Use of APHRODITE Rain Gauge–Based Precipitation for Improving Middle East Seasonal Precipitation Forecasts by the Superensemble Method
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Wednesday, 7 January 2015
A multimodel superensemble developed by the Florida State University combines multiple model forecasts based on their past performance (training phase) to make a consensus forecast. Because observed precipitation reflects local characteristics such as orography, quantitative high-resolution precipitation products are useful for downscaling coarse model outputs. The Asian Precipitation–Highly-Resolved Observational Data Integration Toward Evaluation of Water Resources (APHRODITE) is used for downscaling and as training data in the superensemble training phase. As reference rain gauge-based precipitation, we used APHRO_ME_V1101 (Yatagai et al. 2012), with 0.25-degree resolution over the Middle East (APHRO hereafter). Since winter is rainy season, we only used December, January and February (DJF) data for 10-years (1997/1998 – 2006/2007). First, the winter 30 months starting from December 1997 till February 2007 are selected and monthly mean precipitation are computed. The region that are subjected to the superensemble computation is the same that of APHRO_ME_V1101 (20°E-65°E, 15°N-45°N). We used simulated precipitation of the five coupled general circulation model (CGCM) outputs which are the part of CMIP5 project. The five models are from Centre National de Recherches Meteorologiques (CNRM), two version models from NASA Goddard Institute for Space Studies (GISS), Meteorological Research Institute of Japan (MRI) and Norwegian Climate Centre (NorESM).
For seasonal climate forecasts, a synthetic superensemble (SSE) technique (Krishnamurti et al., 2009, Krishnamurti and Kumar, 2012) was used. The results were compared with ensemble forecasts lacking observation data and singular model outputs. The dense rain-gauge network dataset (APHRODITE) considerably improved winter precipitation forecasts. The superensemble shows highest correlation and lowest root mean square error against the benchmark APHRODITE throughout the simulation period (1997/1998 – 2006/2007). Skill scores of the SSE experiment is much superior to single model simulations and ensemble forecast lacking observation data throughout all rainfall intensity. Only the bias errors when heavy precipitation occurred SSE does not show the best performance. Availability of a dense rain-gauge network is imperative to success of the seasonal forecast as shown for India (Krishnamurti et al., 2009) and for Asian monsoon region (Yatagai et al., 2014).