The term nowcasting is a contraction of “now” and “forecasting”; it refers to the sets of techniques devised to make short term forecasts, typically in the 0 to 12 hour range. Historically, nowcasting techniques have been based on simplified heuristic approaches where the current echo pattern was tracked in the past and extrapolated into the future, sometimes with some modifications to try to account for storm development and decay and/or for decreasing predictability with time. More and more though, nowcasting techniques are becoming more sophisticated, sometimes up to the point of becoming full-fledged numerical weather prediction models.
We pursue three approaches to nowcasting:
1) A “traditional” nowcasting approach, where we try to improve upon the usual extrapolation of radar data from single radars or from national composites. Improvements include the use of ensembles of extrapolations, the tweaking of forecasts by including the diurnal cycle of precipitation, the calculation of probability of precipitation instead of computing the most likely scenario, and the merging with independently derived numerical weather prediction. Our current system, the McGill Algorithm for Prediction by Lagrangian Extrapolation (MAPLE), has been licensed to WDT Technologies.
2) A morphing of model output approach, where the time evolution of past model output is adjusted in time, position, and intensity to match the latest radar data and the forecasting is being done using the modified model output.
3) A mesoscale numerical weather forecasting approach based on the assimilation of radar data (Chung et al. 2010, and references therein).
Independently of the above, we have also attempted to quantify the error associated with traditional nowcasting approaches and attempted to determine if one could find predictors of the magnitude of these errors from the radar data itself (Fabry and Seed 2009).