Rainfall Variability, Return Periods and an ARIMA-Based Rainfall Forecasting Model for River Niger at Lokoja, Kogi State, Nigeria (1986–2020)

Abstract

Flooding is the most damaging hydrometeorological hazard in Nigeria, and Lokoja, seated at the confluence of the Niger and Benue rivers is among the settlements most repeatedly affected by it. Because rainfall is the primary control on every other flood-generating mechanism in the basin, a defensible characterisation of rainfall magnitude, recurrence and short-horizon predictability is a prerequisite for any operational flood early-warning capability. This study examined the variability and return periods of annual rainfall over Lokoja between 1986 and 2020 and developed a univariate time series model capable of forecasting annual rainfall for the succeeding decade. Thirty-five years of monthly rainfall records obtained from the Nigerian Meteorological Agency were aggregated to annual totals, ranked in descending order of magnitude, and analysed for exceedance probability and return period using the Weibull plotting-position method. A Box–Jenkins autoregressive integrated moving average (ARIMA) model was then fitted after logarithmic transformation and first-order non-seasonal differencing, with the autoregressive order selected from the partial autocorrelation function. Results show a strongly asymmetric recurrence structure: rainfall in the 1500–1700 mm band carries an occurrence probability of 9.73 per cent and recurs on average once in about 14.7 years, whereas rainfall of 1097 mm and below carries a probability of 86.10 per cent and recurs roughly once in 1.2 years. The partial autocorrelation function was dominated by a single significant spike at lag 1 (−0.641), supporting an ARIMA (1,1,0) specification with an autoregressive coefficient of −0.621 (t = −4.547, p < 0.001) and a constant of 1.208. Forecast annual rainfall for 2021–2030 converges rapidly towards a long-run mean of approximately 1310 mm, ranging between 1262.8 mm and 1325.0 mm. The findings supply a quantitative recurrence baseline for flood preparedness planning in Lokoja and a parsimonious, operationally reproducible rainfall predictor that can drive a downstream river-stage model.