Remote Sensing Methods for Assessing Abrupt Land Cover Change —A Case Study of the Central Luangwa Valley in Zambia

Abstract

Abstract I examined how abrupt land cover change can be detected, measured, and evaluated via accessible platforms using varied methods applicable to the non remote sensing (RS) specialized analyst. Analysis with free imagery and widely accessible platforms democratizes access to information and research capabilities (Nagaraj, Shears, & de Vaan, 2020). Enabling open access to remotely sensed data can aid in understanding and fostering the sustainability of socioecological systems. As socioecological natural disasters such as abrupt land cover change increase in frequency and intensity, the need for such analysis heightens. Thus, the focus of this thesis is how abrupt land cover change can be measured via readily accessible platforms. As proof of concept for the methods framework, I quantitatively analyzed land cover change with appropriate indices, classification methods, and algorithms. I selected a study area in Zambia’s Luangwa Valley and addressed how land cover changed during the COVID-19 pandemic from prepandemic 2019 to mid-2021. The results supported that the framework of methods I designed could quantify the magnitude of land cover change in km2 and determine whether the rate of change was greater than 10%. Using the random forest algorithm (RF) the magnitude of change in the study area was greater than 10%, with approximately 82% accuracy. With an estimated accuracy of 85%, I concluded that the land cover change extended to an area of 1250 km2. The result was based on the spectral angle (SA) algorithm, which performed with greater accuracy than the minimum distance (MD) and maximum likelihood (MLC) algorithms, respectively. Moreover, analysis via spectral indices and statistics enabled the determination of approximate locations and directions of change. Overall, the devised framework proved worthwhile for conducting a broad land cover change analysis. However, the ability to tailor the methods introduced to a given problem set, such as indices or algorithms, renders this framework worthwhile in studying other socio-ecological problems. The framework of methods introduced has broad applicability.