Extended Abstract (528K)J3.4
An Artificial Neural Network to Forecast Thunderstorm Location: Performance Enhancement Attempts
Waylon G. Collins, NOAA/NWS, Corpus Christi, TX; and P. Tissot
We are attempting to enhance the accuracy of an Artificial Neural Network (ANN) developed to forecast the location of thunderstorms. The inputs to the ANN include selected output from a Numerical Weather Prediction (NWP) mesoscale model, and high-resolution/subgrid scale data that contribute to convective initiation. The basic rationale for the selection of inputs is the following: The mesoscale NWP output provides a forecast of the general mesoscale environment that contributes to convection. However, since individual convective cells develop on the microscale, data are needed that represents the forcing on these scales. With respect to subgrid scale data, research has shown that land surface heterogeneity (including horizontal gradients in soil moisture and vegetation) contributes to micro-scale/meso-alpha scale thermal gradients that ultimately result in convergent wind that can trigger convection. Further, research indicates that aerosols may contribute to convective development by enhancing subsequent storm updraft. The ANN methodology includes a South Texas grid of 13 x 22 equidistant points that create 286 separate 20-km x 20-km box regions. A framework was established to train and test a separate supervised ANN for each box, to predict thunderstorm occurrence.
The goal of the present study is to improve the performance of the ANN. The strategy includes expansion of the ANN training and testing set (dataset) and the incorporation of additional NWP parameters, and additional or alternative subgrid scale parameters. One such subgrid scale parameter to test for performance enhancement is the Antecedent Precipitation Index (API) gradient, which will serve as a proxy for the soil moisture gradient. API will be computed from 4-km Multi-Sensor Precipitation Estimator (MPE) output from the National Weather Service (NWS). The decision to include additional NWP and sub-grid scale parameters will be based on statistical analysis/data mining and/or the results of previous research. Preliminary analysis of the dataset reveals a statistical relationship between Aerosol Optical Depth (AOD) and thunderstorm occurrence for atmospheric conditions conducive to convection, which adds credence to the use of AOD as an ANN input parameter. A metric that will be used to measure ANN improvement is the change in the receiver/relative operating characteristic (ROC) curve.
The implications of this technique are significant. Utility of this ANN model would suggest that the current paradigm of improving thunderstorm prediction by increasing the resolution of NWP models to explicitly predict convection may not be necessary. Instead, a supervised ANN can be developed that inputs selected mesoscale NWP model output parameters and microscale data that contribute to convective initiation. The utility of this method is further enhanced by studies which question the ability of high-resolution NWP models to explicitly predict convection on the microscale. Further, the computing resources necessary for operational NWP forecasts on the micro-scale/meso-alpha scale are enormous. Yet, the ANN model described herein does not require NWP output at such high resolutions.
Joint Session 3, Bridging the Gap between Artificial Intelligence and Statistics in Applications to Environmental Science-I
Wednesday, 23 January 2008, 8:30 AM-10:00 AM, 219
Previous paper Next paper
Recorded presentation