Trend analysis and forecasting of the Gökırmak River streamflow (Turkey)

Water resources are key natural resources for sustainable development. Water resources are also crucial for socio-economic development and healthy ecosystems (Yıldız 2017). Therefore, water resource management is a significant issue for the future of water resources. Changes in the hydrologic environment resulting from climate change and other anthropogenic activities have caused problems in the amount and quality of water resources (Lee & Yeh 2019). A better understanding of the hydrologic environment is necessary for the assessment of water availability. Therefore, the assessment of hydrologic responses to the effects of climate change and anthropogenic activities is of vital importance to water resource managers and policymakers.

Issues related to the impact of climate change on river flows have attracted considerable attention in recent years. Durdu (2010) investigated the effects of climate change on water resources in the Büyük Menderes River Basin and pointed out that climate change has caused a decrease in the streamflow of the river. Bahadir (2011) analyzed changes in the Kızılırmak River streamflow and indicated that a decrease in the streamflow was closely related to climate change. Bozkurt and Sen (2013) examined the effects of climate change on the Dicle and Fırat river basins and stated that climate change has led to a significant decrease in the streamflow of the rivers. Kale et al. (2016a) examined climate change effects on the streamflow of the Karamenderes River and noted that climate change reduced the streamflow. Kale et al. (2016b) studied the impact of climate change on the Bakırçay River streamflow and documented that the streamflow was reduced due to climatic changes. Ejder et al. (2016a) investigated the effects of climate change on the streamflow of the Kocabaş Stream and reported that climate change resulted in a decrease in the streamflow. Ejder et al. (2016b) studied the effects of climate change on the streamflow of the Sarıçay Stream and indicated that a downward trend in the streamflow was due to climate change. Kale et al. (2018) pointed out that those decreasing trends in the runoffs of the Büyük Menderes, Gediz, and Tuzla rivers were observed due to climate change. Sönmez and Kale (2020) investigated climate change effects on the annual streamflow of the Filyos River and concluded that climate change caused a reduction in the streamflow. On the other hand, a comparison matrix related to the key criteria (climate, water, soil, land use and land cover, topography and geomorphology, socio-economy, management) of desertification determined for Turkey was constructed by TÜBITAK-BILGEM-YTE (2015) and the weights of the criteria were estimated. Climate was identified as the criterion with the greatest weight – 35.6%. In addition, the Kızılırmak River Basin was identified as a basin with a high level of moderate risk of desertification (Türkeşet al. 2020).

Researchers have used several methods to understand the impact of climate and anthropogenic factors on rivers. Published papers revealed that anthropogenic and climatic factors play different roles in different river basins. In some river basins, climatic factors have greater impact than the anthropogenic ones, while in others anthropogenic factors were more effective than climatic factors. Some authors reported that the contribution of human factors was greater than that of climatic factors (Jiang et al. 2011; Hu et al. 2012; Zhan et al. 2014; Kanani et al. 2020; Fu et al. 2020). Understanding the factors that trigger changes in water resources and determining the role of each of them in reducing the surface runoffor water resources is important when addressing water resource management issues (Kanani et al. 2020).

Rivers and streams are systems that prevent flooding and other damage caused by climate change and anthropogenic activities. Therefore, accurate assessment of the availability and streamflow of rivers is an important issue for water resource managers. Hydroclimatic changes and anthropogenic activities may cause an important change in streamflow time series (Sönmez & Kale 2019). Past and present trends of water resources should be investigated for their effective management. Villarini et al. (2011) recommended that change point analysis must be employed for a time series of the hydrologic process prior to the trend assessment. Determining the trends in streamflow time series is an important tool to detect changes in hydrologic systems (Chang 2007). Historical monitoring data will significantly contribute to the assessment of future trends (Blöschl & Montanari 2010). Different methods are used to determine trends in water parameters (Sen 1968; Hirsch et al. 1982; Helsel & Hirsch 2002; Şen 2012; Şen et al. 2019). Trend analysis has been used to determine trends in hydrologic and climatic variables (Cigizoglu et al. 2005; Saplıoglu et al. 2014; Ay & Kişi 2017; Tosunoglu 2017; Tosunoglu & Kisi 2017; Kale 2017a,b; Ali et al. 2019). Many scientists have investigated trends in water resources by using trend analysis (Kahya & Kalayci 2004; Cigizoglu et al. 2005; Bahadir 2011; Kale et al. 2016a,b; Ejder et al. 2016a,b; Kale et al. 2018; Ay & Kişi 2017; Ercan & Yüce 2017; Ay et al. 2018; Kale & Sönmez 2018a,b; Myronidis et al. 2018; Kale & Sönmez 2019a,b,c; Kişi et al. 2018; Yildiz et al. 2019; Sönmez & Kale 2020).

Although numerous studies have been carried out on the streamflow of other rivers, there are no studies on streamflow trends for the Gökırmak River. Therefore, this study aimed to determine trends in the streamflow of the Gökırmak River.

Materials and methods

Study area and streamflow data

The Gökırmak River, one of the largest tributaries of the Kızılırmak River, has its source in Ilgaz Mountain in the Kastamonu province, flows through Kastamonu, Taşköprü, Hanönü, Boyabat, Duragan and reaches the Kızılırmak River (the longest river that is completely within the borders of Turkey). The Gökırmak River is 221 km long and the basin is about 7000 km² (Yildirim et al. 2013). The region has a semiarid climate and is used for intensive agriculture, mainly rice cultivation (Dengiz et al. 2015). The Gökırmak River and its tributaries are frequently bordered by fluvial strath terraces that are covered by 3 to 5 m thick gravel deposits, which contain mostly well-rounded pebbles (Yildirim et al. 2013). Furthermore, Dengiz et al. (2015) reported that the soil texture class of the Gökırmak River varies from sandy loam to clay. Precipitation is observed in all seasons, while the mean annual temperature is 13.7°C and maximum evaporation is generally observed in July (Tanatmış 2004). In addition, floods caused by extreme rainfall and sudden snowmelt often occur in the river basin (Baduna Koçyigit et al. 2017).

The streamflow data analyzed in the present study were collected from a streamflow gauging station at the Purtulu Station (E15A045) in the Kastamonu province (Fig 1). The General Directorate of State Hydraulic Works (DSI) has been observing the streamflow at this station since 1 October 1998. In this study, the streamflow data for the period between 1999 and 2014 were processed on the basis of annual, seasonal, and monthly analyses. There is one more station recording flow data for the river. However, that station has been operating since 2013. The present study covers the period between 1999 and 2014. Therefore, data derived from the recently operated station were not included in the study. Due to some lack of monthly data in 2012 and 2013, observed available data were excluded from the streamflow data during the calculation of descriptive percentages.

The Gökırmak River and the location of the streamflow gauging station

Our initial assumption was that the data were normally distributed. Skewness, kurtosis, Kolmogorov–Smirnov and Shapiro–Wilk tests were used to evaluate the normality of the data. If the distribution of data is non-normal, transformations of data are applied to make the distribution of data as normal as possible. As stated by Feng et al. (2014), the log transformation is the most popular transformation used to transform data into data that follow a nearly normal distribution. In our study, the log transformation was used to make the data to conform to normality.

Change point analysis

Change point analysis is a non-parametric test and it was first developed by Pettitt (1979) to detect significant changes in mean values of a time series. In our work, the change point analysis was performed using the R statistical software (R Core Team 2019).

The null hypothesis of this analysis was that variables follow distributions that have a similar position parameter and that there is no change point, contrary to the alternative hypothesis that a change point exists. The following equations were used:

K_{T} = m a x | U_{t, T} |

$${{K}_{T}}=max\left| {{U}_{t,T}} \right|$$

and for

t = 2, \dots \dots, T

$$t=2,\ldots \ldots ,T$$

U_{t, T} = \sum_{i = 1}^{t} \sum_{j = t + 1}^{T} s g n (X_{i} - X_{j})

$${{U}_{t,T}}=\sum\limits_{i=1}^{t}{\sum\limits_{j=t+1}^{T}{sgn}}\left( {{X}_{i}}-{{X}_{j}} \right)$$

The null hypothesis is calculated with K_T and U_t,T confirms whether two examples (x₁,…, x_t and x_t+1,…, x_T) are in the same population or not. The associated probability (p) is used to compute the significance level.

Trend analysis

Trend analysis is a commonly used technique to determine a trend in a hydrologic time series. Box and Jenkins (1976) proposed a technique to find the best fit of a time-series model to historical values of a time series. In the present study, the trend analysis (Box & Jenkins 1976) was used to determine streamflow trends. It is based on linear, discontinuous, and stochastic processes. For stationary processes, autoregressive (AR), moving average (MA), and autoregressive moving average (ARMA) models are performed. On the other hand, the autoregressive integrated moving average (ARIMA) model is efficient for non-stationary processes. The purpose of these models is to find the best fitted model to a time series with the least number of parameters. The AR model requires that the sum of squared errors is minimized using the smallest number of terms, which ensures a good fit to data. The MA model is useful to provide a good fit for various datasets, and changes in multiple exponential smoothing including models can switch periodic components and trends in data. The ARMA model can be developed by merging AR and MA models and can be applied to model a time series with a smaller number of terms that are more extensive than both AR and MA models. ARMA models are rich in terms and above all suitable for stationary and ergodic processes. Mixed models, which can be applied in a wider range of situations, can be created by merging AR and MA models. These well-known mixed models are the ARMA and ARIMA models. In ARMA models, the AR and MA models are merged together as a model of order (p, q), where p is the AR term and q is the MA term. The ARMA model is described below:

X_{t} = ϕ_{1} X_{t - 1} + \dots + ϕ_{p} X_{t - p} + e_{t} + \dots + θ_{1} e_{t - 1} + \dots + θ_{q} e_{t - q}

$${{X}_{t}}={{\phi }_{1}}{{X}_{t-1}}+\ldots +{{\phi }_{p}}{{X}_{t-p}}+{{e}_{t}}+\ldots +{{\theta }_{1}}{{e}_{t-1}}+\ldots +{{\theta }_{q}}{{e}_{t-q}}$$

In this equation, ϕ is an autoregressive parameter to be predicted, θ is a moving average parameter to be predicted, X is the original series, and e is a series of unknown random errors that are expected to follow the normal distribution of the probability. In ARMA models, the null hypothesis assumes that the series is non-stationary as opposed to the alternative hypothesis rejecting the null hypothesis and defining the series as stationary.

On the other hand, while the ARMA model assumes that a time series is stationary, trends and periodicity actually occur in such datasets. Therefore, there is a requirement to reduce these effects before models can be run. Elimination is usually accomplished by adding a primary difference period to the model, which shows that there is no obvious trend or periodicity until the series reaches the lowest level about stationery. The differencing process is defined in the order of differentiation, which is similar to the AR and MA processes. These three components produce multiple forms (p, d, q), which explains the applied model form. At this point, the model is described as an ARIMA model. In ARIMA models, p refers to the number of AR terms, q refers to the number of MA terms and d refers to the order of differencing. The models are ordinarily listed as ARMA (p, q) models if no differencing is performed (d = 0). The letter “I” in ARIMA is the initial letter of the term “integrated” and indicates that a dataset was predominantly differentiated. The dataset must then be combined for final approximations and forecasts when modelling is concluded. The objective of these models is to detect the best model fitting to a time series and to take account of minimum parameters (Box and Jenkins, 1976). The ARIMA model employs a linear combination to forecast a time series. ARIMA supports the selection of the right model to fit a time series. The ARIMA model used in the study is as follows:

X_{t} = c + ϕ_{1} X_{t - 1} + \dots + ϕ_{p} X_{t - p} + θ_{1} e_{t - 1} + θ_{q} e_{t - q} + e_{t}

$${{X}_{t}}=c+{{\phi }_{1}}{{X}_{t-1}}+\ldots +{{\phi }_{p}}{{X}_{t-p}}+{{\theta }_{1}}{{e}_{t-1}}+{{\theta }_{q}}{{e}_{t-q}}+{{e}_{t}}$$

where X_t is the variable that will be defined in t time, e_t is the error in t time, θ is the coefficient of parameter q, ϕ is the coefficient of parameter p, and c is the constant.

Many scientists have frequently applied trend analysis to determine trends in the streamflow of rivers (Ejder et al. 2016a,b; Kale et al. 2016a,b; Myronidis et al. 2018; Kale et al. 2018; Kale & Sönmez 2018a; Khairuddin et al. 2019; Sönmez & Kale 2020). Therefore, this model was implemented to determine the trend in the streamflow data. Trend analyses were performed in SPSS and Minitab statistical software. The identification of ARIMA models is based on calculations of autocorrelation. In addition, autocorrelation analyses were performed to calculate the consistency of trend analysis results.

ARIMA modelling can be divided into three stages. Initially, the order of differencing and degrees of AR and MA was determined using the autocorrelation function (ACF) and the partial autocorrelation function (PACF). It was then predicted that the parameters confirm that the residuals are white noise. Finally, the best-fit model was obtained in the third stage as a result of the analysis of residuals. The Ljung–Box test statistic was used to check the randomness. The normalized Bayesian Information Criterion (BIC), R-squared and p values were compared. The ARIMA models with the minimum normalized BIC, p and R-squared values were selected as the best fit models and used for forecasting. The accuracy of the models was assessed by means of commonly used performance measures, which are the mean absolute deviation (MAD), the mean squared deviation (MSD) and the mean absolute percentage error (MAPE).

Mann–Kendall and Spearman’s rho test

The Mann–Kendall test was initially presented by Mann (1945) and then developed by Kendall (1955). The Mann–Kendall test is a comprehensive test to examine a trend in a time series. One of the advantages of this non-parametric test is that data do not have to follow any specific distribution. The following equation was employed:

S = \sum_{i = 1}^{n - 1} \sum_{k = i + 1}^{n} s g n (x_{k} - x_{i})

$$S=\sum\limits_{i=1}^{n-1}{\sum\limits_{k=i+1}^{n}{sgn}}\left( {{x}_{k}}-{{x}_{i}} \right)$$

In this equation, the time series x_i is derived from i = 1, 2, …, n − 1, and x_k from k =i + 1, …, n.

s g n (θ) = \{\begin{cases} + 1, & θ > 0 \\ 0, & θ = 0 \\ - 1, & θ = 0 \end{cases}

$$sgn\left( \theta \right)=\left\{ \begin{array}{*{35}{l}}+1, & \theta >0 \\\,\,\,\,\,0, & \theta =0 \\-1, & \theta =0 \\\end{array} \right.$$

The normalized test statistic is calculated using the following equation:

Z_{c} = {\begin{array}{l} \frac{S - 1}{\sqrt{v a r (S)}}, & S > 0 \\ \frac{S + 1}{\sqrt{v a r (S)}}, & \begin{array}{l} S = 0 \\ S < 0 \end{array} \end{array}

$${{Z}_{c}}=\left\{ \begin{array}{*{35}{l}}\frac{S-1}{\sqrt{var\left( S \right)}}, & S>0 \\\frac{S+1}{\sqrt{var\left( S \right)}}, & \begin{align}& S=0 \\ & S<0 \\ \end{align} \\\end{array} \right.$$

In this equation, the test statistic is Z_c and when |Z_c|> Z_1–α/2, where Z_1–α/2 is the standard normal variable and α is the significance level for the test, H₀ will be rejected. The size of the trend is calculated using the following equation:

β = M e d i a n (\frac{x_{i} - x_{j}}{i - j}), \forall_{j} < i

$$\beta =Median\left( \frac{{{x}_{i}}-{{x}_{j}}}{i-j} \right),{{\forall }_{j}}<i$$

where 1 < j < i < n.

A positive value of β indicates an increasing trend, while a negative value of β indicates a decreasing trend.

Spearman’s rho test, a non-parametric test, is used to measure the strength of a monotonic relationship between two variables (Lehmann 1975; Sneyers 1990). It was performed to determine the relationship between two variables before the model was produced. Non-parametric Spearman’s rho and Mann–Kendall tests yield more reliable results than parametric tests (Kale & Sönmez 2018a).

Results

The first assumption is that the data are normally distributed. Descriptive statistics of the streamflow data were calculated and are presented in Table 1. The descriptive statistics include: the mean, standard deviation (SD), coefficient of variation (CV), coefficient of skewness (CS), skewness, kurtosis, maximum value, minimum value, and range. CV is the most selective feature among these statistics. If the coefficient of variation is greater than 0.9, the parameter shows high variability, whereas it shows low variability when the CV value is less than 0.1 (Durdu 2010). The calculated value of CV for the annual and seasonal streamflow is less than 0.9 and greater than 0.1. However, the CV value for summer is close to 0.9 (CV = 0.73). These findings show that the statistic can be considered as an indicator to understand the variability in the streamflow. The skewness measures the asymmetry (should be around zero) of a distribution. The kurtosis is a measure of “peakedness” of a given distribution. Skewness and kurtosis values between −2 and +2 are acceptable (George & Mallery 2010), although both values are equal to zero in the normal distribution. Similarly, Kim (2013) remarked that the null hypothesis (assuming that data show a normal distribution) should be rejected when absolute Z scores for either skewness or kurtosis are not between −1.96 and 1.96 for small samples (n < 50). Z scores can be obtained by dividing the skew values or excess kurtosis by their standard errors (Kim 2013). Therefore, skewness and kurtosis were also used to evaluate the normality of data in addition to Shapiro–Wilk and Kolmogorov–Smirnov tests. Table 2 shows the obtained values of skewness, kurtosis, Shapiro–Wilk and Kolmogorov–Smirnov tests. Values are within the acceptable ranges.

Table 1

Descriptive statistics of the streamflow data

Streamflow	Mean	SD	CV	CS	Maximum value	Minimum value	Range
Annual	28.48	3.43	0.45	202.46	49.90	10.46	39.44
Spring	64.89	8.16	0.47	415.31	111.77	25.37	86.40
Summer	21.27	4.17	0.73	78.39	50.71	3.64	47.06
Autumn	8.70	0.89	0.38	69.84	13.90	1.38	12.52
Winter	19.07	2.53	0.50	105.58	34.73	5.00	29.73

Table 2

Results of skewness, kurtosis and normality tests

Streamflow	Skewness	SE_skewness	Z_skewness	Kurtosis	SE_kurtosis	Z_kurtosis	Kolmogorov–Smirnov^*		Shapiro–Wilk
Streamflow	Skewness	SE_skewness	Z_skewness	Kurtosis	SE_kurtosis	Z_kurtosis	Statistics	p-value	Statistics	p-value
Annual	0.04	0.597	0.067	−1.35	1.154	−0.001	0.175	0.200^*	0.935	0.200^*
Spring	0.18	0.597	0.302	−1.53	1.154	−0.001	0.139	0.200^*	0.921	0.200^*
Summer	0.42	0.597	0.704	−1.01	1.154	−0.001	0.182	0.200^*	0.924	0.200^*
Autumn	−0.58	0.597	−0.972	0.55	1.154	0.000	0.129	0.200^*	0.853	0.042
Winter	0.21	0.597	0.352	−1.12	1.154	−0.001	0.137	0.200^*	0.956	0.200^*

Note: ^*indicates the lower limit of the true significance.

^a indicates the Lilliefors Significance Correction

Box and Jenkins (1976) recommended that the autocorrelation function (ACF) and the partial autocorrelation function (PACF) as the major analyses to determine the ARIMA model sequence. The ACF and PACF results obtained using the natural logarithm for the annual and mean seasonal streamflow data are presented in Figure 2. Moreover, the ACF results using the natural logarithm for the mean monthly streamflow data are shown in Figure 3, whereas the PACF results using the natural logarithm for the mean monthly streamflow data are presented in Figure 4. The best fitted model is the model at a certain significance level. Thus, the significance levels of the ARIMA models were compared with each other and the Ljung–Box test statistic was used to check the randomness. The R-squared values were considered when selecting the best fitted model. R-squared values that are closer to zero and lower values of the normalized Bayesian Information Criterion (BIC) indicate a good fit. Therefore, the ARIMA (0,1,1) model was selected to forecast the future trends in the streamflow (Table 3). The accuracy of the models was assessed by employing the commonly used performance measures, which are the mean absolute deviation (MAD), the mean squared deviation (MSD) and the mean absolute percentage error (MAPE).

The autocorrelation functions (ACF) and partial autocorrelation functions (PACF) of the natural logarithm of annual and mean seasonal streamflow data. The lines represent the 95% confidence interval.

The autocorrelation functions (ACF) of the natural logarithm of mean monthly streamflow data. The lines represent the 95% confidence interval.

The partial autocorrelation functions (PACF) of the natural logarithm of the mean monthly streamflow data. The lines represent the 95% confidence interval.

Table 3

Parameters of ARIMA models for annual streamflow data

Parameters	Models
	ARIMA (1, 1, 0)	ARIMA (0, 1, 1)	ARIMA (1, 1, 1)
	AR	MA	AR	MA
Coefficient	−0.637	0.890	−0.126	0.894
SE	0.255	0.314	0.377	344
p-value	0.029	0.016	0.746	0.027
Normalized BIC	15.616	15.409	15.863
R²	−0.565	−0.272	−0.479
Ljung–Box Statistics	25.31	19.11	18.12
Ljung–Box p-value	0.005	0.039	0.034

Note: ARIMA means autoregressive integrated moving average; AR – autoregressive; MA– moving average; SE – standard error; BIC – Bayesian Information Criterion.

The results of the change point analysis indicate that the change point for the mean annual streamflow was 2008. Trend analysis results showed a decreasing trend for the mean annual streamflow (Fig. 5).

Trend analysis results for mean annual streamflow. In variable box; actual is the observed value; forecasts are the predicted values; fits are calculated values that best fitting to forecast. The accuracy of models was assessed by using commonly used performance measures which are mean absolute deviation (MAD), mean squared deviation (MSD), mean absolute percentage error (MAPE).

The change point analysis was carried out to understand seasonal trends in the streamflow. The analysis identified 2005, 2008, 2006, and 2005 as change points for spring, summer, autumn, and winter, respectively. The results of the trend analysis showed that the streamflow tends to increase in summer and decrease in other seasons (Fig. 6).

Trend analysis results for mean seasonal streamflow. In variable box; actual is the observed value; forecasts are the predicted values; fits are calculated values that best fitting to forecast. The accuracy of models was assessed by using commonly used performance measures which are mean absolute deviation (MAD), mean squared deviation (MSD), mean absolute percentage error (MAPE).

For monthly trends in the streamflow, the change point analysis determined 2004, 2004, 2010, 2005, 2005, 2009, 2008, 2005, 2005, 2001, 2006, and 2006 as change points for the monthly streamflow data from January to December, respectively. The results of the trend analysis showed that the mean monthly streamflow tends to increase in May, June, and July. On the other hand, the streamflow showed a decreasing trend for January, March, April, August, September, October, and November (Fig. 7). There is no upward or downward trend in February and December.

Trend analysis results for mean monthly streamflow. In variable box; actual is the observed value; forecasts are the predicted values; fits are calculated values that best fitting to forecast. The accuracy of models was assessed by using commonly used performance measures which are mean absolute deviation (MAD), mean squared deviation (MSD), mean absolute percentage error (MAPE).

Although increasing and decreasing trends were determined for the streamflow of the Gökırmak River by monthly, seasonal and annual analyses, the determined trends were found to be statistically insignificant according to the results of Spearman’s rank correlation test and the Mann–Kendall trend test. Test statistics for both tests are presented in Table 4.

Table 4

Values of non-parametric tests and trend status

Period	Streamflow	Kendall’s tau	p	Trend	Spearman’s rho	p	Trend
Annual	Annual	−0.055	0.784	▼	−0.055	0.852	▼
Seasonal	Spring	−0.209	0.298	▼	−0.297	0.303	▼
	Summer	0.143	0.477	▲	0.240	0.409	▲
	Autumn	−0.297	0.169	▼	−0.437	0.118	▼
	Winter	−0.099	0.622	▼	−0.108	0.714	▼
Monthly	January	−0.209	0.298	▼	−0.262	0.366	▼
	February	−0.165	0.412	▼	−0.204	0.483	▼
	March	−0.055	0.784	▼	−0.099	0.737	▼
	April	−0.209	0.298	▼	−0.288	0.318	▼
	May	−0.143	0.477	▼	−0.143	0.626	▼
	June	0.209	0.298	▲	0.244	0.401	▲
	July	0.011	0.956	▲	0.156	0.594	▲
	August	−0.231	0.250	▼	−0.288	0.318	▼
	September	−0.209	0.298	▼	−0.341	0.233	▼
	October	−0.231	0.250	▼	−0.349	0.221	▼
	November	−0.209	0.298	▼	−0.226	0.436	▼
	December	0.033	0.870	▲	−0.011	0.970	▼

Note: ▼ indicates decreasing trends; ▲ indicates increasing trends

Discussion

Due to climate change and global warming, some countries will face severe water shortages or scarcities of the limited water resources (Hisar et al. 2015). The monitoring of changes in climatic factors is important for predicting possible climate change in the future. It is important to consider the assumptions of different climate change scenarios and to better understand the impact of climate change on river streamflow (Kale et al. 2016a). In this context, research on the past and present trends in water resources provides a valuable contribution to the current knowledge used in water resource management and decision-making processes to mitigate and adapt to possible effects of climate change. Thus, many researchers studied possible effects of climate change on water resources and they reported that climate change can have a serious impact on the availability of water resources (Durdu 2010; Kale et al. 2016a,b; Ejder et al. 2016a,b; Kale et al. 2018; Kale & Sönmez 2018a,b; Kale & Sönmez 2019a,b,c; Sönmez & Kale 2020). Consequently, research on changes in the river streamflow is of major importance.

Annual, seasonal and monthly streamflow data were analyzed in this study using the trend analysis. The analysis showed a decreasing trend in the annual streamflow of the Gökırmak River. Furthermore, decreasing trends were also determined for all seasons except summer. Similarly, many researchers reported decreasing trends in the river streamflow (Kahya & Kalayci 2004; Alcamo et al. 2007; Ozkul 2009; Durdu 2010; Türkeş & Acar Deniz 2011; Saplıoglu et al. 2014; Herawati et al. 2015; Zhou et al. 2015; Li et al. 2016; Pumo et al. 2016). Bahadır (2011) reported a decreasing trend of the Kızılırmak River streamflow. Ejder et al. (2016a) identified a decreasing trend for the Sarıçay Stream. Ejder et al. (2016b) reported a decreasing trend for the Kocabaş streamflow. Kale et al. (2016a) identified a decreasing trend in the Karamenderes River streamflow. Kale et al. (2016b) described a declining trend in the Bakırçay River streamflow. In addition, Ay & Kişi (2017) and Ercan et al. (2017) reported a decreasing trend in the Kızılırmak River streamflow. Kale and Sönmez (2018a) reported decreasing trends for the annual, seasonal and monthly streamflow of the Daday Stream. Kale and Sönmez (2018b) defined a decreasing trend for the Akkaya streamflow. Kişi et al. (2018) found increasing and decreasing trends for the monthly streamflow in three different basins in Turkey. Yildiz et al. (2019) determined a decreasing trend in the long-term streamflow of the Euphrates River. Kale & Sönmez (2019a) indicated a declining trend in the Ilgaz streamflow. Kale & Sönmez (2019b) documented a declining trend in the Araç streamflow. Kale & Sönmez (2019c) reported that the annual and seasonal streamflow of the Devrekani Stream tends to decrease. Sönmez & Kale (2020) stated that increasing temperature and decreasing precipitation due to climate change caused a reduction in the streamflow of the Filyos River. The Gökırmak River is located near the Filyos, Ilgaz, Akkaya, Daday, Araç, Devrekani and Kızılırmak rivers in Turkey. Therefore, the downward trends in the streamflow of rivers may continue in the future.

Both decreasing and increasing trends were observed in the monthly streamflow, whereas no trend was found for February and December. Although some researchers reported decreasing trends in the streamflow of rivers, increasing trends were also reported in the literature. Topaloglu (2006) reported both decreasing and increasing trends for 26 basins in Turkey. Ali et al. (2019) documented that there were both increasing and decreasing trends in the streamflow of the Yangtze River. Lee & Yeh (2019) determined an increase in the streamflow of the Lanyang River, while the streamflow of the Keelung River showed a downward trend. Rani & Sreekesh (2019) described a reduction in the streamflow of the Western Indian Himalaya Watershed due to reduced snow cover resulting from the increasing temperature. Wang et al. (2019) reported that the streamflow in data-scarce mountain basins of Northwest China increased with a warm and wet climate. In this study, the streamflow in the summer season showed great variability. Therefore, this fluctuation in the streamflow in summer could be the main reason for the increasing trend.

In general, variations in the streamflow of rivers could be related to climate change, human impact, population, precipitation patterns, land use and land cover. In this study, both increasing and decreasing trends were determined for the streamflow of the Gökırmak River during the monitoring period. These fluctuations in the Gökırmak River streamflow can be attributed to climatic changes, changes in evaporation and precipitation patterns, snow melting seasons and the amount, volume and time of glacier flow, as well as human impact. Similarly, many scientists have reported that climate change could reduce the streamflow of rivers (Kahya & Kalayci 2004; Bozkurt & Sen 2013; Ligaray et al. 2015; Li & Jin 2017; Naz et al. 2018; Dinpashoh et al. 2019; Hirpa et al. 2019; Lee and Yeh 2019; Yan et al. 2019). Anil & Ramesh (2017) reported an increase in the annual and monthly streamflow of the Harangi Stream due to a reduction in the forest area. Kale et al. (2018) stated that variations in precipitation patterns can have a direct impact on the river streamflow. In addition, Kale and Sönmez (2018b) indicated that the rate of streamflow and water resources may be related to a rise in temperature and evaporation patterns, reduction in rainfall and snowmelt, and additional factors associated with climate change. Furthermore, it was documented that the water surface area (Kale & Acarli 2019a) and shoreline (Kale & Acarli 2019b) of the Atikhisar Reservoir showed spatial and temporal variations due to climate change. On the other hand, Bates et al. (2008) concluded that trends in the streamflow were not always related to changes in precipitation. Furthermore, a number of scientists stated that other factors may affect the streamflow in addition to the effects of climate change, such as agricultural (Durdu 2010; Dügel & Kazanci 2004; Yercan et al. 2004; Chen et al. 2019) and non-agricultural human activities (Li et al. 2007; Gao et al. 2011; Jackson et al. 2011; Jiang et al. 2011; Hu et al. 2012; Wang et al. 2012; Chang et al. 2014; Zhan et al. 2014; Zhou et al. 2015; Guo et al. 2016; Qian et al. 2016; Huang et al. 2018; Saidi et al. 2018; Shahid et al. 2018; Lee and Yeh 2019; Lv et al. 2019; Fu et al. 2020; Kanani et al. 2020).

Conclusions

This is the first study on the streamflow trends for the Gökırmak River. Monthly, seasonal and annual trends in the Gökırmak River streamflow were investigated. Most of the streamflow trends identified during the monitoring period were downward trends. Therefore, trends in the streamflow should be continuously monitored to develop better management strategies for sustainable use of water resources in the future. This paper presents the results of the research on the past and present trends in the streamflow, thus contributing with new knowledge for decision makers and water resource managers.

eISSN:: 1897-3191
Sprache:: Englisch

Zeitrahmen der Veröffentlichung:: 4 Hefte pro Jahr
Fachgebiete der Zeitschrift:: Chemie, andere, Geowissenschaften, Biologie

Zeitschrift RSS Feed

Trend analysis and forecasting of the Gökırmak River streamflow (Turkey)

Article Category: Research Article

Online veröffentlicht: 25. Sept. 2020

Seitenbereich: 230 - 246

Eingereicht: 25. Feb. 2020

Akzeptiert: 20. Apr. 2020

DOI: https://doi.org/10.1515/ohs-2020-0021

© 2020 Faculty of Oceanography and Geography, University of Gdańsk, Poland

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7