The effect of snow-cover area change, precipitation and temperature on streamflow in Tärnaån drainage basin, northern Sweden

(1)

Bachelor’s thesis

Geography, 15 Credits

Department of Physical Geography

The effect of snow-cover area change, precipitation and

temperature on streamflow in Tärnaån drainage basin,

northern Sweden

Kristin Röja

GG 243

2019

(2)

(3)

Preface

This Bachelor’s thesis is Kristin Röja’s degree project in Geography at the Department of Physical Geography, Stockholm University. The Bachelor’s thesis comprises 15 credits (half a term of full-time studies).

Supervisor has been Andrew Frampton at the Department of Physical Geography, Stockholm University. Examiner has been Margareta Hansson at the Department of Physical Geography, Stockholm University.

The author is responsible for the contents of this thesis.

Stockholm, 1 October 2020

Björn Gunnarson Vice Director of studies

(4)

(5)

1

Abstract

Snow cover is a fundamental component of the world’s cryosphere and plays an important role in the hydrological cycle. It is significant as a human water resource but can also be an influencing factor in flood and drought generation. Snow cover shows a great variability and understanding local snow cover and its effect on streamflow is therefore of importance. In this study, the effect of snow-cover area (SCA) change, precipitation and temperature on streamflow and its variability during the season, is studied in Tärnaån drainage basin in northern Sweden with the aim to see whether streamflow levels in Tärnaån drainage basin are mainly controlled by SCA change, precipitation and temperature, or if it is necessary to also consider other influencing factors. This aim will be reached by using MODIS snow-cover data products derived from satellite imagery, meteorological and hydrological data for the drainage basin and by visually analysing SCA changes, streamflow, precipitation and temperature data as well as performing a Pearson moment-product correlation analysis between some of these variables.

The results show that the effect of SCA changes, precipitation and temperature on streamflow vary over the studied time period which is also shown by differing correlation coefficients for different sub-periods. The strongest correlations are shown between SCA change and streamflow and between temperature and streamflow during sub-period 2 and sub-period 3 respectively. It is further clear that other variables than SCA change, precipitation and temperature influence streamflow and need to be considered to correctly predict streamflow levels in Tärnaån drainage basin.

Key words

Snow-cover area change, streamflow, snowmelt, drainage basin, precipitation, temperature, correlation

(6)

2

(7)

3

List of figures

Figure 1. Reference map of the location of Tärnaån drainage basin and an overview map of Tärnaån

drainage basin and the meteorological and hydrological stations. ... 13 Figure 2. SCA and streamflow in 2016 (a) and 2017 (b). Note that the SCA has a resolution of 8-day

periods while the streamflow data has a daily resolution. ... 18 Figure 3. SCA and streamflow in 2016 (a) and 2017 (b). Streamflow values are averaged over each 8-

day period. The temporal resolution is thus the same for both variables. ... 19 Figure 4. SCA change and streamflow change for 2016 (a) and 2017 (b). The two variables have the

same temporal resolution of 8-day periods. ... 20 Figure 5. The graphs show precipitation and streamflow (a), temperature and SCA (b), and

temperature and SCA change (c) for 2016. ... 23 Figure 6. The graphs show precipitation and streamflow (a), temperature and SCA (b), and

temperature and SCA change (c) for 2017. ... 24 Figure 7. Precipitation and streamflow for July and August in 2016 (a) and 2017 (b). ... 27

Figure 8. Scatterplot for SCA change and streamflow for 2016 (a) and 2017 (b). Streamflow values

are averaged over each 8-day period. ... 27 Figure 9. Scatterplot for precipitation and streamflow in 2016 (a) and 2017 (b). ... 28

Figure 10. Scatterplot for temperature and streamflow for 2016 (a) and 2017 (b). ... 29

Figure 11. Scatterplot for temperature and SCA change for 2016 (a) and 2017 (b). Temperature values

are averaged over each 8-day period. ... 29 Figure 12. Scatterplot for precipitation and SCA change for 2016 (a) and 2017 (b). Precipitation

values are averaged over each 8-day period. ... 30 Figure 13. Scatterplot for SCA change and streamflow for 2016 (a) and 2017 (b). The data points are

divided according to sub-period. ... 31 Figure 14. Scatterplot for precipitation and streamflow for 2016 (a) and 2017 (b). The data points are divided according to sub-period. ... 32

(10)

6

Figure 15. Scatterplot for temperature and streamflow for 2016 (a) and 2017 (b). The data points are

divided according to sub-period. ... 32

Figure 16. Scatterplot for temperature and SCA change for 2016 (a) and 2017 (b). The data points are divided according to sub-period. ... 33

Figure 17. Scatterplot for precipitation and SCA change for 2016 (a) and 2017 (b). The data points are divided according to sub-period. ... 34

List of tables

Table 1. 8-day periods and associated dates in 2016 and 2017. Based on Riggs and Hall (2016). ... 15

Table 2. Sub-periods in 2016 and 2017. ... 26

Table 3. Annual Pearson correlation coefficients for 2016 (a) and 2017 (b). ... 28

Table 4. Pearson correlation coefficients for sub-periods 1, 2 and 3 for 2016 (a) and 2017 (b). ... 31

(11)

7

1. Introduction

Snow cover is a fundamental component of the world’s cryosphere (Hock et al., 2019) and affects both natural and social systems in mountain areas as well as downstream of mountain areas (Beniston et al., 2018). It plays an important role in the hydrological cycle, as it delays streamflow during colder periods and releases streamflow during warmer periods (Stewart, 2009).

Snowmelt is an important source of water in the Northern Hemisphere. In many large basins, rainfall alone cannot meet the human water demand in spring and summer. Changes in climate can both decrease these basins’ snowmelt water resources as well as add more basins into the group that does not get their water demand met by rainfall in spring and summer (Mankin et al., 2015).

Beniston et al. (2018) report that studies show that as more precipitation falls as rain instead of snow, both snow depth and duration change. Their compilation of long-term snow depth and duration studies in Europe shows that both snow depth and duration have a declining trend at lower elevations, and that melt happens more often and more intensely (ibid.). Furthermore, with increasing temperatures, snow cover will melt earlier in spring and hence change the timing of the streamflow peak (Barnett et al., 2005). Such a scenario will result in consequences for water management, storage and availability (Beniston et al., 2018). For a better prediction of water availability, the relationship between snowmelt and streamflow needs to be understood (Barnhart et al., 2016).

In Sweden however, the human water demand is met by rainfall (Mankin et al., 2015). Studies of snowmelt generated streamflow is although important for more reasons than water availability for human water demand. Snow cover and its associated streamflow can for instance affect flood generation (Hock et al., 2019) as well as droughts (Xu et al., 2009). Rain on snow (ROS) events, which occur when rain falls on an existing snow cover, can also lead to severe flooding in areas with a seasonal snow cover and in higher elevation areas (Würzer and Jonas, 2018; Wachowicz et al., 2019).

On both a spatial and temporal scale, snow cover shows a great variability (Beniston et al., 2018). Understanding local snow cover and its effect on streamflow is therefore of importance.

This study can contribute to such an understanding in Sweden. It is however important to also consider how other factors, such as precipitation, affect basin streamflow, even in areas with a

(12)

8

seasonal snow cover. For example, for some studied basins in the U.S, one single process alone, such as precipitation or snowmelt, was not able to explain the maximum annual flow (Berghuijs et al., 2016).

1.1 Aims and objectives

Due to the significance of snow for water resources and for its importance in influencing floods and drought generation, the aim of this study is to map snow-cover area (SCA) in Tärnaån drainage basin north of Tärnaby in Storuman municipality in northern Sweden, and to investigate how changes in SCA affect streamflow levels at Solberg gauging station. It will also look at the relationship between precipitation and streamflow, and temperature and streamflow, as well as how precipitation and temperature affect SCA change. By doing this, the aim is to see whether streamflow levels in Tärnaån drainage basin are mainly controlled by SCA changes, precipitation and temperature, or if it is necessary to also consider other influencing factors, which would result in greater costs for these kinds of measurements and research. These aims will be reached by using MODIS snow-cover data products derived from satellite imagery, meteorological and hydrological data for the drainage basin and by visually analysing SCA changes, streamflow, precipitation and temperature data as well as performing a correlation analysis between some of these variables. The questions that will be answered are:

1. During which parts of the studied period do SCA changes affect streamflow levels in Tärnaån drainage basin?

2. Can precipitation explain streamflow levels when or if they cannot be explained by SCA changes?

3. During which parts of the studied period do temperature affect streamflow levels in Tärnaån drainage basin?

4. Can precipitation and temperature explain SCA changes?

(13)

9

2. Background

2.1 Snow hydrology and drainage basin processes

A drainage basin is a defined land area draining into a certain outlet (Mays, 2011; Gupta, 2011).

In a drainage basin, many factors can affect the distribution and characteristics of its snow cover as well as the flow path of meltwater and precipitation through the drainage basin.

As for snow cover, wind can affect the presence of snow in several ways. It can cause an irregular distribution of both snow cover, snow depth and snow properties (Mott et al., 2010), and may also contribute to greater sublimation of snow by increasing the exposed surface area of the snow or ice particles (DeWalle and Rango, 2008).

At times of snowmelt, several factors govern how the meltwater behaves in the snowpack and how long it takes for the water to contribute to streamflow in the basin. Firstly, snow cover characteristics, like snowpack cold content and liquid-water holding capacity, will affect how much and when the meltwater turns into streamflow (DeWalle and Rango, 2008; Jennings et al., 2018). The wetness of a snowpack is defined by its liquid water content (Pérez Díaz et al., 2017), and a snowpack can hold a maximum amount of liquid water. Snowpack cold content can be described as an energy deficit in the snowpack. As long as this energy deficit exists, meltwater may refreeze in colder parts of the snowpack. The refreezing contributes to a warming of the snowpack through latent heat, and hence a decrease of the energy deficit (ibid.).

Before meltwater contributes to streamflow, snowpack cold content and liquid-water holding capacity need to be satisfied. However, it can also take time for the water to flow through the snowpack, and water can additionally be delayed by ice layers in the snowpack (DeWalle and Rango, 2008).

Furthermore, snow cover and rainfall can together contribute to high streamflow during rain on snow (ROS) events. Like snowmelt events, streamflow magnitudes following a ROS event are governed by different factors. In a study of ROS-events in Switzerland, Würzer and Jonas (2018) found that in the beginning of a specific event, rain was kept in the snow which led to a delay in streamflow formation in the area. They saw that a deep snow cover tends to subdue the snowmelt runoff while a shallow snow cover tends to result in more intensive snowmelt runoff.

Some other factors that were seen leading to high streamflow amounts were a spatially homogeneous snow cover, high snow cover fraction and an initially high liquid water content in the snow (ibid.). A high initial liquid water content has also been connected to a short time lag between the onset of rain and streamflow, and vice versa (Würzer et al., 2016).

(14)

10

When meltwater or rainfall reaches the ground, it either infiltrates the ground as subsurface or groundwater flow or contributes to overland flow (DeWalle and Rango, 2008). The permeability of a soil varies depending on soil type. For instance, coarse-grained soils generally have a high permeability and fine-grained soils like clay and silt generally have a low permeability (SNA, 1994). Also, other existing conditions of the soil and ground, such as permafrost, frozen or thawed ground, and soil moisture content, affect the water flow (DeWalle and Rango, 2008; Yang et al., 2009). Frozen ground generally has a low permeability (Burt and Williams, 1976) and can therefore contribute to rapid overland flow and streamflow increase (DeWalle and Rango, 2008), and high soil moisture content can also increase runoff in comparison to low soil moisture content (Würzer and Jonas, 2018).

However, not all water in a drainage basin in the form of rainfall or snowmelt directly contributes to streamflow. Water detained as groundwater may not contribute to streamflow until months or years later (DeWalle and Rango, 2008). Moreover, during the melt season in Yukon basin, meltwater was seen being delayed in ponds, lakes and river valleys before contributing to increased streamflow (Yang et al., 2009). Water may furthermore be lost to evapotranspiration (Kirchner and Allen, 2020).

2.2 Snow-cover and streamflow

In a large part of the Northern Hemisphere, snowmelt dominates streamflow (Barnett et al., 2005) and between 2001-2014, snowmelt was dominating streamflow contribution in four of five studied basins in High Mountain Asia (Armstrong et al., 2019). Rain was the second largest contributor to the streamflow, except in one basin where it was the dominating one. The amount of snowmelt contribution to streamflow varied during the year and between the basins.

Snowmelt was generally the dominating contributor to streamflow but in some of the basins, contribution from rain exceeded the contribution from snowmelt during a few months (ibid.).

The association between snowmelt and streamflow has been the subject of many research studies. However, the proxy that is used for snowmelt can differ between studies. Among these are SCA, SCA change, snow depth decrease and snow water equivalent (SWE).

Azmat et al. (2017) studied the correlation between SCA and streamflow in Jhelum River basin and its sub-basins in Western Himalayas. They found that streamflow increases in Jhelum River basin were related to SCA-decreases, showed by annual Pearson correlation coefficients of - 0.63 for areas between 2001-4000 m a.s.l. and -0.59 for areas above 4001 m a.s.l. However, the correlation for areas below 2000 m a.s.l. was -0.06. The correlation for areas above 2000 m

(15)

11

a.s.l. furthermore varied between seasons, showing strongest correlation during pre- monsoon/snowmelt season and lower correlation during monsoon/extreme rainfall season (ibid.). A similar seasonal relationship was seen in three Siberian watersheds, where seasonal snow cover extent changes affected streamflow, with the strongest association during the melt period (Yang et al., 2003).

Suriano et al. (2019) looked at daily snow depth decreases and their connection to increased river streamflow in the Susquehanna River basin and the Wabash River basin, in the eastern United States, between years 1960-2009. For both studied basins, over 75 % of the snow depth decrease events with a decrease of at least 2,54 cm, were followed by an increase in river streamflow at a 3-day lag time. Although, the snow depth decrease events only occurred from November to April. They also differed in frequency both between different months and between the two basins. Furthermore, the percentage of events that led to an increase in river streamflow also differed between different months. For Susquehanna River basin, a significant relationship between larger (smaller) snow depth decreases and larger (smaller) river streamflow responses could be seen. For the Wabash River basin, the relationship was insignificant. However, for both basins, the relationship was calculated for all months of the year together (ibid.).

SWE and snow cover extent (SCE), both derived from remote sensing products, have furthermore been used together to study the relationship between snow cover and streamflow in Yukon basin in North America (Yang et al., 2009). During the melt season in spring, streamflow was strongly related to changes in SWE and SCE, but also during summer, fall and winter, an association could be seen between streamflow and changes in SWE and SCE (ibid.).

Significant correlations between SCA and streamflow have also been found on an annual scale, as in the Chitral River basin in Pakistan (Ahmad et al., 2018).

2.3 Precipitation, temperature and streamflow

Azmat et al. (2017) found that the correlation between precipitation and streamflow differed between the different sub-basins as well as between seasons. Due to partly solid precipitation in the winter season which delays the streamflow response to precipitation, they found a negative correlation for some of the climatic stations situated more than 2000 m a.s.l. Also, some of the stations located below 2000 m a.s.l. showed an insignificant correlation between precipitation and streamflow during winter. In monsoon or extreme rainfall season on the other hand, stronger correlations were found for most of the stations (ibid.).

(16)

12

Temperature and streamflow correlation also varied between sub-basins and seasons. For stations located >2000 m a.s.l., the correlation was generally stronger than at stations <2000 m a.s.l. and the correlation was overall stronger during pre-monsoon or snowmelt season and monsoon and heavy rainfall season than it was during winter and spring. The authors interpret this as increasing summer temperatures leading to snowmelt which in turn drive the streamflow (Azmat et al., 2017).

It has furthermore been shown that the fraction of precipitation falling as snow versus rain has an impact on a basin’s mean streamflow. Mean streamflow was higher (lower) when the fraction of precipitation falling as snow was higher (lower). Increases in temperature may shift precipitation from snowfall to rainfall and result in decreases in mean streamflow (Berghuijs et al., 2014).

2.4 Temperature, precipitation and snow-cover

Changes is snow cover has seen to be largely governed by temperature. Both SWE and SCE was strongly correlated with temperature in Yukon basin. During melt season, both SWE and SCE decreased during a period of temperatures around 0 ^oC, and increased in fall when temperatures again were around 0 ^oC (Yang et al., 2009).

Azmat et al. (2017) found a correlation between temperature increase (decrease) and SCA decrease (increase) and hence a streamflow increase (decrease). The strongest correlations were found during pre-monsoon or snowmelt season. They furthermore saw an inverse correlation between precipitation and SCA during pre-monsoon and monsoon season, but a positive correlation during the winter season. The correlation was stronger for the higher elevation zones (ibid.).

Temperature was also seen to control snow cover changes in the Gilgit River basin in the Himalayas, with snow melt and subsequent increase in streamflow as a result of high temperatures, and low temperatures resulting in delayed snowmelt (Tahir et al., 2016).

(17)

13

3. Method

3.1 Study area

Tärnaån drainage basin is located north of Tärnaby (65°42'38.7"N 15°15'35.1"E), a town in Storuman municipality in north-western Sweden (Figure 1). The basin is about 1077.2 km²and ranges in elevation from approximately 434 m a.s.l. to 1758 m a.s.l. Several lakes are situated within the basin of which lake Tärnasjön is the largest. It is centrally located in the drainage basin and from this lake, water flows into Tärnaån, where the streamflow for the basin is measured at Solberg gauging station. Tärnaån thereafter flows through Tärnaby town and into lake Gäutan.

Tärnasjön is partially surrounded by mountains, the highest ones are located south-west of the lake. In this area, Norra Sytertoppen at 1768 m a.s.l. is located just outside of the drainage basin border. Within the drainage basin, three glaciers drain into Tärnaån: Tärnaglaciären, Östra Syterglaciären and Måskonåviveglaciären (Bolin Centre for Climate Research, n.d.). In 2008, Tärnaglaciären had an area of 0,15 km². The current information states an area of 0,3 km² for Östra Syterglaciären and 0,36 km² for Måskonåviveglaciären (ibid.).

Figure 1. Reference map of the location of Tärnaån drainage basin and an overview map of Tärnaån drainage basin and the meteorological and hydrological stations.

(18)

14

3.2 Data

3.2.1 Meteorological and hydrological stations

Temperature, precipitation and streamflow data were obtained from the Swedish Meteorological and Hydrological Institute (SMHI). Hemavan Flygplats, Hemavan D and Solberg were the selected stations (Figure 1), due to their proximity to the basin.

3.2.1.1 Temperature data

Temperature data was obtained from Hemavan Flygplats station (65°48'27.72"N 15°5'7.44"E).

The station is located at 458 m a.s.l and 2 meters above the ground surface. It has been in operation since 2008. The temperature is measured in degree Celsius and represents the mean temperature for each day from 00.00 to 00.00 the next day. The quality of the data is classified as ‘suspected or aggregated, roughly controlled archive data and uncontrolled real time data’.

An explanation for this definition does not exist. The station is located west of the drainage basin. Data was collected from 2016-02-26 to 2016-11-15 and from 2017-02-27 to 2017-11-17.

3.2.1.2 Precipitation data

Precipitation data was obtained from Hemavan D station (65°49'20.64"N 15°5'18.6"E). The station is located at 475 m a.s.l. and 2 meters above the ground surface and was operating between 1945 to 2018. Precipitation was not separated into snowfall and rainfall. In the presence of snowfall, the snow was melted and thereafter measured in mm. The amount of precipitation was measured once per 24 hours, at 06.00 a.m. each day. The data is thus representative of the day before the measurement was made. The data is classified as controlled and approved. Hemavan D is located close to Hemavan Flygplats where the temperature data was obtained from. Precipitation data was collected for the same time periods as for the temperature data.

3.2.1.3 Streamflow data

Streamflow data was obtained from Solberg station (65°44'52.8"N 15°23'26.52"E), which measures streamflow in Tärnaån, a river with its outlet in lake Gäutan, central in Tärnaby village. Streamflow has been measured here since 1911. The streamflow data is presented as daily values measured in m³/s.

(19)

15

3.2.2 Snow cover data

Snow cover data was collected from the National Snow and Ice Data Center (NSIDC), using MODIS/Terra Snow Cover 8-Day L3 Global 500m SIN Grid, Version 6 (MOD10A2), Maximum Snow Extent subset (Hall and Riggs, 2016) with a resolution of 500 m. MODIS is short for Moderate Resolution Imaging Spectroradiometer and is an instrument operating on, for instance, the Terra satellite. Thus, the snow cover data is based on satellite imagery. The Maximum Snow Extent subset displays the maximum snow extent during a period consisting of eight daily snow extent tiles. A cell is mapped as snow if snow is present in the cell on any day during the 8-day period. If no snow is present in a cell on any day in the period, the cell is mapped with the land cover category that occurred most frequently, except for clouds, which are only mapped if cloud cover was present in a cell every day in the period (Riggs and Hall, 2016). Categories that occurred in the study area during the studied time period were: no decision, no snow, lake, cloud, lake ice and snow. The 8-day composing periods that were used in this study are periods 8-40 (Table 1).

Table 1. 8-day periods and associated dates in 2016 and 2017. Based on Riggs and Hall (2016).

The data set uses the Normalized Difference Snow Index (NDSI) to classify snow cover (Riggs and Hall, 2016). The index is based on snow cover normally showing high reflectance in visible spectral bands (VIS) and low reflectance in shortwave infrared spectral bands (SWIR) (Riggs

(20)

16

et al., 2015). For MOD10A2, the ratio between the difference of VIS and SWIR is calculated with the following equation:

NDSI = (band 4 – band 6) / (band 4 + band 6)

(Riggs and Hall, 2016). Snow is regarded as being present in pixels with a NDSI value > 0 and not present in pixels with a NDSI value <= 0. However, other land features can also have a NDSI value > 0 (e.g. edges of clouds), and the NDSI value can also differ due to different features of the land surface and viewing conditions such as the angle of sunlight. So called

‘screens’ are applied to each pixel that is considered to have snow present. The purpose of the screens is to discover pixels that are incorrectly considered to contain snow. These pixels are then either classified as a different category or as possibly not snow (ibid.).

3.3 GIS

3.3.1 Drainage basin

Delineation of Tärnaån drainage basin was done using GIS and was performed in ArcMap 10.6.1. The basis for the delineation was a combined digital elevation model (DEM) (Porter et al., 2018; European Environment Agency and Copernicus Land Service, 2015) with a spatial resolution of approximately 7,9 meters. The coordinates for Solberg station were used as the pour point for the drainage basin.

3.3.2 Area calculations

SCA calculations were also performed in ArcMap 10.6.1 The projection of the snow-cover images was changed to North Pole Lambert Azimuthal Equal Area projection. This projection was chosen because of its equal-area attribute. To enable area calculations, the MODIS raster images were converted to vector images. Every 8-day period except for periods 15, 17 and 18 in 2017 were cloud obscured to varying extent. For each 8-day period that was cloud

obscured, the snow-cover fraction of the area that was not cloud covered was calculated.

Thereafter, the snow-cover fraction was multiplied with the cloud covered area and then added to the original snow-cover area. This way, the cloud cover was removed for each period.

3.4 Statistical analysis

Data processing was performed in Microsoft Excel. Correlation analysis was performed with Pearson moment-product correlation coefficient for SCA-change and streamflow, precipitation

(21)

17

and streamflow, temperature and streamflow, temperature and SCA change, and precipitation and SCA-change. Pearson correlation measures the linear correlation between two variables.

The correlation values range from -1 to +1, where -1 is a perfect negative correlation, 0 is no correlation and +1 is a perfect positive correlation.

3.5 Delimitations

Due to time constraints, the time period for this study is limited to two years, 2016 and 2017, and to these years’ melt seasons. The studied time period is 26 February to 15 November for 2016, and 27 February to 17 November for 2017. These periods were chosen based on streamflow data for Solberg gauging station. Since the increase in streamflow begins in the end of April/middle of May, enough time exists between the end of February and the beginning of the streamflow increases to rule out a previous major increase in streamflow during the year. In the middle of November, the streamflow is almost at the same level as in the end of February, although not yet stable. Preferable, the study period would have stretched further until the streamflow levels had stabilised. However, in the following 8-day periods, the drainage basin was partly or completely classified as night in the MODIS snow-cover product which prevented an extended study period.

In this study, the effect of glacier meltwater on streamflow levels has not been included, due to the absence of glacier meltwater data. Moreover, precipitation has not been separated by type of precipitation. Precipitation type data is available from SMHI for Hemavan D, the same station as the precipitation data in this study is obtained from. However, data is not available for every day in the data set and has therefore not been used.

(22)

18

4. Result

4.1 Snow-cover area and streamflow

a)

b)

Figure 2. SCA and streamflow in 2016 (a) and 2017 (b). Note that the SCA has a resolution of 8-day periods while the streamflow data has a daily resolution.

(23)

19

a)

b)

Figure 3. SCA and streamflow in 2016 (a) and 2017 (b). Streamflow values are averaged over each 8- day period. The temporal resolution is thus the same for both variables.

(24)

20

a)

b)

Figure 4. SCA change and streamflow change for 2016 (a) and 2017 (b). The two variables have the same temporal resolution of 8-day periods.

Figure 2 and Figure 3 show SCA and streamflow during the studied periods in 2016 and 2017.

In Figure 2, every day in each 8-day period has the same SCA values, while the streamflow values differ each day. In Figure 3, both SCA and streamflow have the same values every day in each 8-day period. In 2016, the SCA is initially large with only slightly varying values. A larger decrease starts in the end of May, after which the SCA stays around and below 100 km² until the end of September, when the area starts to increase (except for a drop the following 8- day period). The streamflow starts to increase late April and reaches its maximum peak June 2^nd after which it decreases but keeps fluctuating. The following peaks occur in shorter time intervals but are not as pronounced as the maximum peak. After September 12^th, there is a general decrease in streamflow levels.

(25)

21

The 2017 trend for SCA and streamflow is similar to 2016. The more pronounced SCA decrease starts in early June. Except for a couple of larger fluctuations in the end of August and beginning of September, the SCA then stays below 100 km²until the start of a steady increase in the beginning of October. As in 2016, the streamflow values are stable at first, before they start to increase in the middle of May and reach their maximum peak June 10^th. The following decrease is interrupted by multiple streamflow peaks but there is an overall declining trend throughout the studied period.

Figure 4 shows SCA change and streamflow change, where the streamflow change is obtained by calculating the difference between the average streamflow value between two periods and the SCA change is obtained by calculating the difference between two periods. Hence, in Figure 4 both the snow-cover change, and the streamflow change are displayed with the same temporal resolution.

For both years, the gradual increase and first peak in streamflow that occurs before the maximum peak, cannot be explained by changes in SCA, since the SCA does not fluctuate significantly more during this streamflow increase than it did earlier in the season, when no similar increase in streamflow occurred, which can be seen Figures 2, 3 and 4.

The maximum streamflow peak in 2016 seems to be associated with the large SCA decline in the end of May. The SCA decline from around 1038 km² to around 821 km² coincides with higher streamflow levels and the streamflow peak occurs when the SCA declines even further to around 273 km² (Figure 2). In 2017, the day with the highest amount of streamflow during the maximum streamflow peak, coincides with a still large SCA (Figure 2). When the SCA decreases from 1002 km² to 248 km², the streamflow levels are however still high but decreasing. When looking at Figure 3 and 4, where the temporal resolution is the same for both variables, the SCA decrease and streamflow increase although appear highly related.

For the rest of the studied period in 2016, some streamflow increases seem to be associated with SCA decreases. The SCA decrease between periods 21 and 22, and between 28 and 29 (see Table 1), coincides with a streamflow increase. However, the SCA decrease between periods 27 and 28 is larger than the decrease between 28 and 29, but the associated streamflow increase in not as large. Furthermore, several streamflow peaks after period 29 occur when the SCA is rather constant.

In 2017, two streamflow increases are rather clearly linked to SCA decreases, those between periods 28 and 29, and between 30 and 31. In Figure 4, the SCA decrease between periods 32

(26)

22

and 33 also seems to coincide with the streamflow increase, but Figure 2 shows that the streamflow peak occurs slightly before the SCA decrease.

In the middle of July 2017, the streamflow experiences a high peak. The SCA decrease it coincides with is however very small, and another decrease in the same size two periods after is not linked to a similar increase in streamflow.

In the end of the studied period, both for 2016 and 2017, there is an overall trend where the streamflow levels go down when the SCA starts to increase again in the end of September (2016)/beginning of October (2017). Note that the rather large decline in SCA in the beginning of October 2016 does not have a corresponding increase in streamflow. Also, in 2017, the last two peaks in streamflow do not have an apparent corresponding decrease in SCA.

4.2 Precipitation and temperature

a)

(27)

23

b)

c)

Figure 5. The graphs show precipitation and streamflow (a), temperature and SCA (b), and temperature and SCA change (c) for 2016.

(28)

24

a)

b)

c)

Figure 6. The graphs show precipitation and streamflow (a), temperature and SCA (b), and temperature and SCA change (c) for 2017.

(29)

25

As mentioned in section 4.1, there is no significant decrease in SCA during the first streamflow peak in 2016 and 2017. In 2016, precipitation does not seem to be able to explain the streamflow increase either as precipitation amounts are low as well (Figure 5a). However, in 2017, the first streamflow peak is a few days earlier preceded by 23,5 mm of precipitation (Figure 6a), which could explain the first runoff peak in 2017. Approximately the same amount of precipitation is although found later in the studied period but is then not followed by an equally high streamflow peak as the first one. Also, the gradual increase in streamflow leading up to the first peak, is only preceded by 5,4 mm precipitation, which seems unlikely to have caused the streamflow increase, since even larger amounts of precipitation later in the season do not generate similar streamflow increases.

In both years, precipitation amounts are low during the maximum peak in streamflow. After the maximum streamflow peak in both years, the streamflow levels keep fluctuating, showing many peaks of different magnitudes. All these peaks are preceded by quite high or high precipitation amounts. Hence, some of these streamflow peaks coincide with both higher precipitation amounts and SCA decline and some only coincide with higher precipitation amounts.

However, not every high precipitation amount is followed by a streamflow peak. In the beginning of both years’ studied period, there are several precipitation peaks with no succeeding increases in streamflow. Additionally, later in the 2017 season, when the streamflow levels start to decrease, the precipitation still exhibits several peaks.

An explanation for this can be found in the temperature graphs (Figure 5b-c and 6b-c). Both in the beginning and in the end of the studied periods, the temperature is below or around 0 ^oC, which would mean that most precipitation falls as snow and does not directly contribute to the streamflow. On the other hand, when the temperature starts staying above 0 ^oC and rising, in the end of April (2016) and in the middle of May (2017), the streamflow levels also start to increase.

Although the SCA is not linked to this overlapping rise in temperature and streamflow until later. The large SCA decrease begins in late May 2016 and early June 2017, when the temperature rises even more.

In the end of the studied periods, there seems to be a rather clear association between increasing SCA and falling temperatures. Although in 2016, from period 35 to 36, the SCA declines with about 250 km², while the temperature keeps going down. Furthermore, the subsequent increase

(30)

26

in SCA occurs during a period with zero precipitation. In 2017 however, the SCA increase coincides with both falling temperatures and higher amounts of precipitation.

The SCA increase in period 27 in 2016 as well as in periods 28, 30 and 32 in 2017, all coincide with high precipitation amounts, however, the temperature is well above 0 ^oC.

4.3 Correlation analysis

The correlation between SCA change and streamflow, temperature and streamflow, precipitation and streamflow, temperature and SCA change, and precipitation and SCA change, was tested for both years using Pearson product-moment correlation coefficient.

For both years, the correlation was tested for the entire studied periods, 26^th of February to 15^th of November 2016 and 27^th of February to 17^th of November 2017 (“annual” from here on). The studied periods were also divided into three sub-periods to see if the correlation differed between them. The sub-periods are based on temperature data from Hemavan Flygplats station. Sub-period 1 stretches from the end of February until the temperature is no longer below 0 ^oC, then sub-period 2 begins. Sub-period 2 continues until the first drop in temperature below 0 ^oC, which marks the start of sub-period 3.

Table 2. Sub-periods in 2016 and 2017.

The correlation between SCA change and streamflow, temperature and streamflow, and precipitation and streamflow were tested by using the same dates for both variables, but also by offsetting the streamflow with one and two days respectively, to account for the basin lag time.

The purpose was to see if the correlation differed depending on the chosen lag time. The lag time was determined by plotting streamflow and precipitation together during July and August 2016 and 2017, because of the rather small SCA fluctuations during these time periods (Figure 4). The result shows that a peak in precipitation is usually followed by a peak in streamflow one or two days later (Figure 7). Tärnaån drainage basin is therefore interpreted as having a lag time of one to two days.

2016 2017

Sub-period 1 26/2 - 25/4 27/2 - 11/5

Sub-period 2 26/4 - 3/10 12/5 - 7/10

Sub-period 3 4/10 - 15/11 8/10 - 17/11

(31)

27

a b

Figure 7. Precipitation and streamflow for July and August in 2016 (a) and 2017 (b).

4.3.1 Annual

4.3.1.1 Snow-cover area change and streamflow

Since SCA change has a resolution of 8-days and streamflow has a resolution of 1 day, the 8- day average value for streamflow was calculated. Subsequently, for each period, the SCA change and streamflow was plotted together (Figure 8).

a b

Figure 8. Scatterplot for SCA change and streamflow for 2016 (a) and 2017 (b). Streamflow values are averaged over each 8-day period.

The pattern of the data points is similar for 2016 and 2017. In general, there is a negative correlation, meaning that a positive SCA change is associated with a lower streamflow value, while a negative SCA change is associated with a higher streamflow value. Although, the points are rather scattered. For instance, a positive SCA change can have a similar streamflow value as a negative SCA change, and two similarly large negative SCA changes can have very different streamflow values. In 2016, the Pearson correlation coefficient for the scatterplot without any lag time accounted for is -0,57, which is stronger than the one with one-day lag

(32)

28

time (-0,54) and the one with two-day lag time (-0,51) (Table 3a) The same is true for 2017, with correlations of -0,62, -0,59 and 0,54 for no lag time, one-day lag time and two-day lag time respectively (Table 3b). The strength of the correlation coefficients is thus similar for 2016 and 2017.

Table 3. Annual Pearson correlation coefficients for 2016 (a) and 2017 (b).

a b

4.3.1.2 Precipitation, temperature and streamflow

a b

Figure 9. Scatterplot for precipitation and streamflow in 2016 (a) and 2017 (b).

Neither 2016 nor 2017 show any correlation between precipitation and streamflow (Figure 9;

Table 3). Very different precipitation values are associated with similar streamflow values and vice versa. The correlation coefficient for both 2016 and 2017 as well as for the different lag times lies around 0 (Table 3).

2016 ANNUAL

No lag time 1-day lag time 2-day lag time SCA change vs Q -0.57 -0.54 -0.51

P vs Q -0.03 0.04 0.02

T vs Q 0.47 0.47 0.47

T vs SCA change -0.19 P vs SCA change 0.03

2017 ANNUAL

No lag time 1-day lag time 2-day lag time SCA change vs Q -0.62 -0.59 -0.54

P vs Q 0.00 0.07 0.09

T vs Q 0.53 0.53 0.54

T vs SCA change -0.27 P vs SCA change 0.12

(33)

29

a b

Figure 10. Scatterplot for temperature and streamflow for 2016 (a) and 2017 (b).

As for SCA change and streamflow, the pattern for temperature and streamflow is similar for 2016 and 2017. When the temperature is below 0 ^oC, the streamflow values are rather evenly low, but when the temperature is above 0 ^oC, the spread of streamflow values is larger. The correlation coefficient is 0,47 for 2016 and 0,53 for 2017 (Table 3). For both 2016 and 2017, the correlation between temperature and streamflow is very similar for a lag time of one day and of two days compared to no lag time.

4.3.1.3 Temperature, precipitation and snow-cover area change

a b

Figure 11. Scatterplot for temperature and SCA change for 2016 (a) and 2017 (b). Temperature values are averaged over each 8-day period.

(34)

30

a b

Figure 12. Scatterplot for precipitation and SCA change for 2016 (a) and 2017 (b). Precipitation values are averaged over each 8-day period.

Temperature and SCA change only shows a weak correlation of -0,19 for 2016 and -0,27 for 2017 (Table 3; Figure 11) and precipitation and SCA change do not show any correlation (Table 3; Figure 12). For both temperature and SCA change, and precipitation and SCA change, the data points are very scattered and do not display a consistent pattern. It is not possible to say that a low temperature is associated with a positive SCA change or the other way around. The same is true for precipitation and SCA change.

(35)

31

4.3.1 Sub-periods

In the scatterplots for the sub-periods, the data points are denoted a specific colour according to sub-period. In some scatterplots it is difficult to distinguish the sub-periods from each other, why scatterplots for each sub-period is presented in Appendix C-G. Also, the length of the sub- periods differs, which means that the number of data points are not equal for each sub-period.

4.3.2.1 Snow-cover area change and streamflow

Table 4. Pearson correlation coefficients for sub-periods 1, 2 and 3 for 2016 (a) and 2017 (b).

a

b

a b

Figure 13. Scatterplot for SCA change and streamflow for 2016 (a) and 2017 (b). The data points are divided according to sub-period.

The correlation between SCA change and streamflow differs between the three sub-periods.

For sub-period 1, the correlation is -0,42 for 2016 and 0,18 for 2017 (Table 4; Figure 13). This means that for 2016, negative SCA changes are associated with higher streamflow values, and positive SCA changes are associated with lower streamflow values. The opposite relationship applies for 2017. For sub-period 2, the correlation is -0,73 for 2016 and -0,72 for 2017. For sub-

2016

SUB-PERIOD 1 SUB-PERIOD 2 SUB-PERIOD 3

No lag time 1-day lag time 2-day lag time No lag time 1-day lag time 2-day lag time No lag time 1-day lag time 2-day lag time

SCA change vs Q -0.42 -0.41 -0.41 -0.73 -0.68 -0.63 0.19 0.20 0.22

P vs Q -0.01 -0.02 -0.03 -0.14 -0.07 -0.09 -0.32 -0.31 -0.31

T vs Q -0.20 -0.08 -0.02 0.10 0.11 0.10 0.49 0.50 0.49

T vs SCA change 0.17 -0.15 0.62

P vs SCA change -0.54 0.21 0.12

2017

SUB-PERIOD 1 SUB-PERIOD 2 SUB-PERIOD 3

No lag time 1-day lag time 2-day lag time No lag time 1-day lag time 2-day lag time No lag time 1-day lag time 2-day lag time

SCA change vs Q 0.18 0.18 0.17 -0.72 -0.67 -0.60 0.53 0.56 0.57

P vs Q -0.21 -0.21 -0.20 0.05 0.05 0.08 0.19 0.19 0.17

T vs Q 0.32 0.29 0.27 0.17 0.18 0.19 0.64 0.71 0.75

T vs SCA change -0.02 -0.24 0.54

P vs SCA change 0.30 0.15 0.36

(36)

32

period 3, the correlation is 0,19 for 2016 and 0,53 for 2017. Sub-period 1 shows a slightly weaker correlation for a one- and two-day lag time, both in 2016 and 2017. The same is true for sub-period 2, although then the correlation becomes even weaker when accounting for the lag times. For sub-period 3, the correlation is stronger for the lag times. However, since the correlation for sub-period 3 is positive, it means that the association between SCA decline and streamflow increase is weaker when the two lag times are considered.

4.3.2.2 Precipitation, temperature and streamflow

a b

Figure 14. Scatterplot for precipitation and streamflow for 2016 (a) and 2017 (b). The data points are divided according to sub-period.

Precipitation and streamflow showed no correlation for neither the 2016 nor 2017 period, and the same holds true for the sub-periods (Table 4; Figure 14). The correlation value for 2016 is -0,01, -0,14 and -0,32 for sub-period 1, 2 and 3, and in 2017, the value is -0,21, 0,05 and 0,19.

For the lag times, the correlation does not differ much.

a b

Figure 15. Scatterplot for temperature and streamflow for 2016 (a) and 2017 (b). The data points are divided according to sub-period.

(37)

33

In 2016, the correlation between temperature and streamflow is -0,20, 0,10 and 0,49 for sub- period 1, 2 and 3 respectively. In 2017, the corresponding correlations are 0,32, 0,17 and 0,64 (Table 4). Each sub-period is rather clearly separated from each other in the scatterplots (Figure 15). The data points for sub-period 2 show the greatest scatter and this period shows no significant correlation between temperature and streamflow. Sub-period 3 however has about the same correlation as the whole 2016 period and in 2017, the correlation for sub-period 3 is even stronger than it is for the whole 2017 period.

4.3.2.3 Temperature, precipitation and snow-cover area change

a b

Figure 16. Scatterplot for temperature and SCA change for 2016 (a) and 2017 (b). The data points are divided according to sub-period.

Temperature and SCA change only showed a weak annual correlation for 2016 and 2017. The same is true for sub-period 1 and 2 in both years (Table 4; Figure 16). However, for both 2016 and 2017, sub-period 3 has a rather strong correlation of 0,62 and 0,54 respectively. These positive correlation values although imply that when the temperature is high, so is the SCA increase.

(38)

34

a b

Figure 17. Scatterplot for precipitation and SCA change for 2016 (a) and 2017 (b). The data points are divided according to sub-period.

The correlation between precipitation and SCA change is -0,54, 0,21 and 0,12 for the sub- periods in 2016. The correlation for the sub-periods in 2017 is 0,30, 0,15 and 0,36 (Table 4;

Figure 17) Hence, the correlation between precipitation and SCA change during each sub- period does not show any consistent pattern between 2016 and 2017.

5. Discussion

5.1 Data quality

A quick search for ‘MODIS snow-cover’ shows that MODIS snow-cover data have been used in many research studies and are thus a great asset for conducting studies in use of snow-cover data. However, the accuracy of the products is not complete. For instance, MODA102, the data set used in this study, displays a disadvantaged due to its 8-day resolution. Errors apparent in the daily snow cover tiles that the 8-day periods are based on, can be transferred and even aggravated into the 8-day period. If errors vary spatially from day to day in the daily snow cover tiles, errors in the 8-day periods will be displayed in several different locations (Riggs and Hall, 2016).

The pixel resolution of the MODIS snow cover product used in this study can furthermore influence the obtained snow-cover area in each 8-day period. Different land cover types can exist simultaneously within an area of 500 meters, while the 500-meter pixel can only be mapped as one type of land cover. A pixel that is mapped as snow covered, while only being partially snow covered will thus result in an overrepresentation of snow cover, and vice versa.

Both temperature and precipitation data were collected from meteorological stations located west of Tärnaån drainage basin, although in close proximity to the basin. The drainage basin is

(39)

35

about 1077,2 km² and both temperature and precipitation could thus be expected to vary greatly within the area, especially since the area displays a varying elevation with mountain ranges reaching 1768 m a.s.l. Consequently, there is a risk that the temperature and precipitation data collected from Hemavan D and Hemavan Flygplats stations are not representative for the drainage basin. Hence, since the division of the three sub-periods are based on these temperature data, this could have affected the correlation analysis for the different sub-periods.

The temperature data was originally planned to be collected from Hemavan-Gierevarto A station, but data gaps existed in June 2016, why Hemavan Flygplats station was chosen instead.

Hemavan-Gierevarto A is located at 793.504 m a.s.l. compared to 458 m a.s.l for Hemavan Flygplats. The higher elevation of Hemavan-Gierevarto A would likely represent the shifting elevation of the drainage basin more accurately.

5.2 Methodology

The MODIS snow cover raster data was converted to vector data files in order to enable area calculations. However, loss of information always accompanies conversion from raster data to vector data or vice versa (Eklundh & Pilesjö, 2013). A possible demonstration of this can be seen in the slightly different drainage basin areas for each 8-day period. It is about differences in meters and might therefore not have a great impact on the result but shows nevertheless that the result must be analysed with the implications of the method in mind.

The method for cloud cover removal is likely to impact the result even further. Different studies use different cloud removal methods. The method used in this study is not similar to any of the methods mentioned in an overview of cloud removal methods for MODIS snow cover products by Li et al. (2019), but was chosen because of its applicability within the time frame for this study. To remove the cloud cover by replacing it based on the category percentage in the non- cloud covered part of the image could result in both under- and overrepresentation of snow cover. For instance, it is possible that the cloud cover only covers higher elevation areas where snow cover exists and that the land cover classifications in the rest of the image do not consist of any snow cover. By applying the method used in this study would thus result in an underrepresentation of snow cover. The opposite would be true if the snow free areas were cloud covered and the snow-covered areas were not.

(40)

36

5.3 Result discussion

As presented in the results, the association between SCA-change and streamflow varies during both 2016 and 2017. The clearest association is seen when there is a large decrease in SCA and a large increase in streamflow. However, for both years, the streamflow starts to increase well before the SCA starts to decrease and except for the first streamflow peak in 2017, this increase cannot be explained by precipitation either. Both Suriano et al. (2019) and Yang et al. (2009) showed that both snow depth and SWE affect streamflow levels. These are factors that have not been looked at in this study, but they are likely to be able to explain the first streamflow increase in both years, since it cannot be explained by either SCA decrease or precipitation. The temperature increase coinciding with the streamflow increase in both 2016 and 2017 furthermore strengthens this probability since temperature has been showed to have a strong influence on snowmelt generation (Yang et al., 2009; Tahir et al., 2016; Azmat et al., 2017).

However, both the annual correlation and the correlation for the sub-periods, is weaker between temperature and streamflow than it is between SCA change and streamflow. The easiest drawn conclusion from this is that SCA change is a more important factor in streamflow generation.

However, as mentioned above, studies have shown that temperature is an important driving factor of snowmelt, both in decreases of snow depth and SCA (Yang et al., 2009; Tahir et al., 2016; Azmat et al., 2017), and could therefore be said to influence the correlation between SCA change and streamflow. Temperature furthermore governs if precipitation falls as snow or rain, which in turn affects both SCA changes and streamflow.

Furthermore, in 2017, the largest streamflow peak occurs when the SCA is still large. Either, this could also be a consequence of a decrease in snow depth or SWE, or the decrease in SCA is not apparent in the 8-day time resolution. As described in section 3.2.2, the Maximum Snow Extent subset displays the maximum snow cover extent during the eight days that make up the period. The SCA is not necessarily the same every day during an 8-day period and the day to day difference could possibly be large. If a SCA decrease occurred early in the 8-day period and was followed by a SCA increase, the 8-day resolution of the snow cover data prevents the decrease from being showed. This could have impacted the correlation analysis between SCA change and streamflow. If a large SCA decrease that did not show in the SCA data was followed by a large streamflow increase, the resulting correlation is weaker than it would have been if the SCA change had a daily temporal resolution.

(41)

37

5.3.1 Sub-period correlations 5.3.1.1 Streamflow generation

In earlier studies, it is apparent that the division into seasons or sub-periods is important for the correlation analysis between SCA and streamflow (Azmat et al., 2017; Suriano et al., 2019;

Yang et al., 2009). In this study, the correlation between SCA change and streamflow is strongest for sub-period 2 for both 2016 and 2017 (-0,73 and -0,72). For both years, the large SCA decrease and large streamflow increase are included in sub-period 2. The stronger correlation for this period is therefore no surprise. Sub-period 2 is however the longest sub- period and stretches all the way into October. In both 2016 and 2017, precipitation seems to affect streamflow levels more than SCA changes after the large SCA decrease. This is likely to have affected the strength of the correlation between SCA change and streamflow in sub-period 2. If the annual period would have been divided into more than three sub-periods, the correlation may have been different.

For sub-period 1 and 3, the strength of correlation varies between 2016 and 2017. 2016 has a positive correlation in sub-period 3 while 2017 has a negative correlation. The difference between the years could be explained by the drop in SCA that occurs in the beginning of October 2016, which does not have a corresponding increase in streamflow. For 2017 on the other hand, the SCA increase is accompanied with a general decrease in streamflow levels, except for a few peaks caused by precipitation. Sub-period 1 on the other hand shows a negative correlation in 2016 and a positive in 2017. In both years, neither the SCA nor streamflow vary much during this period. It is therefore difficult to say what this difference is due to.

Azmat et al. (2017), showed that also the correlation between precipitation and streamflow is dependent on divisions into seasons or sub-periods, where the strongest correlations could be found in the monsoon/extreme rainfall season. However, in this study, no significant correlation between precipitation and streamflow could be seen in any of the sub-periods (Table 4). In sub- period 1, this is likely due to the generally low temperatures, which probably result in precipitation as snowfall and hence does not generate any streamflow increase. There are however some days with temperatures above 0 ^oC in sub-period 1, both in 2016 and 2017.

Although, as mentioned above, the temperature could have differed within the drainage basin and a higher temperature at the meteorological station does not assure precipitation falling as rain in the entire basin. Also, even if precipitation falls as rainfall, a deep snow cover could

(42)

38

subdue the effect of rainfall on streamflow generation, as explained by Würzer and Jonas (2018).

In sub-period 2, precipitation seems well associated with streamflow after the large decrease in SCA in both 2016 and 2017. Although, the large decrease in SCA and large increase in streamflow is also included in sub-period 2, which would influence the correlation of precipitation and streamflow since precipitation during this time is low while the streamflow is high.

In sub-period 3 in 2017, precipitation is partially related to streamflow increases and partially not. In the end of the period, precipitation fluctuates without any response in streamflow. This coincides with lower temperatures and an increase in SCA and could thus mean that precipitation was falling as snow and contributed to an increased SCA instead of an increased streamflow. In 2016, the correlation is even negative (-0,32) for sub-period 3. There is not a lot of precipitation during this period, but the precipitation that does exist is not associated with any change in streamflow, which is probably also related to the decrease in temperature during this time.

As mentioned above, the temperature increase and the first streamflow increase in both years coincide well with each other. The correlation for sub-period 2 is however low for both years.

This is presumably due to that there is a general increasing temperature trend after the maximum peak in streamflow, while the streamflow decreases. After the maximum peak in streamflow, the SCA is rather small in both 2016 and 2017. Hence, increases in temperature are not able to generate as much snowmelt as it could when the SCA was large, since there is not as much snow available. In sub-period 1 there is a weak correlation in 2017 and even a negative correlation in 2016. During sub-period 1, temperatures are mostly below 0 ôC and is thus unlikely to affect streamflow neither indirectly by generating precipitation as rain nor directly by generating snowmelt. Temperatures can thus fluctuate between e.g. -25 ôC and -5 ôC without impacting streamflow levels. For both 2016 and 2017, the correlation is strongest in sub-period 3. For the 2-day lag time, this period even shows the strongest correlation in this study (0,75).

This period coincides well with a general decrease in both temperature and streamflow.

Moreover, as mentioned in section 5.1, the temperature data might not represent the temperature of the entire drainage basin. A more accurate correlation analysis would probably have been obtained if the drainage basin was divided into elevation areas, as in Azmat et al. (2017), where

(43)

39

they could show that the correlation between temperature and streamflow was stronger for elevations >2000 m a.s.l.

5.3.1.2 SCA change generation

Precipitation and SCA change do not show any significant correlations. This is the case for the annual correlation as well as for the sub-periods. When and if precipitation falls as snow in sub- period 1, the SCA is already large which might explain the absence of a SCA increase. On the other hand, in sub-period 3, the SCA is initially small, which would then lead to an increase in SCA in case of snowfall. Due to the higher temperatures in sub-period 2, precipitation likely falls as rain which does not lead to SCA increases. ROS could possibly even contribute to a decrease in SCA.

According to the correlation coefficients for temperature and SCA change, SCA changes is not generated by temperature either. However, the weak correlation between temperature and SCA change could possibly be explained in the same way as the correlation for temperature and streamflow in sub-periods 1 and 2 above.

These results imply that neither precipitation nor temperature can account for changes in SCA, which seems unlikely due to the general view that changes in snow cover are largely governed by precipitation and temperature (Yang et al., 2009). Instead, the weak correlation between these variables in this study is likely to depend on within what periods of time the correlation analysis is performed. For example, weak correlation coefficients are expected when periods of no significant SCA change are included in the analysis. Moreover, as for SCA change and streamflow, the consequence of the temporal resolution of SCA may also have affected the correlation between temperature and SCA change, and precipitation and SCA change.

5.3.2 Possible influencing factors

There are furthermore many factors that can affect the relationship between SCA change, streamflow, precipitation and temperature which have not been considered in this study. One factor that is significant in the hydrological cycle is evapotranspiration (e.g. Gupta, 2011; Mays, 2011), and is therefore likely to impact streamflow generation in Tärnaån drainage basin by reducing the impact of both precipitation and snowmelt. However, to account for evapotranspiration in a meaningful way would imply difficulties as evapotranspiration most likely varies over the basin. If evapotranspiration was included in this study by extrapolating an evapotranspiration value for the whole basin (as for precipitation and temperature), it would