Spatial interpolation methods are widely used to estimate some ecological and environmental parameters that are difficult to measure. One of these parameters is forest site index, which is a demonstration of forest productivity. The aim of this study was to estimate forest site index in a beech forest ecosystem in Turkey. In this context, soil characteristics, stand parameters, and topographic features were measured in 70 temporary sample plots of beech forest stands. Forest site index of beech forest stands was predicted using different modeling techniques such as multiple regression analysis (MLR), multilayer perceptron (MLP), radial basis function (RBF), multiple regression kriging (MLRK), multilayer perceptron kriging (MLPK), and radial basis function kriging (RBFK). The results showed that the RBFK (R2 = 0.98) and MLRK (R2 = 0.96) outperformed the others to predict forest site index in the study area. The greatest improvement occurred when krigged residual used with MLR, which increase from 0.23 to 0.96. Thus, MLRK method significantly improved the prediction accuracy for site index. The models combined with krigged residuals were more successful than those used without krigged residuals. The results of this study suggest that the combined methods may help obtaining improved site index maps for forest management.
Keywords: Artificial neural networks; Combined methods; Forest site index; Spatial distribution.