Central to this problem is the lack of skillful nowcast guidance, which undermines forecasters’ confidence when issuing warnings. This is in turn partly attributable to the scarcity of reliable surface hail reports for verification, which presents a significant barrier for developing new forecast tools. For example, recent research has found that human-derived hail reports in the US Storm Data archive have a significant negative size bias that amplifies with increasing hail size. These findings almost certainly apply to hail reports made outside the US.
One possibility for mitigating problems associated with human-derived hail reports would be to use radar-derived products such as hail-size estimates from the Maximum Expected Size of Hail (MESH) product as a proxy for surface reports. Radar-derived data have the added advantage of providing high spatiotemporal resolution data and accurate location information in near-real time. Before one can confidently implement such a product, however, it needs to be demonstrated that hail-size estimates from MESH are reliable and accurate over a wide range of sizes and storm environments. A variety of issues make this difficult. How do we verify MESH given the myriad of issues associated with traditional storm data reports? Are there alternative data sources that could help us mitigate problems with traditional severe storm reports? For example, is it feasible to use scalable images of hail posted on social media as a source of accurate hail size measurements?
Here we will address these questions and present findings from a detailed evaluation of the MESH product from eight C-band radars on the Canadian prairies for three summers from 2014 to 2016. Preliminary results indicate that although only a small fraction of images posted on social media are useful for hail verification purposes the sheer volume of reports yields a sizeable database of high confidence and high quality cases. Using these unique data, we find that MESH is superior to the existing hail algorithm available for use at the Meteorological Service of Canada, with a small bias and mean-absolute error. Importantly, we find that the size distribution of MESH data follows a gamma distribution (as is found for hail pad data). We also highlight some of the key challenges and limitations facing users of data posted on social media and provide some potential solutions.