Modelo de pronóstico de escasez y sobreoferta de agua en una cuenca altoandida basado en análisis Big Data

Abstract

Background: Water is one of the most abundant natural resources worldwide; it is also of vital importance for the development of life. In Colombia, more than 11,000 mm of water can fall annually through precipitation, depending on the region of the country, reaching close to 60 l / s per km2 from runoff. However, during the dry season this water performance is severely affected, for this reason, it is necessary to have models that allow anticipating this type of event. In this sense, the prediction of drought events and oversupply of water resources plays an important role for all socio-economic sectors that depend on this resource.

Goals: Develop a model for estimating drought events and oversupply of water resources in a basin, which supports the decisions of environmental authorities and / or entities in charge of Integrated Water Resource Management.

Methods: A flow estimation model is proposed using information from multiple sources at a different spatial scale and a tool for the conformation of a database with these characteristics; The focus of this model is on the use of information in direct and indirect relation with the precipitation and the flow of the tributaries in a high Andean basin. The developed tool performs in a semi-automatic way the necessary data mining processes for the creation of a quality database. The flow estimation model is based on machine learning techniques, which allows a good adjustment between the predicted flow value and the real value.

Results: Two sets of data for precipitation estimation and flow estimation in a high-Andean basin, a semiautomatic tool for the conformation of the multiscale data set, a precipitation estimation model and a flow estimation model in a high Andean basin.

Conclusions: The model based on the machine learning technique Random Forest presented a better performance when estimating the amount of precipitated water and flow compared to other machine learning techniques. More information is required to achieve results were less underestimation of precipitation and flow is observed.