Internal corrosion driven by produced water and dissolved acid gases remains one of the most costly and safety-critical degradation modes in upstream and midstream oil and gas operations. Water co-produced with hydrocarbons carries a complex geochemical matrix and dissolved gases that attack carbon and low-alloy steels in pipelines, flowlines, downhole tubulars and separation equipment [1]. When operating conditions move outside the design envelope, the resulting wall loss and localized attack can lead to leaks and through-wall failures at preferential sites such as bends, weld joints and low points where water accumulates [2].

Carbon dioxide and hydrogen sulfide are the dominant internal corrodents. In sweet service, dissolved CO₂ forms carbonic acid and drives general and localized corrosion whose severity depends strongly on CO₂ partial pressure, temperature and the stability of iron carbonate (FeCO₃) scale; in sour service, H₂S controls the surface chemistry through the formation of iron sulfide (FeS) films that can be either protective or locally damaging. Laboratory studies on line-pipe grades such as API 5L X52 confirm that, at elevated H₂S and CO₂ contents, iron sulfides dominate the corrosion product and the damage morphology shifts accordingly [3]. Engineering prediction of these rates has relied for decades on semi-empirical and analytical models: the de Waard-Milliams correlation relates corrosion rate to CO₂ partial pressure, temperature and pH; a flow-dependent extension couples the reaction rate to mass transfer through liquid velocity; and the NORSOK M-506 model adds wall shear stress and a temperature-dependent pH function. Semi-empirical and analytical corrosion-rate models of this kind have been developed and applied across a range of steel and process settings [4][5]. Complementary mechanistic studies show how steel metallurgy and surface-film formation govern localized attack, and note that H₂S, protective scaling and flow strongly modify the response [6].

Despite the maturity of these models, field corrosion rates are seldom published alongside the quantified partial pressures and water chemistry needed to test them. Most field reports describe service qualitatively as sweet or sour, or quote laboratory exposures performed at gas saturation without stated partial pressures, so the inputs required by the engineering models cannot be reconstructed. Recent work has begun to infer corrosion rates from the dissolution of alloying constituents in produced water, providing rate estimates for dozens of wells, but still without the gas-phase partial pressures that govern the mechanism [7]. In parallel, data-driven and artificial-intelligence frameworks for corrosion-resistant material selection have been proposed, and these explicitly identify data scarcity, especially the absence of field datasets spanning a wide range of service conditions, as the main obstacle to progress [8].

Cracking-related damage in the same steels, including sulfide stress cracking and stress-corrosion cracking, has been documented in case studies of X52 ammonia pipelines [9] and in failure analyses of high-strength X70M line pipe [10], underscoring that the same environments that drive wall loss also raise cracking risk. The present study, however, focuses on quantifying general and localized wall-loss rates and on benchmarking the predictive models against measured field data. The objectives are: 1) to present an anonymized field corrosion-rate dataset for three common oilfield steels, each record paired with the operating drivers; 2) to map the data on the sweet-to-sour spectrum using computed CO₂ and H₂S partial pressures; and 3) to benchmark measured rates against the de Waard-Milliams family of models and quantify where, and by how much, the predictions deviate in mixed and sour service.

In aqueous CO₂ service, the corrosion rate of carbon steel is governed by the partial pressure of CO₂, temperature, solution pH and the formation of protective FeCO₃ scale. The de Waard-Milliams relationship expresses the baseline (worst-case, scale-free) corrosion rate as a function of temperature and CO₂ partial pressure. Above a temperature that depends on CO₂ partial pressure, FeCO₃ precipitates and forms a protective layer that sharply reduces the rate; this is represented by a scaling-temperature correction factor that lowers the predicted rate relative to the worst-case value. Flow influences the rate through mass transfer of corrosive species to, and corrosion products away from, the steel surface; the de Waard-Lotz-Dugstad model combines a reaction-controlled term with a velocity- and diameter-dependent mass-transfer term, while NORSOK M-506 expresses the flow effect through wall shear stress together with a temperature-dependent pH function. Analytical and semi-empirical corrosion models of this kind have been applied to steel equipment and structures in service [4][5].

Hydrogen sulfide changes the picture qualitatively. Even at low partial pressures it can dominate surface chemistry through FeS formation, and the protectiveness of the sulfide film depends on the pCO₂/pH₂S ratio, temperature and flow. None of the CO₂-only engineering models above explicitly represents H₂S, so their accuracy in mixed and sour service is uncertain and must be established empirically; the role of steel metallurgy and surface films in setting localized attack further complicates a purely CO₂-based description [6]. Temperature plays a dual role, accelerating reaction kinetics at low temperature but promoting protective scaling at higher temperature, an effect demonstrated for crude-distillation systems where corrosion factors peak and then decline with rising temperature [11]. Flow can be protective when it is smooth and aggressive when it is turbulent enough to disrupt protective layers, a transition captured by dimensionless-number analyses of flow-induced corrosion [12]. Finally, the intrinsic resistance of the base material varies with composition and microstructure across the ferrous alloys used in oilfield service [13]. The dataset assembled here was designed to span these drivers so that model performance could be evaluated across realistic conditions.

The central result is that CO₂-only engineering models, applied in their uncorrected form, are strongly conservative against real field rates for these steels and services. The de Waard-Milliams 1991 baseline over-predicted by more than an order of magnitude, which is consistent with its design intent as a scale-free worst case and with the well-known observation that protective FeCO₃ scaling lowers real rates at field temperatures. Introducing the scaling-temperature correction moved predictions much closer to the data, in line with the dual role of temperature, which accelerates kinetics but promotes protective scale formation as it rises [11]. The flow-coupled form produced intermediate predictions, reflecting the additional influence of mass transfer that the shear-stress treatment in NORSOK M-506 also seeks to capture [12].

The regime dependence of the deviation is the most informative finding. Because none of the de Waard-Milliams forms includes an H₂S term, the systematic narrowing of the measured/predicted gap from sweet to sour service must originate in the data: H₂S raises field corrosion rates toward, and in some records beyond, the CO₂-only predictions. This is consistent with laboratory evidence that iron sulfides dominate the corrosion product and damage morphology of line-pipe steels at appreciable H₂S contents [3], and with mechanistic findings that steel metallurgy and surface films control localized attack [6]; it cautions against applying sweet models, even scale-corrected, to mixed or sour service without an explicit sulfide treatment. The significant inverse correlation of rate with in-situ pH and the significant positive correlation with H₂S partial pressure reinforce a mechanistic interpretation centred on acidification and sulfide surface chemistry; by contrast, the weak and statistically non-significant correlations with CO₂ partial pressure and chloride indicate that, within this dataset, neither acts as an independent first-order control on the measured rate.

These observations have practical implications for integrity management and prediction. First, corrosion-rate predictions used for inspection planning and remaining-life estimation should incorporate scaling corrections and an explicit H₂S treatment in mixed and sour service; relying on uncorrected sweet models will substantially overstate rates and may misallocate inspection effort. Second, the wide within-regime scatter shows that partial pressures alone do not determine the rate, water chemistry, flow and especially inhibitor availability are decisive, consistent with field experience that inhibitor programs and produced-water management strongly govern outcomes [1][7][14]. The large rate reduction achieved by inhibition in this dataset, and the body of work demonstrating high inhibition efficiencies for oilfield and pipeline steels using both engineered and biomass-derived inhibitors [15]-[18], underline that environmental control is the primary lever once a susceptible carbon steel is in service.

The same CO₂/H₂S environments that drive the wall-loss rates quantified here also elevate the risk of environmentally assisted cracking, including sulfide stress cracking and stress-corrosion cracking, which have been documented in line-pipe case studies and failure analyses [9][10] and which sit alongside alkaline cracking mechanisms treated elsewhere [19]. Cracking susceptibility was outside the scope of the present rate-focused dataset, but the operating windows mapped here identify the high-H₂S, low-pH conditions where cracking assessment should accompany wall-loss prediction. The study has limitations. The dataset, although balanced and broad, derives from a finite set of assets and monitoring campaigns; the records are period-averaged rather than continuous; localized-corrosion records are fewer than general-corrosion records; and the NORSOK M-506 model was used as a reference framework rather than computed point-by-point. Extending the dataset across additional operators and incorporating an explicit, validated H₂S corrosion model are clear directions for further work.

A field dataset of 90 corrosion-rate measurements for three common oilfield steels (API 5L X52, API 5L X65 and ASTM A106 Grade B), each paired with quantified operating drivers and spanning sweet, mixed and sour service, was assembled and analysed.

Measured rates ranged from 0.12 to 14.97 mm/yr and increased significantly with H₂S partial pressure (r = +0.42, p < 0.001) and with decreasing in-situ pH (r = −0.49, p < 0.001), whereas the correlations with CO₂ partial pressure (r = +0.16, p = 0.13) and chloride (r = +0.17, p = 0.12) were weak and not statistically significant.

The uncorrected de Waard-Milliams 1991 model over-predicted field rates by more than an order of magnitude (median measured/predicted ratio 0.05); the FeCO₃ scaling correction improved agreement (median ratio 0.18; 11% within a factor of two) but still over-predicted most records.

The measured/predicted deviation depended systematically on service regime, narrowing from sweet to sour service, which, because the models contain no H₂S term, demonstrates the genuine contribution of sulfide chemistry to field rates.

Among general-corrosion records, inhibited assets corroded approximately 73% slower than uninhibited assets, confirming that environmental control is the primary mitigation lever and supporting the use of explicit scaling and H₂S treatments when predicting rates for inspection planning in mixed and sour service.

The field dataset analysed in this study was obtained from operating oil and gas assets of an anonymous client operator and is reported in anonymized form; operator, field and well identifiers have been removed and assets are referred to by coded labels. Corrosion rates and operating conditions are reported as measured. Aggregated data supporting the findings are available from the author on reasonable request, subject to the data-owner confidentiality agreements.

Graphical abstract: field-measured corrosion rates versus de Waard-Milliams (scale-corrected) predictions across the sweet-to-sour spectrum, with the over-prediction factor shown for each regime.