University of Twente Student Theses


Drought-Induced Tree Mortality Assessment in Mediterranean Ecosystems Using Time Series Analysis

Raunak, Alma (2024) Drought-Induced Tree Mortality Assessment in Mediterranean Ecosystems Using Time Series Analysis.

[img] PDF
Abstract:Remote sensing techniques are crucial in monitoring and assessing forest health, particularly in detecting tree mortality events. This study investigates the effectiveness of spectral indices and Spectral Unmixing (SU) derived from satellite data in identifying tree mortality in Mediterranean forest ecosystems. The study demonstrates the utility of Unmanned Aerial Systems (UAS) imagery as ground truth data, showcasing its high correlation with direct field observations (R² = 0.95, RMSE = 2.97), thus offering a way to complement and/or simplify traditional field campaigns. UAS also verifies the reliability of SU (R² = 0.75, RMSE = 12.4) in estimating vegetation percentage within mixed pixels despite challenges such as variability in vegetation composition and spatial resolution limitations. Time series analysis reveals the minimal influence of a recent single drought event (which happened in 2015-2016) on vegetation indices derived from satellite time series imagery. Tree mortality found in the area is explained by disturbances that occurred in the late 1990s and mid-2000s, highlighting the need for having 2 or more droughts consequently for high tree mortality. Change detection using the LandTrendr algorithm (including using a novel technique of detecting changes in SU’s outputs with LandTrendr) confirms significant vegetation loss, with SU highlighting more pixels with changes than traditional vegetation indices. Despite SU having the lowest overall accuracy, evaluation of the presence of tree mortality accuracy demonstrates SU's superiority (97.96% accuracy) over vegetation indices (i.e., 93.88% NDVI, 93.75% NDWI) while acknowledging challenges in detecting the absence of tree mortality (i.e., 50% SU, 78.57% NDVI, 73.33% NDWI), presumably due to the overestimation of SU comparing to UAS data. Additionally, the research highlights the potential of NDVI in capturing changes in canopy dynamics as well as highlighting areas of drought-resilient tree species and NDWI in identifying areas of high drought susceptibility. The regression analysis between vegetation indices and tree mortality percentage, and between SU and tree mortality percentage, reveals low R² values for NDVI, NDWI, and SU (0.37, 0.18, and 0.35 respectively), and high RMSE values of 56.9, 84.9, and 26.6 respectively, including significantly lower correlations using NDWI. However, removing plots with tree mortality percentages higher than 80% helps to improve it, resulting in R² of NDVI, NDWI, and SU to be 0.42, 0.35 and 0.51; RMSE: 58.2, 80.9, and 20.8 respectively. The study also discusses the importance of spatial and temporal resolutions in remote sensing data to accurately assess tree mortality and the potential of ensemble approaches with higher spatial resolution imagery to improve detection accuracy. Moreover, the findings underscore the significance of early detection of forest decline and the resilience of vegetation to environmental stressors, providing valuable insights for forest management and conservation efforts. Overall, this research advances our understanding of remote sensing applications in monitoring tree mortality and offers recommendations for future studies to enhance detection methodologies and improve forest management strategies.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Subject:38 earth sciences, 43 environmental science
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page