Spatially explicit drivers of forest recovery in the Brazilian Amazon in the last 3 decades

By Ainil Mardhiah

Abstract
As one of the essential ecosystems in the world, the Brazilian Amazon plays a key role in providing a vital ecosystem service and tackling climate change by reducing CO2 . The government and stakeholders have started committing to forest recovery programs to maximize forest function. If there are no activities focus on reforestation, forest functions will continue to be threatened by deforestation, which has occurred massively in recent decades. In order to stimulate large-scale forest recovery in the Brazilian Amazon, it is essential to determine the spatial occurrence of reforestation and its biophysical and socio-economic drivers. This study investigates the spatially explicit relations of biophysical and socio-economic variables to reforestation between 1989-2018. In addition, this study also identified changes in the relationship of each driver of forest recovery over three different periods (1989-1998, 1999- 2008, and 2009-2018). This study uses LIME (Local Interpretable Model-Agnostic Explanation) to determine which biophysical or socio-economic factors are most closely related to forest recovery in the Brazilian Amazon. However, before that, this research used a GIS approach for the pre-processing and processing stages. Pre-processing emphasizes data cleansing and equalizing raster data from different sources and spatial resolution. Furthermore, the processing stage focus on calculating the number of forest recoveries. The amount of forest recovery was fathered from binary-map forest increments from 1989 to 2018. The remaining variables were extracted as raster from multiple spatial resolutions. Further, the clean dataset was trained by the random forest model before it was processed in LIME to explain the relationship between all those variables and forest recovery. Reforestation peaked at 3.1 Mha in 2018, after almost 30 years of fluctuating recovery numerical ratings. The states of Pará, Maranhão and Mato Grosso were the most successful projects of the recovery scheme. It essentially served as a substitute for areas of former anthropogenic activity in which the main pattern of small patches of forest recovery is scattered across roads. Overall, areas akin to successful reforestation shared these traits – moderate population density, travel time to the nearby city, and being surrounded by a high density of trees. Our data analysis was inconclusive regarding whether overarching factors such as topography, Soil Organic Carbon (SOC), and overall temperature impacted a forest recovery scheme. There was hardly a noteworthy change over a sustained period of time.

Keyword: Forest recovery, spatially explicit driver, biophysical and socio-economic variables, GIS approach, random forest, LIME.