Spatial and temporal object-based evaluation of numerical precipitation forecasts

The Method of Object-based Diagnostic Evaluation (MODE) is one of several recently developed object-based methods for the evaluation of spatial structures in numerical weather forecasts. MODE identifies coherent features separately in forecast and observations and quantifies the degree of matching between the two fields. The present study extends MODE by incorporating the time and spatial dimensions simultaneously. In the context of precipitation, this results in three-dimensional objects in (x,y,t), where 't' is time. Using the analogue of the 2-D MODE object identification method, we define 3-D objects in numerical forecasts and observations from the IHOP period and the 2005 Storm Prediction Center (SPC) Spring Program. The advantage of considering 3-D objects is that each forecast or observed dataset can be viewed in terms of a relatively small number of rain systems. Furthermore, the evolution of the predicted systems can be verified using basic geometric properties of the objects. For instance, with time pointing vertically, the angle with respect to the horizontal plane indicates the propagation of the rain system. The temporal centroid, beginning and ending times all pertain to the timing of the system. Using 3-D object attributes, we devise a metric that accounts for spatial, temporal and propagation errors and results in a relative measure of forecast quality that can be used to compare forecasts from different models, or forecasts from a given model on different days.