Projections of Future Cropland Abandonment: Impacts to Biodiversity and Carbon Sequestration

Cropland abandonment occurs when ecological and economic shifts force farmers to retire lands formerly used to grow food. Without proper management, cropland abandonment can lead to soil erosion, inhibit nutrient cycling, increase wildfire risk, and threaten food security. However, if managed strategically, restoration of these abandoned lands can safely store climate-warming greenhouse gasses and enhance biodiversity. While historical cropland abandonment is well documented, patterns of future abandonment are not well-understood. The first stage of our project examines where croplands are projected to be abandoned globally under several future climate change scenarios, and where abandonment overlaps with important areas for biodiversity and climate change mitigation. We find that cropland abandonment will be widespread in 2050, but the amount and location of abandoned croplands vary by climate scenario. Projected abandoned croplands consistently overlap with areas important to biodiversity and carbon sequestration, highlighting valuable conservation opportunities. 

The next stage of our analysis focuses on Brazil, an essential region for global climate change mitigation and biodiversity preservation. Given a conservation budget, our analysis identified regions of Brazil where projected abandoned lands can be managed to maximize benefits for biodiversity and carbon sequestration. Our findings indicate that while patterns of cropland abandonment vary by region, conservation goals can be met by leveraging abandoned lands with existing policy mechanisms. Through our research and analysis, we hope to inform effective policies and management strategies that balance the need for agricultural production with climate and biodiversity goals.

Methods
Rasters of abandoned cropland (globally and in Brazil) were generated by processing raw LULC data from Chen et al., 2021 between the years 2015 and 2050. Abandoned cropland was defined as any area currently classified as cropland that has a different land use classification in future projections, with the exception of land that was urbanized. Urbanized cropland was excluded from the analysis as these areas could not support our project goal of evaluating potential areas for natural restoration. Areas of global cropland abandonment were found using a series of raster calculations between future (2050) and current (2015) LULC data. Within Brazil, any area of cropland projected to be abandoned between 2020-2050 (and not urbanized) served as the input to our spatial prioritization. 

After determining where Brazilian cropland is projected to be abandoned across different climate scenarios (outlined above), we determined which should be prioritized for restoration using the prioritizr package (version 7.2.2) in R. This software uses mixed integer linear programming to flexibily build and solve spatial planning problems. Locations of abandoned cropland served as available planning units for restoration efforts. Along with these planning units, the prioritization software requires the input of the features (biodiversity and carbon) and cost (restoration estimates) to be evaluated for each planning unit, specific targets for how much of each feature should be represented in the solution, and a primary objective for solving the problem. A separate problem was formulated for each SSP scenario and their associated planning units and features data.

Each problem utilized a minimize shortfall objective, in which the solution minimizes the overall target shortfall across all features while not exceeding a specified budget. Biodiversity and carbon were weighted equally, and relative targets were set to 0.5 each (50% feature conservation). Each SSP problem was run under both low and high-budget scenarios. After formulation, each problem was solved using Gurobi Optimizer (version 9.5.2) with a 0.05 “gap to optimality”; this value represents a 5% accepted deviance from the optimal solution, reducing computational requirements. For greater user flexibility in the Shiny app, additional problems and solutions were created weighting biodiversity and carbon differently. These solutions were not evaluated in the final report.
Code for running these analyses are publicly available on the linked GitHub repositories.