Study On Estimation Of Under Forest Fuel Load Based On Remote Sensing Images

This article used the Mao'er mountain 2002 TM images and data collected in 209 forest sample spots to built Ridge estimation equation and neural network model, in order to find out the best way for researching and estimating fuel load in such area. To avoid the negative effects on the working out of pending parameters caused by the near linear correlation between each variables (multi-collinearity), we researched ways for better choosing variables using ridge estimate, we use 20 variables including vegetation index, GIS and remote sensing information to make ridge trace analysis in order to better choose variables. Comprising normalized difference vegetation index (NDVI), ratio vegetation index (RVI), ratio of grayscale value between sample spot and satellite image; also make principle component analysis on sample estimating array. The outcome show that the contribution of the TM3 which indicates red light in 0.630-0.69 um TM3, and 2.090~2.35um TM7 which is sensitive to the plant moisture are greater than others' in the first principle component; in the second principle component, 1.550-1.75um TM5 which indicates the plant moisture and 0.775~0.90um TM4 provides a majority of contribution.We use the filtered variables to build Ridge estimation equation, and calculate the multiple correlation coefficient, at the same time passed the correlation test of the Ridge estimation equation; we built forward neural network model based on SCG method and momentum method under the MATLAB 6.5 situation, when estimating the fuel load of the sampled plots, the average precision of the artificial neural network model is 89.51%.This article studied both the direct method of estimating fuel loads and also the indirect estimating method using stand factors such as biomass, stand average diameter at breast height as middle variables. We compared precisions of each fuel load estimating models, and chose a comparatively better way to build model for estimating forest fuel load of the Maoer mountain area. On the TM images we record the grayscale values according to the sample plot coordinates in the forest distributing area, and estimate the forest fuel load distribution in this region, then Kriging interpolate the outcome, the sum of forest loads in this area: 9.76 t/hm~2and protract the forest fuel load distribution map of the Maoer mountain.