The most known use of wind and wave forecasts is for public safety. In the state of Florida, for instance, hurricane-related forecasts are of crucial importance for the safety of the inhabitants. Thanks to accurate forecasts, appropriate preven-tive measures can be taken to guarantee people safety and avoid casualties.
Wind and wave forecasts are used in the energy sector, especially now that onshore and offshore wind energy has taken a leading role in the development of renewable energy solutions (Pinson et al., 2007, 2012). Energy production monitoring or management requires very precise information about the environ-ment around an energy production site of interest, in near-real time and for the coming minutes to days. For example, the last-minute cancellation of an off-shore wind farm maintenance operation because of rough weather can be highly costly. Weather forecasts are similarly used for energy trading, where traders need visibility into weather changes that will impact demand and prices suffi-ciently in advance. In the actual context of world energy policy changing and led by renewable energy sources, the importance of weather forecasts is growing fast. They will be a necessity for some countries like Denmark that expects to produce 100% of its energy from renewable resources in 2050 (Mathiesen et al., 2009).
Wind and wave forecasts are also used for sailing or maritime transport applica-tions e.g. ship routing (Hinnenthal, 2008), where they generally support finding the “best route” for ships. For most transits this will mean the minimum tran-sit time while avoiding significant risk for the ship. The goal is not to avoid all adverse weather conditions but instead to find the best balance between min-imizing transit time, fuel consumption and not placing the vessel at risk with weather damage and crew injury.
1.2 Ensemble forecasting
Short-term weather forecasts for lead times of a few hours are usually issued based on purely statistical methods e.g. from time series analysis. From 6 hours to days ahead, the most accurate type of weather forecasts are the Numerical Weather Predictions (NWP). Unlike statistical methods, they are flow depen-dent and therefore much more efficient for appraising medium-range weather evolutions. Employing a NWP model entails relying on computer resources to solve fluid dynamics and thermodynamics equations applied to the Atmosphere, in order to predict atmospheric conditions hours to days ahead. An estimated initial state of the Atmosphere, built by collecting observational data all over
the world, is necessary as a starting point for NWP models. However, the At-mosphere can never be completely and perfectly observed due to measurement accuracy and limited observational coverage. Thus the initial state of NWP models will always be slightly different from the true initial state of the At-mosphere. In the early 60’s, Edward Lorenz showed that the dynamics of the Atmosphere were highly sensitive to initial conditions, that is, two slightly dif-ferent atmospheric states could finally result, in the future, in two very difdif-ferent sets of weather conditions. This is also known as the "butterfly effect", naively saying that the motion of the flies of a butterfly could lead to a storm on other side of the world. Such considerations on the sensitivity of initial conditions in numerical weather prediction was the origin for the subsequent development of ensemble forecasting. Ensemble forecasting is based upon the idea that forecast error characteristics, resulting from a combination of initial conditions errors and model imperfections, can be estimated. It is assumed that initial condi-tions uncertainties can be modelled by perturbing the initial state and that model deficiencies can be represented by a stochastic parametrisation of NWP models (see Chapter 2.2.1).
2010−01−01 2010−01−03 2010−01−05 2010−01−07 2010−01−09 2010−01−11 2010−01−13 2010−01−15
0
10 metre wind speed (m s**−1)
Ensemble members Ensemble mean Observation
Figure 1.2: Example of ensemble forecast trajectories of surface wind speed issued at time t for lead times between +06h ahead and +168h ahead. The predicted ensemble mean is represented by the black line, the ensemble members by the grey lines and the observations in red points.
1.2 Ensemble forecasting 5
In practice, an ensemble forecast is composed ofN different forecasts, referred to as ensemble members, issued at the same time t for the same future time t+k (hence with k denoting the lead time). Ensemble members follow their own trajectories. Each of them is relevant in the sense that it obeys to the same physics equations and the same parametrization. Figure 1.2 shows an example with 51 ensemble members of surface wind speed from the European Centre for Medium-range Weather Forecasts (ECMWF). Ensemble forecasts from ECMWF will be further introduced in Chapter 2.2.1. It can be seen from that figure that slightly different initial states and model parametrization can lead to completely different scenarios. For instance for lead time k = +168 h, the different members predict wind speed between 2 m.s−1 and approximately 9 m.s−1.
0 5 10 15
10 metre wind speed
Figure 1.3: Example of ensemble forecast issued at a timetfor a certain time t+k. The vertical black lines represent the ensemble members and the blue one represents the ensemble mean, the curve represents the probabilistic density function computed from the 51 forecasts Ensemble forecasts for a given meteorological variable, location and lead time, can be seen as a sample of a distribution. Then, ensemble forecasts can be translated into a predictive density estimated from the N different forecasts of the ensemble prediction system. However, only dealing with predictive densi-ties for each meteorological variable, location and lead time, implies that the trajectory structure of ensemble members is lost. In more general terms, it is their spatio-temporal and multivariate dependencies that are lost.