Stochastic characteristics and modeling of relative humidity of Ogun Basin, Nigeria

Abstract

Extreme events of atmospheric phenomena are often non deterministic in nature, and this has been a major constraint in achieving agricultural sustainability in most developing countries.To facilitate this study, 29 years information of the observed relative humidity of Ogun basin was obtained from the Federal Ministry of Water Resources, Abeokuta, Nigeria. The data collected covers the periods between 1982 and 2009 and were pre-whitened and aggregated into monthly and annual time series. The stationarity of the time series data was achieved through Mann-Kendal non parametric test and spectral analysis. The Mann-Kendal Z-value obtained is -1.37, which gives no reason to expect the presence of trend in the time series. The spectral density plot showed high variance to lower frequency, and this signifies a positive correlation. No evidence of seasonal effect in the series as clearly depicted by the monthly Periodogram. The autoregressive AR-model, moving average MA-model and autoregressive moving average ARMA-models were fitted for the parameter, with the aid of Akaike Information Criterion (AIC), and error terms of FE, MAE, MSE and MAPE. ARMA model of order (2, 2) was found to be the most parsimonious for predicting relative humidity. Results are highly accurate and promising for all models based on Lewis’ criteria. Prediction scheme applied in this research could be considered in situations where database is a problem during model development.

Keywords: ARMA model, Autocorrelation, Forecasting,Prediction andRelative humidity