Forecasting and Structural-Break Characterisation of Water Cut in a Mature Heavy Oil Field: A Multi-Regime Analysis of the Gorisht-Kocul System (2000–2016)

This study characterises and forecasts monthly water cut in the mature Gorisht-Kocul heavy oil field of southern Albania across 204 monthly observations spanning 2000 to 2016. The record is held in two parallel sets of archival year tables, and a physical-consistency framework built on the reported production identities validates it across the join: water cut equals the ratio of free water to total produced fluid in all 204 months to within 0.00 percentage points, so the target variable is definitionally continuous, whereas the production covariates carry a level step at the transition. A single linear trend is misspecified (R2 = 0.911) with severely autocorrelated residuals. A piecewise-linear model with two structural breaks fits markedly better (R2 = 0.968), and the presence of change in trend is confirmed by a sup-F test against Bai–Perron critical values (sup-F = 229.4, p < 0.001). The breaks are estimated at December 2011 and June 2014, partitioning the trajectory into a rise of 1.51 percentage points per year, an acceleration to 6.47 points per year peaking near 91 per cent, and a reversal of −3.17 points per year. A block bootstrap locates both breaks sharply, but the first falls eleven months after the change in reporting convention and is confounded with it, so it is reported as such and not interpreted in reservoir terms; the second falls well inside a period of unchanged reporting. Rolling-origin cross-validation across six statistical methods and a displacement-based benchmark shows persistence to be the most accurate at every horizon, while a recursive exercise estimated only on pre-2014 data shows covariate-informed hybrids outperforming it at one to two years. Forecasting evaluations conducted on short, single-regime samples may not generalise.