From Rainfall to River Stage: A Parsimonious Regression Model for Flood Forecasting on the River Niger at Lokoja and its Implications for Flood Risk Governance in Nigeria

Abstract

Lokoja, situated at the confluence of the Rivers Niger and Benue, has experienced recurrent and severe flooding, most notably in 2012, 2018, 2020 and 2022. Successive structural interventions — embankments, levees, channelisation and channel widening — have failed to arrest the annual pattern of loss, indicating that the mitigation deficit in the confluence region is one of anticipation rather than construction. This study develops and evaluates a parsimonious statistical model relating meteorological drivers to the water level of the River Niger at Lokoja, and considers what such a model implies for flood risk governance in a resource-constrained setting. Thirty-five years of rainfall, temperature and river stage data (1986–2020) were obtained from the Nigerian Meteorological Agency and the National Inland Waterways Authority, supplemented by ground truthing at the NIWA gauge station to establish stage benchmarks. Multiple regression was used to estimate the joint influence of annual rainfall and mean annual temperature on annual water level, with model adequacy assessed through the mean square error, residual distribution and normal probability plotting. The multiple correlation coefficient was 0.595, indicating that rainfall and temperature together account for approximately thirty-five per cent of the variance in annual stage. Rainfall was the operative predictor, with an unstandardised coefficient of 0.003, implying a 3 mm rise in river level for each additional millimetre of annual rainfall; the temperature coefficient of 0.016 was negligible in operational terms and was consequently omitted, yielding the reduced model y = 4.543 + 0.003x. Mean observed stage was 8.82 m (SD 1.26 m; standard error 0.21 m), and residuals were contained within ±2.6 m. The model is presented not as a substitute for hydrodynamic forecasting but as a deployable first-generation instrument whose principal value lies in its transparency, low data requirement and susceptibility to annual updating. The substantial unexplained variance is interpreted as a substantive finding: upstream reservoir releases from Kainji and Jebba, and Benue inflow, are shown to be indispensable to any adequate forecasting architecture for the confluence.