Work

Reconstructing the alarming fire history of Ankarafantsika National Park in northwestern Madagascar over a 35 year-period

Year: 2025

Type: article

Abstract: The extent of fires in tropical ecosystems is highly dynamic, and responses of tropical dry forests to fire seem to vary and are not well understood. This study reconstructs a 35-year long fire history of the dry forests in the Ankarafantsika National Park (ANP) in Madagascar. In recent years, forest fires have been reported in this region, potentially threatening habitat integrity and connectivity inside ANP. We assessed the fire dynamics with yearly Landsat satellite images from 1988 to 2023 (Landsat 4–5, Landsat 7, Landsat 8-9) by means of the Normalized Burn Ratio (NBR) and the differenced NBR (dNBR) with a 30m resolution. To validate the fire classification in addition to exemplary ground truthing, 350 random points were selected across ANP and visually classified based on yearly Red Green Blue false-color composites and their corresponding dNBR images. Inside ANP, 70% of the vegetation burnt at least once, while 18% burnt even more than three times. Fire distribution was heterogenous with fires being more frequent in the north than in the south of ANP. The extent of burnt areas fluctuated over the years peaking in 2016 with 229 km 2 (22% of the park vegetation). Fires recurred after 1-34 years, but more frequently than expected after 3–5 years. Areas requiring urgent conservation attention are highlighted and a baseline for developing fire management strategies is provided. Finally, this study demonstrates that ground truthing is essential to implement an analytical pipeline to correctly infer burnt and unburnt vegetation parts from series of satellite pictures. • Fire dynamics were studied over 35 years inside a Malagasy dry deciduous forest • Only 30% of the vegetation never burned inside the Ankarafantsika National Park • Up to 22% of the park vegetation was burned per year • Fire analyses from satellite data need to be locally adapted and validated

Source: Remote Sensing Applications Society and Environment

DOI: 10.1016/j.rsase.2025.101521