on the NFCI data,the biomass of sample plots was calculated by biomass expansion factor(BEF)method.Univariate linear regression equation,center of gravity migration,hot spot detection and geographic detector methods were used to analyze the temporal and spatial pattern and driving factors of sample plot biomass in Guizhou Province from 1985 to 2015.The unit biomass of sample plots in Guizhou showed an upward trend,increasing from 9.87 Mg/ha to28.47 Mg/ha from 1985 to 2015,with an average annual growth rate of 6.28%.The unit biomass of arbor forest land decreased from 41.91 Mg/ha to 35.95 Mg/ha from 1985 to 1990,and then gradually increased to 55.46 Mg/ha in 2015.From 1985 to 2015,the biomass of 29.3 % sample plots in Guizhou increased significantly,25.8 % sample plots increased slightly,34.4 % sample plots remained stable,9.1 % sample plots decreased slightly and 1.4 % sample plots decreased significantly.From 1985 to 2015,the center of gravity of biomass in Guizhou sample plot shifted from Qiandongnan to Qiannan,with a total distance of 54.1 km.The biomass hotspots in the sample plot increased from 195 in 1985 to 239 in 2015.The biomass of the sample plot is greatly affected by topography and climate.(3)Based on the NFCI data,the band value,vegetation index and texture features were extracted from Landsat optical and Sentinel-1 radar satellite remote sensing images.The forest canopy density(FCD)was introduced to optimize the deep belief networks(DBN),and the proposed K-DBN model was used to carry out the remote densing inversion modeling of forest biomass in Guizhou,and eight periods of forest biomass inversion maps in Guizhou from 1985 to2020 were produced.The FCD in the study area was divided into three levels: sparse,medium and dense by K-means cluster analysis.FCD was helpful to improve the accuracy of forest biomass remote sensing inversion model.Based on setting the broad-leaved forest(BLF),coniferous forest(CFF)and mixed forest(MXF),K-DBN model achieved better model accuracy than linear regression(LR),random forest(RF)and DBN model.(4)Based on remote sensing inversion mapping of forest biomass and remote sensing classification mapping of rocky desertification,combined with terrain,climate and human factors,the spatio-temporal variation and multi-scale driving analysis of forest biomass in Guizhou were carried out.In this paper,the forward stepwise regression method based on ordinary least squares(OLS)was used to select the driving factors,and gravity-multiscale geographical temporally weighted regression(G-MGTWR)model was proposed to analyze the county scale spatiotemporal driving force of forest biomass in Guizhou.The grid scale spatial driving force analysis was carried out based on geographical weighted regression(GWR)model.From 1985 to 1990,the forest biomass in Guizhou showed a downward trend,and from 1990 to 2020,it showed an increasing trend.Rocky desertification,elevation,sunshine and temperature were the main driving factors of forest biomass change in Guizhou on the county spatial scale.On the spatial-temporal county scale,rocky desertification,population,sunshine and precipitation were the main driving factors.Rocky desertification,elevation,sunshine and temperature were the main driving factors at the grid spatial scale.The driving factors worked powerfully in the southeast of the study area,while the driving factors were relatively weak in the west. |
PDF Full Text Request