Table 3. Root mean square error of ten-year ahead predictions (RMSEP) obtained with two different datasets for several statistical models: linear regression (L), quadratic regression (Q), cubic regression (C), dynamic linear models with and without trend (DLMs, DLM0), and Holt-Winters with and without trend (HWs, HW0).
| Models | |||||||||
| Dataset | Quantity | Units | L | Q | C | DLMs | DLM0 | HWs | HW0 |
| France | RMSEP | t ha−1 | 0.94 | 1.02 | 1.08 | 0.90 | 0.79 | 0.99 | 0.79 |
| Differencea | % | 19.02 | 29.52 | 36.90 | 14.35 | 0.00 | 25.42 | 0.08 | |
| Global | RMSEP | t ha−1 | 0.76 | 1.04 | 1.22 | 0.98 | 0.70 | 1.13 | 0.70 |
| Differencea | % | 8.68 | 47.66 | 74.46 | 39.81 | 0.16 | 60.78 | 0.00 | |
a
The differences with respect to the lowest RMSEP values are expressed as a percentage of the lowest RMSEP values (RMSEPmin); Difference = 100*(RMSEP – RMSEPmin)/RMSEPmin.