Prediction of monsoon rainfall for a mesoscale Indian catchment based on stochastical downscaling and objective circulation patterns

In this study a stochastical approach for generating rainfall time series based on objective circulation patterns (<i>CP</i> is applied to the mesoscale Anas catchment in North West India. This <i>CP</i> based approach was developed and successfully applied in the humid and temperate climate of Central Europe. The objective of the study was to find out whether this approach is transferable to a catchment in North West India with a totally different semi arid climate. For the Anas catchment it was possible to identify a <i>CP</i> classification scheme consisting of 12 <i>CP</i>s defined in a window between 5° N 40° E and 35° N 95° E, which explained the space-time variability of observed rainfall at 10 stations in the Anas catchment. Based on the classification scheme, NCAR pressure data from 500 hPa level were classified into a <i>CP</i> time series for the period of 1964?1994, which was in turn used as meteorological forcing for multivariate stochastical rainfall simulations with a daily time step. On the monthly time scale the model performed well. Except for stations Udaigarh and Bhabra the average annual cycle of monthly rainfall and rainy days in a month was matched well. The frequency distributions of monthly rainfall at different stations were also captured well. Correlation coefficients between simulated and observed monthly rainfall were larger than 0.85 at each station. Within a long term simulation of 30 years the model yielded promising predictions for monthly as well as for seasonal rainfall totals, but showed also clear deficiencies in capturing the very extremes and inter-decadal variability of monsoon strength. In this respect, the introduction of additional predictors such as SST anomalies and wind direction classes promised the most substantial model improvements.