Diagnosing the Distribution of Seasonal Mean Precipitation

While most seasonal mean atmospheric quantities are well-described as Gaussian probability distributions, seasonal mean precipitation remains non-Gaussian in many regions, particularly in the semi-arid drought-prone regions of the world. Understanding the precipitation distribution in such regions is critical to improving prediction of drought and pluvial events, but research into the physical mechanisms giving rise to it has been limited. To investigate the shape of these distributions, satellite/rain-gauge blended observations, land-based gridded rain-gauge observations, and reanalysis precipitation products are used. With these precipitation estimates, the seasonal mean precipitation distributions are investigated in all seasons and around the globe.

We find that the distributions depart significantly from a Gaussian shape mainly in regions of large-scale descent. This non-Gaussianity is associated with significant positive skewness. The Gamma and the Log-Normal distributions have been widely used to describe positively skewed precipitation distributions. We find, however, that both are poor fits to the observed seasonal mean precipitation distributions at many locations around the globe. Both also fail to account for the frequency of no precipitation, a quantity of vital economic interest in many regions of the world. The Gamma fit is poor in regions of strong descent. The Log-Normal fit is poor in regions of strong descent and also of strong ascent, where the observed distributions are largely Gaussian. Comparisons among the precipitation datasets show that the results of this study are robust to the choice of dataset.