Abstract:
Several flood management measures have been undertaken worldwide to mitigate the damage resulting from these events, including the generation of data-based hydrological models to support early warning systems. This article presents the data mining process with the application of filters on water level data in order to select data that represent flood events for future application of level prediction techniques in Machine Learning. To this end, sub-hourly water level data from the Proença river, located in Campinas, SP, for the period from November 2014 to June 2019 were used to form the historical series of data for this work. Python routines were created to select, pre-process, transform and mine such data. Based on the frequency distribution of the water level values of the historical series, three thresholds were tested, which selected 10, 6 and 3% of the total days with available data. The assertiveness of the thresholds in separating dates with occurrence of flooding was evaluated based on reports of floods on news sites and Civil Defense records. The conclusion is that the thresholds that separates 3% of the days has the best assertiveness among the tested thresholds, being thus chosen for the separation of the data to be used in the development of the model in Machine Learning for forecasting floods in the basin.
Reference:
ARAÚJO, Vinícius; RIBEIRO, Paulo Henrique Barreto; ABREU, Ana Elisa Silva de; COSTA, Paula Dornhofer Paro; FALCETTA, Filipe Antonio Marque; BITAR, Omar Yazbek. Correlação de níveis d’água em bacia urbana com registro de inundação no município de Campinas/SP: preparação dos dados para aplicação de machine learning. In: CONGRESSO BRASILEIRO DE GEOLOGIA DE ENGENHARIA E AMBIENTAL, 17, 2022, Belo Horizonte. Anais… 10p.
Document with restricted access. Login to BiblioInfo, Library/IPT-DAIT to access PDF text:
https://escriba.ipt.br/pdf_restrito/177951.pdf
Several flood management measures have been undertaken worldwide to mitigate the damage resulting from these events, including the generation of data-based hydrological models to support early warning systems. This article presents the data mining process with the application of filters on water level data in order to select data that represent flood events for future application of level prediction techniques in Machine Learning. To this end, sub-hourly water level data from the Proença river, located in Campinas, SP, for the period from November 2014 to June 2019 were used to form the historical series of data for this work. Python routines were created to select, pre-process, transform and mine such data. Based on the frequency distribution of the water level values of the historical series, three thresholds were tested, which selected 10, 6 and 3% of the total days with available data. The assertiveness of the thresholds in separating dates with occurrence of flooding was evaluated based on reports of floods on news sites and Civil Defense records. The conclusion is that the thresholds that separates 3% of the days has the best assertiveness among the tested thresholds, being thus chosen for the separation of the data to be used in the development of the model in Machine Learning for forecasting floods in the basin.
Reference:
ARAÚJO, Vinícius; RIBEIRO, Paulo Henrique Barreto; ABREU, Ana Elisa Silva de; COSTA, Paula Dornhofer Paro; FALCETTA, Filipe Antonio Marque; BITAR, Omar Yazbek. Correlação de níveis d’água em bacia urbana com registro de inundação no município de Campinas/SP: preparação dos dados para aplicação de machine learning. In: CONGRESSO BRASILEIRO DE GEOLOGIA DE ENGENHARIA E AMBIENTAL, 17, 2022, Belo Horizonte. Anais… 10p.
Document with restricted access. Login to BiblioInfo, Library/IPT-DAIT to access PDF text:
https://escriba.ipt.br/pdf_restrito/177951.pdf
