Statistical downscaling to 0.25° spatial resolution TRMM monthly precipitation explores the potential and limitations of using the relatively short (here 15 years) but spatially extensive and complete precipitation archive, that is satellite-based and merged with station information. Downscaling models relate reanalysis circulation to TRMM, utilizing principal component regression (PCR) and canonical correlation analysis (CCA). Results are demonstrated for two contrasting regions: Northeastern Brazil (NEB, tropical, distinct wet and dry season) and Central Italy (CIT, mid-latitude/Mediterranean, complex terrain). Models are constructed for individual months (M1) and by pooling three months (M3) to increase model-training sample size when downscaling to the central month of the three. Cross-validated skill with M1 is promising and is noticeably more consistent and higher with M3, e.g., mean skill at the grid-box scale rises from r=0.44 to r=0.59 for CIT (averaged over all months), and from r=0.58 to r=0.68 for the NEB wet season months. Spatial structure of the downscaling models (as revealed by CCA modes) supports a clear expression of orography in the precipitation anomaly fields. Application to a global coupled model climate change scenario (2012-2050) generates plausible downscaled time-series and fields. For CIT, results (skill, spatial structure) are consistent with those produced using a station-only gridded (0.25°) dataset for the extended period 1979- 2012. The overall impression gained is that TRMM data enable estimation of skilful downscaling relationships, at least for some locations. Developments drawing on longer datasets to adjust the downscaled fields will likely further increase the utility of a record like TRMM.
- Keywords: rainfall, regression model, downscaling, correlation, principal component, orography, wind.