8.3 Results and Discussion
8.3.4 The Value of Seasonal Rainfall Outlook Information
in reality, then fluctuations in simulated yields can be expected to exceed fluctuations in the actual production. This problem may be addressed by accepting a more diverse range of agronomic inputs, which will allow for certain portions of the crops in the HCZ to stress, while other portions experience better growing conditions. In the simulations conducted for this study, diverse conditions were simulated when climate data terminated prematurely and different analogue future scenarios had to be assumed. It is suspected that for some HCZs these simulations could indirectly have addressed the lack of diversity in agronomic inputs, since some analogue seasons will allow crops to stress, while others will allow better growing conditions. These suspicions are confirmed for some mills by the increase in forecast errors as the proportion of the crop simulated by analogue data decreases. Negative forecast skills for the Pongola mill in September and December depict that these forecasts were worse than simply assuming the mill’s long-term mean as a forecast. The fact that accuracy at these mills deteriorated over the winter season suggests that simulations of winter growing conditions could be oversimplified. These observations point to the need for further research into the diversification of soil, harvest cycle, water use and irrigation strategy input variables within HCZs.
Even though forecast skills (Table 8.2b) were often less than 30%, the system frequently forecasted the direction of future yields correctly, compared to the previous season (cf. Table 8.2c). For example, the directional skill for the entire industry in the September prior to the milling season was 73%, while the equivalent forecast skill was only 11%. Generally, the directional skill was higher for coastal mills dependent on rainfed sugarcane (e.g. Gledhow, Amatikulu, Darnall, Sezela) rather than for inland mills and mills which were supplied by large areas of irrigated sugarcane.
displays mean actual yields and two forecasts, one based on a neutral outlook and another based on the actual SAWS seasonal rainfall outlook.
Seasonal rainfall outlook information was the most valuable in January, where industry scale skills were improved by an average of 7.9% over the period 1998 to 2002. The value of rainfall outlook information deteriorated from January towards the drier winter months, e.g. May. Also, rainfall outlook information issued in September (early spring) often had a negative effect on forecast skills. Generally, the Eston, Amatikulu and Gledhow mills reflected promising improvements in forecast skills when using the seasonal rainfall outlook (cf. Table 8.3). Also, forecasts based on the seasonal rainfall outlook issued for the Eston and the Umzimkulu mills were consistently more accurate than those based on a neutral outlook (cf. Table 8.3).
The selection of analogue seasons in this study was based on three month accumulated rainfall outlook information. Several other seasonal climate outlooks are also available. These range from monthly accumulated rainfall outlooks, zero to three month mean temperature outlooks and three to six month mean temperature and rainfall outlooks. In addition, several crop response driving parameters, such as the number of rainy days, the number of cold, hot, wet and dry spells and the number of days with temperature and rainfall events exceeding certain threshold amounts may be more valuable than forecasts of means. A close partnership between climate forecasters and the sugar industry (and other sectors) may be necessary to ascertain and make advances on potential synergies.
KwaZulu-Natal Mpumalanga
SEPTEMBER prior to the milling season
30°S
31°E
-30.0 – -25.0 -25.0 – -20.0 -20.0 – -15.0 -15.0 – -10.0 -10.0 – -5.0 -5.0 – 0.0 0.0 – 5.0 5.0 – 10.0 10.0 – 15.0 15.0 – 20.0 20.0 – 25.0 25.0 – 30.0
Change in forecast skill (%)
Legend
-30.0 – -25.0 -25.0 – -20.0 -20.0 – -15.0 -15.0 – -10.0 -10.0 – -5.0 -5.0 – 0.0 0.0 – 5.0 5.0 – 10.0 10.0 – 15.0 15.0 – 20.0 20.0 – 25.0 25.0 – 30.0
Change in forecast skill (%)
Legend
Figure 8.5a Spatial changes in forecast skill as a result of utilising seasonal rainfall outlook information as opposed to assumptions of neutral outlooks for forecasts issued in the September prior to the milling season
KwaZulu-Natal Mpumalanga
30°S
31°E
JANUARY
-30.0 – -25.0 -25.0 – -20.0 -20.0 – -15.0 -15.0 – -10.0 -10.0 – -5.0 -5.0 – 0.0 0.0 – 5.0 5.0 – 10.0 10.0 – 15.0 15.0 – 20.0 20.0 – 25.0 25.0 – 30.0
Change in forecast skill (%)
Legend
-30.0 – -25.0 -25.0 – -20.0 -20.0 – -15.0 -15.0 – -10.0 -10.0 – -5.0 -5.0 – 0.0 0.0 – 5.0 5.0 – 10.0 10.0 – 15.0 15.0 – 20.0 20.0 – 25.0 25.0 – 30.0
Change in forecast skill (%)
Legend
Figure 8.5b Spatial changes in forecast skill as a result of utilising seasonal rainfall outlook information as opposed to assumptions of neutral outlooks for forecasts issued in January
KwaZulu-Natal Mpumalanga
30°S
31°E
MARCH
-30.0 – -25.0 -25.0 – -20.0 -20.0 – -15.0 -15.0 – -10.0 -10.0 – -5.0 -5.0 – 0.0 0.0 – 5.0 5.0 – 10.0 10.0 – 15.0 15.0 – 20.0 20.0 – 25.0 25.0 – 30.0
Change in forecast skill (%)
Legend
-30.0 – -25.0 -25.0 – -20.0 -20.0 – -15.0 -15.0 – -10.0 -10.0 – -5.0 -5.0 – 0.0 0.0 – 5.0 5.0 – 10.0 10.0 – 15.0 15.0 – 20.0 20.0 – 25.0 25.0 – 30.0
Change in forecast skill (%)
Legend
Figure 8.5c Spatial changes in forecast skill as a result of utilising seasonal rainfall outlook information as opposed to assumptions of neutral outlooks for forecasts issued in March
KwaZulu-Natal Mpumalanga
30°S
31°E
MAY
-30.0 – -25.0 -25.0 – -20.0 -20.0 – -15.0 -15.0 – -10.0 -10.0 – -5.0 -5.0 – 0.0 0.0 – 5.0 5.0 – 10.0 10.0 – 15.0 15.0 – 20.0 20.0 – 25.0 25.0 – 30.0
Change in forecast skill (%)
Legend
-30.0 – -25.0 -25.0 – -20.0 -20.0 – -15.0 -15.0 – -10.0 -10.0 – -5.0 -5.0 – 0.0 0.0 – 5.0 5.0 – 10.0 10.0 – 15.0 15.0 – 20.0 20.0 – 25.0 25.0 – 30.0
Change in forecast skill (%)
Legend
Figure 8.5d Spatial changes in forecast skill as a result of utilising seasonal rainfall outlook information as opposed to assumptions of neutral outlooks for forecasts issued in May
KwaZulu-Natal Mpumalanga
30°S
31°E
SEPTEMBER
-30.0 – -25.0 -25.0 – -20.0 -20.0 – -15.0 -15.0 – -10.0 -10.0 – -5.0 -5.0 – 0.0 0.0 – 5.0 5.0 – 10.0 10.0 – 15.0 15.0 – 20.0 20.0 – 25.0 25.0 – 30.0
Change in forecast skill (%)
Legend
-30.0 – -25.0 -25.0 – -20.0 -20.0 – -15.0 -15.0 – -10.0 -10.0 – -5.0 -5.0 – 0.0 0.0 – 5.0 5.0 – 10.0 10.0 – 15.0 15.0 – 20.0 20.0 – 25.0 25.0 – 30.0
Change in forecast skill (%)
Legend
Figure 8.5e Spatial changes in forecast skill as a result of utilising seasonal rainfall outlook information as opposed to assumptions of neutral outlooks for forecasts issued in September
Table 8.3 The change in forecast skill between yield forecasts based on a neutral seasonal rainfall outlook and those based on actual seasonal rainfall outlooks. Positive values depict an improvement in forecast skill when actual outlook information was used. Underlined values are statistically significant (P<0.05) according to a binomial distribution test
Change in forecast skill (%)
No. Mill 1 Sepy-1 1 Jan 1 Mar 1 May 1 Sep Mean
1 Komati -3.49 -7.86 6.64 -4.09 0.03 -1.75
2 Malelane -11.5 6.47 0.81 -1.51 0.05 -1.13
3 Pongola -2.96 1.85 6.16 -1.20 -3.20 0.13
4 Umfolozi 9.29 -1.32 -6.04 -1.66 -0.43 -0.03
5 Entumeni -20.6 7.65 -0.14 -2.38 0.06 -3.09
6 Felixton -2.07 8.35 0.81 0.93 -0.17 1.57
7 Amatikulu -11.8 27.74 15.04 9.36 -0.59 7.95
8 Darnall -12.0 15.67 13.24 6.54 0.84 4.86
9 Gledhow -3.15 17.82 12.79 6.45 0.69 6.92
10 Union Co-op 1.54 -1.03 -2.15 -0.43 -1.32 -0.68 11 Noodsberg 0.92 -1.28 -2.32 -0.37 -1.12 -0.83
12 Maidstone -5.08 4.19 5.66 2.59 -0.45 1.38
13 Eston 10.33 12.38 14.83 3.41 3.91 8.97
14 Sezela -9.50 -0.57 0.92 6.42 0.57 -0.43
15 Umzimkulu 0.53 2.10 0.25 2.06 0.26 1.04
Mean -3.97 6.14 4.43 1.74 -0.06
Industry -0.93 11.57 10.11 8.20 -0.47