The spatial distribution of forest disturbance is commonly calculated using a satellite imagery-driven bi- or tri-temporal change analysis. Working in Colombia’s Cordillera de los Picachos National Natural Park – a region of consistent cloud cover and dramatic topographic relief – a change assessment with such infrequent observations cannot capture long-term trends of vegetative decline (browning) or improvement (greening) nor the drivers associated with these changes. In recognition of the importance of spatio-temporally explicit information for assessing the effects of socio-environmental change and conservation strategy implementation, I developed a rigorous assessment of vegetation change using MODIS and Landsat time-series data and the Breaks For Additive Season and Trend (BFAST) algorithm to identify the timing, trends, and locations of change as well the associated drivers.
First, I measured long-term vegetation trends from 2001-2015 using a Moderate Resolution Imaging Spectroradiometer (MODIS)-based 250m resolution Multi-Angle Implementation of Atmospheric Correction (MAIAC) time-series, and mapped short-term disturbances using all available Landsat images (149 dates from Landsat 5, 7, and 8). BFAST trends based on MAIAC data indicate a net greening in 6% of the park, with a net browning trend of 2.5% in the 10km-wide region surrounding the park. I also identified a 12,500 ha area within Picachos (4% of the park’s total area) that experienced a consecutive vegetative decline or browning during every year of study, a result corroborated with a BFAST Monitor assessment using finer 30m resolution Landsat data. With Landsat, I recorded 12,642 ha (±1440) of disturbed forest within the park at high spatial and temporal accuracy. Spatially, Landsat results had user’s and producer’s accuracies of 0.95±0.02 and 0.83±0.18, respectively. Temporally, a TimeSync-supported temporal validation assessment showed that 75% of Landsat-detected dates of disturbance events were accurate within ± 6 months.
With disturbances identified, I characterized disturbances within Picachos’ southeastern foothills and associated drivers using a set of metrics related to the spectral, pattern and trend properties of disturbance patches derived from Landsat time-series data (1996-2015). A training dataset was initially developed to identify drivers of disturbances using Corine Land Cover maps and high-resolution imagery. A Random Forests classifier was used to attribute disturbances to specific drivers of forest cover change: conversion to pasture, conversion to subsistence agriculture, and non-stand replacing disturbance (i.e., thinning). Attribution of changes had high accuracy at patch and area levels with 1-5% commission and 2-14% omission errors, respectively, for regions that were converted to pasture or experienced thinning. Lower agreement was found for agricultural conversion with 43% omission and 9% commission errors.
I found that conversion to pasture is the main cause of forest cover loss within Picachos at 9901 ha (±72) corresponding to 14.7% of Picachos’ foothills, and that subtle forest alteration contributed to 1327 ha (±92) of forest degradation. Recognizing the diversity of pressures facing conservation strategy implementation in the region, these results have direct relevance for anticipating future land use pressures within Colombia, as well as across similar regions in the Andes-Amazon transition area. Indeed, since these results reveal the possibility to uncover historical disturbances related to human-incursion in protected landscapes, the methods are well suited to enhancing landscape planning particularly where biodiversity richness is quickly diminishing due to anthropogenic presence.