Understanding the relationship between Hg emissions and deposition is essential for assessing future environmental risks and exposure pathways. Researchers have employed various atmospheric Hg models to project Hg deposition patterns under multiple emission scenarios, each with distinct atmospheric chemistry mechanisms, meteorological drivers, and varying assumptions regarding legacy emissions. In particular, numerous studies have projected global anthropogenic Hg emissions up to 2050, using different scenario frameworks (Tables 1–2) and emission inventories (Supporting Information: Table S1). The main differences between scenarios lie in the stringency of Hg control measures and the underlying socio-economic–energy trajectories – ranging from minimal-control, high-growth futures (A1F1, A1B, A2, CPS, Minimal-regulation, SSP5-8.5) (Knightes et al., 2009; Sunderland and Selin, 2013; Vijayaraghavan et al., 2014; Lei et al., 2014; Giang et al., 2015; Chen et al., 2018; Perlinger et al., 2018; H. Zhang et al., 2021; Schartup et al., 2022; Brocza et al., 2024; Geyman et al., 2024, 2025) with rising emissions, through intermediate-policy pathways (B1, B2, PIA, NPS, CLIM, Minamata, SSP5-3.4, SSP2-4.5) (Streets et al., 2009; Sunderland and Selin, 2013; Rafaj et al., 2013; Vijayaraghavan et al., 2014; Lei et al., 2014; Giang and Selin, 2016; Angot et al., 2018; Perlinger et al., 2018; H. Zhang et al., 2021; Y. Zhang et al., 2021; Geyman et al., 2024, 2025) that stabilize or moderately reduce emissions, to maximum-mitigation futures (50 % Reduction, MFR, CLIM + MFR, Aspirational, Zero emissions, Net Zero + MFR, SSP1-2.6) (Pacyna et al., 2016; Angot et al., 2018; Chen et al., 2018; Perlinger et al., 2018; Y. Zhang et al., 2021; Brocza et al., 2024; Geyman et al., 2024, 2025) achieving the deepest cuts. In some studies, e.g., (Giang and Selin, 2016), Minamata and No-Policy scenarios adopt B1- and A1B-type socio-economic assumptions, respectively. To illustrate their origins and policy focus, Figure S1 organizes these scenarios into four thematic families: Hg-focused policy scenarios, SRES-based frameworks, shared socioeconomic pathways, climate-oriented pathways, and hybrid syntheses. Collectively, these studies span the full spectrum from continued growth to deep global reductions in anthropogenic Hg emissions. The acronyms for Hg-focused and SRES scenarios are defined in Tables 1–2, while climate-oriented and hybrid frameworks, often reported only as emission trajectories and, in some cases, not extending to 2050 or lacking deposition outputs – are listed separately in Table S2.

These emission scenarios have been applied in a variety of atmospheric models, which differ in spatial resolution, redox chemistry, and meteorological drivers (Table S7). The oxidation of Hg⁰, primarily by ozone (O3) and hydroxyl radicals (OH), is incorporated in several models, including HYSPLIT-Hg (Cohen et al., 2016), CAM-Chem/Hg (Lei et al., 2014), CTM-Hg (Vijayaraghavan et al., 2014), GLEMOS and ECHMERIT (Pacyna et al., 2016; De Simone et al., 2017; Travnikov et al., 2017). Some models, such as GLEMOS and GEM-MACH-Hg, also incorporate Br oxidation in polar regions, although GEM-MACH-Hg mainly relies on OH chemistry (Travnikov et al., 2017). The older version of GEOS-Chem simulated Hg⁰ oxidation by Br and HgII photoreduction in clouds, following the mechanism proposed by (Holmes et al., 2010). (Horowitz et al., 2017) later refined this chemistry by introducing second-stage oxidation of HgBr by HO₂ and NO₂, based on the mechanisms of (Dibble et al., 2012) and (Shah et al., 2016). More recently, (Shah et al., 2021) proposed a revised mechanism where Hg⁰ oxidation is driven by both Br and OH radicals, with O3 playing a secondary role. HgII is mainly reduced via photolysis of HgII-organic complexes and gas-phase HgI species. In addition, models employ distinct meteorological drivers, ranging from offline simulations forced by reanalysis or prescribed meteorological fields (e.g., GEOS-5, MERRA-2, ECMWF, CCSM in specified-dynamics mode, and NCEP/NCAR) (Bullock et al., 2009; Corbitt et al., 2011; Vijayaraghavan et al., 2014; Lei et al., 2014; Pacyna et al., 2016; Cohen et al., 2016; Travnikov et al., 2017; Angot et al., 2018) to fully online coupling with general circulation or weather prediction models (e.g., NASA GISS GCM ModelE2, ECHAM5, and GEM) (Travnikov et al., 2017; Y. Zhang et al., 2021; H. Zhang et al., 2021) (Table S7). These differences introduce additional variability in simulated atmospheric transport, oxidation, and deposition patterns. Models also rely on different anthropogenic and natural Hg emission inventories, further contributing to variability in projected deposition. Details of the inventories used in the models and the scenarios to which they are coupled are provided in Table S1, while abbreviations for the meteorological drivers are listed in Table S7.

To assess the influence of changes in anthropogenic Hg emissions on future deposition, we analyzed the relationship between annual changes (% yr⁻¹) in global anthropogenic Hg emissions and corresponding annual changes (% yr⁻¹) in Hg deposition projected for 2050 (Fig. 2). Each data point represents a specific combination of atmospheric model, emission scenario, surface type, and region. For each case, the change in emission and deposition estimates are derived from the same study to maintain internal consistency within each scenario. The relationship is strikingly linear (R2 = 0.79, slope = 0.41, p<0.001), meaning that 79 % of the observed variability in Hg deposition is explained by changes in emissions. Yet the slope indicates that a 1 % change in emissions yields about a 0.4 % change in deposition, implying that a substantial fraction of emitted Hg undergoes chemical transformation and/or transport, while also reflecting the influence of re-emissions and the legacy mercury pool.

While these emission projections largely explain future Hg deposition trends, Fig. 2 also illustrates variability in annual changes (% yr⁻¹) in Hg deposition under the same scenarios. These differences, although relatively small, arise from variations in model configurations, including redox chemistry, spatial resolution, meteorological drivers, and treatment of legacy reservoirs. For instance, under the A1B scenario, (Corbitt et al., 2011) applied a coarse-resolution GEOS-Chem configuration with a simplified one-step Br oxidation scheme and static legacy pools, resulting in only a 0.46 % yr⁻¹increase in global Hg deposition (ellipse C, region 3). In contrast, (Y. Zhang et al., 2021) projected a much stronger 2.18 % yr⁻¹increase, reflecting the use of finer spatial grids, updated anthropogenic inventories, a two-step Br oxidation pathway, and fully coupled ocean-terrestrial reservoirs. Even in the relatively stable B1 scenario, where global anthropogenic emissions remain nearly constant, deposition projections for the U.S. diverge slightly. (Corbitt et al., 2011) projected a modest decrease in deposition, while (Giang and Selin, 2016) projected a slight increase (ellipse B, region 4). This contrast can be attributed to differences in model setup: (Corbitt et al., 2011) used a coarse 4° × 5° global grid with fixed meteorology and treated biomass burning as part of the legacy pool, whereas (Giang and Selin 2016) applied a nested 0.5° × 0.667° North American domain with updated emission inventories.

In addition to these differences, the treatment of historical emission inventories further contributes to uncertainty in modeled Hg deposition. While recent work by (Guerrero and Schneider, 2023) challenged previous overestimates of pre-industrial emissions by (Nriagu, 1993, 1994, Streets et al., 2009, 2011, 2017; Streets et al., 2019), they demonstrated that chemically sequestered forms of Hg (e.g., calomel, vermilion) were previously omitted or underestimated in historical emission inventories. Their findings, supported by (Engstrom et al., 2014) and (Outridge et al., 2018), indicate that pre-industrial emissions may have been 2–5 times lower than earlier estimates. However, (Angot et al., 2018) demonstrated that halving historical (1850–1920 CE) mining emissions had no significant impact on the projected benefits of delayed policy action, reinforcing the finding that atmospheric deposition is most sensitive to recent anthropogenic emissions. (Médieu et al., 2024) also emphasized the rapid responsiveness of the atmospheric Hg reservoir to emission changes, in contrast to the slower response of legacy reservoirs like the ocean.

Taken together, these results indicate that contemporary anthropogenic emissions overwhelmingly drive global Hg deposition in the model-simulated world, with diverse modeling frameworks converging on this conclusion despite differences in atmospheric chemistry, spatial resolution, treatment of legacy reservoirs, meteorological drivers, and uncertainties in historical emission inventories. However, an important question remains as to whether real-world systems respond as linearly as suggested by current models.

Model–observation inconsistencies are evident across multiple spatial scales. (Holloway et al., 2012) found that across 31 Mercury Deposition Network (MDN) sites in the Great Lakes region, the CMAQ-Hg model underestimated annual Hg wet deposition by about 21 % on average, with seasonal biases including underprediction in spring–fall and overprediction in winter (approx. 70 %). Similarly, MDN observations in the southeastern U.S. show that the global GEOS-Chem model (4° × 5°) underestimated deposition by about 46 %, largely due to its inability to resolve deep convection, a key driver of Hg wet deposition (Xu et al., 2022). In China, observations indicate faster declines than models predict, with GEM decreasing by approx. 19 % compared to an 11 % modeled reduction over 2013–2017 (Feng et al., 2024; Liu et al., 2019).

Model-observation discrepancies are also apparent at the hemispheric scale. (Travnikov et al., 2017) evaluated four global Hg models (GLEMOS, GEOS-Chem, GEM-MACH-Hg, and ECHMERIT) against observations and found that performance depends strongly on oxidation chemistry and precipitation scavenging. Br-based schemes peak in spring, while OH-based schemes better capture North American and European summer peaks, indicating no single pathway fully explains observed patterns. Furthermore, in the Southern Hemisphere, GEOS-Chem and MITgcm overestimate wet deposition by approx. 5-fold and underestimate dry deposition by 3.5-fold relative to observations, reflecting simplified representations of convective uplift and precipitation scavenging (Leiva González et al., 2022).

Together, these discrepancies highlight the need for more routine and systematic comparisons between modeled and observed deposition trends under evolving policy conditions. Strengthening long-term monitoring networks and further harmonizing emission inventories will be essential to validate model predictions and to assess the real-world effectiveness of Hg control measures under the Minamata Convention.

Atmospheric deposition plays a pivotal role in biogeochemical cycling and fate of Hg in aquatic ecosystems. Studies have shown that atmospherically supplied Hg can rapidly be converted to MeHg in lake ecosystems and enter the base of the food chain, elevating fish MeHg levels within a short period (Harris et al., 2007; Ogorek et al., 2021). Understanding the relationship between atmospheric Hg deposition and MeHg concentrations in fish is essential for predicting future risks under varying emission scenarios, especially in the context of evolving policies.

To estimate MeHg concentrations in fish under future deposition scenarios, researchers have used multiple bioaccumulation models, including SERAFM, Hendrick's Lake Mass Balance Model, D-MCM, BASS, and WASP, as described below. a.

SERAFM: Spreadsheet-based Ecological Risk Assessment for the Fate of Hg model operates under steady-state conditions, estimating Hg concentrations in fish using bioaccumulation factors (BAFs). It employs a three-box system (epilimnion, hypolimnion, and sediments) to model Hg dynamics, incorporating atmospheric deposition of Hg0, MeHg, and HgII with same wet and dry deposition fluxes for lakes and watersheds. However, it does not account for Hg accumulation during ice cover and subsequent release during spring melt. Watershed inputs rely on constant runoff coefficients without seasonal variability, and erosion is modeled using the Revised Universal Soil Loss Equation (RUSLE). Seasonal changes in temperature, stratification, ice cover, and DOC inputs are not simulated, and rate constants for oxidation-reduction, photodegradation, and other processes remain fixed throughout the simulation (Knightes, 2008; Knightes et al., 2009).

To assess how changes in atmospheric Hg deposition influence fish MeHg concentrations, we analyzed the relationship between annual changes (% yr⁻¹) in atmospheric Hg deposition and corresponding changes (% yr⁻¹) in fish MeHg projected for U.S. ecosystems by 2050 (Fig. 3). Each point integrates atmospheric and bioaccumulation models, fish trophic levels, sites, and emission scenario-based deposition. Deposition and fish MeHg changes are derived from the same study, ensuring internal consistency. Despite the diversity of atmospheric models, bioaccumulation frameworks, fish species, and ecosystem types represented, the relationship is strongly linear (R2 = 0.63, slope = 0.87, p<0.01), indicating that roughly 63 % of the variance in the projected change of fish MeHg is explained by scenario-based changes in atmospheric Hg deposition. The slope shows that every 1 % change in deposition yields an approximately 0.87 % change in fish MeHg, a far steeper response than the 0.4 slope found for emissions versus deposition (Fig. 2). This suggests that fish incorporate nearly all deposited Hg into food webs in the model-simulated world. However, the correlation is weaker than that observed between emissions and deposition (Sect. 3.1), reflecting the added complexity of lake-specific characteristics and ecological factors. For example, the modeled decline in fish MeHg for lakes in Michigan's Upper Peninsula (UP) is moderate (0.25 % yr⁻¹). However, two virtual experiments reported by Perlinger et al. (2018) separate the effects of scenario-based atmospheric deposition and watershed sensitivity on fish MeHg responses (Ellipse A; symbol ∇). Applying Adirondack deposition to the Michigan UP lake resulted in increased modeled fish MeHg, whereas applying Adirondack watershed characteristics under the same deposition conditions led to a decrease, highlighting the influence of watershed properties. Similarly, under the high deposition reduction scenario involving a 50 % decrease in Hg deposition (Ellipse B; symbol Δ), five integrated data points show substantial variability in fish MeHg responses under the same policy scenario (Knightes et al., 2009). Lakes with large surface areas, indicating greater direct atmospheric exposure, together with large watersheds and higher wetland percentages (e.g., bay lakes and coastal rivers) showed the greatest declines in fish MeHg. Intermediate declines occurred in seepage and stratified drainage lakes, whereas systems with lower wetland influence and weaker watershed-mediated Hg inputs (e.g., farm ponds) exhibited slower declines in fish MeHg (Knightes et al., 2009). Together, these findings highlight that while deposition strongly governs fish MeHg, lake characteristics can result in site-specific variability, which we elaborate further in Sect. 3.3.

Apart from the lake-specific characteristics, additional ecological and climate-related factors can modulate fish MeHg levels. These have been evident through many individual studies. In particular, the changes in fish MeHg concentrations documented at various lakes in Michigan's Upper Peninsula (UP), U.S. provides a key case study. Comparisons between modeled and observed changes in lake trout MeHg reveal notable discrepancies across different time periods.

Model simulations for lakes in Michigan's UP estimated modest annual fish MeHg declines (-0.25 % yr⁻¹ to -1.48 % yr⁻¹) during 2006–2050 under policy scenarios (Perlinger et al., 2018), whereas, observed lake trout data from lakes in Michigan's UP showed a sharper decline of 3.9 % yr⁻¹ (2004–2015) (Zhou et al., 2017). The reduction was largely driven by a steep drop in fish MeHg before 2008 (-11.7 % yr⁻¹), followed by a subsequent plateau, which was attributed to climate-relevant changes in food web structure and amount of wet Hg deposition (Zhou et al., 2017). More recent work by (Lepak et al., 2025) also reported counterintuitive increases in fish MeHg despite declining local-regional Hg emissions and atmospheric deposition, attributing them to invasive dreissenid mussels that redirected trout toward lower-calorie, benthic prey, coupled with reduced lake productivity and slower fish growth. Stable isotope analyses of Hg (Δ199Hg), together with carbon and nitrogen isotopes (δ13C, δ15N), confirmed these dietary and energy-flow changes (Lepak et al., 2019; 2025). Similarly, (Kerfoot et al., 2018) observed rising fish MeHg (1 %–3 % yr⁻¹, 2001–2013) despite an 81 % regional emission decline, driven by wetland recovery and renewed methylation of legacy Hg.

Despite these ecological complexities, isotopic evidence provides further insight into system responsiveness. Stable-isotope evidence from the Great Lakes indicates that fish Hg isotopic signatures respond more rapidly to changes in atmospheric Hg concentrations than sedimentary records, suggesting that fish act as sensitive biosentinels of recent deposition trends (Lepak et al., 2025). Long-term whole-ecosystem isotope-labeled experiments in a boreal lake support this view, showing that reducing direct atmospheric inputs drives rapid declines in fish MeHg through food webs, even if watershed Hg is retained and mobilized more slowly (Blanchfield et al., 2022). Yet interacting stressors, such as invasive species, climate change, and enhanced wetland inflows, can amplify methylation and delay Hg recovery (Hall et al., 2023; Thompson et al., 2023; Rodríguez, 2023; McCarter et al., 2024; Lepak et al., 2025; Olson et al., 2025).

Collectively, these findings highlight the constraints of assuming that fish MeHg levels will decline in direct proportion to atmospheric deposition. They raise an important question for future assessments: do lake ecosystems truly follow linear dynamics, or do current models oversimplify the complex ecological, biogeochemical, and climate-driven interactions that regulate Hg uptake and bioaccumulation in natural ecosystems?

This section examines how lake-specific characteristics are represented across models when simulating future fish MeHg levels. To evaluate how these characteristics are treated within the model-simulated framework, we incorporated key environmental predictors used in the same studies analyzed in Sect. 3.2. These include lake area, watershed area, wetland percentage, and average depth, together with scenario-specific changes in Hg deposition as well as the corresponding changes in fish MeHg, into our statistical analyses (Knightes et al., 2009; Vijayaraghavan et al., 2014; Angot et al., 2018; Perlinger et al., 2018). The full list of used variables is provided in Table S4. Extensive empirical work has shown that fish MeHg levels respond to lake morphometry (area, depth, wetland extent) (Bodaly et al., 1993; Choy et al., 2009; Ackerman et al., 2019; Knott et al., 2020), watershed size and composition (Sonesten, 2003; Evans et al., 2005; Rypel, 2010; Marusczak et al., 2011; Eagles-Smith et al., 2016; Backstrom et al., 2020), water chemistry (pH, hardness, alkalinity) (McMurtry et al., 1989; Backstrom et al., 2020), hydrologic conditions (drainage pattern) (Grieb et al., 1990), geographical gradient (Morel et al., 1998), seasonal variability (Barletta et al., 2012; Keva et al., 2017; Mills et al., 2018), and fish-specific traits (age, size, trophic level) (Cizdziel et al., 2002; Evans et al., 2005; Marusczak et al., 2011; Backstrom et al., 2020). Together, these studies illustrate that multiple environmental processes regulate MeHg bioaccumulation, providing a basis for examining how Hg bioaccumulation models prioritize these same factors.

We first evaluated whether lake characteristics alone could explain variability in projected changes in fish MeHg across models using MLRM. No strong relationships were identified (Table S5). When scenario-specific changes in Hg deposition were incorporated, as described in Sect. 3.2, only deposition change (p = 0.007) and lake area (p = 0.04) emerged as significant predictors, while other variables contributed weakly (Table S6). These results motivated subsequent PCA analyses to examine how deposition and lake characteristics interact collectively within Hg bioaccumulation model simulations.

PCA was conducted within a scenario-based modeling framework consistent with Sect. 3.2, using lake area, watershed area, wetland percentage, average depth, and projected changes in Hg deposition as explanatory variables, with projected changes in fish MeHg included as a supplementary variable. The resulting biplot (Fig. 4) summarizes dominant patterns of covariation, with the first two principal components explaining 70.82 % of the total variance (PC1 = 38.13 %, PC2 = 32.69 %). PC1 was primarily defined by lake area and wetland percentage, whereas PC2 was driven by average depth and Hg deposition change, with a moderate contribution from watershed area. Consistent with this structure, a MLRM using PC1 and PC2 explained a substantial portion of the variability in the simulated response variable (Change in fish MeHg) and was statistically significant (R2 = 0.57; F = 6.74, p = 0.01). PC1 exhibited a significant negative association (β=-0.26, p = 0.02), indicating that systems with larger lake area and higher wetland coverage tend to show greater declines in fish MeHg in the model simulations. In contrast, PC2 showed a significant positive relationship (β=0.24, p = 0.04). This component reflects a trade-off between atmospheric Hg loading and lake depth versus watershed influence (Fig. 4), such that the bioaccumulation models predict the greatest simulated MeHg increases (or smallest reductions) in deep, atmospherically impacted systems with relatively small contributing catchments. The proximity of the fish MeHg vector in Fig. 4 confirms that modeled future fish MeHg responses are structured along gradients of lake morphometry and atmospheric Hg loading.

Collectively, results from the MLRM and PCA analyses indicate that, within the modeling framework, scenario-specific changes in atmospheric Hg deposition consistently emerge as the dominant predictor of future fish MeHg responses, with lake area exerting a secondary influence, while wetland extent, average depth, and watershed area contributing primarily through multivariate structure in the PCA. Across these statistical approaches, deposition change explains the largest share of variance in modeled fish MeHg, reinforcing the central role of Hg-emission policies in shaping projected bioaccumulation outcomes in the model-simulated world. This finding mirrors the pattern of whole-ecosystem experimental studies showing declines in fish MeHg following reductions in Hg deposition (Blanchfield et al., 2022). However, comparisons between modeled projections and long-term observational records reveal a more complex reality. Case studies from lakes in Michigan's UP demonstrate that fish MeHg concentrations may increase, decrease, or respond nonlinearly despite substantial declines in atmospheric Hg deposition, reflecting the influence of ecological and biogeochemical processes not fully resolved by current models. These include food-web restructuring, invasive species, productivity shifts, wetland-mediated methylation, internal Hg cycling, and climate-driven changes in hydrology and redox conditions. Moreover, the negative association between wetland extent and projected fish MeHg observed here in the PCA differs from many field studies, which commonly report elevated fish MeHg with increasing wetland coverage, although the magnitude and direction of this relationship vary with sulfate availability, dissolved organic matter, and hydrologic setting (Watras et al., 2005; Ackerman et al., 2019; Poulin et al., 2025). This apparent discrepancy likely reflects model assumptions regarding Hg loading pathways and response kinetics. Modeling frameworks derived from or conceptually similar to SERAFM assume relatively direct transfer of atmospherically deposited Hg from the catchment to the lake, with limited representation of hydrologic retention and delayed release from soils and wetlands (Perlinger et al., 2018). Under this structure, wetlands and large watersheds act as highly sensitive receptors of atmospheric Hg inputs, strengthening atmospheric–catchment coupling in wetland-dominated systems. Our supplemental regression analysis supports this interpretation, showing that watershed and wetland area are strong predictors of total Hg loading to the lake (R2 = 0.90 and 0.87, respectively; p<0.0001). In addition, PCA results indicate that lake morphometry controls model sensitivity, with large, shallow lakes responding strongly to deposition changes due to limited dilution. Fish MeHg concentrations are commonly estimated using a steady-state BAF approach, which does not capture the biological and ecological lag times required for fish populations to fully adjust to changes in water-column Hg exposure (Angot et al., 2018). Consequently, declines in modeled water-column Hg are translated into relatively rapid declines in fish MeHg, whereas in natural systems, Hg stored in wetland soils and catchment reservoirs can delay responses to emission reductions. Thus, although wetland-dominated systems often support higher absolute MeHg concentrations in empirical studies, they emerge in these simulations as the most dynamic responders to declining atmospheric inputs because legacy storage and delayed ecosystem adjustment are only partially represented. In contrast, emerging 3-D marine Hg models such as MERCY and OGSTM–BFM–Hg demonstrate how embedding Hg cycling and bioaccumulation in high-resolution physical–biogeochemical frameworks with multi-trophic food webs can resolve climate impacts and food-web–mediated exposure with far greater realism (Rosati et al., 2022; Bieser et al., 2023). Biogeochemical reaction rates and lake-chemistry variables (e.g., DOC, alkalinity, sulfur) could not be evaluated here due to data limitations but incorporating them in future work will be essential for linking model behavior with real-world ecosystem responses.