Many natural disasters in South America are linked to meteorological phenomena. Therefore, forecasting and monitoring climatic events are fundamental issues for society and various sectors of the economy. In the last decades, machine learning models have been developed to tackle different issues in society, but there is still a gap in applications to applied physics. Here, different machine learning models are evaluated for precipitation prediction over South America. Currently, numerical weather prediction models are unable to precisely reproduce the precipitation patterns in South America due to many factors such as the lack of region-specific parametrizations and data availability. The results are compared to the general circulation atmospheric model currently used operationally in the National Institute for Space Research (INPE: Instituto Nacional de Pesquisas Espaciais), Brazil. Machine learning models are able to produce predictions with errors under 2 mm in most of the continent in comparison to satellite-observed precipitation patterns for different climate seasons, and also outperform INPE’s model for some regions (e.g., reduction of errors from 8 to 2 mm in central South America in winter). Another advantage is the computational performance from machine learning models, running faster with much lower computer resources than models based on differential equations currently used in operational centers. Therefore, it is important to consider machine learning models for precipitation forecasts in operational centers as a way to improve forecast quality and to reduce computation costs.

In this study, a method is presented for meteorological convective event forecasting at the terminal control area of the Galeão International Airport, Rio de Janeiro, Brazil, using machine learning, sounding and remotely sensed atmospheric discharge data from 2001 to 2016. A monthly and daily climatology were computed for the atmospheric discharge temporal distribution in the study area. Six machine learning models were trained and cross-validated for 10 years (2001–2010), and a test was produced for 6 years (2011–2016). The results showed that the deep learning fully-connected (dense) algorithm achieved the best results for storm forecast and severity based on the following statistics: probability of detection (0.91 and 0.85), BIAS (1.03 and 1.07), false-alarm ratio (0.12 and 0.20) and CSI (0.81 and 0.69), respectively. The 6-year test analysis showed that the model has increasing performance for high-impact events, and this performance decreases gradually as the events become weaker and more frequent. The models presented here could be useful tools for air traffic management purposes.

Assessing the ability of three global reanalysis products to reproduce South American monsoon precipitation