Abstract:
Surface water bodies in urban areas, such as Bolgoda lake, show complex vegetation
dynamics, typically noticeable by the fluctuating vegetation cover throughout the year.
Primary factors governing these fluctuations include wastewater discharge, anthropogenic
activities (e.g., surface mining), invasive plant growth, and climate change. It is exceptionally
challenging to physically measure and monitor these dynamics over the spatial extent of
these waterbodies consistently over many years. Recent studies have explored the potentials
of employing satellite imagery to quantitatively detect spatiotemporal changes of surface
water vegetation cover. Such attempts have utilised vegetation detection indices, such as the
normalised vegetation index (NDVI), to classify the vegetation cover with significant
statistical accuracy. However, these conventional geospatial analyses require substantial
computational power. They are limited to small timescales and spatial extents. This study
employs the computational power of the google earth engine to address this limitation.
Moreover, it integrates a machine learning classification approach, namely decision tree
regression, to monitor the vegetation cover change over coarser and finer temporal
resolutions using Landsat 8 hyperspectral imagery. Initially, NDVI classification was
performed on 390 Landsat 8 images acquired throughout 2013-2021. Five locations, which
represent different vegetation cover characteristics on the lake, were selected to generate the
time series of the NDVI classified values. The results show that the vegetation cover varies at
two temporal frequencies. The annual variation of the water, vegetation, and non-vegetation
classes are undetectable. However, vegetation dynamics fluctuate rapidly at a finer temporal
resolution (i.e., on monthly cycles). The statistically significant results claimed in this study
will be further explored to support policymakers in optimising environmental resource
management strategies and prioritising eco-preservation that can enhance the health and
productivity of urban surface water bodies.
Citation:
Dassanayake, S.M., Jayawardena, C.L., & Dissanayake, D.M.D.O.K. (2021). Decision tree regression approach for detecting spatiotemporal changes of vegetation cover in surface water bodies [Abstract]. In D.M.D.O.K. Dissanayake & C.L. Jayawardena (Eds.), Proceedings of International Symposium on Earth Resources Management & Environment 2021 (p. 77). Department of Earth Resources Engineering, University of Moratuwa. https://uom.lk/sites/default/files/ere/files/ISERME%202021%20Proceedings_2.pdf