1 ENSO-DRIVEN PREDICTABILITY OF TROPICAL DRY AUTUMNS USING THE SEASONAL ENSEMBLES MULTIMODEL Based on the manuscript ENSO-Driven Skill for precipitation from the ENSEMBLES Seasonal Multimodel Forecasts, in preparation. R. Manzanas rmanzanas@ifca.unican.es Instituto de Física de Cantabria, CSIC-UC, Santander (SPAIN) http://www.meteo.unican.es M.D. Frías, A.S. Cofiño and J.M. Gutiérrez EMS 2010, Zurich, 13-17 September
2 Introduction
3 Introduction Seasonal forecasting: Many applications in different socio-economic sectors such as agriculture, energy, health Several seasonal forecasting systems are run all around the world once or twice a month: NCEP CFS, POAMA, EURO-SIP Appropriate assessment of these systems. Previous studies have focused only on particular models or regions. Assessment of the skill of state-of-the-art seasonal forecasts for precipitation WORLDWIDE, using the longest to date available hindcasts from the ENSEMBLES project MULTIMODEL (MM). Robust statistical validation that permits to obtain the skill scores and their statistical significance, grid point by grid point.
4 Data
5 Data Observations Monthly observed precipitation from VASClimO 1, a worldwide quality controlled gridded dataset spanning the period 1951-2000. Gauge-based, covers only land areas (except Antarctica and Greenland). Seasonal predictions Five global coupled atmosphere-ocean models within the STREAM2 of ENSEMBLES 2, covering the period 1961-2005. Each formed by an ensemble of nine members. - UK Met Office (UKMO). - Météo France (MF). - European Centre for Medium-Range Weather Forecasts (ECMWF). - Leibniz Institute of Marine Sciences (IFM-GEOMAR). - Euro-Mediterranean Centre for Climate Change (CMCC-INGV). Validation period: 1961-2000. Resolution: 2.5º. 1 U. Schneider et al., 2008, Global Precipitation Analysis Products of the GPCC, Internet Publication, 1-12. 2 A. Weisheimer et al., 2009, ENSEMBLES: A new multi-model ensemble for seasonal-to-annual prediction. Skill and progress beyond DEMETER in forecasting tropical pacific SSTs, Geophysical Research Letters, 36.
6 Methodology
7 Methodology Tercile-based For a particular methodology season described M0 by Frías T1 et al. 1, applied Equal weights WORLDWIDE, grid and point year: by grid point. M1 T1 MODEL 1 MODEL 5 M2 M3 M4 M5 M6 M7 M8 M0 M1 M2 M3 M4 M5 M6 M7 M8 T3 1 M.D. Frías et al., 2010, Assessing the skill of precipitation and temperature seasonal forecasts in Spain: Windows of opportunity related to ENSO events, Journal of Climate, 23, 209-212. T2 T3 T1 T3 T2 T1 T3 T1 T3 T1 T3 T1 T2 T3 T1 Probabilistic forecasts P 1 =n 1 /45 P 2 =n 2 /45 P 3 =n 3 /45 n 1 = number of members forecasting T1 n 2 = number of members forecasting T2 n 3 = number of members forecasting T3
8 Methodology Two probabilistic scores to assess the skill of the predictions, the Roc Skill Area (RSA) 1 and the Hit Rate (HIR) 1. RSA takes values in [-1,1]. HIR takes values in [0,1]. In general, the MM outperforms all the single models in all seasons. 1 I.T. Joliffe and D.B. Stephenson, 2003, Forecast verification - A practitioner s Guide in Atmospheric Sciences, John Wiley & Sons.
9 Results
10 Results Global Skill The MM skill was firstly evaluated for the whole period 1961-2000; Global validation. Dry tercile in SON 1-month leadtime 4-months leadtime RSA 1.0 0.8 0.6 0.4 0.2 4-months leadtime 0.0 Confidence level: 95% (bootstrapping with 1000 series). High skill over the tropics, especially in north-eastern South America, the Somali Peninsula, Middle East, the Malay archipelago and Australia. Skill at four months leadtime remains almost constant: Persistence of the predictability signal.
11 Results ENSO-Driven Skill Sensitivity study to determine the ENSO-driven component of the skill. The MM skill was secondly evaluated only in ENSO years; ENSO-conditioned validation. El Niño Autumn years, defined by Pozo-Vázquez et al. 1 based in the SSTs of the El Niño3 region: 1963, 1965, 1969, 1972, 1976, 1982, 1987, 1991 and 1997. Dry tercile in SON 1-month leadtime RSA 1. 0 0. 8 0. 6 0. 4 4-months leadtime Confidence level: 95% (bootstrapping with 1000 series). 1 D. Pozo-Vázquez et al.,2005, El Niño-Southern Oscillation events and associated European winter precipitation anomalies, International Journal of Climatology, 25 (1), 17-31. 0. 2 0. 0
12 ENSO-Driven Skill Results Achieving such a high ENSO-driven skill in the Malay archipelago in SON is an encouraging result, since the most severe environmental and socioeconomic impact of ENSO in Indonesia occurs in this season 1. 1 D.G.C. Kirono and N.J. Tapper, 1999, ENSO rainfall variability and impacts on crop production in Indonesia, Physical Geography, 20 (6), 508-519.
13 Results ENSO-Teleconnections Analysis of ENSO-teleconnections in order to assess the contribution of this phenomenon to the global skill. The frequency of occurrence of each tercile in ENSO years is calculated grid point by grid point and season by season. ENSO-teleconnection: Frequency of occurrence significantly different from 1/3. Dry tercile in SON 1.0 0.8 0.6 0.4 0.2 0.0 Frequency of occurrence Confidence level: 95% (Chi-Square test). ENSO-teleconnected regions are basically those in which skill is highest. ENSO: Skill-driver, especially in north-eastern South America and the Malay archipelago.
14 Attribution of the Global Skill to ENSO Results A large RSA for the dry tercile does not necessarily imply skill in forecasting dry conditions, but it may be the result of correctly forecasting non-dry (normal or wet) conditions. It is important to determine whether the skill comes from the prediction of the occurrence of the dry tercile or from the prediction of its non occurrence. HIR discriminates this origin. The whole period 1961-2000 is split into two subperiods: ENSO years and non- ENSO years. The MM skill is computed, for both subperiods, in terms of the HIR of dry occurrences, in order to check if the origin of the skill is different.
15 Results Attribution of the Global Skill to ENSO Non-ENSO years (neutral and cold phases): Dry tercile observed SON Dry tercile not observed RSA 1.0 0.8 0.6 0.4 0.2 0.0 Threshold considered for hits/mistakes = 1/3. In north-eastern South America and the Malay Archipelago (the most ENSOteleconnected areas), the skill comes to a great extent from the forecast of the non occurrence of the dry tercile.
16 Attribution of the Global Skill to ENSO ENSO years (warm phase): Dry tercile observed SON Dry tercile not observed Results RSA 1.0 0.8 0.6 0.4 0.2 0.0 Threshold considered for hits/mistakes = 1/3. The skill in north-eastern South America and the Malay Archipelago comes primarily from the forecast of the occurrence of the dry tercile. The latter results point out that the global skill found in these areas is fully driven by ENSO, which leads to skilful predictions of dry/normal-wet conditions during its warm/neutral-cold phases, respectively.
17 Conclusions
18 Conclusions High skill for predicting dry conditions over the tropics in SON for the whole period of study 1961-2000, especially in north-eastern South America, the Somali Peninsula, Middle East, the Malay archipelago and Australia. This skill increases during warm ENSO events. Regions exhibiting high ENSO-driven skill are strongly teleconnected with this phenomenon. Skill has a different origin during warm/neutral-cold phases of ENSO.
19 Thank you! R.Manzanas rmanzanas@ifca.unican.es Instituto de Física de Cantabria, CSIC-UC, Santander (SPAIN) http://www.meteo.unican.es