How many rain gauges is a constellation of satellites worth?

Abstract:

Satellite-derived global quantitative precipitation estimation products have been around for several decades. As their accuracy and resolution improves these products are widely used in fields such as climate science, hydrological modeling and even impact studies and risk management. The estimates provided by the satellites are often compared to rain gauge measurements for validation, performance evaluation and uncertainty quantification. However, considering the difference between theses two types of observation systems, in particular in terms of spatial and temporal resolution, comparing gauge and satellite products is far from trivial. Some of the state-of-the-art precipitation products report global precipitation estimations every 15min mapped on spatial grids as fine as 4km. It would however be a hasty assertion to claim that this is equivalent to having a gauge reporting every 15 minutes every 4 km all over the globe.

Pixel-to-gauge comparison is limited by the spatial representativeness of a gauge covering an area smaller than a square meter relatively to an at least 16 million times larger pixel. Additionally, at short time scales, as the number of tips per unit of time gets relatively low, tipping rain gauge measurements may become inaccurate, particularly for low-intensity precipitation. On the satellite side, spatial patterns in estimated precipitation fields may be smooth, misplaced or distorted.

When comparing two gridded precipitation products, an alternative to pixel-to-pixel comparison is to look at the space-time spectral coherence between the two fields as a function of frequency/scale in a transformed domain such as the Fourier or wavelet domains. Here, we use this methodology to compare the performance of several satellite products (IMERG, GSMAP, CMORPH and PERSIANN-CCS) to the performance of a “simulated rain gauge network”, obtained by sub-sampling the MRMS radar-derived precipitation fields over the continental US. We estimate that for hourly rainfall accumulations over the South-Eastern part of the continental US, the best performing satellite products are equivalent to a gauge network with a density of one gauge every 50 km. These results can be extrapolated globally, providing a global estimate of “the worth” of satellite products as compared to ground gauge measurements.