The remote sensing of insect defoliation in Mopane woodland.

Abstract

Mopane (Colophospermum mopane) woodlands are a source of valuable resources that contribute substantially to rural economies and nutrition across Southern Africa. However, a number of factors such as over-harvesting and climate change have brought the sustainability of the mopane woodland resources into question. Insect defoliation remains a major factor contributing to the depletion of woodland resources in rural areas resulting in low vitality and productivity of the woodland. Conventional methods (e.g. visual evaluation) have been used in monitoring insect defoliated areas in the past. These methods are costly and timeconsuming, because of the need to collect data immediately before and after an extreme event. In this regard, remote sensing techniques offer a practical and economical means of quantifying woodland degradation over large areas. Remote sensing is capable of providing rapid, relatively inexpensive, and near-real-time data that could be used for monitoring insect defoliation especially in semi-arid areas where data collection may be difficult. The present study advocates the development of techniques based on remotely sensed data to detect and map defoliation levels in Mopane woodland. The first part of the study provides an overview of remote sensing of insect defoliation, the implications for detecting and mapping defoliation levels as well as the challenges and need for further research especially within Mopane woodland. Secondly, the study explored whether Mopane species can be discriminated from each of its co-existing species using remote sensing. This was done as a prerequisite for classifying defoliation on mopane trees. Results showed that, with limited training samples, especially in semi-arid areas, Mopane trees can be reliably discriminated from its co-existing species using machine learning algorithms and multispectral sensors with strategic bands located in sensors such as RapidEye. These positive results prompted the need to test the use of ground based hyperspectral data and machine learning algorithm in identifying key spectral bands to discriminate different levels of insect defoliation. Results showed that the random forest algorithm (RF) simplified the process and provided the best overall accuracies by identifying eight spectral wavelengths, seven of which belongs to the red-edge region of electromagnetic spectrum. Furthermore, we tested the importance of the red-edge region of a relatively cheaper RapidEye imagery in discriminating the different levels of insect defoliation. Results showed that the red-edge region played an important role in mapping defoliation levels within Mopane woodland with NDVI-RE performing better than the traditional NDVI. Thirdly, the study tested the reliability and strength of the internal validation technique of RF in classifying different defoliation levels. It was observed that the bootstrapping internal estimate of accuracy in RF was able to provide relatively lower error rates (0.2319) for classifying a small dataset as compared to other validation techniques used in this study. Moreover, it was observed that the errors produced by the internal validation methods of RF algorithm was relatively stable based on the confidence intervals obtained compared to other validation techniques. Finally, in order to evaluate the effects of insect defoliation on the biophysical properties of mopane canopies at different defoliation levels, the study estimated leaf area index (LAI) of different defoliation levels based on simulated data. This was done using PROSAILH radiative transfer model inverted with canopy spectral reflectance extracted from RapidEyeRapidEye imagery by means of a look-up-table (LUT). It was observed that the significant differences exist between the defoliation levels signifying reduction in the LAI as a result of the defoliation. Furthermore, results showed that the estimated LAI was in the range of those reported in literature. The NDVI-RE index was the most strongly correlated with the estimated LAI as compared to other variables (RapidEye bands and NDVI). Overall, the study demonstrated the potential of remote sensing techniques in discriminating the state of Mopane woodland after insect defoliation. The results are important for establishing an integrated strategy for managing defoliation processes within Mopane veldt, thereby satisfying both the needs of local populations for Mopane trees and the worms.