4 Value of long-term inflow forecasts
4.4 Reliability of simulated annealing
4.5.3 Utilization of the results
The results of the two parts of this study can be combined. The usability of the new model in real time forecasting can be evaluated based on the results of the second part of this study. It was shown that if the distribution of the relative forecast errors is too wide, forecasts should not be used. Although in the first part of the study the forecasts were made for periods as long as six months, it was discovered in the second part that in the optimal operation of the River Kymijoki system only forecasts of lead-times up to 2-3 months are necessary. It is, however, of hydrologic interest to analyse the possibilities to forecast as far as six months ahead in these study basins. The accuracy of the new model for forecasting a period of three months ahead was about =0.16 (16%) on April 1 and =0.2 (20%) on October 1 and for forecasting a period of two months ahead about the same. These figures can be compared with the results in Figure 44, Table 43 and Table 44. The optimal forecast length for operation of the system by using forecasts of this accuracy is about 2-3 months. If compared with the optimal operation (4 months of perfect forecasts available), the use of the forecast model with this accuracy would cause losses less than 1% in hydropower production annually. Neither would the number of absolute water level violations increase dramatically, compared with the optimal operation. An improvement in the accuracy
of the inflow forecasts from =0.2 to =0.1 in Lake Päijänne (a1=0.0) would increase
hydropower production by about 0.6% (6.0 GWh annually) in the River Kymijoki basin.
In Lake Pyhäjärvi the accuracy of the model was at best about =0.5 on April 1 and close to =1.0 on October 1, although these figures were slightly unreliable because of
a few large errors. The optimal forecast length by using this kind of forecast would still be around 5-6 months. However, the losses in hydropower production caused by inaccurate forecast could be as much as 10%. By improving the forecast accuracy in Lake Pyhäjärvi from =0.5 to =0.3, hydropower production would increase about 2- 3%.
Because the value of the inflow forecasts is dependent on the characteristics of the reservoirs, the usability of the forecasts models should not be evaluated purely based on the goodness-of-fit criteria. Also the system for which the forecast model is planned should be presented. For a very large reservoir, indicative forecasts are adequate and neither the maintenance of expensive forecast systems nor the expensive projects aiming at improvements in forecast accuracy is justified if the expected improvement in accuracy is small. For example, a model can be sufficiently good with
a low value of R2 if long-term forecasts are used and the target system has a large live
capacity. An approach given in this study would be valuable in approximating the usefulness of the forecast model.
In Finland, the live capacities of the regulated lakes are relatively small. In some of the rivers that are most vulnerable to floods (e.g. River Kyrönjoki), the live capacity is only about 5% of the annual discharge. Normally, the live capacities of the watercourses are less than 100% of the annual runoff. The live capacities of the two most important Finnish rivers for hydroelectric power production, River Kemijoki and River Oulujoki, are around 40% and 60% of the annual discharge downstream, respectively. In 2005, a report was published on the possibilities to increase hydropower production in Finland (The Ministry of Trade and Industry, 2005). The potential energy lost as spill was approximated to be as much as 750 GWh annually. The potential increase in hydropower production by improving the accuracy of the inflow forecasts and thereupon regulation was not discussed. The issue was also ignored in the latest report on the possibilities to increase hydropower production in Finland (Oy Vesirakentaja, 2008)
Let us assume that the results of the study could be generalized. By using a conservative approximation that the results of the River Kymijoki system are valid all over Finland, it might be possible to increase hydroelectric power production by a minimum of 90 GWh (0.7%·13000 GWh) annually, if perfect inflow forecasts were available. Because the live capacities of the most important lake-river systems in Finland are larger than the live capacity of the River Kymijoki system, and because the whole outflow of Lake Päijänne can not be regulated, the additional value of a perfect forecast would probably be at least few percentages. In Lake Pyhäjärvi, where the live capacity is 57% of the annual inflow, it was approximated to be as much as 9%. Thus, the percentage might be as large as 5%. This would increase hydropower production by about 650 GWh annually. Thus it is possible that a great part of the spillage might be avoided without updating the regulation licenses if the accuracy of the long-term forecasts could be improved. At the same time flood and drought problems would decrease.
In reality, perfect forecasts are a utopia, and thus the potential increase in hydropower production by improving forecast accuracy lies below these approximations. As a matter of fact, improvement in accuracy is very difficult to achieve. As the case studies showed, the improvement in forecast accuracy in Lake Päijänne, for example, from =0.2 to =0.1 would benefit hydropower production in River Kymijoki system by about 0.6 % annually. In Lake Pyhäjärvi, the improvement in forecast accuracy, for
example, from =0.5 to =0.3 would benefit hydropower production by about 2-3% annually. Thus if this is the accuracy of the current real-time forecast models, the possibilities of improving hydropower production in Finland by improving forecast accuracy might be as much as 0.5-2% (65 GWh-260 GWh). At the same time, flood and drought problems would, of course, decrease. These figures can be compared with approximations of the increase in production if the man-made reservoir in Vuotos in the River Kemijoki basin or the reservoir in Kollaja in the River Iijoki basin were built. The potential production increase based on the Vuotos reservoir is 325 GWh/a and on the Kollaja reservoir 200 GWh/a (Oy Vesirakentaja, 2008). Thus the possibilities for improving the forecast accuracy and operation policies should be further studied.