In this paper, we analyze the surface winds of ECMWF ERA5 reanalysis in the Atlantic Ocean. The first part addresses a reanalysis validation, studying the spatial distribution of the errors and the performance as a function of the percentiles, with a further investigation under cyclonic conditions. The second part proposes and compares two calibration models, a simple least-squares linear regression (LR) and the quantile mapping method (QM). Our results indicate that ERA5 provides high-quality winds for non-extreme conditions, especially at the eastern boundaries, with bias between −0.5 and 0.3 m/s and RMSE below 1.5 m/s. The reanalysis errors are site-dependent, where large RMSE and severe underestimation are found in tropical latitudes and locations following the warm currents. The most extreme winds in tropical cyclones show the worst results, with RMSE above 5 m/s. Apart from these areas, the strong winds at extratropical locations are well represented. The bias-correction models have proven to be very efficient in removing systematic bias. The LR works well for low-to-mild wind intensities while the QM is better for the upper percentiles and winds above 15 m/s—an improvement of 10% in RMSE and 50% for the bias compared to the original reanalysis is reported.

An index for frost prediction is proposed and calibrated against observations. It takes into account: (1) the main meteorological variables that favour or oppose to frost; (2) weights attributed to these variables; and (3) means and standard deviations of these variables, only for cases in which frost occurs, as defined by observation of temperatures that are ≤ 6°C. The meteorological variables used for the frost index IG (from the Portuguese, Índice de Geada) are numerically predicted by a regional weather forecast model. An outcome of the calibration processes results is that temperature has the largest contribution, followed by pressure and winds, while the other variables were adjusted to obey the restriction that the sum of weights are equal to 1. After index calibration and threshold determination, the method was applied for the 2017 winter season, and a case study for May 2018 was also considered. In order to verify whether the new index can satisfactorily contribute to the weather forecasting, the results using the IG were compared with the temperature outputs of the numerical regional model. It was found that for three selected areas, and for all the forecasted hours, the IG produces better results than the model's direct temperature forecasts. Thus, it was concluded that the use of the IG in an operational environment potentially provides considerable improvement in the predictive skill of frost events.