Table 3. Summary statistics of jackknife test of environmental variable importance, evaluation measures, and maximum training sensitivity plus specificity threshold results for MaxEnt model 1 (sites with CFA≥5%) and model 2 (sites with CFA prevalence ≥15%).
| Model 1 (CFA≥5%) | Model 2 (CFA≥15%) | |
| Variable contribution to model training gain (%) | ||
| Land cover | 34.3 | 23.8 |
| Human Influence Index (HII) | 20.9 | 1.5 |
| LSTday* (hot-dry season) | 19.6 | 22.4 |
| Distance to water bodies | 6.1 | 11.7 |
| NDVI** (hot-dry season) | 5.4 | 1.0 |
| LSTday (rainy season) | 2.4 | 13.7 |
| Altitude (DEM) | 0.2 | 9.1 |
| Model evaluation measures | ||
| AUC (SD)*** | 0.866 (0.045) | 0.892 (0.074) |
| CORprev**** (p-value) | 0.117 (0.234) | 0.355 (<0.001) |
| Threshold dependent sensitivity | 68.8% | 76.9% |
| Threshold dependent specificity | 46.6% | 64.5% |
| Threshold cut-off probability value | 0.412 | 0.465 |
Only the 7 predictors that were ranked in the top three of at least one of the two models are included. The top three predictors for each model are highlighted in bold.
*LST; Land Surface Temperature.
**NDVI; Normalized Difference vegetation Index.
***AUC; the area under the Receiver Operating Characteristic curve (and standard deviation).
**** CORprev is the Pearsons product moment correlation between model logistic probability and the measured CFA prevalence at survey sites.