Remote Sensing
Although activities such as cattle ranching have traditionally been the most significant drivers of landscape change in Colombia, the global biofuel boom and its many national and regional socioeconomic benefits have catalyzed additional landscape change through the expansion of oil palm since the 1970s. The cultivation of oil palm remains an appealing industry for Colombia due to its ability to create jobs and to generate economic growth both nationally and regionally, which therefore could potentially be reinvested into communities in place of what may be weak existing institutions, especially in rural areas.
Because of these benefits, the Colombian government continues to support the oil palm industry through government subsidies, tax exemptions, research funding, and a biofuel program established in 2001. However, researchers note that this support often gives preference to large shareholders rather than owners of smaller farms, therefore contributing to the creation of an elite class whose concentrated power reduces participation in policy-making decisions among non-elites and exacerbates the impacts of poverty. Additionally, an increase in violence and illegal land-grabbing has been attributed to oil palm expansion in certain regions. The negative socioeconomic implications of oil palm expansion in Colombia are relatively well-known, but existing research analyzing the environmental implications, especially regarding deforestation and biodiversity loss, is more contentious.
Colombia is recognized globally as a sustainable producer of oil palm, as the majority of Colombian oil palm expansion allegedly occurs on already degraded land (such as former pastures) rather than contributing to deforestation or other natural landscape conversion. Government reports and numerous research studies support this sustainable production claim; however, not all existing studies agree, and many attempts to quantify the natural landscape conversion that may have occurred as a result of oil palm expansion present contradicting results. To gain a more accurate understanding of the role Colombian oil palm expansion plays in deforestation and other land cover changes, additional research exploring the changes in landscapes over time is necessary.
Our research adopts a spatial and temporal approach which employs remote sensing methods and GIS analysis to map agricultural land covers and their contributions to deforestation over a period of ten years (2010-2020). The study area is the Magdalena River Valley in northern Colombia, which has been identified as having the second largest amount of land suitable for oil palm conversion in the country. This analysis includes the production and comparison of two supervised land cover classifications in order to quantify and characterize land use change and deforestation in the study region. The outputs of this analysis will be combined with and compared to existing global spatial data on forest cover change to assess the suitability of global forest change products in characterizing forest lost in this region.
The overarching goals of this study are to:
1. quantify deforestation between 2010-2020,
2. determine the extent to which oil palm contributes to the deforestation of natural forest in the study region,
3. assess remote sensing analysis’ potential contributions to informing environmental conservation and planning efforts, and to
4. determine the suitability of global forest change products in characterizing forest loss.
The 2020 land cover classification uses three separate datasets: 1) mosaicked Landsat Analysis Ready Data (ARD) images that were captured throughout 2020, 2) a digital elevation model (DEM), and 3) HH and HV radar images from the ALOS-PALSAR satellite, which were acquired in 2018. The DEM was obtained from the United States Geological Survey (USGS)’s EarthExplorer platform, and the mosaicked ARD images were obtained using the University of Maryland's Global Land Analysis & Discovery (GLAD)image downloading protocol.
Because our study focuses primarily on the lowlands of the study area where much of the regional agricultural expansion is occurring, areas of high elevation or dissimilar biotic components were excluded from the initial study area. Using the Landsat 8 images as well as Google Earth imagery as references, spatial training polygons were drawn in QGIS to represent patches of different land cover classes; these polygons (772 total) were divided into 18 different classes, which all correspond to different land cover types found in the study area. These classes include bare soil, new and old burn scars, burned forest, dense (natural) forest, grains, flooded and unflooded grasslands, lakes, three stages of oil palm maturity, tilled soil, sandbanks, river, planted tree plantations, and urban areas.
The normalized difference vegetation index (NDVI) was calculated using the ARD mosaic’s visible and near-infrared wavelengths to measure vegetation density, and speckle filtering was applied to the HH and HV bands of the 2018 ALOS-PALSAR image to reduce noise and inaccuracies associated with SAR images’ granularity. The NDVI, both PALSAR bands, and the DEM were stacked to the mosaicked and pre-processed ARD image to form a total of 11 bands.
A supervised classification (Random Forest method) was run with the raster stack and trained polygons as inputs. The outputs include a thematic map of the study area’s 18 land cover classes and a trained model to classify the 2010 mosaic.
This project is ongoing. Check back again in January 2022 for a more complete summary of the project.
Next steps include:
1. Using the 2020 classification model to classify the 2010 mosaic,
2. Running accuracy assessments for both classifications,
3. Using spatial filtering to ensure the accuracy of classes such as oil palm, which tend to be more clustered,
4. Conducting a change analysis to measure the amount of natural forest loss that can be attributed to oil palm expansion, and
5. Comparing the results from the change analysis to global and national forest products such as Hansen's global forest map or the national IDEAM forest/non-forest map to assess the value of each.