MSc Thesis

Remote Sensing of Tree Health: Evaluating time-series MODIS NDVI data for monitoring forest health and detecting forest felling in Wales

MSc Thesis (PDF 9MB)

MSc details :
Remote Sensing and GIS MSc. Dissertation
Institute of Geography and Earth Sciences, Aberystwyth University


Forests have played an integral role in human history and are the most widely distributed ecosystem on the earth, affecting the lives of most humans daily, either as an economic good or an environmental regulator. In Wales forests are an important commercial resource, fundamental to the economy, contributing ~£372 million to the economy. New approaches to woodland management and forest health monitoring are required in Wales to address and quantify the current and potential future threats of diseases (e.g. Phytophthora genera), pests and climate change on timber stock health. In addition, there is a growing need to identify precise forest felling date to aid the management of insidious pests such as the Pine Weevil (Hylobius abietis), the principal pest of young conifers throughout the U.K. This poses a major challenge for an area the extent of Wales in terms of the required labour, data handling and ultimately cost, encouraging the application of Remote Sensing technologies. This study explores the capability of MODIS-derived 16-day composite NDVI time-series data (2000-2013) to monitor forest health change and detect forest felling on a regional scale in Wales. MODIS data from 2000-2013 were re-projected, temporally stacked and processed using the Continuous Monitoring of Forest Disturbance Algorithm (CMFDA) proposed by Zhu et al. (2012) to model MODIS pixel phenological response. The spatial and temporal variation in the modelled NDVI and phenological (seasonality) parameters were assessed to create a forest felling classification in ENVI 4.8 to identify felling sites each individual year. The outputs were then assessed against ground-truth felling sites derived from Landsat imagery. The modelled NDVI MODIS time-series was effective at identifying subtle temporal changes in forest stand condition and phenology; including forest thinning, forest stress and disease, forest re-growth, and abrupt changes such as forest felling. Errors in classification and forest felling identification were dependant on the geometry of the felling site, its size and inherent issues with regards to the coarse resolution of MODIS data (e.g. mixed pixel problem, mis-registration). An overall detection accuracy of 62.37% was achieved for the forest felling classification; with a reasonable balance between change commission errors (37.02%) and omission errors (30.7%). The results suggest that MODIS-derived NDVI time-series data provide an innovative insight into forest health and condition globally and show promise for monitoring and assessing forest health temporally in Wales. More work is needed to improve the felling classification and explore the use of the MODIS SWIR band (500m) to detect forest clearing. Keywords: MODIS, NDVI, Forest Health, Time-Series, Phenology