Spatial patterns of land use & land cover changes surrounding Colombian water reservoirs

Abstract

<p>Hydropower dams can induce spatial and temporal changes in land and water systems in terms of resource access and use. Regardless of the main purpose of these dams (that is, to produce energy), analyzing these changes by the lenses of the water-energy-food nexus helps us to identify the synergies and trade-offs between these components in the watershed context. Changes  in land & water systems happen in different times (dam’s construction and operation), spatial scales and, in many cases, are also influenced by national political-economic context. Colombia is moving towards peace agreements in recent years, and this process already showed impact in the patterns of land and water uses, especially by agricultural systems. This new scenario can create or consolidate some local and national socioeconomic characteristics, adding inequalities beyond the dam’s construction. This work focuses on the land cover and land use transitions surrounding two dams in the Magdalena basin, Betania-Quimbo and Hidrosogamosso. This basin is responsible for 70% of the national energy production, it concentrates the production of important food/energy value chains for the local and global market and possess high biodiversity. Both dams were built in 2009 and started operations after 2015, so Landsat satellite images were used to build the land use & cover maps in 2009, 2015 and 2021 for 7 classes (rice, palm oil, pasture, forest, water surface, temporary and permanent crops and others). Due to the intensity of clouds and high altitudes, Colombia is one of the most difficult regions in the world to build these maps, and for this reason global, or even local, mapping available turn out to be unrealistic. A random forest model was chosen, and, as variables, indexes and spatial temporal metrics using 176 bands in total. To extract the pixel information for training and testing the model, a stratified random sample was run using different secondary maps, and after that, we used Google Earth for visual verification (1196 observation for all years). For the accuracy assessment just the sample from the current year was used. The model was run in Google Earth Engine. The model achieved a similar overall accuracy in all years (79%), and for certain agricultural systems a high accuracy, such as the case of palm oil with100% accuracy in 2021 and 2015, despite reaching 70% in 2009. Accuracy for rice was also high: 95%, 84% and 88% in 2021, 2015 and 2009 respectively. Pasture achieved a medium accuracy: 78% (2021): 83% (2015); and 71% (2009). Water surface achieved a high accuracy.For water surface: 100% in 2021 and 2015, and 97% in 2009. The Forest category reached a medium-high accuracy: 81%, 88% and 77% in 2021, 2015 and 2009 respectively. Land use and Land cover maps of areas impacted by dams is of high importance to support decisions that will be implemented via instruments such as Basin Management Plans or compensatory schemes.</p>