Linear Regression Analysis and Statistical Models Using Oceanic and Atmospheric Variables for Streamflow Forecasting in Chixoy River, Guatemala

The objective of this study is to forecast streamflows in the Chixoy River Basin, Guatemala. The basin covers an area of about 5,000 km², it is located upstream of the Usumacinta River which drains into the Mexican Gulf, and it was selected as case study because it supplies the main hydropower plant of the country. Based on teleconnection principles and linear multiple regression analysis, statistical models were performed. The independent variables were oceanic and atmospheric indices used for monitoring El Niño South Oscillation (ENSO) by the National Weather Service/Climate Prediction Center of the National Oceanic and Atmospheric Administration (NOAA). These variables include sea level pressures (SLP), southern oscillation index (SOI), sea surface temperatures (SST), and trade winds (TW) in different regions of the Tropical Pacific Ocean, as well as autocorrelation of the river flows. The dependent variable was the seasonal (three-monthly average) river flow into the hydropower reservoir Chixoy for the period 1980–2009, which was operated by the Guatemalan National Electrification Institute (INDE in Spanish). The models were assessed through multiple step-wise regression analysis and cross correlations lagged from 1 to 12 months. To appraise the forecast potential of the model, cross validation was performed where the correlation coefficient, the difference between the forecast and the observed river flow, and the standard error were evaluated. In addition, the Durbin-Watson test was applied to verify that there was no autocorrelation on residuals. Six models were developed for the trimesters September-October-November (r²=0.803); October-November-December (r²=0.748); November-December-January (r²=0.818); December-January-February (r²=0.800); January-February-March (r²=0.820); and February-March-April (r²=0.855).