Regression model optimization using least square algorithms for streamflow data transposition in tropical humid Water Basin

Abstract

Several projects are under consideration within the Anambra-Imo River Basin (AIRB). However, the location of these projects differs from established gauging stations, necessitating transposition of hydrological information for ungauged sites. This study investigates the transposability of maximum monthly stream records using regression models, employing neighboring stream records as predictors. Ten-year records of five rivers (Adada, Ajali, Otanmiri, Imo, and Ivo Rivers) were deployed to predict one another using a simple regression with five trends (linear, quadratic, cubic, logarithmic, and power). A metric based on compromise programming reveal that Imo records best predict Adada ($R^2=0.636$), Ajali ($R^2=0.777$), and Ivo ($R^2=0.403$) records, while Ajali and Ivo records are the best predictors of Imo ($R^2=0.703$) and Otanmiri ($R^2=0.349$) records, respectively. Linear trends appear to capture relationships among neighboring rivers most effectively. Lack of correlation among neighboring rivers, despite sharing the same basin, suggests that transposition models are catchment-specific and do not support regionalization.