Future river exports of nutrients, plastics, and chemicals worldwide under climate-driven hydrological changes

LETTER • OPEN ACCESS Future river exports of nutrients, plastics, and chemicals worldwide under climate-driven hydrological changes To cite this article: Mirjam P Bak et al 2025 Environ. Res. Lett. 20 094033 View the article online for updates and enhancements. You may also like Permafrost vulnerability to climate change: understanding thaw dynamics and climate feedback of permafrost degradation Jing Tao, Anna K Liljedahl, Christopher R Burn et al. - Modelling future coastal water pollution: impacts of point sources, socio-economic developments & multiple pollutants Mirjam P Bak, Carolien Kroeze, Annette B G Janssen et al. - Advancing water quality model intercomparisons under global change: perspectives from the new ISIMIP water quality sector Maryna Strokal, Rohini Kumar, Mirjam P Bak et al. - This content was downloaded from IP address 147.125.55.115 on 15/12/2025 at 16:43 Environ. Res. Lett. 20 (2025) 094033 https://doi.org/10.1088/1748-9326/adf860 OPEN ACCESS RECEIVED 10 March 2025 REVISED 9 July 2025 ACCEPTED FOR PUBLICATION 6 August 2025 PUBLISHED 19 August 2025 Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. LETTER Future river exports of nutrients, plastics, and chemicals worldwide under climate-driven hydrological changes Mirjam P Bak1,7,∗, Ilaria Micella1,7,∗, Edward R Jones2, Rohini Kumar3, Albert Nkwasa4,5, Ting Tang6, Michelle T H van Vliet2, Mengru Wang1and Maryna Strokal1 1Earth Systems and Global Change Group, Wageningen University & Research, Wageningen, The Netherlands 2Department of Physical Geography, Faculty of Geosciences, Utrecht University, Utrecht, The Netherlands 3Department Computational Hydrosystems, Helmholtz Centre for Environmental Research GmbH—UFZ, Leipzig 04318, Germany 4Water Security Research Group, Biodiversity and Natural Resources Program, International Institute for Applied Systems Analysis (IIASA), Schlossplatz 1, A-2361 Laxenburg, Austria 5Department of Water and Climate, Vrije Universiteit Brussel (VUB), 1050 Brussel, Belgium 6Biological and Environmental Science and Engineering Division, King Abdullah University of Science and Technology, Thuwal, Saudi Arabia 7Equal first authorship. ∗Authors to whom any correspondence should be addressed. E-mail: [email protected] and [email protected] Keywords: water quality, multiple pollutants, building trust, climate, uncertainty, hydrological changes Supplementary material for this article is available online Abstract Future climate-driven hydrological changes may strongly affect river exports of multiple pollutants to coastal waters. In large-scale water quality (WQ) models the effects are, however, associated with uncertainties that may differ in space and time but are hardly studied worldwide and for multiple pollutants simultaneously. Moreover, explicit ways to assess climate-driven uncertainties in large-scale multi-pollutant assessments are currently limited. Here, we aim to build trust in future river exports of nutrients (i.e. nitrogen and phosphorus), plastics (i.e. micro and macroplastics), and chemicals (i.e. diclofenac and triclosan) under climate-driven hydrological changes on the sub-basin scale worldwide. We used a soft-coupled global hydrological (VIC) and WQ (MARINA-Multi) model system, driven by five Global Climate Models (GCMs), to quantify river exports of selected pollutants to seas for 2010 and 2050 under an economy-driven and high global warming scenario. Subsequently, we developed and applied a new approach to build trust in projected future trends in coastal water pollution for the selected pollutants. Results reveal that in arid regions, such as the Middle East, East Asia, and Northern Africa, climate-driven uncertainties play a key role in future river exports of pollutants. For African sub-basins, high increases in river exports of pollutants are projected by 2050 under climate-driven hydrological uncertainty. Nevertheless, over 80% of the global sub-basin areas agree on the direction of change in future river exports of individual pollutants for at least three GCMs. Multi-pollutant agreements differ among seas: 53% of the area agrees on increasing river exports of six pollutants into the Indian Ocean by 2050, whereas 17% agrees on decreasing trends for the Mediterranean Sea. Our study indicated that even under climate-driven hydrological uncertainties, large-scale WQ models remain useful tools for future WQ assessments. Yet, awareness and transparency of modelling uncertainties are essential when utilising model outputs for well-informed actions. 1. Introduction Nutrients, plastics, and chemicals enter rivers and then are exported to coastal waters [1–4]. Rivers export these pollutants often from common sources such as agricultural runoff and sewage systems [2, 5–8] impacting the aquatic environment. Nutrient pollution, for example, triggers harmful algal © 2025 The Author(s). Published by IOP Publishing Ltd Environ. Res. Lett. 20 (2025) 094033 M P Bak et al blooms [9,10], while plastics and chemicals disrupt ecosystems [11,12]. Today, many rivers and coastal waters are exposed to multi-pollutant issues [13–16]. In the future, water pollution is likely to increase due to socio-economic developments like urbanisation and population growth [15–18]. Climate change is expected to influence river exports of pollutants because of long-term changes in runoff, river discharge patterns [16,19,20], and water storage [21,22]. In large-scale water quality (WQ) models, this may affect flows of pollutants as well as their retention in river systems [17,18]. In addition, climate change affects terrestrial pollutant sources and biogeochemical processes. For example, rising temperatures can alter nutrient cycling and pollutant emissions [23]. This study, however, focuses specifically on the effects of climate-driven hydrological changes on river exports of multiple pollutants. The meteorological forcings (e.g. air temperature, precipitation) from global climate models (CGMs) are often used by global hydrological models to project runoff and river discharges, which are further used as input to global WQ models. Large-scale WQ models such as MARINA-Multi (Model to Assess River Inputs of pollutaNts to seAs) [18], IMAGE-GNM (Integrated Model to Assess the Global EnvironmentGlobal Nutrient Model) [24], SWAT+(Soil and Water Assessment Tool) [25], WorldQual [26], and DynQual (Dynamical Surface WQ model) [27] are the most suitable tools to study water pollution issues on regional to global scales. They account for hydrological flows driven by GCMs. Yet, many GCMs depend on climate forcings that differ largely in space and time [28,29], adding uncertainties to hydrological projections [30]. The effects of these uncertainties, particularly on river exports of nutrients, plastics, and chemicals, are hardly studied worldwide in a spatially explicit way (knowledge gap 1). Building trust under uncertainties associated with climate-driven hydrological changes is important for WQ assessments. Yet, to date, there is no comprehensive assessment of the uncertainties associated with hydrological drivers in large-scale WQ models. Traditional evaluation methods, such as model validation at the catchment scale, are inadequate for the complexities of large-scale, climate-driven models: e.g. large diversity in pollutants and limited observation data [31,32]. Hence, large-scale models need thorough evaluation to ensure accuracy and reliability, especially for policymaking and environmental management. Gleeson et al [33] and Strokal et al [32] emphasise the need for new evaluation methods that go beyond validation, especially for emerging pollutants lacking observations [31]. Strokal et al [32] presented a building trust approach for large-scale WQ models (see SI appendix B) with 13 strategies to evaluate model inputs, outputs, and structures via comparisons, sensitivity analysis, innovations, expert knowledge, and local models [32]. However, those strategies focus on individual models rather than propagating uncertainties through modelling chains for multiple pollutants. Hence, explicit ways to assess climate-driven hydrological uncertainties in global multi-pollutant assessments of coastal waters are limited in current building trust approaches (knowledge gap 2). Our study aims to build trust in future river exports of nutrients (i.e. nitrogen and phosphorus), plastics (i.e. micro and macro), and chemicals (i.e. triclosan and diclofenac) under climate-driven hydrological changes on the sub-basin scale worldwide. We define coastal water pollution as river exports of pollutants to seas (in loads). We used a softcoupled water quantity (variable infiltration capacity model; VIC [34,35]) and WQ (MARINA-Multi [18]) model system, driven by five GCMs, to simulate river exports of six pollutants in 2010 and 2050. We followed an economy-driven and high global warming scenario: Shared Socioeconomic Pathway 5 [36] and Representative Concentration Pathway 8.5 [37] (SSP5-RCP8.5) [38]. Then, we developed and applied a new approach to build trust in projected trends in coastal water pollution across GCMs and pollutants. Focusing on multiple pollutants simultaneously is important for two main reasons. First, since pollutants may have different sources and pathways, they may respond differently to climate-driven hydrological uncertainties [39]. Second, real-world exposures are typically to multiple pollutants [17,18], highlighting the need to understand the robustness of multipollutant trends under climate-driven hydrological uncertainties. This could support the development of environmental policies that are resilient to climatedriven uncertainties. Our study contributes to the first global-scale WQ model intercomparison effort as proposed by the WQ sector of the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP) which is an international collaborative effort that assesses climate change impacts. (www.isimip.org/). 2. Methodology 2.1. A soft-coupled model system We used a soft-coupled water quantity (VIC) and WQ (MARINA-Multi) model system: i.e. outputs of VIC were used as inputs to MARINA-Multi (figure 1). VIC provided hydrological data driven by five GCMs (figure 1). The MARINA-Multi model aims to analyse trends and sources of water pollution. Hence, it simulates annual river exports of dissolved inorganic (DIN, DIP) and dissolved organic (DON, DOP) nitrogen (N) and phosphorus (P), micro- (MIP), and macroplastics (MAP), triclosan (TCS), and diclofenac (DCL). We combined inorganic and 2 Environ. Res. Lett. 20 (2025) 094033 M P Bak et al Figure 1. Overview of the soft-coupled water quantity (light-grey dotted box) and water quality (dark-grey dotted box) model system. Global climate models (GCMs) provided inputs (forcings) to the hydrological VIC model that simulated river discharges for the water quality MARINA-multi model. MARINA-multi model outputs included river exports of nutrients (TDN, TDP), plastics (MIP, MAP) and chemicals (TCS, DCL) in loads (kg yr−1). We used five GCMs resulting in five model runs. We used the results of the five model runs to analyse future inter-GCM and multi-pollutant agreements for sub-basins worldwide (see table 1 for definitions). RCP8.5 is short for Representative Concentration Pathway 8.5. Source: see section 2.2 for references to the GCMs and model descriptions. organic nutrients into total dissolved N and P: TDN and TDP. This is done by source for 8890 sub-basins for 2010 and 2050 in three steps. First, the model simulates inputs of pollutants to rivers from point and diffuse sources (kg yr−1). Point sources include sewage systems and direct discharges of animal manure (only for China in 2010) and untreated human waste. Diffuse sources are distinguished between anthropogenic and nonanthropogenic (natural). Anthropogenic sources include synthetic fertilisers, animal manure, atmospheric N deposition on agricultural areas, biological N2fixation by crops, leaching of organic matter, weathering of P-contained minerals from agricultural areas and mismanaged plastic waste. For natural sources, the model includes atmospheric N deposition on non-agricultural areas, biological N2 fixation by natural vegetation, leaching of organic matter, and weathering of P-contained minerals from non-agricultural areas. Inputs of pollutants from land (diffuse sources) to rivers are corrected for the retention and losses in the soil. Second, the model simulates inputs of pollutants reaching the outlets of sub-basins (kg yr−1). These inputs are corrected for retention and losses during the export (e.g. river damming, water removals, denitrification). Third, the model simulates river exports of pollutants to the river mouths (coastal waters) (kg yr−1) while considering retention and losses. For our model runs, we used socio-economic and climate drivers following the combined storylines of SSP5-RCP8.5. This economy-driven scenario assumes high emissions and moderate population growth, with continued reliance on fossil fuels and a reactive approach to environmental challenges. Input data related to socio-economic aspects like population, urbanisation, land use, human development, wastewater treatment, agriculture, and waste management were directly taken from Micella et al [18] (see SI appendix A). VIC provided five runs for drivers namely river discharges based on five GCMs (figure 1and SI (appendix A). We selected five different GCMs following the Coupled Model Intercomparison Project 5 (CMIP5) [40] and ISIMIP2b [41] (www.isimip.org/): (1) MIROC-ESM-CHEM [42], (2) IPSL-CM5A-LR [43], (3) HadGEM2-ES [44], (4) NorESM1-M [45], and (5) GFDL-ESM2M [46]. This selection covered a variety of features: e.g. their components differ in their resolutions and interaction levels [29]. Each GCM was used by VIC (version 4.1.2 [34,35]) to simulate annual natural river discharges under RCP8.5. VIC is a widely used process-based hydrological model [19, 47–51] that provided data at the 0.5-degree grid scale. We averaged the data over 2005–2015 (for 2010) and 2045–2055 (for 2050) and processed it to the subbasin scale for MARINA-Multi [5,15,18], separately for all five GCMs (figure 1, see SI appendix A for details). We chose VIC because of its earlier integration into the MARINA-Multi model [19,47–51] (SI appendix A and SI appendix C) and VIC solves both surface energy and water balances [34,35]. The GCM forcing data for VIC were downscaled and bias-corrected following the trend-preserving ISIMIP approach [41]. 2.2. Building trust under climate-driven uncertainties We developed a three-stage approach to build trust under climate-driven uncertainties. This approach 3 Environ. Res. Lett. 20 (2025) 094033 M P Bak et al Table 1. Agreement classes on the direction of change (increases or decreases) between the year 2010 and the year 2050 in river exports of pollutants to seas. The agreement classes are used to assess inter-GCM agreement for individual pollutants (Stage 2 in section 2.2) and the multi-pollutant agreement for areas with moderate to very high inter-GCM agreements (Stage 3 in section 2.2). Our study includes five GCMs and six pollutants at the sub-basin scale. GCM is short for global climate model. Agreement classes Inter-GCM agreement (number of GCMs agreeing on the direction of change for individual pollutants out of the five GCMs) Multi-pollutant agreement (number of pollutants agreeing on the direction of changes out of the six pollutantsa) Very high 5/5 6/6 High 4/5 4-5/6 Moderate 3/5 3/6 Diverging — 2/6 or 3/6b Disagreement <3/5 ⩽2/6c aOnly applicable for areas with moderate to very high inter-GCM agreement. bEqual agreement among pollutants. This applies to two situations: (A) three pollutants agree on an increasing trend, three pollutants agree on a decreasing trend; (B) two pollutants agree on an increasing trend, two pollutants agree on a decreasing trend, two pollutants show disagreements in trend (i.e. inter-GCM disagreement). cThe majority of the pollutants disagree due to inter-GCM disagreements. Hence, the multi-pollutant agreement remains inconclusive. This applies to six situations: (A) two pollutants agree on an increasing trend, one pollutant shows a decreasing trend, three pollutants show disagreements in trends (i.e. inter-GCM disagreement); (B) two pollutants agree on a decreasing trend, one pollutant shows an increasing trend, three pollutants show disagreements in trends (i.e. inter-GCM disagreement); (C) one pollutant shows an increasing trend, one pollutant shows a decreasing trend, four pollutants show disagreements in trends (i.e. inter-GCM disagreement); (D) one pollutant shows an increasing trend, five pollutants show disagreements in trends (i.e. inter-GCM disagreement); (E) one pollutant shows an increasing trend, five pollutants show disagreements in trends (i.e. inter-GCM disagreement); (F) all six pollutants show disagreement in trends (i.e. inter-GCM agreement). complements the 13 alternative strategies as identified by Strokal et al [32]. In Stage 1, we analysed the ensemble mean and coefficient of variation (CV) in river exports of pollutants from five GCMs for 2010 and 2050. The ensemble mean was calculated by averaging the annual river exports of each pollutant per unit sub-basin area (kg km−2yr−1or g km−2yr−1) over five GCMs. The CV, calculated as the ratio of the standard deviation to the mean, indicates the spread in projected river exports of each pollutant among the five GCMs by sub-basin. In Stage 2, we evaluated inter-GCM agreement (table 1) for changes in river exports of single pollutants, focusing solely on 2010 and 2050. Agreements are associated with trust in model projections for individual pollutants, whereas disagreement indicates higher climate-driven uncertainty in projections. We first calculated the percentage change in river exports of pollutants by sub-basin between 2010 and 2050 per GCM (SI, figure H.1). Second, we set a 5% threshold for changes in river exports of pollutants to determine the direction of change per GCM: >5% indicates an increase in river export, <−5% indicates a decrease in river export, and changes between −5% and 5% are deemed inconclusive. Finally, we assessed the agreement across GCMs for individual pollutants by subbasin using the agreement classes as defined in table 1. In Stage 3, building on the outcomes of Stage 2 we analysed the multi-pollutant agreement (table 1) by sub-basin and by sea. Agreements are associated with strong multi-pollutant trends, whereas disagreements indicate uncertainty in multi-pollutant trends. We estimated the area share of multi-pollutant agreement classes for five large seas in the world: the Arctic Sea, Mediterranean Sea, Atlantic Ocean, Pacific Ocean, and Indian Ocean (SI, figure E.1). 3. Results 3.1. Ensemble means and variability for individual pollutants (Stage 1) Pollutant loads are projected to be high in many sub-basins of Asia, Europe, and Central America in 2050 (figure 2). This holds for most pollutants: >900 kg km−2yr−1for TDN, >50 kg−2yr−1for TDP, >1.5 kg−2yr−1for MIP, >3 g−2yr−1for TCS, and >0.9 g−2yr−1for DCL. Exceptions are many subbasins of Africa and Asia where rivers are projected to export more MAP (>6 kg−2yr−1) compared to sub-basins elsewhere in the world. Generally, rivers are projected to export much TDN (50 Tg yr−1globally) compared to other pollutants (e.g. 2.7 Tg yr−1 for TDP, 0.6 Tg yr−1for MAP, and 0.2 Tg yr−1for DCL globally). Climate-driven uncertainties play a key role in water pollution in (highly) arid areas in 2050 (figure 2). For example, the spread in river exports of all pollutants among the five GCMs (measured by CV) is generally large in the Middle Eastern, East Asian, South Asian, Northern African, and some 4 Environ. Res. Lett. 20 (2025) 094033 M P Bak et al Figure 2. Ensemble means of river exports of individual pollutants in loads at the sub-basin scale worldwide (left panels, kg km−2y−1or g km−2yr−1) and their coefficient of variation associated with climate-driven hydrological changes (right panels, unitless) for the year 2050. The ensemble mean is estimated over five model runs, each of which is based on hydrology simulated using climate forcings from one of the five global climate models. CV is short for the coefficient of variation, which is the ratio of standard deviation to the mean. ∗=bins of CVs are different for nutrients (TDN, TDP) compared to plastics (MIP, MAP) and chemicals (TCS, DCL) to show spatial variability. Pollutants include total dissolved nitrogen (TDN), total dissolved phosphorus (TDP), microplastics (MIP), macroplastics (MAP), triclosan (TCS), and diclofenac (DCL). 2050 is based on Shared Socioeconomic Pathway 5 (rapid urbanisation and high economic development) and Representative Concentrative Pathway 8.5 (high global warming). NS (Not part of the study area) denotes sub-basins that are not part of our study area as they do not drain into the seas or are part of Greenland (see SI, figure E.1 for details). Source: the MARINA-Multi model [52] (see section 2for the model and scenario descriptions). North and Central American sub-basins (figure 2). For those sub-basins, CVs are highest for nutrients (>0.45), but also relatively higher for plastics and chemicals (>0.05) compared to other regions. For nutrients, the spread is also projected to be large in Australian sub-basins. Conversely, the spread is generally small for all pollutants in sub-basins across (sub)arctic regions, Southeast Asia, and South America. Among pollutants, the spread is approximately four times larger for nutrients than for plastics and chemicals (figure 2). For Africa, mean river exports of pollutants are projected to increase largely under high climatedriven uncertainties. To illustrate, pollution levels in rivers are projected to rise under global change by 2050, ranging from 29%–206% across pollutants. 5 Environ. Res. Lett. 20 (2025) 094033 M P Bak et al Figure 3. Inter-GCM agreements for the direction of change (increases or decreases) in river exports of individual pollutants between 2010 and 2050 at the sub-basin scale. Maps show sub-basins for which three (moderate agreement), four (high agreement), or five (very high agreement) GCMs agree on the direction of change or sub-basins for which less than three GCMs agree (disagreement) on the direction of change. For more details regarding agreement classes see table 1. Horizontal bars show the share of the global sub-basin area for each agreement class (table 1). This shows the results using a 5% threshold for changes in river export of pollutants to determine the direction of change (see SI appendix I, for results using a 1% and 10% threshold). GCMs are short for global climate models. Pollutants include total dissolved nitrogen (TDN), total dissolved phosphorus (TDP), microplastics (MIP), macroplastics (MAP), triclosan (TCS), and diclofenac (DCL). 2050 is based on Shared Socioeconomic Pathway 5 (rapid urbanisation and high economic development) and Representative Concentrative Pathway 8.5 (high global warming). NS (Not part of the study area) denotes sub-basins that are not part of our study area as they do not drain into the seas or are part of Greenland (see SI, figure E.1 for details). Source: the MARINA-Multi model [52] (see section 2for the model and scenario descriptions). These increases in river exports of pollutants are often accompanied by a larger spread in future projections, suggesting the importance of climate-driven uncertainties in Africa: e.g. 48%–60% of the sub-basin areas show higher CVs for pollutants in 2050 compared to 2010 (figures 2and SI appendix G). Asian sub-basins show similar patterns, though less extreme, with pollution levels ranging from a 9% decrease to a 93% increase and higher CVs in 14%–38% of the areas. Other regions displayed varying trends depending on the pollutant and projection (see SI appendix G). 3.2. Inter-GCM agreements (Stage 2) Our results show inter-GCM agreements for over 80% of the sub-basin areas globally (figure 3). These areas agree on the direction of change (increases or decreases) in future river exports of individual pollutants between 2010 and 2050 for at least three GCMs (table 1). For increases in future pollution levels, 45%– 84% of sub-basin areas globally show moderate to very high inter-GCM agreement (figure 3, table 1 for definitions of the agreement classes). This 6 Environ. Res. Lett. 20 (2025) 094033 M P Bak et al range depends on pollutants. For increases in river exports of nutrients, approximately two-thirds of sub-basin areas show moderate to very high interGCM agreements, especially for many sub-basins of Southeast Asia, Sub-Saharan Africa, and parts of North America. For increases in river exports of plastics, very high inter-GCM agreements range from 47% (MIP) to 84% (MAP) of the sub-basin areas globally. For chemicals, these ranges are 45%– 50% (figure 3). Many North American, Sub-Saharan African, and Southeast Asian sub-basins show very high inter-GCM agreements for increases in river exports of MAP and chemicals. Anthropogenic sources are expected to play an important role in subbasins with moderate to very high inter-GCM agreements (on increases or decreases). This is because agricultural activities (e.g. fertilisers, animal manure, sewage) are projected to contribute 50% of TDN and 70% of TDP in coastal waters globally in 2050 (SI, figure J.1). By 2050, household sources, such as laundry and dust, are expected to dominate MIP export by rivers (SI, figure J.1). For decreases in future pollution levels, 10%– 47% of sub-basin areas globally show moderate to very high inter-GCM agreements (figure 3, table 1). For nutrients, 32%–34% of the sub-basin areas show moderate to very high inter-GCM agreements, particularly along the west coast of North America, Europe, Eastern Asia, and the east coast of Australia. Results show very high inter-GCM agreement on decreasing trends in river exports of plastics (10%–47% of the area) and chemicals (35%–42% of the area). This especially holds for many sub-basins of Europe and the east coast of North America. While many regions project increases in river exports of MIP by 2050 (see the previous paragraph), some sub-basins, especially in Eastern Asia, are expected to export less. Disagreements among GCMs on future trends in river exports of pollutants are estimated for 5%–20% of the sub-basin areas globally (figure 3). For example, the lowest inter-GCM disagreement is estimated for future river exports of MIP (5% of the area) because of the greater effects of anthropogenic sources (e.g. household dust in sewage) compared to hydrology. For nutrients, the disagreements are estimated for over 15% of the sub-basin areas. This is primarily due to the large contribution of natural sources to future nutrient pollution in those areas (SI, figure J.1). The highest disagreement is, however, estimated for future river exports of chemicals (15%–20% of the global surface areas). 3.3. Multi-pollutant agreements (Stage 3) For most of the global sub-basin areas, multipollutant agreements (⩾3 GCMs for ⩾3 pollutants, table 1) are estimated for increases or decreases in river exports by 2050 (figure 4). For increases, very high multi-pollutant agreements cover regions such as Sub-Saharan Africa, South Asia, and Subarctic North America (table 1for agreement classes). High agreements predominantly cover regions like Eastern South America, parts of the United States, and Europe. For decreases, very high agreements appear in scattered locations and high agreements are prevalent in large parts of Asia, Mexico and parts of Europe and South America. For both directions (increases and decreases), areas of moderate agreement are scattered. Diverging trends or disagreements cover parts of Northern Asia and Southern America or parts of North America. This implies that, although interGCM agreements exist for individual pollutants, their responses to urbanisation and climate change vary among areas. Those regions are often characterised by increases in river exports of nutrients, and MIP, whereas MAP and chemicals are projected to decrease by 2050. Contrarily, areas of multi-pollutant disagreement are often associated with prominent inter-GCM disagreements for individual pollutants, indicating the presence of climate-driven uncertainties. Multi-pollutant agreements on future trends for 2050 differ among seas (figure 4, pies). This specifically holds for coastal waters of the Indian Ocean and the Mediterranean Sea, which show opposite trends. Projections for the Indian Ocean show multi-pollutant agreements on increases for 81% of its drainage area (figure 4). In contrast, for the Mediterranean Sea, multi-pollutant agreements are on decreases for 64% of its drainage area. This differs from other coastal waters. For the coastal waters of the Atlantic Ocean, multi-pollutant agreements on increases in future pollution are estimated for nearly two-thirds of its drainage area. In the Pacific Ocean, this is for 41% of the drainage area, whereas 33% agrees on decreases and 18% shows diverging trends. The Arctic Ocean has a mix of multi-pollutant agreements (figure 4). 4. Discussion 4.1. Water Quality in a changing climate Climate change affects the water cycle, and in turn, the WQ. Although hydrological changes remain uncertain [53,54], inter-GCM agreements highlight hotspots of future wetter and drier conditions [55] (selected examples of agreement approaches in SI appendix D). Under RCP 8.5 (high emissions), approximately five billion people could experience substantial shifts in precipitation patterns by 2100 [55]. Comparing Trancoso et al’s [55] agreements in water quantity trends with our agreements on WQ trends (figures 3,4and SI appendix I), we find that some wetting regions (e.g. Northern Europe, and Northern America) show increasing pollution levels. Yet, this does not apply to all areas, implying that socio-economic drivers play an important role in WQ trends [17,18,39]. Generally, arid areas (e.g. Saharan Africa or Australia) show relatively low wetting or 7 Environ. Res. Lett. 20 (2025) 094033 M P Bak et al Figure 4. Multi-pollutant agreements on the direction of change (increases or decreases) in their river exports between 2010 and 2050. The map shows the multi-pollutant agreement (as defined in table 1) in the direction of change in river exports for at least three GCMs across six pollutants in a spatially explicit way. The pies show the area share of multi-pollutant agreement classes for five large seas in the world (i.e. see SI, Figure E.1 for a specification of the drainage areas by sea). Multi-pollutant agreement classes include: moderate (3/6 pollutants agree) high (4–5/6 pollutants agree), and very high (6/6 pollutants agree), disagreement (<3 pollutants agree) and diverging (an equal number of pollutants, i.e. 2/6 or 3/6 pollutants, agree on each direction). See table 1 for details on agreement classes. NS (Not part of the Study area) denotes sub-basins that are not part of our study area as they do not drain into the seas or are part of Greenland (see SI, Figure E.1 for details). Pollutants include total dissolved nitrogen (TDN), total dissolved phosphorus (TDP), microplastics (MIP), macroplastics (MAP), triclosan (TCS), and diclofenac (DCL). Source: the MARINA-Multi model [52] (see section 2for the model and scenario descriptions). drying agreements [55,56]. This aligns with our findings of greater variability in natural river discharge (higher CV, see SI appendix F) and pollutant exports (higher CV in figure 2). As we used 10-year averaged hydrological inputs for five GCMs, our sample size was relatively small and may have introduced biases in our CV results. While averaging reduced the effect of cascading uncertainties (SI, figure L.3), we may have underor overestimated uncertainties related to dry and wet years (SI, figures L.1 and L.2). The main message remains unchanged when accounting for yearly hydrological inputs (55 model runs, SI, figure L.2), while the results require careful interpretation. Our results show that in an economically driven future with reactive environmental management and high-emissions (SSP5-RCP8.5), river exports of studied pollutants will increase globally, with greater climate-driven uncertainty in model simulations across all analysed regions. This highlights the need to act. This especially holds for areas like Sub-Saharan Africa, where monitoring data are lacking [17], pollution levels are projected to increase substantially, and model uncertainty is greatest (figure 2). In regions with high river exports of pollutants, investments might be useful to focus on greater political and public awareness of WQ issues [31], along with identifying and implementing effective solutions to tackle regional pollution challenges. This requires an understanding of climate-driven hydrological uncertainties in WQ models (this study), drivers of pollution in hotspot areas [57], technological developments [58, 59], alternative treatment pathways (e.g. constructed wetlands) [60], and awareness campaigns [61– 63]. In areas with higher climate-driven uncertainties, investments could be useful in mitigating climaterelated WQ risks via monitoring strategies that ensure accessible and transparent outputs. For example, accessible monitoring data could help to enhance our understanding of prominent issues today (i.e. evaluation of current status and supporting decisionmaking), while preparing for arising issues in the decades to come (i.e. reduce uncertainties in global water quantity and quality models) [64]. In figure M.1 of the SI, we show that an alternative future with proactive environmental management and low-emissions (SSP1-RCP2.6) can substantially limit pollution and 8