Predictability and skill of boreal winter forecasts made with the ECMWF Seasonal Forecasting System II

The performance of boreal winter forecasts made with the European Centre for Medium-Range Weather Forecasts (ECMWF) System 11 Seasonal Forecasting System is investigated through analyses of ensemble hindcasts for the period 1987-2001. The predictability, or signal-to-noise ratio, associated with the forecasts, and the forecast skill are examined. On average, forecasts of 500 hPa geopotential height (GPH) have skill in most of the Tropics and in a few regions of the extratropics. There is broad, but not perfect, agreement between regions of high predictability and regions of high skill. However, model errors are also identified, in particular regions where the forecast ensemble spread appears too small. For individual winters the information provided by t-values, a simple measure of the forecast signal-to-noise ratio, is investigated. For 2 m surface air temperature (T2m), highest t-values are found in the Tropics but there is considerable interannual variability, and in the tropical Atlantic and Indian basins this variability is not directly tied to the El Nino Southern Oscillation. For GPH there is also large interannual variability in t-values, but these variations cannot easily be predicted from the strength of the tropical sea-surface-temperature anomalies. It is argued that the t-values for 500 hPa GPH can give valuable insight into the oceanic forcing of the atmosphere that generates predictable signals in the model. Consequently, t-values may be a useful tool for understanding, at a mechanistic level, forecast successes and failures. Lastly, the extent to which t-values are useful as a predictor of forecast skill is investigated. For T2m, t-values provide a useful predictor of forecast skill in both the Tropics and extratropics. Except in the equatorial east Pacific, most of the information in t-values is associated with interannual variability of the ensemble-mean forecast rather than interannual variability of the ensemble spread. For GPH, however, t-values provide a useful predictor of forecast skill only in the tropical Pacific region.

Interannual to decadal climate predictability in the North Atlantic: A multimodel-ensemble study

Ensemble experiments are performed with five coupled atmosphere-ocean models to investigate the potential for initial-value climate forecasts on interannual to decadal time scales. Experiments are started from similar model-generated initial states, and common diagnostics of predictability are used. We find that variations in the ocean meridional overturning circulation (MOC) are potentially predictable on interannual to decadal time scales, a more consistent picture of the surface temperature impact of decadal variations in the MOC is now apparent, and variations of surface air temperatures in the North Atlantic Ocean are also potentially predictable on interannual to decadal time scales. albeit with potential skill levels that are less than those seen for MOC variations. This intercomparison represents a step forward in assessing the robustness of model estimates of potential skill and is a prerequisite for the development of any operational forecasting system.

Moist singular vectors and the predictability of some high impact European cyclones

The ECMWF full-physics and dry singular vector (SV) packages, using a dry energy norm and a 1-day optimization time, are applied to four high impact European cyclones of recent years that were almost universally badly forecast in the short range. It is shown that these full-physics SVs are much more relevant to severe cyclonic development than those based on dry dynamics plus boundary layer alone. The crucial extra ingredient is the representation of large-scale latent heat release. The severe winter storms all have a long, nearly straight region of high baroclinicity stretching across the Atlantic towards Europe, with a tongue of very high moisture content on its equatorward flank. In each case some of the final-time top SV structures pick out the region of the actual storm. The initial structures were generally located in the mid- to low troposphere. Forecasts based on initial conditions perturbed by moist SVs with opposite signs and various amplitudes show the range of possible 1-day outcomes for reasonable magnitudes of forecast error. In each case one of the perturbation structures gave a forecast very much closer to the actual storm than the control forecast. Deductions are made about the predictability of high-impact extratropical cyclone events. Implications are drawn for the short-range forecast problem and suggestions made for one practicable way to approach short-range ensemble forecasting. Copyright © 2005 Royal Meteorological Society.

Progress on “Changing coastlines: data assimilation for morphodynamic prediction and predictability”

The task of assessing the likelihood and extent of coastal flooding is hampered by the lack of detailed information on near-shore bathymetry. This is required as an input for coastal inundation models, and in some cases the variability in the bathymetry can impact the prediction of those areas likely to be affected by flooding in a storm. The constant monitoring and data collection that would be required to characterise the near-shore bathymetry over large coastal areas is impractical, leaving the option of running morphodynamic models to predict the likely bathymetry at any given time. However, if the models are inaccurate the errors may be significant if incorrect bathymetry is used to predict possible flood risks. This project is assessing the use of data assimilation techniques to improve the predictions from a simple model, by rigorously incorporating observations of the bathymetry into the model, to bring the model closer to the actual situation. Currently we are concentrating on Morecambe Bay as a primary study site, as it has a highly dynamic inter-tidal zone, with changes in the course of channels in this zone impacting the likely locations of flooding from storms. We are working with SAR images, LiDAR, and swath bathymetry to give us the observations over a 2.5 year period running from May 2003 – November 2005. We have a LiDAR image of the entire inter-tidal zone for November 2005 to use as validation data. We have implemented a 3D-Var data assimilation scheme, to investigate the improvements in performance of the data assimilation compared to the previous scheme which was based on the optimal interpolation method. We are currently evaluating these different data assimilation techniques, using 22 SAR data observations. We will also include the LiDAR data and swath bathymetry to improve the observational coverage, and investigate the impact of different types of observation on the predictive ability of the model. We are also assessing the ability of the data assimilation scheme to recover the correct bathymetry after storm events, which can dramatically change the bathymetry in a short period of time.

Northern hemisphere snow: measurement, modelling and predictability

Northern hemisphere snow water equivalent (SWE) distribution from remote sensing (SSM/I), the ERA40 reanalysis product and the HadCM3 general circulation model are compared. Large differences are seen in the February climatologies, particularly over Siberia. The SSM/I retrieval algorithm may be overestimating SWE in this region, while comparison with independent runoff estimates suggest that HadCM3 is underestimating SWE. Treatment of snow grain size and vegetation parameterizations are concerns with the remotely sensed data. For this reason, ERA40 is used as `truth' for the following experiments. Despite the climatology differences, HadCM3 is able to reproduce the distribution of ERA40 SWE anomalies when assimilating ERA40 anomaly fields of temperature, sea level pressure, atmospheric winds and ocean temperature and salinity. However when forecasts are released from these assimilated initial states, the SWE anomaly distribution diverges rapidly from that of ERA40. No predictability is seen from one season to another. Strong links between European SWE distribution and the North Atlantic Oscillation (NAO) are seen, but forecasts of this index by the assimilation scheme are poor. Longer term relationships between SWE and the NAO, and SWE and the El Ni\~no-Southern Oscillation (ENSO) are also investigated in a multi-century run of HadCM3. SWE is impacted by ENSO in the Himalayas and North America, while the NAO affects SWE in North America and Europe. While significant connections with the NAO index were only present in DJF (and to an extent SON), the link between ENSO and February SWE distribution was seen to exist from the previous JJA ENSO index onwards. This represents a long lead time for SWE prediction for hydrological applications such as flood and wildfire forecasting. Further work is required to develop reliable large scale observation-based SWE datasets with which to test these model-derived connections.

Case studies in interannual to decadal climate predictability

The predictability of ocean and climate variables is investigated, using a perfect model-based case study approach that recognises that predictability is dependent on the initial climate state. In line with previous studies, large scale ocean variables, show predictability for several years or more; by contrast, the predictability of climate variables is generally limited to, 2 years at most. That predictability shows high sensitivity to the initial state is demonstrated by predictable climate signals, arising in different regions, variables and seasons for different initial conditions. The predictability of climate variables, in the second year is of particular interest, because this is beyond the timescale that is usually considered to be the limit, of seasonal predictability. For different initial conditions, second year predictability is found in: temperatures in southeastern, North America (winter) and western Europe (winter and summer), and precipitation in India (summer monsoon) and in the tropical, South Atlantic. Second year predictability arises either from persistence of large-scale sea surface temperature (SST) and, related ocean heat content anomalies, particularly in regions such as the North Atlantic and Southern Ocean, or from mechanisms, that involve El Nino Southern Oscillation (ENSO) dynamics.

Time series multistep-ahead predictability estimation and ranking

A predictability index was defined as the ratio of the variance of the optimal prediction to the variance of the original time series by Granger and Anderson (1976) and Bhansali (1989). A new simplified algorithm for estimating the predictability index is introduced and the new estimator is shown to be a simple and effective tool in applications of predictability ranking and as an aid in the preliminary analysis of time series. The relationship between the predictability index and the position of the poles and lag p of a time series which can be modelled as an AR(p) model are also investigated. The effectiveness of the algorithm is demonstrated using numerical examples including an application to stock prices.

Processes governing the predictability of the Atlantic meridional overturning circulation in a coupled GCM

The processes that govern the predictability of decadal variations in the North Atlantic meridional overturning circulation (MOC) are investigated in a long control simulation of the ECHO-G coupled atmosphere–ocean model. We elucidate the roles of local stochastic forcing by the atmosphere, and other potential ocean processes, and use our results to build a predictive regression model. The primary influence on MOC variability is found to come from air–sea heat fluxes over the Eastern Labrador Sea. The maximum correlation between such anomalies and the variations in the MOC occurs at a lead time of 2 years, but we demonstrate that the MOC integrates the heat flux variations over a period of 10 years. The corresponding univariate regression model accounts for 74.5% of the interannual variability in the MOC (after the Ekman component has been removed). Dense anomalies to the south of the Greenland-Scotland ridge are also shown to precede the overturning variations by 4–6 years, and provide a second predictor. With the inclusion of this second predictor the resulting regression model explains 82.8% of the total variance of the MOC. This final bivariate model is also tested during large rapid decadal overturning events. The sign of the rapid change is always well represented by the bivariate model, but the magnitude is usually underestimated, suggesting that other processes are also important for these large rapid decadal changes in the MOC.

The impact of large scale atmospheric circulation patterns on wind power generation and its potential predictability: a case study over the UK

Over recent years there has been an increasing deployment of renewable energy generation technologies, particularly large-scale wind farms. As wind farm deployment increases, it is vital to gain a good understanding of how the energy produced is affected by climate variations, over a wide range of time-scales, from short (hours to weeks) to long (months to decades) periods. By relating wind speed at specific sites in the UK to a large-scale climate pattern (the North Atlantic Oscillation or "NAO"), the power generated by a modelled wind turbine under three different NAO states is calculated. It was found that the wind conditions under these NAO states may yield a difference in the mean wind power output of up to 10%. A simple model is used to demonstrate that forecasts of future NAO states can potentially be used to improve month-ahead statistical forecasts of monthly-mean wind power generation. The results confirm that the NAO has a significant impact on the hourly-, daily- and monthly-mean power output distributions from the turbine with important implications for (a) the use of meteorological data (e.g. their relationship to large scale climate patterns) in wind farm site assessment and, (b) the utilisation of seasonal-to-decadal climate forecasts to estimate future wind farm power output. This suggests that further research into the links between large-scale climate variability and wind power generation is both necessary and valuable.

International evidence on the predictability of returns to securitized real estate assets: econometric models versus neural networks

The performance of various statistical models and commonly used financial indicators for forecasting securitised real estate returns are examined for five European countries: the UK, Belgium, the Netherlands, France and Italy. Within a VAR framework, it is demonstrated that the gilt-equity yield ratio is in most cases a better predictor of securitized returns than the term structure or the dividend yield. In particular, investors should consider in their real estate return models the predictability of the gilt-equity yield ratio in Belgium, the Netherlands and France, and the term structure of interest rates in France. Predictions obtained from the VAR and univariate time-series models are compared with the predictions of an artificial neural network model. It is found that, whilst no single model is universally superior across all series, accuracy measures and horizons considered, the neural network model is generally able to offer the most accurate predictions for 1-month horizons. For quarterly and half-yearly forecasts, the random walk with a drift is the most successful for the UK, Belgian and Dutch returns and the neural network for French and Italian returns. Although this study underscores market context and forecast horizon as parameters relevant to the choice of the forecast model, it strongly indicates that analysts should exploit the potential of neural networks and assess more fully their forecast performance against more traditional models.