Geospatial Assessment of Current and Future Land Suitability for Peruvian Amylaceous Maize (Zea mays L.) Using Random Forest Modeling

Sivmny V. Valqui-Reina , Cleyver Rivera, Carlos I. Arbizu , Alex J. Vergara 

Climate change poses an increasing threat to crop suitability and food security, particularly for varieties of great cultural and economic importance, such as Peruvian starchy maize (Zea mays L.), whose optimal growing areas remain poorly characterized at the national level. This study presents the first comprehensive geospatial assessment of current and future land suitability for starchy corn in Peru, integrating bioclimatic, edaphic, and topographic variables using a machine learning approach. A dataset of georeferenced accessions, combined with pseudo-absence data, was used to train and validate a model using the Random Forest algorithm based on 29 bioclimatic, edaphic, and topographic predictors. The model demonstrated high statistical reliability (AUC-training = 0.98; AUC-validation = 0.99; precision = 0.971; recall = 0.962; F1 score = 0.966; accuracy = 0.966), with temperature-related variables (BIO09, BIO05, BIO15, and annual mean temperature) identified as the most important predictors. Projections made using the MIROC6 general circulation model under four shared socio-economic scenarios (SSP1-2.6, SSP2-4.5, SSP3-7.0, SSP5-8. 5) for the periods 2061–2080 and 2081–2100 indicate a net expansion of highly suitable areas ranging from 20.76% to 45.46%, depending on the scenario. The most optimistic scenario (SSP1-2.6) projects the greatest gain (approximately 45% by 2070), while the most pessimistic (SSP3-7.0) shows an initial contraction followed by a marked recovery by 2100. These findings provide a robust geospatial framework for climate-smart agricultural planning, breeding programs, and sustainable land-use policies for Peruvian starchy maize in the context of future climate change.