Statistical inferential techniques for approaching forest mapping. A review of methods

Abstract

The increasing availability of remote sensing data at no or low costs can be used as ancillary data in order to spatialize and improve the estimation of forest attributes and without increasing the sampling effort and costs. In this review paper, a description of the main statistical inferential techniques for approaching forest mapping is proposed. This article reviews the most used forest mapping methods based on the sole spatial information as well as techniques exploiting auxiliary information from remotely sensed data. The advantages and drawbacks of each method have been described on the basis of several factors, such as the aims of the investigation and the area under examination. Two main groups were here discussed with model-based methods on one side and model-assisted methods on the other, moving the attention from the model used to interpolate surfaces to the sampling scheme. Model-based methods include kriging, locally weighted regression, K-NN, decision trees and neural networks, while the inverse distance weighting interpolator is presented in the model-assisted group. Reliable and up-to-date information on forest characteristics are mandatory tools for any decisional process. The main input data of such systems are wall-to-wall maps depicting the spatial structures of forests and additional elements. Actually, if the original aim of forest inventories was to estimate harvestable timber amounts, a general interest towards multipurpose surveys is mandatory. Such information must deal with increased costs and more time-consuming procedures.