Evaluating different strategies for machine learning training applied to flow forecasting based on clustering of flood events

ABSTRACT This paper presents a new hydrological modeling approach for discharge prediction based on flood clustering. Combined with Machine Learning techniques, river flow simulation is optimized through increased data similarity within clusters. Using daily mean discharge from 1964 to 2015 in União...

Full description

Saved in:
Bibliographic Details
Main Authors: Patrícia Cristina Steffen, Júlio Gomes, Eloy Kaviski, Daniel Henrique Marco Detzel
Format: Article
Language:English
Published: Associação Brasileira de Recursos Hídricos 2025-05-01
Series:Revista Brasileira de Recursos Hídricos
Subjects:
Online Access:http://www.scielo.br/scielo.php?script=sci_arttext&pid=S2318-03312025000100218&lng=en&tlng=en
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850142425002016768
author Patrícia Cristina Steffen
Júlio Gomes
Eloy Kaviski
Daniel Henrique Marco Detzel
author_facet Patrícia Cristina Steffen
Júlio Gomes
Eloy Kaviski
Daniel Henrique Marco Detzel
author_sort Patrícia Cristina Steffen
collection DOAJ
description ABSTRACT This paper presents a new hydrological modeling approach for discharge prediction based on flood clustering. Combined with Machine Learning techniques, river flow simulation is optimized through increased data similarity within clusters. Using daily mean discharge from 1964 to 2015 in União da Vitória (Iguaçu River basin, Paraná State, Brazil), the Fuzzy C-Means algorithm clustered flood events into three groups. So, five models were trained: one for the complete series, one for all flood events, and one for each cluster. The Support Vector Regression algorithm was used to develop Artificial Intelligence (AI) models, that had better performance in predicting discharge for each group they were trained and showed similar efficiency to the model trained for the entire series for a 1-day forecast time. The present paper discusses only the results from the training and testing phases. A future paper (in elaboration) will present the development and evaluation of the flow forecast models based on the proposed methodology.
format Article
id doaj-art-9e5024b51cff4a8a85095a4af0debb97
institution OA Journals
issn 2318-0331
language English
publishDate 2025-05-01
publisher Associação Brasileira de Recursos Hídricos
record_format Article
series Revista Brasileira de Recursos Hídricos
spelling doaj-art-9e5024b51cff4a8a85095a4af0debb972025-08-20T02:29:04ZengAssociação Brasileira de Recursos HídricosRevista Brasileira de Recursos Hídricos2318-03312025-05-013010.1590/2318-0331.302520240087Evaluating different strategies for machine learning training applied to flow forecasting based on clustering of flood eventsPatrícia Cristina Steffenhttps://orcid.org/0000-0002-2672-4619Júlio Gomeshttps://orcid.org/0000-0003-0746-9875Eloy Kaviskihttps://orcid.org/0000-0002-3618-7976Daniel Henrique Marco Detzelhttps://orcid.org/0000-0003-2841-6502ABSTRACT This paper presents a new hydrological modeling approach for discharge prediction based on flood clustering. Combined with Machine Learning techniques, river flow simulation is optimized through increased data similarity within clusters. Using daily mean discharge from 1964 to 2015 in União da Vitória (Iguaçu River basin, Paraná State, Brazil), the Fuzzy C-Means algorithm clustered flood events into three groups. So, five models were trained: one for the complete series, one for all flood events, and one for each cluster. The Support Vector Regression algorithm was used to develop Artificial Intelligence (AI) models, that had better performance in predicting discharge for each group they were trained and showed similar efficiency to the model trained for the entire series for a 1-day forecast time. The present paper discusses only the results from the training and testing phases. A future paper (in elaboration) will present the development and evaluation of the flow forecast models based on the proposed methodology.http://www.scielo.br/scielo.php?script=sci_arttext&pid=S2318-03312025000100218&lng=en&tlng=enArtificial Intelligence techniquesData-driven intelligent modelsFuzzy C-Means algorithmHydrological modelingSupport Vector Regression algorithm
spellingShingle Patrícia Cristina Steffen
Júlio Gomes
Eloy Kaviski
Daniel Henrique Marco Detzel
Evaluating different strategies for machine learning training applied to flow forecasting based on clustering of flood events
Revista Brasileira de Recursos Hídricos
Artificial Intelligence techniques
Data-driven intelligent models
Fuzzy C-Means algorithm
Hydrological modeling
Support Vector Regression algorithm
title Evaluating different strategies for machine learning training applied to flow forecasting based on clustering of flood events
title_full Evaluating different strategies for machine learning training applied to flow forecasting based on clustering of flood events
title_fullStr Evaluating different strategies for machine learning training applied to flow forecasting based on clustering of flood events
title_full_unstemmed Evaluating different strategies for machine learning training applied to flow forecasting based on clustering of flood events
title_short Evaluating different strategies for machine learning training applied to flow forecasting based on clustering of flood events
title_sort evaluating different strategies for machine learning training applied to flow forecasting based on clustering of flood events
topic Artificial Intelligence techniques
Data-driven intelligent models
Fuzzy C-Means algorithm
Hydrological modeling
Support Vector Regression algorithm
url http://www.scielo.br/scielo.php?script=sci_arttext&pid=S2318-03312025000100218&lng=en&tlng=en
work_keys_str_mv AT patriciacristinasteffen evaluatingdifferentstrategiesformachinelearningtrainingappliedtoflowforecastingbasedonclusteringoffloodevents
AT juliogomes evaluatingdifferentstrategiesformachinelearningtrainingappliedtoflowforecastingbasedonclusteringoffloodevents
AT eloykaviski evaluatingdifferentstrategiesformachinelearningtrainingappliedtoflowforecastingbasedonclusteringoffloodevents
AT danielhenriquemarcodetzel evaluatingdifferentstrategiesformachinelearningtrainingappliedtoflowforecastingbasedonclusteringoffloodevents