Assessing the health of Pinus radiata plantations using remote sensing data and decision tree analysis

Forest health monitoring is essential to sustainable management of Pinus radiata D. Don plantations. Conventional survey techniques such as aerial sketch mapping are qualitative and subjective, their effectiveness depending on the skill of the surveyor. In contrast, digital remote sensing has the potential to provide quantitative and objective data on the location, extent, and severity of crown damage at a range of spatial scales. Decision tree analysis can incorporate both categorical and continuous data and is inherently non-parametric. Decision trees were used to model the crown condition of P. radiata plantations in southern New South Wales in three situations involving discoloured leaves, stunted crowns, and transparent crowns associated respectively with the Diplodia pinea (Desm.) Kickx fungus, nitrogen deficiency, and the pine aphid Essigella californica Essig. Spectral indices and fraction images derived from linear spectral mixture analyses of remote sensing scenes were used to classify crowns into either two or three condition classes. The best performing model was obtained for D. pinea with a two-class classification of crown discoloration (overall accuracy [OA] 92%; Kappa 83%). The three-class model of crown transparency from E. californica defoliation was moderately accurate (OA 68%, Kappa 28%), while the weakest model was a two-class model of crown volumes affected by soil nitrogen levels (OA = 58%, Kappa = 21%). Correlation between modelled outputs and environmental spatial variables revealed a high correlation between net solar radiation levels and crown transparency classes due to E. californica defoliation.