Impact of precipitation extremes on energy production across the São Francisco river basin

Abstract

The Brazilian electrical system is predominantly hydrothermal, with hydroelectric power plants (HPP’s) dependent on rainfall variability. The São Francisco river basin plays a fundamental role in the country's electricity production with HPP’s in the Northeast and Southeast regions. However, climate extremes events have affected the energy production. To manage the use of HPP’s to avoid energy shortage during dry periods and the activation of thermal power plants, is a challenge as it increases production costs and may result in water wastage during rainy periods. The main variables influencing operational decisions are Stored Energy (STE) and Affluent Natural Energy (ANE), used to calculate the Marginal Cost of Operation (MCO) and the Settlement Price of Differences (SPD). The current study investigates the relationships between these variables and climate precipitation extremes events in the São Francisco river basin. Spatial distribution and trends of 11 extremes precipitation indices are analyzed. The seasonality, trends, and correlation between the energy variables and the extreme indices are also investigated. Three machine learning algorithms (Random Forest, Artificial Neural Networks, and k-Nearest Neighbors) were applied as regression models to estimate the energy variables (ANE, STE, MCO, and SPD). Correlations between energy variables show the impact of changes in ANE and STE availability in the São Francisco river basin on MCO and SPD, in the Northeast and Southeast/Midwest subsystems. ANE and STE showed downward trends, while MCO and SPD experienced an upward trend. Furthermore, the seasonal behavior throughout the year was demonstrated for STE and ANE, influenced by extreme precipitation rates in different time scales. Trends indicate a reduction in total precipitation (PRCTOT) and the number of wet days (CWD), as well as an increase in the number of dry days in the basin (CDD). Results based on machine learning algorithm indicate that it is reasonable to efficiently estimate ANE and STE using extremes precipitation data. These findings have significant implications for the planning and management of the Brazilian electricity sector, contributing to strategic decision-making and the formulation of public policies that ensure the country's energy security. Keywords: Affluent Natural Energy. Stored Energy. Prediction.