Evaluating medium-range tropical cyclone forecasts in uniform- and variable-resolution global models

An evaluation of medium-range forecasts of tropical cyclones (TCs) is performed, covering the eastern North Pacific basin during the period 1 August-3 November 2014. Real-time forecasts from the Model for Prediction Across Scales (MPAS) and operational forecasts from the National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS) are evaluated. A new TC-verification method is introduced that treats TC tracks as objects. The method identifies matching pairs of forecast and observed tracks, missed and false alarm tracks, and derives statistics using a multicategory contingency table methodology. The formalism includes track, intensity, and genesis. Two configurations of MPAS, a uniform 15-km mesh and a variable-resolution mesh transitioning from 60 km globally to 15 km over the eastern Pacific, are compared with each other and with the operational GFS. The two configurations of MPAS reveal highly similar forecast skill and biases through at least day 7. This result supports the effectiveness of TC prediction using variable resolution. Both MPAS and the GFS suffer from biases in predictions of genesis at longer time ranges; MPAS produces too many storms whereas the GFS produces too few. MPAS better discriminates hurricanes than does the GFS, but the false alarms in MPAS lower overall forecast skill in the medium range relative to GFS. The biases in MPAS forecasts are traced to errors in the parameterization of shallow convection south of the equator and the resulting erroneous invigoration of the ITCZ over the eastern North Pacific.