An Objective Validation Approach to GOES-R Based Convection Initiation Forecasting Algorithms

The enhanced temporal and spatial resolution of the new GOES-R series will allow for cloud top cooling based convection initiation (CI) forecasting algorithms to be used, including the University of Wisconsin cloud top cooling algorithm (UWCTC) and the University of Alabama-Huntsville's satellite convection analysis and tracking algorithm (SATCAST). Use of the initial forms of CI algorithms on GOES East raised questions about how environmental variables could change CTC based detection. An area of study over the Great Plains was selected to analyze CTC algorithm performance with respect to the environmental conditions. An objective validation approach was designed with data from the National Quantitative Precipitation Estimation mosaic radar dataset (NMQ) to determine if and where convection occurs. The warning decision support services-integrated information w2segmotion tools are used with NMQ to identify and cluster radar objects into detections of convection based on quantitative and spatial thresholds. The identified clusters are then tracked using the thunderstorm observations by radar (ThOR) algorithm, which utilizes North American Regional Reanalysis storm motion data to track active cells. UWCTC pixels with cooling values greater than -4˚C 15 min-1 and all SATCAST pixels are compared to ThOR evaluated initiation points. Both CTC algorithms are tracked through time and grouped using similar principles to ThOR. Any pixel group that corresponds with (without) an evaluated initiation point within an hour of the first detection is considered a positive detection (false alarm). A missed detection is a ThOR evaluated initiation point without an indication forecast that was not originally masked out by the UWCTC algorithm cirrus shield. The positive and false detections from this study are compared to rapid refresh numerical weather prediction model (RAP) data to evaluated pre-convective environments with respect to performance. RAP based convective variables are calculated and a principle components analysis (PCA), discriminant analysis (DA) with an analysis of covariance (ANCOVA) are performed to filter out data points based on convective variables. The DA filter is applied to several case study days. Performance of the filtered datasets are then evaluated and compared to original dataset values to see if environmental filters would improve the products. Results are presented with respect to the convective variables, showing which in particular had the most statistical impact on the algorithms. The results will provide forecasters with a means to use numerical weather prediction data with GOES CTC based CI forecasting algorithms to establish areas that may change algorithm performance. Pre-filtered datasets presented here can also be used with the statistical results to improve fuzzy logic algorithms for future forecasting products.