Statistical case study analysis of the quantity of water resources in Azerbaijan: Absheron Peninsula

Karimov E., Sundu M., Tufekci M., Pir İ., Gunes Durak S., Turkoglu Demirkol G., ...More

ENVIRONMENTAL RESEARCH COMMUNICATIONS, vol.8, no.7, 2026 (SCI-Expanded, Scopus)

Abstract

This study develops long-term water supply and demand forecasting models for the Absheron Peninsula of Azerbaijan, a semi-arid region with 4.0 million inhabitants entirely dependent on external water sources. Daily production and consumption data from 2016-2024 are analysed using three complementary specifications: ARMA(2,1) on the level series as a conservative parametric baseline, ARIMA(2,1,1) as a comparison model for assessing the effect of first differencing, and Prophet with automatic trend and seasonality decomposition. A seasonal-naive baseline ( yt=yt-365) serves as a non-parametric reference, and forecasts are validated through rolling-origin cross-validation across five hold-out years (2020-2024). Strong annual seasonality is present, with 30%-35% amplitude variation between summer peaks and winter troughs. On the 2024 hold-out, Prophet attains the strongest near-term accuracy (RMSE 0.75 m3 s-1), followed by the seasonal-naive baseline (1.07 m3 s-1); ARMA(2,1) on level (2.30 m3 s-1) outperforms ARIMA(2,1,1) (3.20 m3 s-1) among the ARIMA-family models and is therefore used as the conservative parametric baseline. Long-term projections to 2050 diverge sharply: ARMA(2,1) yields a stable annual surplus of approximately 4.9 Mm3 yr-1 (approximate to 0.87% of supply), ARIMA(2,1,1) approximately 2.8 Mm3 yr-1, and Prophet an expanding surplus reaching 14.8 Mm3 yr-1 by 2050. When the documented 35%-37% transmission and distribution losses are incorporated as a stress-test scenario, all three models project structural deficits of approximately 173-283 Mm3 yr-1 by 2050. The findings underscore the need for multivariate forecasting frameworks, integrated hydrological modelling, and adaptive management strategies for long-term water security.