Artificial intelligence to model the potential distribution of Agave durangensis

We used four artificial intelligence algorithms; MaxEnt, climate space model (CSM), back propagation neural network (BPNN) and vector support machine (VSM) to model the potential distribution of Agave durangensis. In the field, 300 georeferenced records of agaves were obtained, for which information on 18 climates and three topographic variables was retrieved from geospatial databases. With the presence records and the variables, the 80% of the data was used for modeling and the remaining 20% was used to validate the model by estimating the receiver operating characteristic (ROC). Two models had an acceptable performance with ROC> 0.9. We observed that MaxEnt predicted agave distributions in canyons that did not correspond to the distribution of this species. The BPNN model predicts 95% of the areas that coincide with the natural distribution of the agaves. Therefore, the BPNN algorithm was the most accurate for predicting areas for agave repopulation.