Utility Week - authoritative, impartial and essential reading for senior people within utilities, regulators and government
Issue link: https://fhpublishing.uberflip.com/i/378571
24 | 12th - 18th September 2014 | UtILItY WeeK Operations & Assets Market view B efore numerical modelling was invented in the 1950s, weather fore- casting was subjective and based on empirical rules. If the barometer was drop- ping, low pressure was on the way. Thank- fully things have moved on considerably since then. The development of numerical forecast- ing systems made it possible to model the movement of the atmosphere, theoretically providing a deeper understanding of what weather events were likely in the near future. The problem is that numerical forecasts are oen wrong. This is mainly because of poor under- standing of the natural volatility (chaos) inherent in the atmosphere and the relatively limited num- ber of observations used to understand the natural state of the atmosphere. In order to create any numerical forecast, you first need to create a set of initial condi- tions – a three-dimensional approximation of the atmosphere "right now". To do this, a model starts with standard weather observa- tions, which it superimposes on to a grid. Once the forecasting system has an initial state, it uses the equations of motion and other time-dependent equations to deter- mine how the atmosphere will evolve over time. In other words, it creates a forecast. This is called a deterministic forecast: a single set of inputs results in a single set of outputs. However, this approach leaves the user exposed to volatility that it is impossible to estimate from the output of the model. All a user sees is, for example, an expected tem- perature of 22ºC. This is where ensemble forecasting come in, to significantly improve the understand- ing of the volatility in the atmosphere. The initial conditions described above are altered slightly or "perturbed". Then the forecast is run multiple times. Imagine a particular observation point is capable of observing temperature to a granu- larity of 0.5ºC. This is good enough for most uses, but for a forecast system a difference of around 1ºC from the coolest to warmest pos- sible temperatures is substantial. In order to estimate within this potential difference, ensemble forecasting systems will tweak the most likely observation slightly a number of times, running the forecast model in the same way. The difference is that now, instead of the system producing a single answer, it produces a range of possibilities. Satellite data is used to create the ini- tial conditions for models, including cloud top temperatures and atmospheric water vapour information. Specialised satellites can also use the ripples on water surfaces to estimate wind speeds. These remotely sensed observations are secondary to "on site" observations and are used when these "normal" observations are not available. The primary benefit of ensemble forecast- ing is that it provides the user with a meas- ure of objective forecast certainty. Because the model is run multiple times, the output is a range of possibilities. This measure of certainty is particularly useful for severe weather prediction, because the costs associ- ated with these events can be significant. Imagine that a windfarm's safe operating window is up to a wind speed of 40km/h. If the forecast is for 35km/h, no alert would be triggered, but in reality there may be a 30 per cent probability of winds exceed- ing this safety threshold. That is, in three out of every ten similar circumstances the winds would damage the turbines. A deterministic forecast would not provide warning of this. For windfarm operators, understanding this risk and making contingency plans can save a great deal of time and money. Similarly, if a user is particularly con- cerned about the probability of heavy rain, the forecast range can be compared with this threshold to give a probability of it being exceeded. This helps the user to clearly understand what the risk is for them. During winter when icing and snow is important, a user can use range data to pro- vide the worst-case scenario forecast. When looking at longer range forecasts, say over five or ten days, range data also allows us to track each of the possible solutions, to see whether they agree and if or where they dif- fer significantly. Another application of ensemble forecast- ing is in cases where a very accurate "single answer" is needed. Ensemble models can help users understand the probability den- sity function, or spread of what is likely. This calculation usually results in a single most likely solution, as in the graphs, right. Ensemble forecasting has primarily been adopted by users interested in the weather forecast more than five days ahead, mostly organisations involved in the energy indus- try, including generation and trading. However, it is becoming increasingly pop- ular with a broader range of industries from logistics planners to stock co-ordinators, to Gauging the weather Deterministic forecasts do not factor in the chaotic nature of weather systems. Ensemble forecasting offers more value, security and savings to utilities, explains Byron Drew. Ensemble forecasting allows a user to fully understand the risks they are being exposed to data visualisation