Links between Nordic and Arctic hydroclimate and vegetation changes: Contribution to possible landscape‐scale nature‐based solutions

In Nordic and Arctic regions, the rapidly warming climate sustains hydroclimatic and vegetation changes in the landscape. There is evidence for an increase in vegetation density in some regions, a trend that is expected as a response to increasing temperature and precipitation. If the hydroclimatic changes are linked to vegetation response, it could be viewed as a landscape‐scale nature‐based solution (NBS) that could moderate the runoff response, as denser vegetation should lead to increased evapotranspiration and lower runoff. In this paper, we investigate and compare hydroclimatic changes over a set of basins in the Nordic region and northwest America and compare with changes in vegetation density, analyzed using the normalized difference vegetation index (NDVI) for three time periods: 1973–1978, 1993–1998, and 2013–2016.

. Both temperature and precipitation are increasing, because the capacity to hold water vapor increases with higher temperatures (Francis et al., 2009).
In cool temperate, subarctic, and Arctic climates, higher temperature and precipitation typically enhance vegetation productivity and diversity (Francis et al., 2009). Therefore, with ongoing warming, the vegetation distribution is expected to shift northward in latitude and upward in altitude (Kapfer, Grytnes, & Hédl, 2017). Furthermore, general greening, or increased vegetation density, can be expected in northern and Arctic landscapes, when not considering other drivers that modify vegetation, such as anthropogenic land management. These changes have been investigated throughout the Arctic (e.g., Myers-Smith et al., 2015;Naito & Cairns, 2011). For tundra landscapes, the term 'shrubification' is used to describe the increasing propensity for shrub vegetation growth in landscapes that were previously dominated by open tundra (Pearson et al., 2013;Tape, Sturm, & Racine, 2006). However, changes in vegetation type and density are not only a response to climate change, they also feed back to the local landscape and the climate system. One example is the feedback between forest canopy and the energy and water cycle (Betts, Falloon, Goldewijk, & Ramankutty, 2007;Dale, 1997;Jia et al., 2014). Vegetation canopy cover has a major impact on transpiration (the less canopy, the less transpiration). Evapotranspiration, in turn, is responsible for the link between the water and energy cycle and connects the terrestrial and the atmospheric system (Bring et al., 2016). This link between landscape development and hydrology has also been investigated in northern and Arctic regions (e.g., Jaramillo, Prieto, Lyon, & Destouni, 2013;Karlsson, Jaramillo, & Destouni, 2015). In general, increasing vegetation density should lead to less runoff, as more water is stored and transpired back to the atmosphere by plants. Although effects are often more complex, with local controls on hydrology determining the actual response, there is some evidence for such links, both in northern and Arctic regions (Mård et al., 2017) and elsewhere (Buendia, Batalla, Sabater, Palau, & Marcé, 2016;Cox, Sarangi, & Madramootoo, 2006;Yang & Lu, 2017).
With strong climate warming in the past and in the projected future, this general tendency for increased vegetation density in northern and Arctic basins could be considered a nature-based solution (NBS) at the landscape scale, in that it modulates flow and alters the partitioning of precipitation into runoff and evapotranspiration. This could partly offset the effect of increasing temperature and precipitation (Mård et al., 2017). Thus, ignoring other changes in the landscape, the climate-driven change in northern and Arctic basins should lead to less runoff with increasing vegetation density in the basin. This large-scale NBS could consequently help to moderate effects of climate and anthropogenic change in northern and Arctic basins.
Although NBS have been increasingly promoted as a way of adapting human environments to climate and environmental hazards, they are typically investigated at small or local scales, such as urban regions, cities, or parts of cities (Keesstra et al., 2018). However, considering that future climate warming will also bring hydroclimate shifts that act on basin scales, it is relevant to consider the potential role of NBS on landscape scales. This potential role in modulating the increasing precipitation and runoff on drainage basin scales is an issue that has been much less explored.
Projected changes to northern and Arctic hydroclimate are already leading to substantial effects on ecosystems and human societies (Wrona et al., 2016). For instance, flows are generally increasing in response to increasing precipitation and moisture transport (Bring et al., 2016), which in turn affects infrastructure development (Instanes et al., 2016) and freshwater resources for both human, environmental, and industrial needs (Evengard, Berner, Brubaker, Mulvad, & Revich, 2011). If there is an overall effect of runoff modulation that accompanies an observed vegetation increase, it would be a relevant factor to consider when attempting to understand, predict, and plan for effects of climate change in northern and Arctic regions.
In this study, we investigate three research questions that pertain to northern and Arctic basins and the potential importance of vegetation changes as NBS at the landscape scale. First, we investigate hydroclimatic changes over a suite of basins with reliable hydrological records to determine changes in temperature, precipitation, and runoff. Second, we investigate changes in vegetation density over the same set of basins in order to determine whether there is a pattern of increasing vegetation density over time. Finally, we compare the vegetation density observations with hydroclimate data to determine whether the observed vegetation patterns are compatible with expected hydroclimate drivers and responses to landscape change.
2 | MATERIAL AND METHOD

| Study areas
To explore hydroclimatic changes in relation to vegetation changes, we selected river basins in Nordic and Arctic regions that have long-term records of discharge data available from the Global Runoff Data Centre (GRDC; based in Koblenz, Germany). We included basins comprising at least 50 years of data with less than 10% missing values. In the Nordic region, we investigated 36 river basins across Finland, Sweden, and Norway, located between approximately 57-70°N and 6-32°E and with a total area of 491,362 km 2 ( Figure 1a). In northwest America, we investigated five river basins with a total area of 115,790 km 2 located between approximately 60-65°N and 136-159°W (Figure 1 b). We grouped the 41 basins into six study areas: northern Finland (five basins), southern Finland (eight), northern Sweden (seven), central Sweden (five), southern Sweden/Norway (11), and Alaska/Canada (five).

| Data and method
We investigated patterns in runoff and climate concurrent with vegetation observations over three time periods (1970s, 1990s, and early 2010s). These three time periods were chosen on the basis of the availability of satellite data for each river basin. The periods allowed a separation of the entire length of record into three distinctively different periods, separated from each other by 20-year intervals. In this way, we aimed to capture longer-term changes rather than shorterterm variability between years. For the 1970s, 1990s, and early 2010s, we used hydroclimate data for 1969-1978, 1989-1998, and 2007-2015, respectively. These years were selected on the basis of the maximum availability of runoff data. Furthermore, we chose to use slightly longer periods for hydroclimatic data in comparison with vegetation data, and we also chose to let these hydroclimatic data periods start somewhat earlier than the vegetation data periods. The reason for this was because changes in hydroclimate not only have an immediate impact of vegetation in the same year but can also affect the vegetation in the following years, for instance, through prolonged drought. In general, the subpolar, boreal, and cool temperate regions we investigate here have been shown to have the longest lag times from hydroclimatic change to vegetation response (Vicente-Serrano et al., 2013). We downloaded global gridded monthly average temperature and precipitation time series from the CRU TS 4.00 dataset (Harris, Jones, Osborn, & Lister, 2014) and extracted spatially averaged data for all basins. Runoff data were obtained from the GRDC (see Table S1 for a complete list of stations and GRDC codes). For temperature and precipitation data, mean annual values were calculated for each period and basin. For detection of changes in runoff, mean monthly river discharge was used together with basin area to calculate average annual runoff, which was averaged for each study area.
Although it would have been possible to gather a wider set of climatic data and perform a full sensitivity and robustness analysis of the choice and quality of data, we considered such a comparison out of scope for the present study and chose to use the commonly investigated and widely acknowledged datasets listed above.
To identify vegetation changes, Landsat images were used to calculate the normalized difference vegetation index (NDVI) for each basin. The NDVI is a standard vegetation index based on the reflectance properties of areas with a vegetation cover (Rujoiu-Mare & Mihai, 2016) and has shown good accuracy for Nordic mountainous regions (Nordberg & Evertson, 2005). Satellite images from Landsat 2 and 3 were used to cover the study area for the 1970s (images selected from However, the magnitude of errors should be limited and tend to average out over basins, which were then also further aggregated into study areas. As cloud cover is a major limitation when using satellite data, we introduced a threshold to make sure we rejected images with high cloud cover, in order for these images not to confound our analysis. This threshold was set to 10%; thus, we only included satellite data with cloud cover of less than that number. We noted a relatively higher degree of cloud cover for the 1970s than for the other periods, but we were nevertheless still able to analyze a sufficient number of images to cover all basins. Here, we note also that the cloud cover threshold itself may be sensitive to the resolution of the sensor (Wielicki & Parker, 1992, Zhu & Woodcock, 2012, but this is an issue outside the scope of this study. To capture the main growing season, we selected imagery acquired between June and August and with a major proportion in July. The red and near-infrared (NIR) bands of the satellite image packages were processed in ArcMap (ESRI) version 10.4 where NDVI was calculated as follows: which is based on the reflectance in the red (R RED ) and near-infrared (R NIR ) bands of the electromagnetic spectrum by the green parts of the plants. The NDVI layers were combined into a mosaic and clipped to generate one NDVI layer for each basin and period. The NDVI values ranged from −1 to +1, with water, snow, and ice having a negative value, bare soil having a value around zero, and plants having a positive value (the greener the plants, the higher the NDVI value due to larger reflectance in the NIR band) (Lillesand, Kiefer, & Chipman, 2015). Five NDVI classes (I-V) were defined on the basis of this classification and a previous study by Al-doski, Mansor, and Shafri (2013) where Class I represents water, snow, and ice with a negative NDVI value (NDVI <0), Class II represents bare soils and areas with almost no vegetation
To further explore the robustness of these changes and put them in a longer-term perspective, we also investigated changes when con-

| Vegetation changes
Vegetation changes were detected for all 41 basins during the observation period (Figures 3 and 4)

| DISCUSSION
With this study, our overarching aim was to explore links between hydroclimate and vegetation changes in Nordic and Arctic regions.
Principally, we aimed to investigate whether vegetation changes act as a landscape-scale NBS that moderates the proportion of precipitation turning into runoff.  relationship between increasing temperature and increasing precipitation varies between seasons (Francis et al., 2009), an aspect we did not take into consideration here.

| Hydroclimate changes
In terms of runoff, we observed an increase for all study areas from the 1970s to the 2010s. Similar to precipitation, the increase in runoff from the 1970s to the 1990s was higher than the increase from the 1990s to the 2010s. An increase in precipitation should generally lead to an increase in river discharge (Hartmann et al., 2013), and we found that the two variables were in fact strongly connected when considering all regions and periods. In this study, we did not consider human influences such as dam construction and land use changes, which can have strong impacts on the streamflow (Hartmann et al., 2013;Kalantari et al., 2014;Kalantari, Ferreira, Walsh, Ferreira, & Destouni, 2017), in addition to climate change. However, this influence is generally much stronger on seasonal than on annual scales, as dams and reservoirs primarily redistribute water between seasons rather than years (Di Baldassarre, Martinez, Kalantari, & Viglione, 2017). Furthermore, the periods investigated here did not overlap with the years of the most intense hydropower development schemes in the study regions, and therefore, we expect that the effect of dam construction during or between the observed periods to be limited.
Overall, the observed changes in hydroclimate were as expected in terms of drivers, both when we considered changes between periods and when we investigated the long-term trend over the entire length of record. This was particularly the case for temperature, which increased steadily between all three periods and for all study areas. Precipitation also increased in general, but the change was more variable over time and less pronounced from the second period to the third. Also, some study areas did not show significant changes over the entire period. The runoff response of the basins studied was also largely as expected when

| Vegetation changes
Our second question concerned the observed patterns in vegetation density, and whether they agree with a general expected greening, or increase in density. Overall, we found that changes in vegetation were quite different between the 1970s to 1990s and the 1990s to the 2010s, and there was no consistent pattern of increasing vegetation density when considering all three periods. In contrast to the observed changes between the 1970s and 1990s, a strong decline in moderate vegetation and a substantial increase in sparse vegetation was observed between the 1990s and the 2010s across all study areas. In four of the six study areas, no dense vegetation was observed in the 2010s. Therefore, the three periods differed quite markedly, with changes between the first two showing a general greening or increasing vegetation density, whereas change between the second and third period showed an opposed tendency. In the next section, we discuss possible explanations for these overall patterns when considering hydroclimate change, but first, we address here two other factors that could have influenced our vegetation observations.
In addition to the cloud cover and resolution limitations discussed above, the vegetation observations could have been affected by timing, as vegetation density increases during the growing season.
We attempted to select images from the same time of season, but tested this effect by comparing the date of satellite image acquisition for all study areas and periods with the corresponding vegetation density. For the proportions of sparse, moderate, and dense vegetation, the correlations with acquisition dates were next to zero, ranging from −0.12 to −0.02 (p > 0.6 in all cases). For water/snow/ice and bare soil, the correlations were higher (0.26 and 0.23, respectively) but still not significant (p = 0.29 and p = 0.35, respectively). Thus, although we cannot rule out an effect of seasonal differences between periods, we consider it unlikely that differences in image acquisition times during the growing season were responsible for the observed differences in vegetation density patterns.
Finally, there could be other nonhydroclimate factors affecting the observations. For instance, we did not consider forestry practices, logging, or other human influences that could alter vegetation density, nor other natural disturbances such as wildfires. In Canada, a decline in forest area of about 0.33% was noted from 1990 to 2010, mostly due to conversion from forest to agricultural or urban land. However, this is a very small value in relation to the observed changes. Similarly, the forest area in Europe did not decrease but rather increase between 1990 and 2015 (Barreiro, Schelhaas, McRoberts, & Kändler, 2017), and for both the Nordic countries and most of Alaska, an increase in biomass was observed during the 1980s and 1990s (Myneni et al., 2001). Therefore, although land use and forestry practices can certainly have an effect, they are unlikely to have contributed substantially to the observed results.

| Links between hydroclimate and vegetation changes
Our final question concerned the links between hydroclimate and vegetation changes and whether the observed patterns support a possible landscape-scale NBS.
Considering the change between the first two periods, when both temperature and precipitation increased in step with vegetation density, we observed an expected pattern whereby increases in temperature and precipitation may have supported an increase in biomass.
This also aligns with general observations of increased vegetation cover and plant biomass (greening), with the treeline moving upward in altitude and northward in latitude, and forest biomass increasing (Finstad et al., 2016). However, for the change between the second and third periods, the hydroclimate observations did not support the expected process, as vegetation density decreased while temperature increased and precipitation also increased or was mostly unchanged.
In general, our observations indicate that changes in precipitation had a substantially higher influence on river discharge than changes in vegetation, as there was a strong relationship between precipitation and runoff ( Figure S1). Regarding the relationship between vegetation and runoff, we observed an increase in runoff with reduced vegetation density (more sparse vegetation) and a decrease in runoff with increasing vegetation density (more moderate and dense vegetation), which is compatible with the expected effect landscape-scale NBS.
However, compared with the influence of precipitation, this relationship was much weaker and not statistically significant ( Figure 5). We note, however, that the relationship was substantially stronger when considering all individual basins instead of groups of basins, although it was still not significant. This conclusion is strengthened by the fact that the satellite observations that underlie results for the two latter periods were more reliable than those supporting the opposing greening tendency between the earlier two periods.
We also note that investigating the data in different ways support this interpretation: Both on the study area level and individual basin level, there was no significant statistical relationship between the share of vegetation in the sparse, moderate, and dense categories and the runoff during the corresponding period. Similarly, the hydroclimatic patterns we observe were also salient when investigated as changes between periods and when investigated as longterm trends.
However, despite these indications, there are also some arguments that support further research to provide a definitive answer.
In this study, we have selected study areas on the basis of hydrological data availability and found that there is relatively weak evidence of a landscape-scale NBS moderating runoff across these basins. Thus, our starting point, in terms of the study areas, was based on the possibility to evaluate patterns on the basis of solid hydrological data.
Another approach that could possibly be helpful in improving our understanding of hydrology and vegetation interactions on landscape scales could be to start from satellite data and carefully select periods and areas to achieve a contrast in vegetation patterns-for instance, isolating a number of cases with clear increases or decreases in vegetation. Then these study areas could be evaluated in terms of hydrological data availability, and an investigation of the NBS effect could be carried out from that end.
In addition to this alternative approach, observing vegetation and evapotranspiration on the ground could also be explored. Although Thus, future work could involve water balance analysis for regions where there is available ground truthing data, to support and complement remote sensing investigations.
Finally, there are also some limitations in our study that could have influenced the results and that could motivate further investigation. For example, we found large changes in vegetation density in several cases, which cannot fully be explained neither by hydroclimate drivers nor possible confounding factors. Although we attempted to select image dates during the same time of the season, we cannot rule out the possibility that seasonal differences had some other effects that were not possible to detect here. Furthermore, the increase in precipitation and runoff was smaller from the 1990s to the 2010s than from the 1970s to the 1990s, which could have affected vegetation growth negatively over these periods and thus contributed to the result of decreasing vegetation density. Additionally, climate warming can ultimately have negative effects on vegetation even in northern climates under certain conditions. Increasing temperature and increasing vegetation lead to higher evapotranspiration rates that can cause drying of soils. Drier summer months further increase the risk of forests fires (ACIA, 2004). A decrease in vegetation could also be caused by an increasing number of extreme events and autumn and winter warming, as warmer autumns can trigger a reduction in winter hardening and rain-on-snow events can lead to plant ice encasement (Phoenix & Bjerke, 2016). In summary, although we consider a runoff moderating effect unlikely for these basins, other aspects of the large vegetation density changes observed in this study could merit further investigation, in order to identify and understand their drivers.
are more pronounced. This is especially important in northern high-latitude regions, where the rate of climate change will remain high over the coming decades.