Sign in Subscribe
Machine learning

Machine Learning Improves Daily Precipitation Estimations in Brazil

Strategic Insights

— 1 min read

Researchers at Federal University Paraiba have developed a machine learning model that estimates daily precipitation in Brazil with high accuracy, without relying on ground-based data. The IMERG BraMaL-D model uses satellite-based precipitation data and 53 re-analysis variables from MERRA-2 to produce accurate daily precipitation estimations. The model outperformed other global satellite-based precipitation products, including IMERG Final Run, PERSIANN-CDR, MSWEP, and CHIRPS, in terms of Kling-Gupta Efficiency (KGE) and data dispersion.

Key Takeaways:

  • The IMERG BraMaL-D model uses a combination of 6 regression and 8 classification models to estimate daily precipitation in Brazil, outperforming other global satellite-based precipitation products.
  • The model has a KGE of 0.70 for daily estimations, compared to values ranging from 0.05 to 0.66 for other analysed global products.
  • The monthly accumulated estimations of IMERG BraMaL-D also presented better performance, with smaller data dispersion and KGE rising from 0.86 to 0.95 compared to IMERG BraMaL-M.
  • The model's advantages include non-dependency on ground-based datasets after calibration, improvement of precipitation estimations where satellite-based products underestimate rain gauge data, and faster availability to end-users.
  • The research was funded by Universidade Federal da Paraiba, Conselho Nacional de Desenvolvimento Cientifico e Tecnologico (CNPQ), Coordenaco de Aperfeicoamento de Pessoal de Nivel Superior, and Fundaco de Apoio a Pesquisa do Estado da Paraiba.

Statistics:

  • KGE of 0.70 for daily estimations by IMERG BraMaL-D.
  • Data dispersion reduction of 0.09 point for monthly accumulated estimations.
  • KGE rise of 0.09 points for monthly accumulated estimations.
  • 3227 rain gauges used for evaluation of IMERG BraMaL-D.
  • IMERG BraMaL-D outperformed PERSIANN-CDR, MSWEP, and CHIRPS in terms of KGE.

Sources:

  • NewsRx. Studies from Federal University Paraiba Reveal New Findings on Machine Learning (Evaluation of Machine Learning Models To Improve Daily Precipitation Estimations From Orbital Remote Sensing and Reanalysis Data). Journal of Engineering. August 4, 2025; p 4532.
  • Evaluation of Machine Learning Models To Improve Daily Precipitation Estimations From Orbital Remote Sensing and Reanalysis Data. International Journal of Climatology, 2025.

Read more