Utilization of a high-resolution daily rain-gauge based precipitation product for an improvement of the Asian monsoon forecasts

Quantitative estimate of latent heat release over Asian monsoon region is primarily important for understanding the monsoon mechanism and for improving the forecasts. Over a complex terrain, simulating latent heat release by 4-dimensional assimilation technique is one of the most challenging topics. Hence, we started developing daily rain-gauge based grid precipitation dataset as the main activity of the Asian Precipitation – Highly Resolved Observational Data Integration Towards the Evaluation of Water Resources (APHRODITE) project. The product, APHRO reproduce precipitation of Monsoon Asia, especially over Himalayas well, since off-line data of Pakistan, India, Nepal, Bhutan, Bangladesh, China and other Asian country's data were involved. Hence, APHRO is being used for improving precipitation forecasts as well as for validating estimated precipitation by models and satellites.

Multimodel superensemble developed by Florida State University (FSU, Krishnamurti et al., 1999) combines multiple model forecasts based on their past performance (training phase) to make a consensus forecast. Since observed precipitation reflects local characteristics such as orography, quantified high-resolution precipitation products are useful for downscaling the coarse model outputs. In this study, APHRO_MA_V0902 (Yatagai et al., 2009) and TRMM3B43 (Huffman et al., 2007) are used for downscaling and as teacher data in the training phase of superensemble. We used 7 years (1998-2004) monthly precipitation for June, July and August over Asian monsoon region (60-150E, 0-50N), and results of four coupled climate models, Geophysical Fluid Dynamics Laboratory (GFDL), National Centers for Environmental Prediction (NCEP), Seoul National University (SNU) and University of Hawaii (UH). TRMM3B43 was adjusted by a linear regression for the 7 years (modified TRMM).

First, coarse resolution precipitation data from the four GCMs were interpolated to 0.25 degree grid and regression coefficients a and b were obtained by least square linear fit of the model interpolated precipitation with that of the high resolution observation datasets (APHRO and modified TRMM). For the seasonal climate forecasts, synthetic forecasts for each of the member models are created from their original forecast time series using linear regression with their observed counterpart in the Empirical Orthogonal Functions (EOF) space (Krishnamurti et al., 2002). A cross-validation technique was adopted in which the year to be forecasted was excluded from the calculations in obtaining the regression coefficients. These regression coefficients were applied to the forecast year to obtain downscaled model forecasts for that year. The above steps were repeated for each year of the period 1998 to 2004.

After getting 0.25 degree forecast precipitation, skill scores, such as the correlation of modeled rain to the observation (APHRO) seasonal estimate, corresponding root mean square (RMS) errors, and equitable threat scores (ETS) are calculated. We also compute the ensemble mean of seasonal forecasts of our suite of multimodels that do not use any TRMM or gauge data sets. The main results derived from seasonal superensemble forecast are as follows.

1. The seasonal forecasts of Asian monsoon precipitation were considerably improved by using APHRO rain-gauge based data or the modified TRMM product. These forecasts are much superior to those provided by the best model of our suite and the ensemble mean.

2. Use of a statistical downscaling and a synthetic superensemble methodology for multimodel forecasts of seasonal climate provide a major improvement in the prediction of precipitation at higher resolution.

3.The availability of a dense rain-gauge network based analysis was essential for the success of this work.

We are preparing for daily summer monsoon forecast by using The Observing System Research and Predictability Experiment (THORPEX) Interactive Ground Global Ensemble (TIGGE) database. To use such a huge and common database starting from October 2006, APHRO should be updated till 2008, and “period of the day” should be adjusted. The results over East Asia will be shown in the presentation.