Data-informed operational optimization of a small island SWRO plant: Field-validated energy reduction and pretreatment simplification over 22 months

This paper reports two field-validated engineering outcomes obtained at a 600 m 3 /d seawater reverse osmosis (SWRO) plant on Jibei Island, Taiwan, through data-informed review of 22 months of SCADA records (May 2024–February 2026) and 14 months of utility-billing corroboration (through March 2026). First, redistributing production from two to three SWRO trains reduced whole-plant billing specific energy consumption (SEC) from 5.24 ± 0.26 to 4.61 ± 0.38 kWh/m 3 —a 12.0% reduction before adjustment, or about 8–9% after adjusting for feed-water temperature as a seasonal covariate (n = 14 monthly bills; 8 baseline and 6 post-change)—while the billing-independent, SCADA-derived SWRO high-pressure-pump hydraulic SEC proxy was approximately flat (a small decrease, about 1.96 to 1.90 kWh/m 3 , Δ ≈ −0.07). The divergence of the two metrics supports fixed-load dilution as the largest modelled contributor rather than a membrane-transport efficiency gain; because the comparison is a single-site before/after change, residual time-related confounding cannot be fully excluded. The observed, unadjusted billing difference corresponds to approximately 72,100 kWh/yr, equivalent to about 35.6 t CO₂/yr at Taiwan's 2024 grid emission factor (0.494 kg CO₂/kWh); using the temperature-adjusted estimate the corresponding saving is approximately 50,000 kWh/yr. Second, ferric chloride (FeCl₃) pretreatment dosing was discontinued after 5 months of archived ultrafiltration (UF) filtrate data confirmed sustained low turbidity from the beach-well intake; 17 months of subsequent monitoring recorded median UF filtrate turbidity of 0.025 NTU (P95 0.092 NTU), with no degradation in downstream SWRO normalized differential pressure or salt passage, and all three trains operating 7.7–15.4 months post-withdrawal before their first clean-in-place session. Both decisions resulted from human interpretation of digitally archived plant data, not automated algorithms. Prospective analytics modules for adaptive antiscalant dosing and predictive maintenance scheduling are described but explicitly classified as unvalidated. The study contributes a three-tier evidence framework that separates field-validated engineering outcomes, including an energy reduction externally verified against utility billing, from installed digital infrastructure and prospective analytics reported as unvalidated, providing a reporting template directly applicable to other small-island and remote-community SWRO installations.