Interbasin river mapping between Californian and Turkish rivers for climate change assessment Open Access

Climate change due to anthropogenic activities extends beyond natural climate variability and has led to adverse impacts and associated losses for both ecosystems and human populations. This change is characterized by more frequent and severe weather and climate events (IPCC 2022). Projections indicate that alterations in the magnitude, timing, and frequency of streamflow, along with associated extremes, will have adverse effects on freshwater ecosystems and water availability for agriculture, hydropower generation, and urban areas in the mid- to long-term, particularly affecting snowmelt-dependent water systems (IPCC 2022).

The Mediterranean Basin, characterized by low annual precipitation and high interannual variability, with warm and dry summers and precipitation falling mostly in winter, is key for meeting agricultural and urban water demands (Tuel & Eltahir 2020), making it vulnerable to climatic changes. With rising global temperatures, more extreme and intense precipitation events are expected over mid-latitudes, including the Mediterranean region (Srivastava et al. 2020). In this region, climate change is expected to cause prolonged and stronger heat waves and intense droughts in an already dry climate (Ali et al. 2022). Mediterranean climates can be observed in northern Iran, Chile, Australia, South Africa, California and the coastal regions of the Mediterranean Sea, including Turkey (Türkeş et al. 2020). According to the latest statewide climate assessment reports of both Turkey (GDWM 2020) and California (Bedsworth et al. 2018), precipitation will mostly fall as rainfall rather than snowfall due to increased temperatures, and snowmelt runoff will be observed in the earlier months.

Located in the Mediterranean region of the subtropical zone, Turkey is susceptible to changes in climatic conditions (Bağçaci et al. 2021). Turkey experienced an increase in average temperature during the 40-year period between 1960 and 2000 (Türkeş et al. 2002). In an early study, Fujihara et al. (2008) evaluated the climate change impacts on the Seyhan River Basin in Turkey and found up to a 60% decrease in annual runoff under a warm and dry climate scenario. Climate projections for Turkey suggest warmer and drier summers and wetter winters with increased climate variability (Bağçaci et al. 2021), while Turkes et al. (2020) and Seker & Gumus (2022) predicted an increase in temperature and a decrease in precipitation, with an increase in drought severity, especially in the southern part of Turkey under climate change. Aziz et al. (2020) found that increased temperature might contribute to less snowpack, shifting snowmelt runoff to earlier months and decreasing water availability in the summer, especially in the eastern part of Turkey. Sönmez & Kale (2020) found a decreasing trend in streamflow and precipitation for the Filyos River in northern Turkey.

Historical and future projected climates simulated by General Circulation Models (GCMs) are the fundamental source of information for climate change assessment studies (Fowler et al. 2007; Carvalho et al. 2021). These models generate climate outputs, such as temperature and precipitation, at coarse spatial scales, with grid sizes ranging from 100 to 600 km (Kumar et al. 2021). Thus, it is necessary to downscale the coarse global-scale climate outputs to the finer catchment scale to evaluate climate change impacts on water resources (Wilby et al. 1998). Dynamical and statistical downscaling methods are used to extract information for local regions from global-scale GCM outputs (Wilby et al. 1998; Chen et al. 2011, 2020; Hundecha et al. 2016; Shamir et al. 2019; Ishida et al. 2020; Tapiador et al. 2020; Onarun et al. 2023; Trinh et al. 2023), in addition to empirical downscaling (Walker et al. 2022). Statistical methods are computationally cheap and relatively easier to implement compared to dynamical methods (Fowler et al. 2007). However, obtaining a stable relationship between observed local climate variables and GCM outputs is critical for statistical methods (Chen et al. 2011). Machine learning algorithms and regression are commonly used statistical approaches (Ghosh & Mujumdar 2008; Chen et al. 2012; Kirdemir et al. 2022; Zhang et al. 2022; Nguyen et al. 2023; Polasky et al. 2023). Index Basin Mapping (IBM) is another statistical method for transferring downscaled climate information from one river basin to another that is hydrologically and climatologically similar. Zhu et al. (2005) and Medellín-Azuara et al. (2008) used IBM for climate change assessment studies on California's water supply system, where 37 reservoir inflows are mapped to six index basins to transfer hydrologic information. Connell-Buck et al. (2011) mapped 13 out of 18 available index basins to 37 rim inflow locations to analyze climate change effects and explore adaptation strategies in California.

While previous studies have primarily focused on projecting future temperature and precipitation trends in Turkey (GDWM 2016), there has been a notable lack of systematic and comprehensive evaluations regarding the impacts of climate change on streamflow timing and availability across various hydrological systems. This study aims to fill this gap by examining the effects of climate change on eight rivers situated in diverse hydrological regions of Turkey. A wide variety of climate scenarios, ranging from wetter to drier streamflow projections, are analyzed using the IBM approach, which bridges Turkish and Californian river systems. Leveraging data from five out of 11 available index basins in California, this study maps these basins onto eight river basins in Turkey to assess changes in streamflow timing and magnitude. The study incorporates 20 CMIP5 climate scenarios from 10 GCMs under two Representative Concentration Pathway (RCP) cases (RCP 4.5 and RCP 8.5). Monthly normalized hydrographs are employed to identify hydrologically similar basins, with performance indicators of mean squared error (MSE) and Pearson's correlation coefficient. Subsequently, monthly perturbation ratios are calculated for the index river basins and applied to their corresponding rivers in Turkey. The study addresses several key assessment questions, including the potential impacts of climatic changes on streamflows, shifts in monthly and seasonal streamflow availability, and the implications for both rain-fed and snowmelt-fed systems.

The studied rivers in Turkey were selected based on historical streamflow data availability between January 1, 1990, and December 31, 2013. In addition, streamflow locations on the main branch and away from downstream large reservoirs were preferred. Thus, eight river locations were selected: Ceyhan River at Misis, Çoruh River at İspir, Eşen River near Kınık, Göksu River near Karahacılı, Kızılırmak River near Söğütlühan, Meriç River near Kirişhane, Yeşilırmak River near Kale and Zamantı River at Fraktin. Their average flow rates and drainage areas at the river locations are summarized in Table 1. Historical streamflow observations are obtained from the streamflow almanac of the State Hydraulic Works (https://www.dsi.gov.tr/Sayfa/Detay/744).

Downscaled CMIP5 streamflow projections under 20 climate scenarios from 10 GCMs and two RCPs were generated as part of California's Fourth Climate Assessment and are available from Cal-Adapt (https://cal-adapt.org). The selected 20 climate models reflect likely future climate scenarios and include various hydrologic cases, ranging from wetter to drier than existing conditions (Herman et al. 2018). These downscaled streamflow projections are available for 11 river locations: Sacramento River near Red Bluff, Feather River near Oroville, Yuba River near Smartville, Bear River near Wheatland, American River at Fair Oaks, Mokelumne River at Pardee, Calaveras River at Jenny Lind, Stanislaus River at New Melones, Tuolumne River at Don Pedro, Merced River at Exchequer and San Joaquin River at Millerton. Historical observations are available from January 31, 1950, to December 31, 2013, at monthly time steps. Future streamflow predictions under 20 CMIP5 scenarios for index basins are used from January 2014 to December 2099, where the years between 2014 and 2049 are called early century, 2050 and 2074 are called mid-century, and 2075 and 2099 are called late century.

A systematic approach was employed to identify the suitable index basins for each surveyed river location in Turkey by examining the correlation and temporal distribution between index basin flows and studied river locations. Monthly correlation coefficients between the historical runoff of the surveyed river basins and index basins from 1990 to 2013 were computed. MSE and Pearson's r were separately calculated for the wet season (December through May) and dry season (June through November) between index and corresponding basins. The best index basin or basins were determined by selecting the index basin with the lowest MSE and the highest Pearson's r. Partitioning a water year into wet and dry seasons aids in identifying the most suitable fit for snowmelt or rainfall–dominant runoff regimes. Out of the 11 available index river basins in California, five are used to match the studied river basins in Turkey. Two index schemes are evaluated: single index and wet and dry season index, as shown in Table 2. The single season index method maps to only one index river basin. MSE values range between 0.011 and 0.048, and Pearson's r values range between 0.79 and 0.96 with the single season index scheme. However, better mapping can be achieved when different rivers are assigned in wet and dry seasons, a method also used by Zhu et al. (2005). Wet season index rivers match the single index scheme, but dry season index rivers vary. In the wet and dry season index scheme, MSE values range between 0.011 and 0.039, and Pearson's r values range between 0.84 and 0.97. For the Eşen and Meriç Rivers, the same index river (Sacramento) gives the lowest MSE and highest Pearson's r. Therefore, the wet and dry season index scheme with lower MSE and greater Pearson's r is used throughout the study.

All evaluated GCMs are part of the CMIP5 climate models (Taylor et al. 2012), as shown in Table 3. These CMIP5 models predict warmer air temperatures for the Mediterranean region. Precipitation predictions, however, vary depending on the scenario. For example, some models like CanESM2 and CNRM-CM5 predict a wetter future, while others like ACCESS1-0, HadGEM2-ES and MIROC5 predict a drier climate. These 10 GCMs were run under medium (RCP 4.5) and high (RCP 8.5) carbon emission scenarios, resulting in a total of 20 downscaled climate projections available for index river basins. Additionally, results from ensembles of RCP 4.5 and RCP 8.5 scenarios are provided, and projected climate results are compared to historical climate. These precipitation and temperature projections are downscaled and routed through the Variable Infiltration Capacity (VIC) hydrologic model (Liang et al. 1994) to generate streamflow for 11 index river basin locations (Herman et al. 2018).

An advantage of employing the perturbation ratios approach is that it can be applied to other locations and branches within the same river system to assess climate change impacts. The main limitation of this approach is that it lacks explicit representation of interannual variability (Vicuna et al. 2007). The method relies on the same degree of interannual variability that occurred historically. To overcome this to some extent, different reference years, such as early, mid, and late centuries, are considered while calculating perturbation ratios and perturbed hydrology, as discussed later. Additionally, while the best possible match is achieved between index and mapped rivers, each river system and its hydrograph are unique due to its topography and river basin characteristics. Uncertainties can also arise from the reliability of climate projections.

Climate change effects vary depending on the region and river system. These effects can increase, reduce, or shift water availability to earlier months. Table 4 summarizes the percent change in overall water availability in selected rivers under projected climates compared to historical climate. Some wet scenarios, for example, CanESM2 and CNRM-CM5, predict a 36–54.3% increase in overall annual water availability for the Göksu River, whereas RCP 4.5 and 8.5 ensembles predict a 5.2 and 5.5% increase, respectively. The RCP 4.5 ensemble projects more overall water availability for Ceyhan, Eşen, Göksu, Kızılırmak and Meriç Rivers. On the other hand, the RCP 8.5 ensemble projects more overall water availability only for Ceyhan and Göksu Rivers. Both ensemble scenarios project less water availability for Çoruh, Yeşilırmak and Zamantı Rivers. MIROC5 with RCP 4.5 and ACCESS1-0 with RCP 8.5 scenarios predict 16.7 and 31.7% less water availability than historical conditions for the Çoruh River. Preparing for and taking adaptation measures for droughts and reduced water deliveries to agricultural, environmental, and urban users in rivers with less projected water availability, and for floods in rivers with increased projected water availability, will be important water management issues.

This study investigated the impacts of climate change on streamflow availability and timing for eight selected rivers in Turkey under 20 projected climate scenarios, ranging from wetter to drier conditions. The IBM method was employed to transfer climate change information from index rivers in California, where downscaled projections are available, to regions in Turkey with similar hydrological characteristics but limited data for climate change assessment. This statistical method first identifies hydrologically similar river basins by analyzing their monthly hydrographs under comparable climatic conditions (specifically, the Mediterranean climate), employing MSE and Pearson's correlation coefficient as performance metrics. Then, monthly perturbation ratios from index rivers, conveying projected climate change information, are applied to mapped rivers. As a result of climate change, winter water availability generally increases in most rivers, while spring and summer water availability significantly decreases due to reduced snowmelt runoff, leading to heightened monthly streamflow variability. The Çoruh River is particularly impacted by these climatic changes, experiencing overall water availability reductions of 12.1 and 14.6% under RCP 4.5 and 8.5 ensemble scenarios, due to its dependence on snowmelt. The increased monthly streamflow variability, projected to intensify toward the end of the century, could pose challenges for water management and allocation, especially for environmental and agricultural users. Higher winter water availability may result in floods, while reduced water availability in spring and summer could exacerbate drought conditions. To mitigate the adverse effects of climate change, exploring adaptation measures such as enhancing irrigation efficiency, implementing water conservation strategies, and expanding surface water storage is needed. Additionally, managed aquifer recharge, involving the diversion of excess winter flows to groundwater for pumping during periods of low surface water availability, is another useful adaptation measure.

This study used streamflow projections developed from the CMIP5 database. When CMIP6 streamflow projections become available, the proposed method can be applied and the analysis can be expanded to other streamflow locations.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.