Extraction Of Remote Sensing Information And Spatiotemporal Simulation Of NPP Of Bamboo Forest In China

Bamboo forests play an important role in carbon cycle and carbon sequestration in subtropical forest ecosystems.As of the eighth(2009-2013)forest resource inventory,The area of bamboo forest in China has reached 6.01 million hectares,accounting for 3% of theforestarea of our country and 20% of the bamboo forest area of the world,which is a truly ?bamboo kingdom?.At present,the bamboo forest area information is mainly obtained by the combination of samples of bamboo stands,remote sensing images anddifferent classification methods.The research area is mostly concentrated on the scales of country or province and lacks of the rapid and accurate extraction of the temporal-spatial distribution information of bamboo forests in the whole country.The carbon cycle of forest ecosystems is of great significance in improving global climate.Among them,net primary productivity(NPP)is indicator parameter in the process of carbon cycle in forest ecosystems.However,there are few researches on bamboo forest NPP in China and abroad.Based on the extraction of remotelysensedinformationfrom forest land,the studyfirstconstructed C5.0 algorithm decision tree to obtain initialbamboo forest information.Secondly,the bamboo forest abundance information was obtained by the least squares mixed pixel decomposition.Finally,the NPP of the national forest ecosystem was simulated by using bamboo forest abundance as an input parameter to drive the Boreal Ecosystem Productivity Simulator(BEPS)model.The main study contents include as follows:1.Maximum likelihood forestland information extraction.Comprehensive MODIS remote sensing,ground survey data and other basic data,combined with Google earth,the maximum likelihood method was usedto divide the country's land use types into water bodies,forests,farmland,towns,bare land,and through the mask to obtain the spatial distribution of national woodland.2.The spatial distribution information of bamboo forest was obtain by the C5.0 decision tree and the mixed pixel decomposition method.Firstly,bamboo forest,broad-leaved forest,and coniferous forest stand samples were obtained by forestland MODIS products,Landsat8 OLI,Zhejiang land use thematic maps and bamboo forest sample data.Secondly,the bamboo forest information extraction model was constructed byusing the C5.0 algorithm decision tree.Finally,the abundance of bamboo forest was extracted by mixed pixel decomposition,and the spatial distribution of bamboo forest was delineated.3.Estimation oftemporal-spatial pattern of NPP in bamboo forest ecosystem.NPP of bamboo forest ecosystem in China was simulated by using bamboo forest abundance information as aninput parameter to drive the BEPS model.The statistical method was used to analyze the temporal-spatial dynamic changes of NPP of bamboo forest ecosystem in China.There are several conclusions from this studyas follows:1.Based on the maximum likelihood method,the extraction accuracy distribution information offorestland information was high.The producer‘s and user‘s accuracy of the three phases,in 2003,2008 and 2014,were all above 90%,and the Kappa coefficient was 0.91,0.90,and 0.97,respectively,which laid the foundation for further extraction of bamboo forest information form forest land.2.The bamboo forest remote sensing information was extracted by the C5.0 decision tree model based on forestland information,which has an average accuracy of 80.07%.The results can accurately reflect the spatial-temporal distribution characteristics of bamboo forests on the large scale.However,because of the low resolution of MODIS remote sensing,thebamboo forest information still has mixed pixel.Therefore,the bamboo forest abundance information was obtained by using the least squares pixel mixed decomposition.The research results have a high consistency with the bamboo forest resource inventory area,the coefficient of determination(R2)is greater than 0.95,indicating that bamboo forest area precision base on the decision tree model combined with the the mixed pixel decomposition was relatively satisfactory.Itprovides a guarantee for the accurate estimation of bamboo forest area and the study of the spatial-temporal dynamic carbon cycle of large-scale bamboo ecosystems.3.The annual/monthly scale NPP of bamboo forest ecosystem in Chain was obtained based on the parameter optimized BEPS model.Compared with the existing research results and sample survey data,the average R2 of the third phase was 0.73.In terms of spatial distribution,the high value of NPP in bamboo forests mainly gathered in Zhejiang,Fujian,Jiangxi,and Hunan.In addition,the distribution of Yunnan,Shanxi and other places wasrelatively scattered.On the timescale,the NPP of bamboo forest changed significantly.On the annual scale,the total NPP wasonascent.On the monthly scale,the NPP of China‘s bamboo forests first increased and then slowed down,the value of May was the highest and the value of December was the lowest.On the seasonal scale,the contribution rate of the four seasons to the whole year wasspring(32.94%)> summer(26.72%)> winter(22.19%)> autumn(21.64%).