Water resources of the Desna river basin under future climate

Water resources management must take into account climate change because global warming evidently accelerates (Allen et al. 2018). In Ukraine, the highest warming rate for the period 1981–2010 was observed for the northern and northeastern parts with the rates of annual mean temperature 0.58–0.66 °C/10 years (Balabukh & Malitskaya 2017). Annual precipitation has not significantly changed, but monthly values have smoothed out, and heavy rain frequency has risen (Grebin 2010; Balabukh et al. 2018; Palamarchuk & Shedemenko 2020). In the river runoff, the spring flood and flood peak have decreased, and the annual minimum discharge has increased along with the groundwater ratio in total runoff (Grebin 2010).

Ukraine belongs to the countries with insufficient water supply and uneven distribution of water flow over the territory (Stashuk et al. 2014). Water resources of the Dnipro, which is the main waterway of Ukraine, are mostly formed in the north-western part of the country, Belarus, and the Russian Federation by the transboundary rivers Prypiat (30% of the Dnipro discharge), Upper Dnipro (31%), and Desna (27%). Then, the water flow is accumulated by six reservoirs and redistributed by 15 artificial channels to households, heavy industry, and agriculture in water-scarce central and south Ukraine.

The Desna mean water runoff decreased by only 6% in 1991–2020 compared with 1961–1990. However, in the last decade (2011–2020), the mean annual discharge was the lowest since observation started in 1895. If the trend continues, it could be a risk for the Dnipro downstream water supply infrastructure. Secondly, a decrease in river discharge causes a higher concentration of water pollutants from municipal wastewater plants and agriculture practices (Osadcha et al. 2020). Water quality is critical because the Desna river outlets supplies the capital Kyiv with fresh water.

For now, only continental- or country-scale studies cover the projected changes in water yield, discharge, and soil water within the Desna basin (Krakovska & Gnatiuk 2013; van Vliet et al. 2015; Rouholahnejad et al. 2017; Didovets et al. 2020; Loboda & Kozlov 2020; Mentaschi et al. 2020; Snizhko et al. 2020). The results show a wide spread of the annual discharge change even in the near future: from −35% (Loboda & Kozlov 2020) to +55% (Mentaschi et al. 2020). The studies omit the intra-annual changes in water balance components. Moreover, zooming in on specific catchment from large-scale maps is not recommended for climate impact assessment (Krysanova et al. 2018).

In the last decades, the ecohydrological process-based river basin models are used globally to support decision-making processes in a changing environment (Fatichi et al. 2016). These models provide complex insights into ecosystem-climate behavior and could be used to quantify the effectiveness of possible adaptive measures. In Ukraine, the SWIM (Soil and Water Integrated Model) was used to assess climate change impact on flood risk and water resources in several mesoscale catchments (Didovets et al. 2017, 2019), and the SWAT (Soil and Water Assessment Tool) model was used to estimate future water budgets in the Upper Western Bug catchment (Fischer et al. 2014).

This study aims to provide a quantitative background for the water resources management within the Desna basin, predict trends of environmental issues, and highlight risks of climate change. To achieve robust outcomes, the research calibrates the daily discharge of 12 gauges SWAT model and wide spread of possible future scenarios: eight high-resolution (∼12 km) climate models from the Euro-CORDEX initiative (Jacob et al. 2014). The multi-model ensemble couples three regional climate models (RCMs) forced by five general circulation models (GCMs) from CMIP5 (Coupled Model Intercomparison Project 5) and includes two representative concentration pathways (RCPs), the RCP4.5 and RCP8.5 scenarios (Moss et al. 2010). The past-performance approach was used to verify the ability of climate models to represent long-term monthly variations of temperature and precipitation. The results summarize the annual and intra-annual spatial changes in temperatures, water balance components, and weather extremes for two future periods 2021–2050 and 2071–2100 over the baseline 1991–2020. In the discussion, the results are addressed to the vulnerabilities of the Densa basin over climate change.

The Desna river basin is located in the central part of the East European Plain: the upstream part – in the Russian Federation (62%) within Smolenskaya, Bryanskaya, Kurskaya, Kaluzhskaya, and Tulska Oblasts; and the downstream part – in Ukraine (38%) within Chernihivs'ka, Sums'ka, and Kyivs'ka Oblasts (Figure 1). The Tulskaya mixed forest zone dominates across the basin, except for the forest-steppe zone in the southeast part of the basin. The landscape is plain, mainly cut by gullies and ravines.

Eutric podzoluvisols cover 42% of the basin territory; their gleyed species are formed in the depressed areas (4.5%) (Figure 1(b)). The rest of the territory is covered by greyzems (24%), chernozems (10%), phaeozems (10%), fluvisols over floodplains (10%), and histosols (4%).

For the Desna SWAT model, we used the merged land cover layer from GlobeLand30 (http://www.globallandcover.com/, 2010 land cover) and Global Forest Change (https://earthenginepartners.appspot.com/science-2013-global-forest, 2012 forest state) (Hlotka 2017). Agricultural lands cover 55% of the basin territory, mixed forests – 29%, meadows – 9%, residential areas – 3%, wetlands – 2%, and water – 2%. (Figure 1(c)). Crop production is the leading economic sector of the Chernihiv and Sumy regions.

The Desna basin has a moderate continental climate with both warm humid summer and winter. According to the E-OBS dataset (ver. 23.1) (Cornes et al. 2018), the mean annual temperature increased by 1.3 °C compared with the period of 1961–1990. The highest growth was in winter, by 2 °C, followed by summer by 1.3 °C, then spring by 1.2 °C, and autumn by 0.8 °C (Table 1). Summer precipitation decreased by 14%, while precipitation in spring and autumn increased by 7 and 8%, respectively, and did not change in winter.

Snowmelt drives a hydrological regime inducing spring flood. But the spring flood peak tremendously decreased in the last decades (Figure 2). Instead, the cold season discharge increased due to more frequent thaws and higher precipitation in October, January, and February (Grebin 2010) (Table 1).

The SWAT is an open-source semi-distributed process-based ecohydrological model that is globally used to support the decision-making process in a complex environment (Arnold et al. 2012).

The SWAT assimilates weather and landscape information to calculate the hydrological cycle as well as sediment and nutrient transport. The input data include daily weather parameters (maximum and minimum air temperature, precipitation, relative humidity, wind speed, and solar radiation), DEM, soils, land cover, and agricultural operation schedule (e.g., planting of crops, fertilization, tillage, harvest). The watershed is delineated into subbasins where areas with similar land cover, soils, and slope are grouped into hydrological response units (HRUs). The water balance is simulated for each HRU calculating evapotranspiration, infiltration, capillary flow, shallow and deep aquifer recharge, surface, lateral, and groundwater flows. In comparison with previous versions, the SWAT2012 includes deep groundwater flow in total water yield (https://swat.tamu.edu/media/116557/rev681_source.zip, file subbasin.f, lines 411–412). Then, the water is routed through the river network calculating the discharge of the reach in each subbasin.

In a previous study, we calibrated and validated the SWAT model of the Desna basin for the period 2008–2014 (Osypov et al. 2018). For this research, we extended the modeling period to 2008–2019. The calibration was based on daily river discharge (12 gauges), snow cover depth (13 stations), and crop yield. The performance criteria were good (NS > 0.7, R² > 0.75, PBIAS < 10%) for all five gauges located in Ukraine.

To assess the impact of climate change, the SWAT model is sequentially fed by the climate variables of each RCM. The results are aggregated for the subbasins of the Desna watershed, the main channel, and the tributaries more than 450 km². The geospatial data are presented using Python packages: geopandas (https://geopandas.org/), matplotlib (https://matplotlib.org/), and mapclassify (https://pypi.org/project/mapclassify/).

For Europe, the Euro-CORDEX initiative provides high-resolution (∼12 km) bias-adjusted temperature and precipitation data of RCMs (Jacob et al. 2014) that are based on the five CMIP5 global climate projections (Taylor et al. 2012) and the new RCPs (Moss et al. 2010). Bias correction is commonly performed for precipitation and temperature as the major hydrological drivers. It improves the reliability of simulated hydrological process and projected mean river discharge (Huang et al. 2014).

Regional climate impact studies usually reduce the number of climate models because of computational reasons trying to keep the range of possible future scenarios. We selected eight RCMs that have climate variables required for the SWAT simulation and have both RCP4.5 and RCP8.5 emission scenarios (Table 2).

A climate model should be able to reproduce the behavior of hydrology-relevant parameters (e.g., precipitation and temperature) in the past for the studied watershed (Yuan et al. 2020). On the other hand, the selection of only the ‘best’ performance model may omit possible future scenarios (Lutz et al. 2016). There is no evidence that the best performing models in the past are the most realistic with regard to a climate signal. Therefore, it seems reasonable that the model performance in the past should rather be used to detect and remove severely unrealistic models (Mendlik & Gobiet 2016).

The RCA4 model forced by HadGEM2-ES and MPI-ESM-LR showed the best past performance, but the criteria of other models were close to each other (Table 3). However, RCA4 forced by CNRM-CM5 and EC-EARTH showed the worst results and substantially missed the long-term monthly cycle of precipitation (Appendix A), therefore, it was excluded from the analysis of the SWAT output.

Climate impact studies should provide traceable findings to evaluate the relevance of possible adaptation measures. To communicate the degree of certainty of the results, the calibrated language developed in (Mastrandrea et al. 2010) is widely used in climate projection studies, including IPCC reports. The core vocabulary includes:

• a.

  Consistency of evidence (e.g., quality of models): ‘limited’, ‘medium’, ‘robust’.

• b.

  Degree of agreement (e.g., between modeling results): ‘low’, ‘medium’, ‘high’.

• c.

  Level of confidence in the validity of a finding: ‘very low’, ‘low’, ‘medium’, ‘high’, ‘very high’. Confidence increases along with more robust evidence and higher agreement.

• d.

  Likelihood: ‘very likely’ (>90%), ‘likely’ (66–100%), ‘about as likely as not’ (33–66%), ‘unlikely’ (0–33%), ‘very unlikely’ (0–10%). Likelihood can be assigned in case of high agreement and(or) robust evidence.

The evidence of this study could be expressed as robust because it is based on the well-calibrated SWAT model of the Desna basin and the high-resolution bias-adjusted RCMs that capture the wide spread of climate signals. The likelihood is assigned as ‘very likely’ when all six models agree on the direction of change or show no change (±5%), ‘likely’ – 4–5 of six models, ‘unlikely’ – 1 of 6 models and ‘very unlikely’ – any models.

The comparison between climate model data and observations is given in Table 4. Additionally, the exceedance probability of mean annual temperature and precipitation is presented in Appendix D. All models were right predicting noticeable warming against 1961–1990. Considering temperature and precipitation trend together, we could say that RCA4 forced by HadGEM-ES and MPI-ESM-LR and RACMO22E forced by HadGEM-ES better agree with the observed trend.

Projections of annual temperature and precipitation by all eight RCMs are presented in the diagram of Figure 3 and their ensemble mean with standard deviation in Figures 4 and 5. Temperature and precipitation signals mostly fall in one area in the near future (2021–2050): warming around 1 °C and precipitation increase by 0–10%. In the far future (2071–2100), RCP4.5 scenarios also fall nearby – around 2 °C warming and 8% precipitation surplus. RCP8.5 scenario spreads wider – up to 3–5 °C warming and 5–25% precipitation surplus.

Across the Desna basin, projected temperature increases in the direction from west to east (Figure 4). Precipitation increases more strongly for the northern part of the basin; uncertainty is higher for the southern part (Figure 5).

The models predict the increase in the intensity of heavy precipitations by 12–20% and the number of the periods (three consecutive days) with extreme temperature by 0.3–1.7 per year in the far future (Appendix C).

In the 21st century, annual water yield might stay almost constant (±10%) or prominently increase (16–53%) under both RCP4.5 and RCP8.5 (Table 5). The increase is explained by higher precipitation, especially during the cold period when the soil becomes highly saturated stimulating the formation of water runoff. Only HIRHAM5 forced by EC-EARTH projects water yield decrease by 24% in the near future because of precipitation loss by 6%.

Groundwater and lateral flows will likely increase, mostly due to higher water infiltration during the cold period rather than accumulation in the snow cover. In the near future, the surface flow could moderately decrease as likely as increase under RCP4.5 and likely decrease more than 15% under RCP8.5. In the far future, the surface flow will likely fall by more than 20% under RCP4.5 or 40% under RCP8.5. A decrease in surface runoff is caused by two main reasons: first, lower soil moisture in June–October due to higher evapotranspiration; second, soils in the catchment do not freeze because of higher winter temperatures, while frequent thaws contribute water infiltration into deeper horizons.

A soil water change will likely fluctuate near zero due to simultaneous effects of higher precipitation and higher evapotranspiration. However, intra-annual changes are significant.

The discharge of the Desna river will likely slightly increase in the near future and will continue to rise slowly during the 21st century, mostly as a result of the noticeable increase of the discharge in the northern part of the basin (Figure 6). The same is true for the tributaries. Note that the possible future range also includes the slight decrease of discharge.

Precipitation will likely be increasing in November–March and May in the near future, then all months between October and May (Figure 7). In June–September, models disagree about the direction of precipitation change, and the mean signal is close to zero.

Water yield mostly repeats the dynamics of precipitation, except for March and April because of lower snowmelt runoff (Figure 7). Furthermore, water yield might increase even under precipitation loss due to higher groundwater discharge during the year (Appendix B).

At the Desna outlet, the water discharge reflects the water yield across the whole basin with a delay of around 1.5 months (Figure 8). Like previous years, the discharge will very likely continue to increase during the cold period and likely decrease during the spring flood peak (April or May). For the low water period (late June–October), the trend is unclear, primarily because precipitation change in June–September is also unclear. There is a high agreement that the discharge will rise in July, but the surplus might be as small as significant (>30% in the near future and >50% in the far future). In the near future, the discharge will likely increase in August and slightly decrease in October. Therefore, September and October will be the months with the lowest water levels. In the far future, the models disagree on the direction of discharge change in August–October, but the possible surplus (70%) is higher than the possible loss (20%).

The predicted discharge trends are very similar under RCP4.5 and RCP8.5, but the most extreme decrease and increase are possible under RCP8.5. In particular, HIRHAM5 forced by EC-EARTH predicts discharge loss by 30–36% at the Desna outlet in August–October in the near future. RCA4 forced by MPI-ESM-LR predicts a rise by 30–48% in all months except February, May, and June in the near future and rise by near 50% in April and June, near 60% in October–December, and near 80% in January–March and July–September in the far future.

Soil water will likely slightly decrease during the vegetation period in the near future and very likely more seriously decrease in the far future (Figures 7 and 9).

We suggest that climate change may exacerbate temperature and water stresses for crops, fire risk, and eutrophication (Table 6). Soil water will be decreasing during the warm period because of higher evapotranspiration and likely lower precipitation in July and(or) August. This negative effect will be compensated partially by precipitation surplus in October–May.

On the positive side, the surface runoff will likely be decreasing in the 2050s preventing flood risk, soil erosion, and diffuse pollution. The lower snow accumulation will smooth peak discharge in spring. Moreover, the warming in winter intensifies infiltration that will increase groundwater inflow during the year.

The uncertainty of climate models prevents achieving a confident conclusion in some cases. In particular, the disagreement on the direction of precipitation change in summer leads to three possible scenarios of the lowest water period (August–October): no change, medium-high surplus, or medium loss. If the negative trend of summer precipitation continues, it will be a signal of the higher risk of point source pollution.

Adaptive planning should also take into account ‘low likelihood’ scenarios with high risks. One of eight models, HIRHAM5 forced by EC-EARTH, predicts a slight precipitation loss that will lead to the soil water decrease by 20% in summer even in the near future and a prominent decrease in water discharge during the low water period in June–October (by 36% in October).

Among other studies, the results better agree with Didovets et al. (2020), who have used the global hydrological model WaterGAP calibrated on long-term annual river discharge at 1319 basins all over the world, including the Desna river (Hunger & Döll 2008) (Table 7). Despite the lower calibration efficiency in comparison with the current study (Didovets et al. 2020), the results also have good agreement on monthly signals for the highest water discharge (April–May) and the lowest one (August–September). This supports the conclusion that annual discharge will rather increase, and only a slight fall (5–10%) is possible.

Large-scale studies commonly have low confidence for a specific basin because they lack comprehensive calibration since computational capabilities and limited access to the observations (Krysanova et al. 2018). As mentioned above, this research is based on the SWAT model calibrated on daily discharge at 12 gauges across the basin, snow observations at 13 stations, and crop yield of the main crops (Osypov et al. 2018). Therefore, we believe that our results have higher credibility because of a better representation of the processes, especially for the tributaries.

However, the current Desna model omits comprehensive adjustment of nutrient transport, land-use change scenarios, and future agricultural practices. These factors obstruct the decisive prediction about the diffuse pollution trend. Nonetheless, we justify that climate will favor the reduction of nitrogen and phosphorus loads, especially in the far future (Table 6). Van Vliet et al. (2015) concluded the same for Ukraine based on the HYPE model for Europe.

Uncertainty in climate change projection comes from three different sources: internal variability, climate model response, and emission scenario. Lehner et al. (2020) estimated absolute and relative uncertainties for GCMs from the CMIP5 project. They showed that the emission scenario is the main source of uncertainty for temperature projection and climate model for precipitation. The ensemble used in this study includes a wide range of possible climate projections (e.g., from 6% decrease to 23% increase in precipitation) that ensure coverage of the major part of uncertainties.

The findings could help to reach environmental objectives established in the river basin management plan (RBMP), which Ukraine should launch in 2024. In the second and third cycles of RBMP, the experts should develop the monitoring program of climate change detection, indicate expected climate pressure and impacts, and propose an adaptation strategy (European Communities 2009). As a sign of climate change, here we demonstrate that winter discharge has been rising since the end of the 20th century and very likely will continue rising according to the model prediction. Another important finding of our study is the projected higher temperature and soil water deficit in the vegetation period that will definitely require adaptation measures for agriculture in the Desna basin.

We presented the novel projections for the water resources of the Desna river basin that are based on Euro-CORDEX high-resolution RCMs and the process-based SWAT model. In comparison with the period 1991–2020, the mean annual discharge will likely increase by 8% (possible range from −7 to 26%) under RCP4.5 and by 4% (from −23 to 33%) under RCP8.5 in the near future (2021–2050), and by 9% (from −1 to 30%) under RCP4.5 and by 17% (from −6 to 60%) under RCP8.5 in the far future (2071–2100) at the Desna outlet. The flood risk will stay at the same level or go down because of lower snow accumulation before the spring thaw. Soil water will strongly decrease in summer from the middle of the 21st century – by 16% (possible range from −26 to −7%) under RCP4.5 and by 18% (from −35 to 5%) under RCP8.5 in the far future (2071–2100) for the Ukrainian part of the basin.

The model results disagree on the sign of the Desna river discharge change in the low water period (August–October). The surface runoff also might change both ways in the near future. However, in the far future for both RCPs, the surface runoff will likely decrease, minimizing diffuse pollution.

We consider that strong attention should be paid to the best management practices in agriculture (e.g., irrigation strategies, new crop rotations, etc.) because of possible higher water stress for crops and temperature impact under future climate.

The authors are grateful to Serhii Filipchuk for the text editing.

All relevant data are available from https://drive.google.com/drive/folders/1NmXhC-dDQn0mDSzjZ1qsZfAt0bZ7Ng98?usp=sharing.