Thanks Firoz Ahmad. SDSM can only downscale daily data. But I need 1 hourly data if any model can downscale which I do not know. Some people use GEV distribution, like in one of above paper, but that is only for single day. I need long time series like 2020-2040 etc.
This may be a helpful link http://onlinelibrary.wiley.com/doi/10.1029/2010WR010342/full
I would estimate that the success at downscaling may vary with type of storms. Usually it is best to collect detailed precipitation data from the beginning, but I recognize that many historic records are daily readings and some sites lack funding for a recording raingauge. I am not experienced in the daily to hourly conversion, but had an idea that under some circumstances may be useful.
Usually we try to predict streamflow from rainfall, and various approaches to do that. If you happen to have some detailed stream gauging stations such as at 15 minute or hourly intervals in the vicinity of your daily rainfall records, you might be able to look at the stream response, time to peak, etc. and sort of reverse calculate or apportion the hourly rainfall estimates from the timing and changes in streamflow rates. Since rainfall is not perfectly distributed anyway across a watershed or basin, the stream is not going to reflect just the point value of a rainfall station, but is the composite across the hydrologic unit. If you are familiar with unit hydrograph approach, if the effective unit of rainfall has a unit of flow signature, then why not consider taking a unit of flow changes over the day to reallocate the daily rainfall into hourly? Since there is often a time lag between stream response and the daily rainfall, adjusting for that time lag would be needed.