Use of Remote Sensing for Multitemporal Analysis of the Severity of the Fontibón Fire in Pamplona, Colombia

Abstract

Wildfires in tropical mountain ecosystems are difficult to assess because steep terrain and persistent cloud cover constrain field surveys and satellite observation. This study developed an integrated workflow combining Unmanned Aerial Vehicle (UAV) photogrammetry, Sentinel-2 imagery, and Google Earth Engine to quantify burn severity and early vegetation recovery after the January 2024 Fontibón fire in Pamplona, Colombia. RGB imagery acquired with a DJI Mavic 3E was processed in Agisoft Metashape to delineate the affected area and produce 3-cm orthophotos. Sentinel-2 Normalized Burn Ratio (NBR) and differenced Normalized Burn Ratio (dNBR) were used to classify severity, while Normalized Difference Vegetation Index (NDVI) images from January 2024 and January 2026 characterized greenness trajectories. UAV orthophotos supported image-based validation through 500 purely random checkpoints. The fire affected 175 ha, of which 61.9% showed moderate-high or high severity. Classification agreement with UAV visual interpretation was 87.4% (Cohen's κ = 0.82). Mean NDVI declined from 0.62 ± 0.09 before the fire to 0.28 ± 0.13 immediately afterward and increased to 0.38 ± 0.06 by 2026 (paired t-test, p < 0.001). Although 98.7% of the area showed moderate greenness by 2026, no zone returned to the pre-fire high-NDVI class. These results indicate widespread spectral greening without evidence of complete structural or ecological recovery. The workflow provides operational inputs for restoration planning, but UAV image interpretation is not independent ecological ground truth, the validation sample was not stratified by severity, and the two-year observation window limits long-term ecological inference.