Inter-annual variation of statistically downscaled February precipitation from 1982-2009 (red bar) and the corresponding observations (blue bar) at Elon, Israel. The black bar on the far right is the multi-year climatology (mean of the 1982-2009 observations).
Fine-scale spatial variation of the statistically downscaled February 2009 precipitation based on the CFS forecasts issued in November 2008 (red) and corresponding station observations (blue). The reference line is bias-corrected CFS forecast taken at a grid point (32.5oN, 32.5oE) over Israel.
Advancements in scientific understanding of the climate system and in climate modeling have promoted real-time seasonal climate prediction at several national centers. Such seasonal prediction provides reasonable global perspectives and outlooks of the climate in few months advance. However, its usefulness has been limited due to its coarse resolution (~200 km). To support regional water resource study and management, RAL had been developing statistical and dynamical algorithms to downscale the global seasonal precipitation predictions with additional fine-scale information driven by the specific local forcing. This project is sponsored by the Israel Water Authority.
Description
To bridge the gaps between the needs and the global scale seasonal forecasts by major national/international centers, RAL, in collaboration with the Institute of Biological Research of Israel, has developed an integrated statistical and dynamical downscaling system. The downscaling technology is based on the “k-nearest neighbor” (KNN) algorithm and a WRF-based four-dimensional data assimilation model. The system is driven by the NCEP Climate Forecast System (CFS) seasonal forecasts. The initial system has been accomplished recently. After passing intensive testing and evaluation, the system has been transferred to Israel Water Authority for real-time operation seasonal precipitation forecasting. Retrospective forecast experiments using 1982-2009 NCEP/DOE reanalysis and historic rain-gauge measurements have been conducted at 18 rain-gauge stations in the region to test statistical algorithm. Figure 1 shows the retrospective forecasts and corresponding observations for Februarys from 1982-2009 at Elon (33.06oN, 35.21oE). It is apparent that the interannual variation in February rainfall is well reproduced. The downscaling algorithm performs really well in some years, such as 1983, 2004, 2005, 2009 and the extreme wet situation in 2003. On the other hand, the test also exposes unsatisfactory performance in years of 1984, 1996, 1997 and 1999. Figure 2 is the downscaled forecasts for February 2009 at ten stations based on the CFS prediction issued in November 2008. Overall the forecasts in three months advance compares pretty well with the measurements. The downscaling algorithm has proven to add valuable information for water resources planning and climate-associated risk managements to the original coarse grid global model forecasts. Further study to overcome the deficiencies of the algorithm is under way.
