Confusion matrices for model predictions using EFM data 60 min prior to any observed lightning within the central cape lightning warning circle for 28,872 observations during 10–30 June 2013
| Observed | Observed | ||||||
|---|---|---|---|---|---|---|---|
| 0 | 1 | 0 | 1 | ||||
| Predicted | 0 | 27,597/27,708 | 111/27,708 | Predicted | 0 | 27,578/27,708 | 130/27,708 |
| 99.5% | 0.5% | 99.5% | 0.5% | ||||
| 1 | 87/1,164 | 1,077/1,164 | 1 | 51/1,164 | 1,113/1,164 | ||
| 7.5% | 92.5% | 4.4% | 95.6% | ||||
| Mean imputation | Spatiotemporal imputation | ||||||
| Observed | Observed | ||||||
|---|---|---|---|---|---|---|---|
| 0 | 1 | 0 | 1 | ||||
| Predicted | 0 | 27,597/27,708 | 111/27,708 | Predicted | 0 | 27,578/27,708 | 130/27,708 |
| 99.5% | 0.5% | 99.5% | 0.5% | ||||
| 1 | 87/1,164 | 1,077/1,164 | 1 | 51/1,164 | 1,113/1,164 | ||
| 7.5% | 92.5% | 4.4% | 95.6% | ||||
| Mean imputation | Spatiotemporal imputation | ||||||
Note(s): A prediction or observed value of “0” corresponds to no lightning, whereas a “1” denotes observed or predicted lightning within the lightning warning circle. Results indicate sizable improvements in the positive identification of lightning when spatiotemporal imputation is used to complete the EFM dataset
Source(s): Table courtesy of authors