大理州红花大金元种植适宜性评价方法研究OA

[Objective]The study aims to compare the performance of the Maximum Entropy(MaxEnt)model and machine learning(ML)algorithms in predicting the potential suitable planting areas for Honghuadajinyuan tobacco in Dali Prefecture,so as to reveal the applicability and advantages of different modeling strategies.[Method]Using 309 field-surveyed occurrence records and environmental covariates related to climate,soil and topography,the differences in potential distribution patterns generated by the two approaches were assessed.[Result]The MaxEnt model showed strong overall predictive performance,with an AUC of 0.941 for the training set and 0.873 for the test set.Among the machine learning models,Bagging-MLP performed well in suitability type classification,with an AUC of 0.937 for the training set and 0.852 for the test set.However,in suitability level classification,the test set AUC decreased to 0.703,indicating reduced generalization ability.The RF model performed best in unsuitable level classification,achieving an AUC of 0.992 for the training set and 0.846 for the test set.Variable importance rankings differed across methods,with MaxEnt highlighting topographic and climatic factors,while ML emphasized the combined influence of soil nutrients and vegetation indices.Both approaches revealed a high degree of spatial overlap in predicted suitable areas,prima-rily located in northeastern Weishan,northeastern Eryuan and northwestern Midu.[Conclusion]Machine-learning algorithms outperform MaxEnt in assessing the suitability of Honghuadajinyuan tobacco cultivation in Dali Prefecture,delivering more continuous and stable predic-tions that can scientifically guide regional planning for this characteristic crop.