Study of GOES-R Thermodynamic Indices for Short-Term Forecasting of Convective Weather Events Using Machine Learning

Soares, Caroline Menegussi; França, Gutemberg Borges; Almeida, Manoel Valdonel de; Almeida, Vinícius Albuquerque de

doi:10.1007/s00024-021-02889-7

Cited by 5 publications

(1 citation statement)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…POD, FAR, BIAS, KAPPA, and F-measure statistics values for 5-h CME prediction (and in parentheses if it is severe when the occurrence of CME AD is greater than or equal to 1,000) are 0.91 (0.85), 0.05 (0.06), 0.92 (0.89), 0.74 (0.77), and 0.88 (0.95), respectively. Similarly, Soares et al (2021) proposed 6-to 8-h CME forecast models, for area I and II corresponding to the Rio de Janeiro -São Paulo air route, using ve TIIs determined using data extracted from atmospheric pro les from the GOES-R satellite. The results showed that the Multilayer Perceptron and Simple Logistic algorithms performed the best in areas I and II, with POD (0.86 and 0.94), FAR (0.14 and 0.08), bias (1.01 and 1.01), F-measure (0.86 and 0.94), and KAPPA (0.72 and 0.85).…”

Section: Introductionmentioning

confidence: 99%

Application of Machine Learning Models for Convective Meteorological Events

Pereira-Nunes,

França,

de Almeida

2024

Pure Appl. Geophys.

Self Cite

View full text Add to dashboard Cite

This research developed models, based on machine learning (MA), for forecasting 16 hours and 4 hours of occurrence of a convective meteorological event (CME), 4 hours for forecasting severity and evaluating the applicability of the optimal models of 12 UTC using thermodynamic instability indices (TII) data extracted from the WRF model with two different types of parameterization con guration in an attempt to develop a30-hour CME forecast model. In the training and testing of the MA algorithms, the classic TIIs (input) were used, obtained from the atmospheric pro les of the Brasilia upper air sounding and atmospheric discharges (output) detected in the study area for the characterization of CME, considering the period from 2012 to 2017. The optimal models applied to the modeled TIIs were evaluated through statistical metrics with con guration II obtaining signi cant results. For CME detection, the results showed that the best models obtained POD, 1-FAR, F-MEASURE and KAPPA with values respectively greater than 0.90, 0.80, 0.90, 0.80 and BIAS ranging from 0 .89 and 1.12. For the detection of event severity, the model presented the following statistical values (in parentheses): POD (0.82), 1-FAR (0.78), F-MEASURE (0.82), KAPPA (0.59 ) and BIAS (0.97). The results of 16-h and 4-h CME prediction hindcasts (30 days) with developed models demonstrated acceptable performance in identifying the occurrence or non-occurrence of CME and its severity for the study area.

show abstract

Section: Introductionmentioning

confidence: 99%