A Precipitation Satellite Downscaling & Re-Calibration Routine for TRMM 3B42 and GPM Data Applied to the Tropical Andes

Tuesday, 16 December 2014: 5:00 PM
Bastian Manz1, Wouter Buytaert2,3, Conrado Tobón4, Marcos Villacis2 and Fernando García5, (1)Imperial College London, Civil and Environmental Engineering, London, SW7, United Kingdom, (2)Escuela Politécnica Nacional, Quito, Ecuador, (3)Imperial College London, Civil and Environmental Engineering and Grantham Institute for Climate Change, London, SW7, United Kingdom, (4)Universidad Nacional de Colombia, Medellín, Colombia, (5)Instituto Nacional de Meteorología e Hidrología - INAMHI, Quito, Ecuador
With the imminent release of GPM it is essential for the hydrological user community to improve the spatial resolution of satellite precipitation products (SPPs), also retrospectively of historical time-series. Despite the growing number of applications, to date SPPs have two major weaknesses. Firstly, geosynchronous infrared (IR) SPPs, relying exclusively on cloud elevation/ IR temperature, fail to replicate ground rainfall rates especially for convective rainfall. Secondly, composite SPPs like TRMM include microwave and active radar to overcome this, but the coarse spatial resolution (0.25°) from infrequent orbital sampling often fails to: a) characterize precipitation patterns (especially extremes) in complex topography regions, and b) allow for gauge comparisons with adequate spatial support. This is problematic for satellite-gauge merging and subsequent hydrological modelling applications.

We therefore present a new re-calibration and downscaling routine that is applicable to 0.25°/ 3-hrly TRMM 3B42 and Level 3 GPM time-series to generate 1 km estimates. 16 years of instantaneous TRMM radar (TPR) images were evaluated against a unique dataset of over 100 10-min rain gauges from the tropical Andes (Colombia & Ecuador) to develop a spatially distributed error surface. Long-term statistics on occurrence frequency, convective/ stratiform fraction and extreme precipitation probability (Gamma & Generalized Pareto distributions) were computed from TPR at the 1 km scale as well as from TPR and 3B42 at the 0.25° scale. To downscale from 0.25° to 1 km a stochastic generator was used to restrict precipitation occurrence to a fraction of the 1 km pixels within the 0.25° gridcell at every time-step. Regression modelling established a relationship between probability distributions at the 0.25° scale and rainfall amounts were assigned to the retained 1 km pixels by quantile-matching to the gridcell. The approach inherently provides mass conservation of the downscaled pixels at the 0.25° gridcell scale. Validation was performed at 1 km pixel scale with synchronous instantaneous TPR images when available and by pixel-gauge comparison considering original error surfaces. The method is designed to be directly applicable to the standard Level 3 gridded time-series GPM product once this becomes available.