Estimation of Tree Species Proportions Using Ranging Scatterometer

Tree species proportions of forest stands were estimated using ranging scatterometer called HUTSCAT. Employed estimation method was multilayer perceptron neural network with error backpropagation training algorithm. Two methods for feature extraction were tested; other based on intensity and shape of measured profiles, other only shape. Different neural network configurations were tried and it was found that differences between them were rather small. Shape−features performed little better because the differences between different networks were smaller. The best classification accyracy of the main tree species was about 86% and the mean error of estimation for tree species proportions was 0.30. Largest errors are due to low stem volume, because stand is in the beginning of its production cycle (clearing or sapling) or soil is poor.