Spatial structure of the raindrop size distribution at small scale

Rainfall is a physical process highly variable in time and space, which makes its quantification difficult especially with the problem of the spatial representativity (point, volume) of the measurements. While investigations on the spatial variability of the raindrop size distribution (DSD) are becoming more and more common in the recent years, most of the experimental setups did not allow to fully characterize the spatial variability of the DSD for distance lags below 1 km (which is the typical size of a weather radar pixel).

In this context, a network of 16 optical disdrometers (OTT-Parsivel) has been deployed in March 2009 over a typical weather radar pixel (about 1x1 km²) in Lausanne, Switzerland. The network has been running for 16 months and has collected data (i.e., DSD spectra) at 30 s temporal resolution corresponding to 540 hours of rain and to an average rain amount of about 820 mm.

The variogram, which quantifies the spatial structure of a random field, has been estimated for different types of rain events (convective, frontal or mixed) which have been identified from visual inspection of MeteoSwiss weather radar maps. According to the geometry of the network, 120 disdrometer pairs are available with interdistances ranging from 85 to 800 m which allows investigations of the spatial variability of the DSD at small scale (below 1 km). Overall, the analyses show that independently of the quantity of interest (e.g., total concentration of drops, mass-weighted diameter, rain rate), there is a spatial structure, that is the DSD fields are "organized" and not randomly varying. In addition, the decorrelation distance appears to be larger than 1 km (hence not seen by the network). Convective events exhibit a higher absolute variability than frontal ones. For example, the absolute variability of the total concentration of drops for a distance lag of about 200 m (600 m) is in the order of 45 (70) m^-3 for convective events while it is in the order of 30 (45) m^-3 for frontal ones.

The influence of the temporal resolution on the different variograms is investigated as well. As expected, temporal integration smooths the fields and for long time steps, there is no more a clear spatial structure at the 1 km² scale and the variogram corresponds to a white noise (related to the sampling uncertainty of the instruments). The quantification of the spatial structure of the DSD will be helpful to improve the comparison between gauge and radar, to better understand the uncertainties associated with radar estimates, and improve the estimation of the spatial error covariance for data assimilation.