MODIS Image Based Estimation Of Zhejiang Provicial Forest Carbon Storage

Forest carbon storage is an important indicator reflecting the forest productivity.Estimation of forest carbon with traditional method is usually labor intensive and time consuming.Remote sensing technology with the characteristics of real time,rapid,continuous,large-area coverage and many other advantages,has become an important means of forest carbon estimation.MODIS is suitable for large-scale carbon storage study by wide coverage,high temporal resolution and hyperspectral resolution these advantages,but the problem of data matching,series mixed pixel,same spectrum with different objects,uncertainty and so on propblems are also existed in the application.Therefore,Zhejiang province is chosen as the representative of large-scale area in this thesis,by using MOD09A1/MOD13Q1(MODIS-NDVI)image,DEM data,National Forest Inventory data in 2009,combined with multiple linear stepwise regression and Sequential Gaussian co-simulation model to estimate forest carbon storage and its distribution in the study area,meanwhile introduced geographic factor to the NDVI time-series stepwise regression make a further modeling.The part of water has a lower pixel reflectivity;it is confused with the low albedo feature of buildings,mountains,other low vegetation shadows etc.easily,that affect extraction of vegetation components fully,and lead to the results error.Hence on the basis of model united to the two linear unmixing vegetation coverage of removing water or not respectively to estimate the above-ground carbon storage and distribution of the study.By contrasting estimation accuracy and analyzing the error evaluate pros and cons of 9 estimation methods,results showed that:1.Comparing with various estimation methods.Nine methods are applied to estimate the aboveground forest carbon storage and distribution of Zhejiang province.Based on measured carbon,the estimation accuracy of NDVI time-series stepwise regression model,the stepwise regression of geographical factors introducted and spatial simulation relative to measured value are 94.89%,95.13%,57.15%,the estimation accuracy of combining with non-excluded water vegetation coverage are 48.08%,47.51%,67.99%,the estimation accuracy of combining with excluded water vegetation coverage are 73.60%,74.19%,96.21%.The results showed that spatial simulation combined with vegetation coverage removed water get the best accuracy.Regression model caused distortion of local study area by smoothness,simulation remedy this defect and add the uncertainty of expression in carbon distributed results,better than the regression model significantly;NDVI time series spectral information are used by stepwise regression model to improve the simulation accuracy;Geographic information are added in the stepwise regression model can enrich the pixel information,improve discrimination of same spectrum with different objects and enhance estimation accuracy;the result of combination the vegetation coverage by removing water was better.2.The estimation results of forest carbon storage in study area.By comparing estimation results of forest carbon storage from nine methods,the best estimation method is Sequential Gaussian co-simulation with the vegetation coverage extracted after removing water area.The above-ground forest carbon storage is 1.589×108Mg,carbon density ranged from 0 to 150.715Mg/hm~2 with a mean value of 15.6123Mg/hm~2 estimated by this way,the precision of estimation reach 96.21%,the carbon density distribution of estimation is consistent with the actual situation,local difference obviously,estimation variance describes variability of forest carbon storage distribution clearly.3.Excluding water is conductive to improve unmixing accuracy.In this study,2D scatter plot and PPI are integrated to select endmember from MODIS,and extract vegetation coverage by fully-constrained linear decomposition method.The results showed that RMS averages of whether or not the water removed under two projectors are all between 0.007-0.009,much less than 0.02 the requirement error,but total vegetation coverage before removing water is 38.68% and the coverage after removing water is 38.68%,the latter one approach to actual forest coverage rate 59.07%,accuracy reach to 94.55%.Therefore,water removed in favor of improving vegetation coverage extraction accuracy of MODIS in this study.