Seasonal predictability of heavy precipitation events using the CMCC-SPS system
Alessio Bellucci, Andrea Borrelli, Stefano Materia and Silvio Gualdi - CMCC, Italy
The seasonal predictability of daily precipitation extremes over Europe is assessed using the Centro EuroMediterraneo sui Cambiamenti Climatici (CMCC) Seasonal Prediction System (hereafter CMCC-SPS; Borrelli et al., 2012; Materia et al.,2013).
Extreme precipitation is defined using a percentile based counting method applied to daily gridded observations and corresponding model forecasts. Here we focus on moderate extremes, with a 10% probability of occurrence (i.e., 1 in 10 day events) so as to ensure that the selected events feature a sufficiently large frequency of occurrence and, at the same time, a sizeable impact on the environment. More specifically, we base our analysis on an index defined as the number of wet days exceeding the 90th percentile (precip90) of daily total precipitation. This type of events can be more appropriately referred to as “heavy rain” (rather than extreme) events.This analysis is restricted to the European region, using the E-OBS (Haylock et al., 2008) gridded precipitation data as observational reference.
The predictability of the extreme precipitation (i.e., the fraction of heavy rain days in a given 3-month period) is quantified against the predictability of the mean precipitation anomalies (over the same period as for the wet days) and against an empirical model based on persistence.
We analyse daily model output from 6-month long hindcasts (retrospective forecasts) for the the 1989-2005period. Each hindcast consists of 9 ensemble members, initialized with perturbed states of the atmosphere. This approach tries to sample the uncertainty in the initial conditions, using a realistic estimate for the initial states of atmosphere and oceans, based on global reanalyses (Materia et al., 2013). The model is initialized every year at four different start dates, corresponding to the first of February, May, August and November.
The forecast skill of the model ensemble mean is assessed for the extreme wet events, counted as the number of days exceeding the precip90 percentile. All seasons are concatenated starting at lead-month 1 of the forecast (DJF, MAM, JJA and SON). The anomaly correlation coefficient (ACC; Wilks, 2006) between forecast and observed anomalies is computed in order to quantify the predictive skill associated with the representation of heavy rain events in the CMCC-SPS system. The predictability of wet extremes is evaluated against mean precipitation anomalies and persistence (shown in Fig.1).
The ACC pattern from CMCC-SPS hindcasts (top panel in Fig. 1), features positive, but overall modest, values over most of the European area, highlighting an enhanced forecast skill (compared with northern-central Europe) in the Mediterranean region, including most of the Italian and Iberian peninsulas, Turkey, Middle-East and Northern Africa. On the other hand, negative ACC values (corresponding to no skill) are found over the Balkan area, and over north-western Europe.
Compared with the mean precipitation and persistence, the skill associated with the forecast of heavy rain days is clearly higher over most of Europe (see Fig. 1), although the improvement are particularly significant over the areas surrounding the Mediterranean basin. Persistence outperforms the dynamical forecasts over the Balkans and north-western Europe.
These results are broadly consistent with a similar analysis performed by Eade et al. (2012) using the UK Met Office Decadal Prediction System, where a similar enhancement of wet extremes skill (with respect to mean precipitation) is found over Europe, except for the eastern European region (see Fig. 2 in Eade et al., 2012).
Fig. 1: Skill of seasonal hindcasts for precipitation. Top: anomaly correlation patterns for (left) heavy rain days (precip90), (middle) mean precipitation and (right) persistence. Bottom-left: ACC of heavy rain days minus ACC for predictions of the seasonal mean precipitation. Bottom-right: ACC of heavy rain days minus ACC of persistence.
Concerning the North Adriatic region, relatively low skill is found, with anomaly correlations ranging from 0.1-0.2 over north-eastern Italy, to negative values over parts of Slovenia and Croatia (shown in detail in Fig.2). Therefore, the potential for using seasonal predictions as a tool to accurately forecast extreme precipitation events in advance is fairly limited for this region at the moment.
The analogies found between the present analysis and the results in Eade et al. (2012) work (based on a similarly low-resolution dynamical model) corroborate the indication that reliable predictions of rainfall extremes on seasonal timescales are still a challenge for state-of-the-art seasonal prediction systems. The poor representation of regional scale features in the low resolution models which are currently used to perform seasonal forecasts, and the limited predictability associated with mid-latitude processes (particularly pronounced for precipitation) contribute to the overall modest skill found for heavy rain events. Interestingly, while generally low, the correlations found for wet extremes are mostly higher than for the mean precipitation, pointing to a higher signal-to-noise ratio associated with this type of event as a possible explanation (Eade et al., 2012).With specific reference to the North Adriatic case study, possible ways to enhance our current seasonal forecast capabilities may rely on the use of statistical downscaling techniques, allowing for a better representation of regional scale features, based on an optimal use of observations and model results.
Fig. 2: Skill (anomaly correlation) of seasonal hindcasts for heavy rain days (fraction of days exceeding the 90th percentile) in the North Adriatic region.
Borrelli, A., Materia S., Bellucci A., Gualdi S., and A. Alessandri 2012. Seasonal Prediction System at CMCC. Research Paper, 147, Centro Euro-Mediterraneo sui Cambiamenti Climatici, Bologna, Italy. Available at http://www.cmcc.it/publications-meetings/publications/research-papers/rp0147-seasonal-prediction-system-at-cmcc.
Eade R., E. Hamilton, D. M. Smith, R. J. Graham, A. A. Scaife, 2012: Forecasting the number of extreme daily events out to a decade ahead, J.of Geophys. Res., 117, D21110.
Haylock, M.R., N. Hosfra, A.M.G. Klein Tank, E.J. Klok, P.D. Jones, and M. New, 2008: A European daily high-resolution gridded data set of surface temperature and precipitation. J.of Geophys. Res., 113, D20119.
Materia, S., A. Borrelli, A. Bellucci, S. Gualdi and A. Alessandri, 2013: Land-atmosphere initial state influences surface temperature forecast in dynamical seasonal predictions, Quarterly Journal of Royal Met. Soc., under review.
Wilks, D. S., 2006: Statistical Methods in the Atmospheric Sciences. Second Edition, International Geophysics Series, Vol. 91, Academic Press.