The prediction of hail. Part II: the movement of hail cells

Abstract This part of the two-part paper presents prediction models of the location and areal extent of hail cells. The data sets and evaluation procedures are described in Part I. A new elliptic model for the approximation of the radar contours has been developed which considers the apparent asymmetry of hail cells in a high-shear environment. The parameters of the elliptic model are analysed in detail in order to find a regression model for predicting these parameters. The regression model also contains predictor variables which characterize the early stage of new convective developments above the 0°C level. A comparison of the observed and predicted elliptic contours allow an estimate of the predictability of hailfall location as a function of the forecast period. Small-area hailfalls (hailed area ∼ 1 km2) are found to be predictable up to a few minutes only. Large-area hailfalls (hailed area ∼ 10 km2) and extremely large-area hailfalls (hailed area ∼ 100 km2) are predictable up to about 15–20 min and 30 min, respectively. It was also found that the gain in time for hailfall prediction is 1–3 min only when using a sophisticated statistical prediction model for prediction instead of a simple persistence model. This latter finding is explained by the long-tailed frequency distribution of the parameters of the elliptic approximation.