Study area
The drainage basin of the Baltic Sea includes all or part of 14 countries (Fig. 1a in ESM). We do not further discuss the Czech Republic, Norway, Slovakia, or Ukraine, because these countries occupy < 3% of the basin area. We defined sub-national regions using the EU’s data collection system (Nomenclature of Territorial Units for Statistics, NUTS2) and oblasts (Russia and Belarus) within the drainage basin of the sea. All countries have multiple sub-national regions, except for Estonia, Latvia, and Lithuania where the NUTS2 delineation is the same as the entire country.
The Baltic Sea has seven sub-basins: Baltic Proper, (BP) Bothnian Bay, Bothnian Sea, Danish Straits, Gulf of Finland (GF), Gulf of Riga (GR), and Kattegat (Fig. 1b in ESM). Land areas that drain to each sea sub-basin are detailed in Online Resource 1.
Net anthropogenic nitrogen and phosphorus inputs
Our work is based on the NANI-NAPI nutrient accounting approach that is described in detail by Swaney et al. (2012). Briefly, NANI was estimated as the sum of imported inorganic fertilizer, net import or export of N embedded in food and feed, biological N-fixation (BNF) by agricultural plants, and oxidized atmospheric N-deposition. NAPI was estimated as the sum of imported fertilizer, and import or export of P embedded in food and feed, and imported P detergent. Nitrogen and P that were converted to livestock manure represented recycled nutrients, not new sources. As such, manure is not typically calculated under the NANI-NAPI approach; however, Hong et al. (2017) explicitly calculated manure in order to estimate agricultural nutrient surpluses and explore NUE in crop production.
Previous research found strong, linear relationships between NANI and NAPI and riverine fluxes of N and P, respectively, for regions in North America, Europe, and Asia (Howarth et al. 2012). As a result, riverine N and P inputs resulting from human activities can be estimated as the product of NANI and NAPI, respectively, and land-to-sea transfer efficiencies for each sea sub-basin. Transfer efficiency, the proportion of NANI and NAPI exported from land to sea for the Baltic Sea as a whole, was 14 and 4%, respectively, in 2010. Transfer efficiencies did not differ significantly between 2000 and 2010, despite substantial changes in regional NANI and NAPI over the same period (− 40 to + 58 and − 25 to + 19%, respectively, depending on sub-basin) (Hong et al. 2017). These results suggest that we can estimate potential reductions in nutrient loads as product of the reduction in NANI-NAPI and the transfer efficiency; as a result, we held transfer efficiencies constant across scenarios (Online Resource 1).
Nutrient use efficiency in crop and forage production
We estimated NUE for the 5-year periods centered on 2000 and 2010 as
$$ {\mathrm{N}\mathrm{UE}}_{\mathrm{N}}={\mathrm{Plant}}_{\mathrm{N}}/\left({\mathrm{Fert}}_{\mathrm{N}}+\mathrm{BNF}+{\mathrm{Dep}}_{\mathrm{N}}+{\mathrm{Man}}_{\mathrm{N}}\right) $$
(1)
and
$$ {\mathrm{NUE}}_{\mathrm{P}}={\mathrm{P}\mathrm{lant}}_{\mathrm{P}}/\left({\mathrm{Fert}}_{\mathrm{P}}+{\mathrm{Man}}_{\mathrm{P}}\right). $$
(2)
PlantX was nutrients removed in harvested crops and by grazing livestock, FertX was imported inorganic fertilizer, and ManX was livestock manure applied to crops and deposited on grazed areas by livestock, where X was N or P (all in kg per hectare of utilized agricultural area (UAA)). BNF was biological N-fixation by agricultural plants. DepN was atmospheric deposition of oxidized-N. ManX was less than the amount excreted by animals because of leaching and volatilization losses (Online Resource 1).
One can estimate N and P surpluses on agricultural land from these same components as the difference between the denominator and the numerator in Eqs. 1 and 2, respectively (i.e., inputs minus outputs). All components of NUE were obtained from Hong et al. (2017), who used statistical databases for the EU, Russian Federation, and Belarus and published conversion parameters (e.g., nutrient content in excreta of different livestock) for NANI-NAPI calculations (Online Resource 2). We also calculated livestock units from data provided by Hong et al. (2017) using standard coefficients from Eurostat (2013) to facilitate analysis.
Future scenarios
We constructed scenarios to explore the potential to meet BSAP nutrient reduction targets by reducing the imported synthetic and mineral fertilizer component of NANI-NAPI (HELCOM 2013). Our approach assumed that these reductions were achieved by redistributing manure nutrients within a region, from areas that focus on livestock production, where over-application of nutrients occurs, to areas that focus on crop production, where it is substituted for some of the imported fertilizer. Specialization and spatial separation of crop and livestock production systems can result in over-application of nutrients in regions that have large amounts of manure in relation to arable land (Nesme et al. 2015).
Our approach to estimating potential reductions in NANI-NAPI iteratively increased the minimum NUE for a region (as shown in Fig. 1a in ESM) until certain stopping conditions were met. In concept, regions with low NUE bore greater NANI-NAPI reductions than regions with already high NUE. Through this process, NUE for a region could only increase, not decrease. Starting with the regions with the lowest NUE, we reduced the fertilizer component (thus reducing NANI-NAPI and increasing NUE) until one of three conditions was met: (1) the new regional NUE met a theoretical limit, (2) fertilizer import in a region was reduced to zero (in such cases, crop nutrient needs equal to PlantX were met by ManX), or (3) the BSAP reduction target for the sea sub-basin was met. There was no transfer of manure between regions.
We capped regional NUEN at 0.75, a theoretical limit for Europe that is below the 90% established by the EU Nitrogen Expert Panel (2015), because using manure-N efficiently in crop production in mixed crop-livestock systems is challenging (Zhang et al. 2015).
Identifying a limit for NUEP was more difficult, because farmers are often advised to apply sufficient P to compensate for the amount removed in crop harvest once a certain level of soil P-availability is established and to apply less than the amount removed in crop harvest if soil P levels are above the recommended range (Jordan-Meille et al. 2012). If soils P levels are very high, crop yields can be maintained even at very low or no P application (Le Noe et al. 2018). It was beyond the capability of our approach to consider existing soil reserves. For simplicity, we assumed that NUEP does not exceed 0.9, recognizing that some inefficiency is inevitable.
Where regional NUE for 2010 exceeded the theoretical limit, we increased fertilizer imports (and thus, NANI-NAPI and river loads to the sea) for that region until NUEN of 0.75 or NUEP of 0.90 was achieved. We made this adjustment in the scenarios in order to acknowledge that NUE above theoretical limits is not sustainable for long periods.
The three scenarios were designed to compare outcomes if reductions in nutrient loads aligned with country-allocated reduction targets (CART) under the BSAP or if reductions were made in regions where there is greater opportunity to improve agronomic practices, as indicated by low NUE:
-
OptCART: NANI-NAPI reductions were optimized to meet CART, and no country-level reduction exceeded their BSAP commitment.
-
NoCART: NANI-NAPI reductions were prioritized for regions with low NUE, irrespective of CART. The intent of this scenario was to explore “best-case” nutrient reductions; thus, it was possible that a country could bear greater nutrient reductions than its BSAP commitment.
-
NoCART-EU: Same as NoCART scenario but for EU countries only. The intent of this scenario was the same as for NoCART, but we excluded Russia and Belarus, because common policy instruments exist to address nutrients from agriculture within the EU. There was no change in NANI-NAPI from year 2010 for Russia and Belarus.
We assumed that reducing nutrient application rates did not affect crop yield. We recognize this to be a simplification, because the fertilizer-yield curve shows that, all things being equal, reducing nutrient application rates reduces crop yield. However, in the low-response segment of the yield curve, crop yield responses to nutrient inputs tend to diminish when nutrient application rates are high, suggesting that reducing “over-application” will have minimal impact on yield.
Historical nutrient budgets have found that for many countries, decreases in N inputs have been accompanied by constant or increasing crop yields (Lassaletta et al. 2014). Such results have been credited to changes in agricultural practices, such as the placement and timing of nutrient application and the use of improved seed varieties (Cassman et al. 2002). For the scenarios, we assumed that reductions in N inputs were accompanied by changes in, for example, how, when, and where nutrients were applied (Kirchmann et al. 2002). In the case of P, a synthesis of field-scale studies found negligible to low yield responses to fertilization on soils with medium or high soil P levels (which is the case for much of the Baltic Sea region per Tóth et al. (2014)) (Valkama et al. 2009).
We only explored land-based nutrient reductions to BP, GF, and GR, because the latest HELCOM assessment found that further nutrient reductions are needed to achieve BSAP targets for these sub-basins, while targets for other sub-basins have been met or nearly met (Svendsen et al. 2015a).
Eutrophication responses to reduced riverine nutrient inputs
We used the Simple as Necessary Baltic Long Term Large Scale (SANBALTS) model to explore the effect of scenarios of reduced nutrients on eutrophication conditions in the sea in comparison to 1900, 2010, and BSAP-target conditions. This marine model was used to develop nutrient reduction targets for the BSAP in 2007. For the scenarios, we only altered data inputs related to riverine nutrient loads to BP, GF, and GR and held other data inputs constant (Online Resource 3).
SANBALTS simulates steady-state, coupled-N and coupled-P cycles in the seven sub-basins of the sea. The model is driven by external nutrient inputs, biogeochemical fluxes in sub-basins, and transport between sub-basins (http://apps.nest.su.se/nest/). Wulff et al. (2013) provide the model documentation. Briefly, the rate of N-fixation by phytoplankton depends on the ratio of water-column N/P concentrations and the stoichiometric P surplus. When the N/P ratio is < 7 (mass basis), there is a surplus of P relative to N under to the Redfield ratio, which creates conditions that favor N-fixation. The modeled Secchi depth is based on empirical relationships between mean water clarity and water-column N and P concentrations. Savchuk and Wulff (2009) reported that SANBALTS output compared well with measurement-derived data for a number of ecosystem components, such as sub-basin average nutrient concentrations, Secchi depth, and hypoxic area.
We used SANBALTS-derived estimates for water-column concentrations of TN and TP (both in μmol L−1) and Secchi depth (m) as response variables for the scenarios. We also used the size of the hypoxic area (km2) for the BP; this is the only sub-basin for which SANBALTS estimates hypoxia, because most hypoxic areas are located there.