Comparison of land cover mapping methodologies to detect ecosystem change in conservation areas in the Eastern Savannas in Colombia
Jeronimo Rodriguez-Escobar, Temple University, jeronimo.rodriguez@temple.edu (Presenter)
Victor Hugo Gutierrez-Velez, Temple University, tug61163@temple.edu
Mary Blair, American Museum of Natural History, mblair1@amnh.org
Land cover mapping is a critical tool for monitoring and managing protected areas, particularly in regions with high biodiversity. In this study, we assessed two land cover mapping methodologies and their ability to detect change in an area of 1,120 km2 in the influence area of El Tuparro National Park in Colombia.
The first methodology, used by the National Institute for Environmental Studies and Meteorology (IDEAM), applies visual interpretation of optical satellite imagery to produce land cover classifications using an adapted Corine Land Cover classification legend. This methodology has been used to produce national land cover maps for the years 2001, 2007, 2011, and 2018.
The second methodology applies a hybrid machine learning-based approach, integrating calibrated multispectral and synthetic aperture radar satellite imagery, terrain data, and harmonized global and national forest cover products with training data collected through visual interpretation of a single reference year to train and extrapolate a supervised land cover classification model. We incorporate a landscape metrics-based spatial cluster and multi-temporal analysis to enhance land cover class separability. We produced land cover maps for 2003, 2007, 2011, 2015, 2016, 2019 and 2021.
We tested both methodologies in terms of the magnitude and type of change mapped, finding that the second one consistently reported a larger amount of change both as a share of total area and in terms of the transitions between natural and planted land covers and between types of savanna. We attribute the differences to a higher thematic and temporal resolution in the automated method that translates into a more nuanced representation of transitions between grassy land covers and shrublands, and permanent crop development stages.
Our findings can facilitate the timely delivery of land cover maps for land management and conservation due to the reduced human workload of the automated classification. The finer thematic representation can also inform targeted decisions associated with different types of land cover transitions while providing temporal consistency, improving the temporal comparability of land cover maps. This can facilitate a consistent, accurate, and detailed understanding of change, key for conservation purposes in protected areas.
Associated Project(s):
Poster Location ID: 1-13
Presentation Type: Poster
Session: Poster Session 1
Session Date: Tue (May 9) 5:00-7:00 PM
CCE Program: LCLUC