Machine Learning Approach for Correcting Missing Rainfall Data in CEMADEN's Monitoring System

Fernando Humberto de Almeida Moraes Neto; Danielle Silva de Paula; Paulo Henrique Ferreira da Silva; Marcio Roberto Magalhães de Andrade; Daniel Metodiev; Sebastian Quintanilla Terminel; Cassiano Antonio Bortolozo

doi:10.1590/0102-778641250015

Machine Learning Approach for Correcting Missing Rainfall Data in CEMADEN's Monitoring System

Fernando Humberto de Almeida Moraes Neto
, Danielle Silva de Paula
, Paulo Henrique Ferreira da Silva
, Marcio Roberto Magalhães de Andrade
, Daniel Metodiev
, Sebastian Quintanilla Terminel
, Cassiano Antonio Bortolozo

Department of Mathematical Sciences

Universidade Federal de Mato Grosso
General Coordination of Research and Development
Federal University of Bahia
University of São Paulo

Research output: Contribution to journal › Article › peer-review

Abstract

Rainfall data gaps and sensor anomalies can critically compromise the effectiveness of early warning systems for natural disasters in Brazil. This study investigates precipitation data from CEMADEN's monitoring network, focusing on detecting and correcting faulty or missing measurements. Using data from 34 rain gauges in Blumenau over a two-year period, we applied both classification and regression models to predict rainfall occurrence and intensity. Exploratory analysis revealed stations with abnormal behavior, while machine learning models - especially XGBoost and Random Forest Regressor - demonstrated strong performance in correcting inaccuracies. Our findings highlight the potential of predictive modeling to enhance the reliability of rainfall data and support disaster risk reduction efforts.

Original language	English
Article number	e41250015
Journal	Revista Brasileira de Meteorologia
Volume	40
Early online date	27 Apr 2026
DOIs	https://doi.org/10.1590/0102-778641250015
Publication status	Published - 27 Apr 2026

Data Availability Statement

The data used in this study are publicly available at the following interactive portal: https://mapainterativo.cemaden.gov.br.

Funding

Sebastian Quintanilla Terminel is supported by a scholarship from the EPSRC Centre for Doctoral Training in Statistical Applied Mathematics at Bath (SAMBa), under the project EP/S022945/1. Paulo H. Ferreira also acknowledges support from CNPq (Grant Number 302620/2022-2). Danielle Silva de Paula also acknowledges support from CNPq (Grant Number 383911/2025-7). We want to thank the Coordination of Superior Level Staff Improvement (CAPES), for funding this work. This study is supported, in part, by the São Paulo Research Foundation (FAPESP), Brazil. Process Number 2024/16888-5.

Keywords

CEMADEN's monitoring network
predictive modeling
rain gauges
rainfall data

ASJC Scopus subject areas

Atmospheric Science

Access to Document

10.1590/0102-778641250015Licence: CC BY

Cite this

@article{e6b7430ed8f24388ad3942c43aabffce,

title = "Machine Learning Approach for Correcting Missing Rainfall Data in CEMADEN's Monitoring System",

abstract = "Rainfall data gaps and sensor anomalies can critically compromise the effectiveness of early warning systems for natural disasters in Brazil. This study investigates precipitation data from CEMADEN's monitoring network, focusing on detecting and correcting faulty or missing measurements. Using data from 34 rain gauges in Blumenau over a two-year period, we applied both classification and regression models to predict rainfall occurrence and intensity. Exploratory analysis revealed stations with abnormal behavior, while machine learning models - especially XGBoost and Random Forest Regressor - demonstrated strong performance in correcting inaccuracies. Our findings highlight the potential of predictive modeling to enhance the reliability of rainfall data and support disaster risk reduction efforts.",

keywords = "CEMADEN's monitoring network, predictive modeling, rain gauges, rainfall data",

author = "\{de Almeida Moraes Neto\}, \{Fernando Humberto\} and \{de Paula\}, \{Danielle Silva\} and \{da Silva\}, \{Paulo Henrique Ferreira\} and \{de Andrade\}, \{Marcio Roberto Magalh{\~a}es\} and Daniel Metodiev and Terminel, \{Sebastian Quintanilla\} and Bortolozo, \{Cassiano Antonio\}",

year = "2026",

month = apr,

day = "27",

doi = "10.1590/0102-778641250015",

language = "English",

volume = "40",

journal = "Revista Brasileira de Meteorologia",

issn = "0102-7786",

publisher = "Sociedade Brasileira de Meteorologia",

}

TY - JOUR

T1 - Machine Learning Approach for Correcting Missing Rainfall Data in CEMADEN's Monitoring System

AU - de Almeida Moraes Neto, Fernando Humberto

AU - de Paula, Danielle Silva

AU - da Silva, Paulo Henrique Ferreira

AU - de Andrade, Marcio Roberto Magalhães

AU - Metodiev, Daniel

AU - Terminel, Sebastian Quintanilla

AU - Bortolozo, Cassiano Antonio

PY - 2026/4/27

Y1 - 2026/4/27

N2 - Rainfall data gaps and sensor anomalies can critically compromise the effectiveness of early warning systems for natural disasters in Brazil. This study investigates precipitation data from CEMADEN's monitoring network, focusing on detecting and correcting faulty or missing measurements. Using data from 34 rain gauges in Blumenau over a two-year period, we applied both classification and regression models to predict rainfall occurrence and intensity. Exploratory analysis revealed stations with abnormal behavior, while machine learning models - especially XGBoost and Random Forest Regressor - demonstrated strong performance in correcting inaccuracies. Our findings highlight the potential of predictive modeling to enhance the reliability of rainfall data and support disaster risk reduction efforts.

AB - Rainfall data gaps and sensor anomalies can critically compromise the effectiveness of early warning systems for natural disasters in Brazil. This study investigates precipitation data from CEMADEN's monitoring network, focusing on detecting and correcting faulty or missing measurements. Using data from 34 rain gauges in Blumenau over a two-year period, we applied both classification and regression models to predict rainfall occurrence and intensity. Exploratory analysis revealed stations with abnormal behavior, while machine learning models - especially XGBoost and Random Forest Regressor - demonstrated strong performance in correcting inaccuracies. Our findings highlight the potential of predictive modeling to enhance the reliability of rainfall data and support disaster risk reduction efforts.

KW - CEMADEN's monitoring network

KW - predictive modeling

KW - rain gauges

KW - rainfall data

UR - https://www.scopus.com/pages/publications/105038223325

U2 - 10.1590/0102-778641250015

DO - 10.1590/0102-778641250015

M3 - Article

AN - SCOPUS:105038223325

SN - 0102-7786

VL - 40

JO - Revista Brasileira de Meteorologia

JF - Revista Brasileira de Meteorologia

M1 - e41250015

ER -

Machine Learning Approach for Correcting Missing Rainfall Data in CEMADEN's Monitoring System

Abstract

Data Availability Statement

Funding

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this