Climate and socioeconomics in the 2050s
The new scenario framework developed by the climate change research community over recent years consists of two sets of pathways: Representative Concentration Pathways (RCPs) that describe the extent of climate change and SSPs that depict plausible socioeconomic developments during the 21st century (Riahi et al. 2017). We selected RCP 8.5 (Riahi et al. 2007), one of the more severe climate change pathways, together with SSP1 (Sustainability), SSP2 (Middle of the road), and SSP5 (Fossil-fueled development). The use of SSPs is preferred over extending current trends when evaluating future impacts since future policies, political developments, or other unexpected events may drastically change the direction of current development.
We investigate the importance of these societal developments under one climate pathway (RCP8.5). By mid-century, the differences between RCP4.5—a less extreme pathway—and RCP8.5 are not as pronounced as they become by the end of the century (Hawkins and Sutton 2009). While higher greenhouse gas concentrations (GHC) scenarios are typically not used with SSP1, this combination is plausible by mid-century when taking into account emerging major emission sources (e.g. melting arctic peatland) from positive feedback loops in the natural system (Schuur et al. 2008).
Climate forcing data
Full climate ensemble data were collected from Coupled Model Intercomparison Project Phase 5 (CMIP5) projections, downscaled within the Coordinated Regional Downscaling Experiment (CORDEX, www.cordex.org; Jacob et al. 2014). In a pragmatic attempt to maximize the ensemble spread and to minimize the required number of simulations, four climate model (CM) projections with the highest and lowest changes in mean summer temperature and precipitation during a 30-year period around mid-century (2041–2071) were selected (Fig. 1). The changes were determined for a square region encompassing the BSDB south of 60 degrees latitude for each Global Climate Model (GCM) in combination with the Regional Climate Model (RCM) that was used to regionally downscale the GCM. The summer months (June, July and August) were deemed to be the most important for changes in nutrient concentrations due to the importance of plant nutrient uptake. However, the four selected CMs also encapsulate the range of changes seen in the annual averages (Fig. 1b).
Because the GCM/RCM combination CCLM-MPI-ESM-LR (see Supplementary Materials S1 for the lengthy acronyms in this section) represented both the lowest precipitation change and the lowest temperature change, a fourth GCM/RCM combination showing the 2nd lowest temperature change was chosen (RCA4-CNRM-CM5). Also, three models showed very similar changes in temperature but different changes in precipitation. Thus, we chose RCA4-CanESM2 instead of Arpege-CNRM-CM5 to further diversify the GCMs in the ensemble included in the modelling (Table 1).
The selected CMs were then bias-adjusted and downscaled to the required resolution using Distribution Based Scaling (DBS) (Yang et al. 2010). Here, we applied DBS to daily precipitation and temperature using Watch ERA-Interim Forcing Data (WFDEI, Weedon et al. 2011) as a reference dataset. The reference period for the calibration of the bias-adjustment parameters was set to 1991–2010.
Shared Socioeconomic Pathways (SSPs)
The changes in the model forcing data were interpreted through three regionally extended SSPs covering socioeconomic drivers that affect nutrient loading to the Baltic Sea (Zandersen et al. 2019). SSPs are used in the global climate research community to explore impacts associated with alternative climate and socioeconomic futures (van Vuuren et al. 2011; O’Neill et al. 2017). SSPs are quantitative and qualitative narratives of plausible socioeconomic futures up to the end of the century. The three SSPs used here are as follows:
SSP1 (Sustainability) describes a world making relatively good progress towards the United Nation’s (UN) Sustainable Development Goals (SDGs) while reducing resource intensity and fossil fuel dependency. The goals of the EU WFD and management plans for reducing nutrient loadings from agriculture would be fully implemented. Consumption trends would change towards less demand for meat. More sophisticated and comprehensive sewage treatment technologies would be adopted. Atmospheric deposition of nitrogen would be reduced following cleaner energy production and use of electric vehicles.
SSP2 (Middle of the road) describes a world where trends typical of recent decades continue with some progress towards achieving SDGs, including reductions in resource and energy intensity and a slow decrease in fossil fuel dependency. Larger farms, intensive farming, and industrialized and more effective agriculture would increase. Management plans (WFD) would be only partly implemented. Sewage treatment technology development and increased urbanization would lead to reduced nutrient loadings. Atmospheric deposition would follow the decrease in NOx emissions as hybrid and electric cars become more widely used.
SSP5 (Fossil-fueled development) is a world that stresses conventional development oriented towards economic growth with a high energy demand mostly met with carbon-based fuels. A global market for agricultural products combined with an increasing global demand for animal products for growing populations would lead to an increase in agriculture and livestock production in the Baltic Sea region. There would be less regulation of agricultural nutrient loadings, but innovations in production technologies may reduce nutrient emissions in relative terms. Increased urbanization and population growth would lead to a higher amount of wastewater, but with higher removal efficiencies due to improved sewage treatment technologies. New technologies to reduce NOx emissions would continue to expand but at a reduced rate compared to SSP1 and SSP2.
SSP1, 2, and 5 were implemented by quantifying changes in nutrient sources (Table 2) based on a spatial interpretation of qualitative narratives (Engardt et al. 2017; Zandersen et al. 2019) and the numerical projections available at the IIASA SSP Database (IIASA 2017). Point source loads reflect changes in the population size, efficiency of sewage treatment, and new investments in infrastructure. Land use and agricultural practices follow changes in population, urbanization, and food production. Note that the SSPs should not be interpreted as “the best case”, “the worst case”, or even as “the most likely case” scenarios as they only represent plausible future developments.
Nutrient impact modelling
Daily discharges and nutrient loads for current and future conditions were simulated with E-HYPE for the BSDB. HYPE is an integrated hydrological and nutrient transport model code developed by SMHI (Lindström et al. 2010). E-HYPE, a pan-European model built using the HYPE software, simulates rainfall, runoff, riverine processes, and nutrient processes in hydrologically delineated catchments with a median size of 215 km2 for all of Europe. We used E-HYPE v.3.1.4 (Hundecha et al. 2016; Bartosova et al. 2017).
The E-HYPE model v.3.1.4 includes deeper soils with active groundwater (European Hydrogeology map; BGR & UNESCO 2014) and reflects more recent crop distributions (Eurostat 2013) and point source discharges (Urban Wastewater Treatment Directive 2016). While these updates did not significantly affect model performance, we used a stepwise, representative gauged basin (RGB) approach (Strömqvist et al. 2012; Donnelly et al. 2016) to recalibrate selected model parameters that affect nutrient processes. In addition to the typical calibration approach where the outputs are compared to observed concentrations, we also reviewed model performance with respect to three sets of data associated with fundamental hydrological and biogeochemical processes: baseflow fraction in streamflow gauges, nitrogen leaching (Andersen et al. 2016), and the rate of nitrogen reduction in groundwater (Højberg et al. 2017). However, the latter two datasets were based on model analyses and expert judgement with varying spatial resolution and tools and data inputs used to produce the estimates. For example, the rate of nitrogen reduction in groundwater for Sweden was estimated from two different national hydrological models with a median catchment size of 7 km2, but for Germany it was estimated as a one constant value for all contributing catchments. Thus, nitrogen leaching and the rate of nitrogen reduction in groundwater were used to guide larger spatial patterns rather than for calibrating individual catchments.
The recalibration focused only on nutrient concentrations; model parameters that affect stream flow remained the same as in the previous E-HYPE version. At the pan-European domain, 46% of the 1015 stream gauges had Nash–Sutcliffe model efficiency (NSE, Nash and Sutcliffe 1970) greater than 0.5 and the relative error (RE) was within 50% at 85% of the gauged locations. The proportion of sites with NSE greater than 0.5 was the same in the BSDB (46% of 368 stream gauges), but a higher proportion of the sites (93%) had a relative error within 50%.
The E-HYPE model was recalibrated using 89 sites at the full pan-European scale (Fig. 2) because insufficient nutrient observations were available to capture the southern agricultural parts of the BSDB adequately. The calibration period was from January 1, 2001 to December 31, 2010 (10 years) with an initial warm-up period from 1979 to allow the model to achieve stable conditions. The remaining sites were used to validate the model using the same time period.
The relative error (RE) in the recalibrated model varied among the monitoring sites, with 65% and 66% of the sites having RE within 50% for total phosphorus (TP) and total nitrogen (TN) concentrations, respectively, across Europe. The median RE was 8% for TP and − 10% for TN. The recalibration is described fully in Bartosova et al. (2017).
Recalibration resulted in a considerable change in the internal representation of the biogeochemical processes in some areas with a model performance comparable to the original calibration. This is expected to increase the plausibility of the overall model results, especially with respect to nitrogen reduction and retention processes.
Current conditions were simulated with the calibrated model using a 30-year period from 1981 to 2010 (with a 10-year warm-up period from 1971). Future conditions were also simulated using a 30-year period representing the 2050s (from 2036 to 2065 with a 10-year warm-up period from 2027). Time slices were preferred over transient runs as long-term changes in nutrient storage within soils are difficult to validate and can have a large impact on scenario results.
Both current and future conditions were simulated using all four selected CMs. Seven E-HYPE v.3.1.4 model runs were executed with each CM: (1–2) current and 2050s periods with current land use and nutrient sources, (3–5) 2050s period with land use and nutrient sources representing SSP1, 2 and 5, and (6–7) current and 2050s periods with current land use and nutrient sources, but with model parameters prior to recalibration. The last set of model runs was used to test the robustness of the E-HYPE model simulations and the dependency of the results on certain model parameters relevant in nutrient processes.
The model output data were processed in R utilizing the R-package HYPEtools (SMHI 2018). The results from the CMs were then averaged and a relative change from the average of the current period with current land use and nutrient sources (i.e. the first of the seven runs) was calculated. Ensemble-average values for main river basins that discharge directly to the Baltic Sea were summarized to obtain total fluxes to the Baltic Sea.
We also looked closely into how different sectors in SSP2 affect the nutrient loads by simulating four additional sub-scenarios where only one of the following sectors was modified at a time: (1) atmospheric deposition, (2) land use and agricultural practices, (3) point source effluents, and (4) contributions from the rural population (Table 2). SSP2 was selected for this analysis because it is most closely aligned with recent trends in development. In order to enhance insights into the origin of nutrient loads across the BSDB, we conducted a source apportionment analysis for long-term average annual loads from a number of source groups under current and 2050s climate conditions. Modelled loads in HYPE were traced from their origin to user-defined outlet points within the model domain and aggregated as net loads to the Baltic Sea from agriculture, forests, pasture, mixed-use and semi-urban lands, non-forested (semi-)natural lands, rural households, wastewater treatment plants, and industrial effluents.