Table 3 Districts in Sri Lanka for which inclusion of a covariate in the mean term of the best (S)ARIMA model tested improved the mean absolute relative error of out of series prediction at forecasting horizons of 1 to 4 months ahead.
District | Horizon (months) | Lag (months) | covariate | Improvement (%) |
|---|---|---|---|---|
Badulla | 4 | 4 | rainy day index, with a separate coefficient for each calendar month | 6.5 |
Gampaha | 3 | 4 | logarithmically transformed total monthly rainfall (mm) | 3.8 |
Gampaha | 4 | 4 | logarithmically transformed total monthly rainfall (mm) | 4.5 |
Mannar | 1 | 2 | logarithmically transformed total monthly rainfall (mm) | 5.2 |
Moneragala | 2 | 2 | monthly rainfall factored into quintiles | 4.1 |
Moneragala | 2 | 3 | rainy day index | 4.6 |
Moneragala | 3 | 3 | rainy day index | 3.2 |
Mullaitivu | 1 | 1 | monthly rainfall factored into quintiles | 2.6 |
logarithmically transformed total monthly rainfall (mm), with a separate | ||||
Ratnapura | 3 | 4 | coefficient for each calendar month | 3.9 |
logarithmically transformed total monthly rainfall (mm), with a separate | ||||
Ratnapura | 4 | 4 | coefficient for each calendar month | 3.6 |
logarithmically transformed total monthly rainfall (mm), with a separate | ||||
Trincomalee | 2 | 2 | coefficient for each calendar month | 8.4 |
logarithmically transformed total monthly rainfall (mm), with a separate | ||||
Trincomalee | 3 | 3 | coefficient for each calendar month | 9.2 |
Vavuniya | 4 | 4 | logarithmically transformed total monthly rainfall (mm) | 2.5 |