This work presents a novel approach for simulating visibility (Vis) and ceiling base height (H_c) in up to 1 h using several machine learning (ML) algorithms. Ten years of meteorological data at 15 min intervals for Santos Dumont airport (SDA), Rio de Janeiro, Brazil were used in the ML method training and testing process. In the investigation, several categorical and regressive algorithms were trained and tested, and the results were verified with observations. The forecast results reveal that the categorical methods produced satisfactory results only up to 15 min for visibility prediction with the probability of detection greater than 85%. On the other hand, the regressive methods were found to be more capable of generating an accurate prediction of Vis and H_c compared to categorical method up to 60 min. The forecast evaluation metrics for Vis and H_c had correlation coefficients of 0.99 ± 0.00 and 0.96 ± 0.00, with mean absolute errors of 324 ± 77 m, and 167 ± 21 m, respectively. Results suggested that ML methods can improve the prediction of Vis and H_c up to 1 h when accurate observations are used for the analysis.

This paper presents a novel model, based on neural network techniques, to produce short-term and local-specific forecasts of significant instability for flights in the terminal area of Galeão Airport, Rio de Janeiro, Brazil. Twelve years of data were used for neural network training/validation and test. Data are originally from four sources: (1) hourly meteorological observations from surface meteorological stations at five airports distributed around the study area; (2) atmospheric profiles collected twice a day at the meteorological station at Galeão Airport; (3) rain rate data collected from a network of 29 rain gauges in the study area; and (4) lightning data regularly collected by national detection networks. An investigation was undertaken regarding the capability of a neural network to produce early warning signs – or as a nowcasting tool – for significant instability events in the study area. The automated nowcasting model was tested using results from five categorical statistics, indicated in parentheses in forecasts of the first, second, and third hours, respectively, namely proportion correct (0.99, 0.97, and 0.94), BIAS (1.10, 1.42, and 2.31), the probability of detection (0.79, 0.78, and 0.67), false-alarm ratio (0.28, 0.45, and 0.73), and threat score (0.61, 0.47, and 0.25). Possible sources of error related to the test procedure are presented and discussed. The test showed that the proposed model (or neural network) can grab the physical content inside the data set, and its performance is quite encouraging for the first and second hours to nowcast significant instability events in the study area.