A century of variations in extreme flow across Finnish rivers

Most Finnish rivers flow to the Baltic Sea through the Gulf of Bothnia, the Gulf of Finland, and the Archipelago Sea. However, some rivers flow to Russia and Norway, ending their path in the Arctic Ocean. We studied 16 headwater stations on Finnish rivers: regulated, pristine, flowing down to lakes, or beginning from lakes. These stations were well distributed across the country's geographic locations (figure 1(B)). The selected regulated river stations are located upstream of hydropower stations and dams. Therefore, no selected station was significantly influenced by anthropogenic drivers. Some selected rivers are known as Arctic rivers, including the Tana, Torniojoki, Paatsajoki, and Kemijoki rivers. The daily flows from 1911 to 2020 were obtained to quantify the long-term changes in extreme flow in the selected stations. Due to world economic growth in the 1960s, extreme flows during 1911–1960 and 1961–2020 were also evaluated to monitor the flow variation in the selected rivers before and after the industrial revolution.

Zoom In Zoom Out Reset image size

Moreover, four additional 30 year periods, including 1911–1930, 1931–1960, 1961–1990, and 1991–2020 were selected to quantify the hydrologic changes. The lowest and highest mean discharge values were at the Lestijoki @ Lestijärvi-luusua (7.3 m³ s⁻¹) and Kemijoki @ Kemihaara (556 m³ s⁻¹) rivers.

To address the current gap in existing metrics for flow regime (interannual variabilities and temporal shifts) assessment, this study introduced four novel indices, including a summer–winter low flow ratio, spring-absolute high flow ratio, time-to-peak (TPI) index, and increasing rate (IRI) index during the snowmelt period, to assess the interannual variabilities and temporal shifts in extreme hydrology regimes (figure 1(A)).

The summer–winter low flow ratio (L_SW) was applied to assess the spatiotemporal changes in the low flow regime (especially the shift in the timing of low flow):

$$\begin{equation*}{L_{{\text{SW}}}} = \frac{{{Q_{{\text{LS}}}}}}{{{Q_{{\text{LW}}}}}}\end{equation*}$$

where Q_LS and Q_LW are the summer and winter low flows. Q_LS and Q_LW were extracted as the recorded minimum flow.

The spring-absolute high flow time ratio, , was defined as a new ratio to quantify the high flow regime alteration over time:

$$\begin{equation*}\varepsilon = \frac{{{T_{{\text{HS}}}}}}{{{T_{{\text{HA}}}}}}\end{equation*}$$

where T_HS and T_HA are the associated time (number of days after 1 October according to hydrological year) to spring peak flow (Q_HS, due to snowmelt occurring between 1 February and 1 June) and the absolute high flow (Q_HA), respectively. The expected value of for typical Nordic rivers is '1', which reveals that annual maximum flow is caused by snowmelt in spring, while the unequal value of 1 means that peak flow occurred between 1 June and 1 December, due to intensive rainfall. The > 1 (and < 1) means the second peak flow occurred before (and after) the first of October, as T_HS and T_HA count from the first of October. This index can assess possible changes in Nordic rivers with a dual peak flow regime. The number of events and their return periods were calculated over the studied periods. In this study, the return period was defined as the number of > 1 (or < 1) events divided by the number of years in the corresponding period.

The TPI and IRI indices were designed to estimate temporal changes in the snowmelt rate

$$\begin{equation*}\;{\text{TPI }} = ({T_{{\text{HS}}}} - {T_{{\text{LW}}}})\end{equation*}$$

$$\begin{equation*}{\text{IRI}} = \frac{{\left( {{Q_{{\text{HS}}}} - {Q_{{\text{LW}}}}} \right)}}{{{\text{TPI}}}}\end{equation*}$$

where T_HS and T_LW are the corresponding times of Q_HS and Q_LW. In addition, the date of summer daily low flow (T_LS) was defined to assess possible changes in the timing of low flow. Moreover, the mean flows during fall–winter (1 October–31 March), and spring–summer (1 April–3 September ) were considered to analyze seasonal flow trends in selected gauges (figure 1(A)). Generally, the low/peak flows and their associated times were directly extracted from the observed daily flow over the study period. The MATLAB code was used to calculate the mentioned extreme flow indices.

The nonparametric Mann–Kendall (MK) test (Mann 1945, Kendall 1975) was applied to assess the trend of introduced indices. In addition, The pre-whitening procedure was used when a significant serial correlation was detected through a Durbin–Watson test (Durbin and Watson 1950, 1971) at the 5% level. This statistical test uses the residuals from regression analysis to detect the temporal relationship between values (autocorrelation). The Durbin–Watson test was applied to the original time series when the autocorrelation was significant at the 5% level; the data must then be pre-whitened before the MK test. Finally, trend magnitude (change per unit of time) was calculated using a non-parametric linear Sen's slope estimator (1968). A Sen's slope estimator was applied here if a time series presented a significant (P < 0.05) linear trend.

Despite Tana, Kymijoki, Kokemäenjoki, Oulujoki, and Iijoki, Q_LW experienced an upward trend during 1911–2020. Out of this, 43.75% of stations indicated a significant (P < 0.05) positive trend (figure S2). This tendency for Q_LW intensified between 1991 and 2020. Additionally, winter low flow occurred earlier (0.04–0.67 d yr⁻¹) in all rivers except Arctic rivers. Due to the important implications of climatic conditions in headwaters' hydrology (Peterson et al 2002, Ashraf et al 2016, Vihma et al 2016), these increasing and earlier tendencies can be attributable to accelerated warming in higher latitudes over recent decades. Increasing permafrost thawing accelerates infiltration and enhances deeper flow paths, resulting in broad-scale mobilization of subsurface water into rivers (St. Jacques and Sauchyn 2009, Evans et al 2020). Moreover, an increased rainfall over snow during the fall and winter seasons may significantly increase the fall–winter stream flows (Reihan et al 2007, Adam and Lettenmaier 2008, Klavins and Rodinov 2008, Rawlins et al 2009).

The Q_HA followed a negative trend in the Kalajoki river, increased in the Kymijoki River and showed no significant trend in 87.5% of reference stations (figure S3) from 1911 to 2020. About 62.5% of stations (Kyrönjoki, Iijoki, Kalajoki, Kiiminginjoki, Kuivajoki, Pyhäjoki, Simojoki, Oulujoki, Kokemäenjoki, and Kymijoki) indicated decreasing tendencies in Q_HA during 1961–2020 which is likely to be dominated by climate change, while during 1911–1960 and 1911–2020 showed weak negative or even positive trends. The results also showed that in 1911–1960, 43.75% of stations experienced increasing tendencies in Q_HA, while only 12.5% and 18.75% of stations showed similar trends during 1961–1990 and 1981–2010, respectively.

The T_HA showed a negative trend in 68.75% of stations during 1961–2020, while it significantly decreased in 50% of stations (i.e. Paatsjoki, Kyrönjoki, Iijoki, Pyhäjoki, Simiojoki, Oulujoki, Vuoksi, and Kymijoki) (figure S2). The selected gauges in Arctic rivers experienced a shifting back in the T_HA between 1911–1960 (−0.04 to −0.37 d yr⁻¹), 1911–2020 (−0.03 to −0.11 d yr⁻¹), and 1961–2020 (−0.11 to −0.20 d yr⁻¹) periods (figure S2). Likewise, other studies confirmed similar tendencies in Q_HA and T_HA for Arctic rivers across the permafrost regions of Canada, Russia, Sweden, and Finland (Matti et al 2016, Ahmed et al 2020, Ploeg et al 2021, Shrestha et al 2021).

The Q_HS indicated a significant decreasing tendency (P < 0.05) in the Kalajoki (−0.78 m³ s yr⁻¹) river, with significant increasing tendencies (P < 0.05) in the Tornionjoki (5.66 m³ s yr⁻¹), Vouksi (0.23 m³ s yr⁻¹), and Kymijoki (0.08 m³ s yr⁻¹) rivers, and an insignificant (P > 0.05) trend in the other rivers during the 1911–2020 period (figure S3). In all stations, T_HS occurred around 0.05–0.17 d yr⁻¹ earlier from 1911 to 2020. This backward shift in T_HS was significant (P < 0.05) in the Tornionjoki (−0.05 d yr⁻¹), Lestijoki (−0.12 d yr⁻¹), Tana (−0.04 d yr⁻¹), Oulujoki (−0.08 d yr⁻¹), Vouksi (−0.04 d yr⁻¹), Kokemäenjoki (−0.06 d yr⁻¹), Kymijoki (−0.17 d yr⁻¹), and Kyrönjoki (−0.12 d yr⁻¹) rivers (figure S2). The Q_HS indicated decreasing tendencies in Kemijoki, Kyrönjoki, Iijoki, Kuivajoki, Kiiminginjoki, Simojoki, Kalajoki, and Oulujoki during 1961–2020, while it increased in 1911–1960. In contrast, negative tendencies in Tana, Tornionjoki, and Lestijokirivers during 1911–1960 followed by positive trends in 1961–2020.

In Kalajoki, Kyrönjoki, Kokemäenjoki, and Kymijoki, the results indicated the occurrence of Q_HA during the summer and fall due to extensive rainfall in this period or increasing temperature and reduced snow cover in spring. The increasing trends of fall–winter average flow in Kyrönjoki and Kymijoki, as well as the positive tendency in fall–winter base flow, confirm this high flow pattern. Given no significant changes in Q_HS values for 75% of the studied stations, temperature rise due to global warming can be introduced as the primary factor in the acceleration of snowmelt (Makarieva et al 2019, Suzuki et al 2020), leading to an earlier spring peak of the selected rivers.

During 1991–2020, 75% of stations showed negative trends in T_HS, with the largest decreasing rates in Kymijoki (−0.56 d yr⁻¹), Kokemäenjoki (−0.40 d yr⁻¹), Kyrönjoki (−1.1 d yr⁻¹), Vouksi (−0.30 d yr⁻¹), Kiiminginjoki (−0.27 d yr⁻¹), and Pyhajoki (−0.46 d yr⁻¹). Several previous studies confirmed this tendency towards earlier snowmelt runoff in Arctic rivers and attributed it to climate change (Semmens and Ramage 2013, Shen et al 2018).

The TPI showed no trend in the Oulujoki Kymijoki, and Paatsjoki rivers, while it was shortened in the Tornionjoki (−0.13 d yr⁻¹), Kemijoki (−0.08 d yr⁻¹), and Tana (−0.18 d yr⁻¹) rivers and lengthened (0.04–0.71 d yr⁻¹) at the rest of the stations in the 1911–2020 period (figure 2). The increasing (1911–1960) trends in TPI reversed in the Tana, Paatsjoki, Kemijoki, Simojoki, Kokemäenjoki, and Kymijoki during 1961–2020 which is likely to be dominated by global warming. In the 1991–2020 period, the declining tendencies in TPI weakened for the Tana and Paatsjoki rivers and reversed in the Tornionjoki, Oulujoki, Kiiminginjoki, and Kemijoki rivers (figure S4). The earlier winter and spring warming trends lead to slower snowmelt rates (Musselman et al 2017, Alonso-González et al 2020), which induce snow melting extension (0.11–0.89 d yr⁻¹) in the reference stations located in sub-Arctic regions and southern Finland. As snowmelt shifts earlier in a warmer climate, the snowmelt rate is significantly reduced through the contraction of the snowmelt to a time of lower available energy. The Sun could not provide enough energy under the lower sun angles of late winter and early spring to drive high snowmelt rates (Musselman et al 2017). Therefore, earlier snowmelt occurs at slower rates, leading to relatively lower spring peak flow, consistent with the observed changes in the 1991–2020 period in 68.75% of the studied rivers. However, fast warming can accelerate snowpack ablation during spring and summer in very cold climates (Wu et al 2018), which is in accordance with our results (−0.08 to −0.18 d yr⁻¹) in Arctic rivers, such as the Tana, Paatsjoki, and Kemijoki (except the last period) rivers, located in higher latitudes.

Zoom In Zoom Out Reset image size

According to figure 3, the IRI showed a positive trend in Arctic rivers, while it significantly (P < 0.05) increased in the Tornionjoki (0.21 m³ s⁻¹ d⁻¹ yr⁻¹) Tana (0.10 m³ s⁻¹ d⁻¹ yr⁻¹), and Kymijoki (0.001 m³ s⁻¹ d⁻¹ yr⁻¹) and reduced at the rest of the stations (except the Oulujoki) over the 1911–2020 period. These declining tendencies in IRI are significant (P < 0.05) in the Kyrönjoki (−0.03 m³ s⁻¹ d⁻¹ yr⁻¹), Kalajoki (−0.03 m³ s⁻¹ d⁻¹ yr⁻¹), Kuivajoki (−0.02 m³ s⁻¹ d⁻¹ yr⁻¹), Iijoki (−0.06 m³ s⁻¹ d⁻¹ yr⁻¹), Simiojoki (−0.04 m³ s⁻¹ d⁻¹ yr⁻¹), and Kokemäenjoki (−0.001 m³ s⁻¹ d⁻¹ yr⁻¹) rivers.

Zoom In Zoom Out Reset image size

The negative (1911–1960) trends in IRI reversed in the Kemijoki, and Simiojoki during 1961–2020, which is likely to be dominated by global warming. However, the increasing trends in Iijoki and Oulujoki were followed by negative trends after 1961. In Arctic rivers, except for Paatsjoki and Tana, the positive trends in IRI reversed to significant (P < 0.05) negative tendencies during 1991–2020 (figure S7). The significant reduction in Q_HS and time to peak extension due to slower snowmelt rates is the dominant factor in decreasing IRI tendencies over the recent decades.

The L_SW index exhibited decreasing tendencies in 1911–2020 in low-latitude rivers (i.e. Tornionjoki (−0.01 year⁻¹), Kyrönjoki (−0.01 yr⁻¹), Kalajoki (−0.03 yr⁻¹), Kuivajoki (−0.01 yr⁻¹), Simiojoki (−0.006 yr⁻¹), Kymijoki (−0.005 yr⁻¹), Vouksi (−0.002 yr⁻¹), Oulujoki (−0.004 yr⁻¹), Iijoki (−0.004 yr⁻¹), Lestijoki (−0.002 yr⁻¹), Kiiminginjoki (−0.001 yr⁻¹), and Paatsjoki (−0.0004 yr⁻¹) rivers), and increased in the Pyhajoki (0.02 yr⁻¹), Tana (0.01 yr⁻¹), Kemijoki (0.002 yr⁻¹), and Kokemäenjoki (0.0005 yr⁻¹) rivers (figure 4). The increasing tendencies in Q_LS can be understood to be the leading cause of the positive trends in L_SW, reinforcing the main hydrologic feature of the Arctic flow regime in Northern Finland, except for in the Tornionjoki. In contrast, from 1991 to 2020, the Arctic rivers (Tana, and Kemijoki) indicated negative trends of L_SW due to significant reductions in Q_LW compared with earlier study periods (figure S8).

Zoom In Zoom Out Reset image size

The increasing (1911–1960) trends in L_SW reversed in Iijoki, Kalajoki, Kiiminginjoki, Lestijoki, and Oulujoki during 1961–2020, which is likely to be dominated by global warming (figure 4). In contrast, the negative tendencies in Kymijoki Pyhäjoki and Vuoksi were followed by positive trends from 1961 to 2020.

This is a clear signal of change in the timing and magnitude of the absolute low flow regime in the studied rivers. The detected decreasing tendencies of L_SW in the Tornionjoki, Kemijoki, and Kymijoki rivers are directly tied to an increase in Q_LW, while the Lestijoki and Vouksi rivers showed a significant positive trend in mean fall–winter flow only.

The natural flow regime (L_SW > 1) in Finnish rivers is characterized by low winter flows during extensive permafrost coverage (Arheimer et al 2017). However, the detected decreasing trend in the L_SW index confirms the significant alteration of low flow pulses in Finnish rivers (except for Tana, Paatsjoki, and Pyhajoki), suggesting the frequent occurrence of an absolute low flow in summer instead of winter (L_SW < 1), which is likely to be dominated by climate change (table 1). The results showed that the absolute low flow in Tornionjoki and Iijoki would usually occur more in late summer than winter in the 1991–2020 period. The number of experienced flow regimes with L_SW < 1 decreased in the Oulujoki, Lestijoki, Kiiminginjoki, Kalajoki, and Kymijoki rivers for the last period (1991–2020) and remained relatively constant in the other rivers (table 1).

The indicated that Q_HS (caused by snowmelt) did not necessarily match the maximum absolute flow for southern rivers (i.e. the Lestijoki, Vouksi, Kalajoki, Kokemäenjoki, Kymijoki, and Kyrönjoki) from 1911 to 2020 (figure S10). The number of years with = 1 (normal flow regime) in the Lestijoki, Vouksi, and Kokemäenjoki rivers exhibited a significant (P < 0.05) decline from 1911 to 2020. Among southern rivers, the most extreme flow (greater return period) happened in the Kymijoki, Kalajoki, and Kokemäenjoki, with 3.3, 3.7, and 5 year return periods, respectively. However, the Lestijoki, Vouksi, and Kyrönjoki experienced the lowest return periods of = 1 events, ranging from 2.3 (Vouksi) to 3 (Lestijoki) years.

During 1911–2020, the number of > 1 events (occurrence of absolute high flow during fall) showed a declining trend in the Kokemäenjoki, Vouksi, and Kymijoki rivers located in southeastern Finland (table 2). However, this pattern reversed after that, and this number exhibited a positive trend in the Vouksi and Kymijoki rivers during the 1991–2020 period. Additionally, the Lestijoki, Kyrönjoki, and Kalajoki rivers experienced a positive trend in the number of > 1 events from 1911 to 2020. According to table 2, the Vouksi, Lestijoki, Kymijoki and Kyrönjoki rivers exhibited the lowest return periods of > 1 events, ranging from 3 to 6 year during 1991–2020 period. In contrast, the Kyrönjoki and Kalajoki had the highest return periods, ranging from 25 to 50 years during the 1911–2020 period.

Increasing base flow and/or average flow has contributed significantly to annual maximum flow in southeastern Finland during the fall and winter.

The number of < 1 events (occurrences of absolute high flow in summer) showed a positive trend in the Kokemäenjoki and Kalajoki rivers during 1911–2020. Moreover, the Kyrönjoki river experienced a decreasing trend in the number of this event, while this pattern reversed to an increasing trend during 1991–2020. The Kalajoki and Kyrönjoki rivers showed positive tendencies in the occurrence of annual maximum flow in the mid-summer ( < 0.9) throughout the 1911–2020 period. For the Lestijoki, Vouksi, and Kymijoki rivers, this pattern reversed to a negative trend during the recent decades (1991–2020). Given that increasing summer flow caused by higher rainfall can cause a major delay in Q_HA, the observed variation in flow regimes can be attributed to a changing climate.