Abstract
The issuance of gridded temperature forecasts by NWS Weather Forecast Offices requires an assessment of the accuracy of the various guidance products available to the forecasters. In this presentation we will report the results of an investigation of the maximum and minimum temperature guidance and forecasts made from late winter through early spring for the Miami Weather Forecast Office (WFO) County Warning Area. Cold frontal passages are most common in South Florida during that time of year, with the resulting shifts in wind flow regimes as well as temperatures. The performance of numerical model guidance during easterly, northwesterly, and southwesterly wind regimes was analyzed for both the short- and extended-period forecasts using the BOIVerify software package. The numerical guidance used in this study included direct model output from several numerical models, statistical (i.e., post-processed) model output, bias-corrected versions of the direct and statistical model output, and various weighted and unweighted combinations of the model output. The official forecasts issued by the Miami WFO were also used in the comparisons. Special attention was given to the performance of the statistical model called ADJMEX, since it is used as a reference metric by the NWS. The results of this local study were compared to those of previous studies from other parts of the country. In general, the bias-corrected direct model output, and several combinations of the dynamical and bias-corrected guidance, were the most accurate overall. They were also somewhat more accurate, on average, than the official forecasts. The errors in the minimum temperature guidance, and also those of the official minimum temperature forecasts, grew more rapidly than did those for maximum temperature. In most cases the official forecasts were more accurate than the ADJMEX statistical model. This was most evident for the short term forecasts, with an improvement in mean absolute error of 0.5-0.75 of for the first five forecast periods. Although the bias-corrected direct model output tended to perform slightly better than the corresponding statistical model guidance for each wind regime, the greatest differences between them were in the maximum temperature guidance during times of northwesterly flow, when the bias-corrected statistical models performed much worse. For the easterly flow regime, which was the most dominant flow, the bias corrected models substantially outperformed their uncorrected counterparts. The bias-corrected dynamical models also performed better than the bias-corrected statistical models. This improvement was more noticeable for the northeasterly flow than the southeasterly flow. For the southwesterly flow regime, the findings were less clear, due in part to a smaller sample size. The official forecast did perform better relative to other models in this regime. There were no apparent differences between statistical and dynamical models or between bias-corrected and non-bias-corrected models.