Terminal-Level Production Decline Analysis and Forecasting: A Dual-Model Machine Learning Approach for Nigeria's Upstream Export Infrastructure

Nigeria's crude oil production has been characterised by severe volatility between 2020 and 2026, shaped by OPEC+ quota agreements, widespread pipeline vandalism across the Niger Delta, and the gradual onset of reservoir depletion in maturing deepwater fields. While national production aggregates have received considerable attention in the literature, the structural divergence occurring within Nigeria's upstream export terminal portfolio remains largely unexamined. This paper presents a terminal-level production decline analysis and forecasting study applied to 30 Nigerian export terminals using 76 months of publicly available NUPRC monthly production data spanning January 2020 to April 2026. To the best of the author's knowledge, no prior published study has applied disaggregated machine learning forecasting at the export terminal level using NUPRC's public production reports, nor classified Nigerian export terminals by trajectory type using data-driven methods across this time horizon. A dual-model approach is employed, comparing Facebook's Prophet time-series decomposition model against an XGBoost gradient boosting model trained as a panel across all terminals simultaneously using 17 engineered temporal features. Both models are trained on January 2021 to December 2023 data and evaluated on a held-out test set covering January to December 2024. XGBoost substantially outperforms Prophet on every terminal in the evaluation set, achieving a mean MAPE of 7.88 percent against 88.90 percent for Prophet, with a median MAPE of 4.46 percent and R-squared values above 0.96 across all terminal tiers. The residual analysis confirms negligible systematic bias at the portfolio level with a mean residual of negative 0.002 million barrels across 321 test observations. Each terminal is classified as Growing, Recovering, Stable, or Declining based on its production trajectory. Key findings include the identification of a tightly coupled deepwater decline cluster, the strength of the Forcados recovery to near pre-crisis levels by early 2026, and the emergence of UTAPATE and NEMBE as significant new contributors to Nigeria's production base. The analysis demonstrates that national aggregate statistics conceal structural divergence within the terminal portfolio that has material implications for investment, regulatory, and operational decision-making. An interactive dashboard deployed on Streamlit makes the full analysis accessible to non-technical audiences.