J7.3 A Bin-Classifier Model to Diagnose and Forecast Deep Convection

Thursday, 26 January 2017: 2:00 PM
Conference Center: Tahoma 4 (Washington State Convention Center )
Christina Bonfanti, NOAA/ESRL and CIRES, Boulder, CO; and A. E. MacDonald

In this study, a first-version classifier model was designed to diagnose deep convection from a variety of data sources available when and where radar is either scarce or unavailable so that the model could be implemented globally without dependence on radar. While composite radar data is used to train the model, after it is trained the model runs solely with the following data inputs: 1) GOES-R east and west infrared (IR) satellite data, 2) National Lightning Detection Network (NLDN) lightning data, 3) The High-Resolution Rapid Refresh (HRRR) derived cloud-top pressure data, 4) HRRR Convective available potential energy (CAPE) data, and 5) Convective inhibition (CIN) data. These data are fed into a classification array that sorts each data grid point into a bin that estimates its radar reflectivity value based off of the bin it was sorted into. The bins and their boundaries are created dynamically for each type of data based off of the corresponding measured values of reflectivity that are determined realistic approximations based from graphical analysis. Once the data are in the classification array, the model estimates the average radar reflectivity per grid point and in the future it can be used in conjunction with weather model data to determine a 3D heating function associated with deep convection. The bin-classifier deep convection model was tested over a section of the Eastern United States where there was rich composite reflectivity radar data to train and compare results to over two selected time periods, 7 and 23 UTC, on June 2nd, 2015. Preliminary results indicate that with further statistical analysis over a greater range of time, the bin boundaries should be modified to produce more realistic reflectivity approximations since the first version is too dependent on certain data inputs, such as CAPE, but that this model and methods show good potential and have provided new insight on relations between satellite, lightning, and model data grid point values to their composite radar reflectivity grid point measurements.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner