Analysis of hydro climatological time series using data mining and artificial intelligence for river flood prediction: A case study of the Consotá River.

Abstract

<p>Within the climate change and variability context, the need for continuous monitoring of climate and hydrological variables becomes evident. Due to Colombia being located within an equatorial region, the country is dominated by a tropical climate where precipitation is the aspect most affected by climate variability. Precipitation fluctuations are responsible for the frequency and intensity of extreme events such as flooding, drought, and landslides. In response to the need for frequent meteorological information, the state of Risaralda has created a hydro climatological monitoring network that gathers data every five minutes. This network has been in operation since 2005 and it consists of 40 weather stations; however, there is currently no hydrological model operating as an early warning system for local floods. This study aimed to implement data mining and artificial intelligence methodologies to estimate the probability of flood occurrence within the Consotá River watershed. A database was created using a five-minute time series of precipitation and water level data of five rain gauges and two water level sensors. A total of six classification algorithms were implemented: k-nearest neighbor (KNN), support vector machine (SVM), random forest, AdaBoost, decision tree, and neural network through Orange data mining software. The results indicate that it is possible to estimate the probability of flood occurrence 30 minutes prior to the flood occurring in the study area, with an accuracy of 88% using the random forest algorithm. This information could be valuable for relief agencies and decision-making institutions when utilized for flood preparedness and response planning.</p>