Experimental Quantification of the Variability of the Raindrop Size Distribution at Small Scales

Reliable quantitative precipitation estimation is crucial to better understand and eventually prevent water-related natural hazards (floods, landslides, avalanches, ...). Because rainfall is highly variable in time and space, precipitation monitoring and forecasting is a complex task. In addition, the variability of rainfall at small scales (for instance within the radar pixel) is still poorly understood. Knowledge of the rain drop size distribution (DSD) is of primary concern for precipitation estimation and in particular weather radar. To better understand the variability of the DSD at small scales, a network of optical disdrometers (Parsivel) has been designed and set up. The instruments are fully autonomous in term of power supply and data transmission. The network of 16 disdrometers has been deployed over a typical operational weather radar pixel (1 × 1 km2) in Lausanne, Switzerland, for 16months collecting DSD data at a high temporal resolution (30 s). The sampling uncertainty associated with Parsivel measurements has been quantified for different quantities related to the DSD, using a 15-month data set collected by two collocated disdrometers. Using a geostatistical approach, and in particular variograms, the spatial structure of quantities related to the DSD has been quantified. The analyses have been conducted on 36 rainfall events that have been grouped according to three types of rainfall (i.e., convective, transitional and frontal). It shows a significant variability, i.e., larger than the one induced by the sampling process, of the different quantities of interest. The observed spatial structure is significant for temporal resolution below 30 min from which it is difficult to distinguish between the natural variability and the one induced by the sampling process. The impact of the observed variability of the DSD on radar rainfall estimators is investigated focusing on two different radar power laws (the classical Z-R law for conventional radar and the R-Kdp law for polarimetric radar). The parameters of the power laws are estimated at different spatial scales: at the single station scale, at the aggregate of stations scale (aggregate of point measurements) and at the pixel area scale (average over all the stations). First, it shows clear distinct groups of power law parameters according to the type of rainfall. Moreover, the observed variability of these parameters is significantly larger than the variability induced by the sampling process of the instruments. The observed variability in power law parameters can be responsible for deviation in terms of rain amounts at the single station scale ranging from -5 to +15% the one estimated at the pixel (average) scale. The original contributions in this work are: (1) the design and deployment of an innovative network of autonomous disdrometers Parsivel over a typical operational weather radar pixel (1 km2), (2) the quantification of the sampling uncertainty associated with Parsivel measurements, (3) the quantification of the spatial variability of different quantities related to the DSD within this typical radar pixel and (4) the quantification of the influence of this spatial variability of the DSD on radar rainfall estimation (for conventional and polarimetric radar). For illustration, the results are important: (i) to illustrate the added value of the network of disdrometer that has been designed and the interest for various environmental fields (meteorology, hydrology, risks of water-related natural hazards, ...), (ii) for a better knowledge of the spatial variability of the DSD at different scales which should helps improving radar rainfall estimation, (iii) for the quantification of the errors associated with the extension of relationships derived at a specific location to larger domains (e.g., pixel) and (iv) for the ground validation of numerical weather model.