| Popis: |
Background Machine learning (ML) methods and remote sensing data were used to build multi-level multi-scale resource selection models and predictive maps onto the extended landscape for jaguars (Panthera onca) in the Brazilian Pantanal. Objectives were to compare multiple predictive modeling and exploratory modeling approaches. Included in the analysis, multi-scale raster grains (30m, 90m, 180m, 360m, 720m, 1440m), GPS collaring temporal levels (point, path, and step) and model data structural levels (group, individual, case-control).Methods Multi-scale multi-level data subsets were fit with explanatory and predictive statistical methods. Conditional logistic regression, generalized additive modeling (GAM), and classification regression trees, such as random forests (RF) and gradient boosted regression tree (GBM) were compared for their utility to the study. Model evaluation, using training and testing data in a k-fold cross-validation approach, determined the AUC, Kappa, and TSS for model evaluation and comparison. · Results Results indicated that the multi-level, multi-scale techniques improved model outputs. Overall, larger level models and those that used multi-scale raster grains showed the best model evaluation. The highest-ranked model was the multi-scale path selection function GBM and was one of the broadest levels of data. ·Conclusions Results indicated that multi-level, multi-scale models produced mixed results of applicability across models and levels. The identification of the appropriate temporal scale and statistical model needs careful consideration in predictive mapping efforts. |
