H33D-1651
Bias correction of daily precipitation in south-central Chile using NCEP CFSv2

Wednesday, 16 December 2015
Poster Hall (Moscone South)
Lina Mabel castro Heredia1,2, Tamara Maass1, Francisco I Suarez1 and Bonifacio Fernandez1, (1)Pontifical Catholic University of Chile, Santiago, Chile, (2)Pontificia Universidad Católica de Valparaiso, Valparaiso, Chile
Abstract:
Hydroelectric power plant operations are heavily influenced by the streamflow forecasts on their basins. In Chile, these forecasts are based on historical observations. However, this approach has reached its limit of quality and reliability, being difficult to adapt to current weather conditions (climate change), to extreme weather conditions, and to ungauged basins. In this work, we evaluated the bias correction of NCEP-CSv2 daily precipitation with the aim of incorporating this forecast into a real-time hydrological forecasting system. Bias correction was performed using two approaches of the Quantile Mapping (QM) method: a) a polynomial fit (APo) applied to the differences between the forecasted and observed cumulative distribution functions (CDFs) for the training period; and b) using a Gamma probability distribution (APb) to fit the forecasted and observed CDFs. The bias correction was applied at two locations in south-central Chile: over the valley and in the Andes mountains. To estimate the CDFs and the QM fitting models in the training period, historical records and data from the CFSv2 Reforecast model (between 1995 and 2009) were used. The bias correction evaluation was done between 2011 and 2014 with the forecast of the CFSv2 model. The uncorrected CFSv2 results show that the mid-term forecasts (six months) have a high correlation (r>0.5) for the first days of the forecast (2 weeks), but an important underestimation in the observed data from both the valley and the mountain. After applying the bias correction (APo or APb), the errors of the corrected forecasts decrease in relation to the uncorrected CFSv2 forecasts, with a noticeable improvement for the first forecasted days (being the APo errors lower than those of the APb). In the long term, and as might be expected, the errors increase: the peak precipitation is underestimated and the null rainfall is overestimated.