Table 4.
Zonal ARIMAX models’ variables, parameters and forecasts
| Model and parameters ( p,d,q ) | AIC-fit | MAE-test |
|---|---|---|
| High-ARIMAX( 6,1,5) |
13.34 |
6.4% |
| Medium-ARIMAX (4,0,3) |
80.61 |
16.2% |
| Low-ARIMAX (4,0,4) |
137.44 |
28.6% |
|
High zone variables |
Coefficient |
p-value |
| L7 DWP |
0.033 |
p < 0.01 |
| L7 LST |
0.14 |
p < 0.001 |
| L5 NDVI |
0.83 |
p = 0.19 |
| L7 NDVI |
1.38 |
p < 0.01 |
|
Medium zone variables |
Coefficient |
p-value |
| L5 DWP |
0.076 |
p < 0.001 |
| L8 NDVI |
3.51 |
p < 0.001 |
|
Low zone variables |
Coefficient |
p-value |
| L3 LST |
0.024 |
p = 0.348 |
| L8 NDVI |
3.79 |
p < 0.001 |
| L5 DWP |
0.073 |
p < 0.001 |
|
Forecast w37-41 2013 |
MAE |
Predicted
vs.
actual incidence |
| HIGH-ARIMAX |
-29.40% |
63.0 vs. 89.9 |
| MEDIUM-ARIMAX |
40.27% |
3.77 vs. 3.48 |
| LOW-ARIMAX* | -29.59% | 0.91 vs. 1.48 |
Akaike’s information criteria (AIC) and mean average prediction error (MAE) are shown for the best-fit models of the high-, medium- and low-transmission zone ARIMAX models. MAE is calculated for the testing period of the models, i e, 2011 w19 through 2012 w18. Normalized differentiated vegetation index (NDVI), nocturnal land-surface temperature (LST) and nocturnal dew point (DWP) are shown with specific lag-times with its coefficients as well as p-values. Four-week forecasts MAE and predicted cases vs. actual cases for the four weekly forecasts of the zonal models are shown.
*ARIMAX LOW forecast only three weeks ahead, due to using LST lagged at three weeks.