Global Agricultural Expansion Potential Maps at 300m Resolution

Global maps at a 10 arc-second resolution model the potential for land conversion into three categories of cropland. Artificial Neural Network models trained on conversion data from 2007-2020 achieved cross-validation AUC scores of 0.88-0.93. The data, from a study by Mirza Čengić at Radboud University Nijmegen, is intended to downscale projections for global environmental assessments.

Use Cases

• Downscaling coarse global land change model projections based on high-resolution potential maps.
• Assessing environmental impacts of agricultural expansion based on topography, climate, soil, and accessibility variables.
• Hindcasting historical land conversion patterns using models validated on data from 1992-2006.
• Mapping conversion potential for specific cropland categories like cropland-only or mosaic landscapes.

Strengths

• High spatial resolution of approximately 300 meters at the equator.
• Model performance validated with AUC values between 0.88 and 0.93.
• Covers three distinct agricultural land categories derived from ESA CCI land cover data.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The maps provide potential for conversion, assuming drivers of expansion remain constant from the training period.

Provenance

Source

Radboud University Nijmegen

Collection Method

Artificial Neural Network models trained on locations of past conversions (2007-2020) from ESA CCI land cover data, using predictor variables for topography, climate, soil, and accessibility.

Time Range

Model training based on conversions from 2007-2020; hindcast validation from 1992-2006.

Geography

Global