The estimation of Eucalyptus plantation forest structural attributes using medium and high spatial resolution satellite imagery.

Abstract

Sustaining the socioeconomic and ecological benefits of South African plantation forests is challenging. A more systematic and rapid forest inventory system is required by forest managers. This study investigates the utility of medium (ASTER 15 m) and high (IKONOS 1-4 m) spatial resolution satellite imageries in an effort to improve the remote capture of structural attributes of even-aged Eucalyptus plantations grown in the warm temperate climatic zone of southern KwaZulu-Natal, South Africa. The conversion of image data to surface reflectance is a pre-requisite for the establishment of relationships between satellite remote sensing data and ground collected forest structural data. In this study image-based atmospheric correction methods applied on ASTER and IKONOS imagery were evaluated for the purpose of retrieving surface reflectance of plantation forests. Multiple linear regression and canonical correlation analyses were used to develop models for the prediction of plantation forest structural attributes from ASTER data. Artificial neural networks and multiple linear regression were also used to develop models for the assessment of plantation forests structural attributes from IKONOS data. The plantation forest structural attributes considered in this study included: stems per hectare, diameter at breast height, mean tree height, basal area, and volume. In addition, location based stems per hectare were determined using high spatial resolution panchromatic IKONOS data where variable and fixed window sizes of local maxima were employed. The image-based dark object subtraction (DOS) model was better suited for atmospheric correction of ASTER and IKONOS imagery of the study area. The medium spatial resolution data were not amenable to estimating even-aged Eucalyptus forest structural attributes. It is still encouraging that up to 64 % of variation could be explained by using medium spatial resolution data. The results from high spatial resolution data showed a promising result where the ARMSE% values obtained for stems per hectare, diameter at breast height, tree height, basal area and volume are 7.9, 5.1, 5.8, 8.7 and 8.7, respectively. Results such as these bode well for the application of high spatial resolution imagery to forest structural assessment. The results from the location based estimation of stems per hectare illustrated that a variable window size approach developed in this study is highly accurate. The overall accuracy using a variable window size was 85% (RMSE of 189 trees per hectare). The overall findings presented in this study are encouraging and show that high spatial resolution imagery was successful in predicting even-aged Eucalyptus forest structural attributes in the warm temperate climates of South Africa, with acceptable accuracy.