Study On Estimation Of Aboveground Carbon Storage Of Moso Bamboo Forest Based On LANDSAT TM Image

The study is part of the National Natural Science Foundation (grant, 30700638) and'948'item of national forestry bureau (grants, 2008-4-49). Bamboo is a special forest type mainly distributed in semi-tropical area of China, which plays an important role in regional ecosystem and in carbon dioxide (CO2) absorbability and global carbon balance. However, the studies in abroad and home show that the traditional forest mensuration is still the common approach used to estimate the aboveground biomass (AGB) of bamboo, and the studies are lacking in map of spatial distribution of AGB, and it is still difficult to analyze and evaluate the AGB of bamboo in spatially while the carbon and net flux resulted from land use change need to be estimated from the biomass map.Therefore, the study is very significant. The moso Bamboo forest has advantage in area. The total area of moso Bamboo forest in China is 3.37 million hectare (ha) and approximately accounts for 70% of bamboo area. The area of moso Bamboo forest in Zhejiang province about accounts for 9.8% of whole forest area and in Anji county of Zhejiang province about accounts for 38% of forested land.Taken moso bamboo for an example, the objective of this study is to estimate the aboveground carbon (AGC) of moso bamboo forest based on remote sensing data and supported by geographic information system (GIS) and global position system (GPS).The main contents are as follows:1. Classification method of moso bamboo, including selecting the optimum bands used for moso bamboo forest classification and comparision of different classification methods;2. Developing the estimated models using the traditional linear regression analysis, partial least-squares regression (PLS) and gaussian error function back-propagation (Erf-BP) neural network based on field data and remote sensing data, and selecting the best model to estimate the AGB of moso bamboo forest;3. Analysis and map of AGC of moso bamboo forest based on Geographic information system (GIS). The main results are as follows:1. The feature selection result shows that the combination of band1, band3, band4, band5, NDVI, and IIVI is suitable for moso bamboo forest classification.2. The total accuracy and Kappa coefficient of BP neural network are higher than maximum likelihood method.The production and user accuracry of moso bamboo forest based on BP neural network are 88.25% and 91.94% respectively.3. Through accuracy evaluation, the Erf-BP model provides the best estimation accuracy, and the determination coefficient (R2) of training and validation set equal to 0.997 and 0.822 respectively. The formula of Erf-BP model is as follows:The relationships between AGB and EVI, Tass4, Vari1, Homo2, Homo4, Diss2, and Diss3 are positive while for Tass5, Prin4, Mean5, Homo3, and Entr5 the relationships with AGB are negative. The variables of EVI, Tass5, Mean5, Homo2, Homo3, and Diss3 have significant contribution to AGB estimation.4. The total AGC storage by moso bamboo forest in Anji county are estimated to be 1.252 Tg C and 1.238 Tg C in 2004 and 2008 respectively based on Erf-BP model.