Climate Change Effects on Precipitation and Streamflow in the Mediterranean Region

Abstract

This study investigates the impact of climate change on the Mudurnu Stream Basin in northwest Türkiye by analyzing climate parameters in the Mediterranean region. Historical data from EC-Earth2, HadGEM2-ES, and MPI-ESM-MR GCMs from the CMIP5 Euro-CORDEX archive were assessed, and future precipitation and temperature data were derived using five statistical bias correction methods for the selected EC-Earth2 model under RCP4.5 and RCP8.5 scenarios. The SWAT model was employed to simulate future runoff amounts for the Mudurnu Stream Basin. The findings reveal notable changes in precipitation and temperature. The annual and seasonal variations of total precipitation and average, maximum, and minimum temperatures for the RCP4.5 and RCP8.5 scenarios in the Sakarya and Mudurnu regions were analyzed and determined. The projections for future river flow indicate a significant increase in precipitation during the rainy seasons. The Mudurnu Stream mainstem will experience an increase in flow of between 70 and 140% under RCP4.5 and between 80 and 160% under RCP8.5. In the Dinsiz Stream tributary, a 32–55% increase is observed for the spring and summer months. In this context, the rainfall and runoff projections required for the estimation of potential drought and flood risks in the near and distant future were calculated.

1. Introduction

Understanding the future impacts of changes in climate parameters on water resources, agricultural production, and ecosystems is of critical importance for achieving a sustainable world. The Mediterranean region is among the most vulnerable areas globally to both climatic and anthropogenic changes [1]. Situated in a climatic transition zone, this region exhibits significant temporal and spatial variability in precipitation patterns, typically characterized by wet winters and dry summers [2]. Assessments of climate change in the Mediterranean Basin consistently indicate continuous warming of both the atmosphere and the sea. Projected changes include rising temperatures and altered precipitation regimes [3,4]. Due to the region’s complex morphology and high climatic variability, examining these transformations and conducting hydro-climatic modeling remains a significant challenge [5,6,7]. To address this challenge, the use of high-resolution and high-quality climate data is essential for comprehensively evaluating local and regional impacts [8].

Although global climate models (GCMs) can successfully simulate large-scale atmospheric processes, they generally lack the spatial resolution necessary for decision making at local and regional levels [9,10]. Therefore, downscaling techniques are required to render GCM outputs meaningful at finer spatial scales [11]. These techniques are broadly classified into two main categories: dynamical and statistical downscaling [12]. Dynamical downscaling, which employs regional climate models (RCMs), is a robust approach due to its ability to explicitly represent physical processes in greater detail [10]. However, it has significant drawbacks, such as high computational demands and the generation of large volumes of data [13]. On the other hand, statistical downscaling methods are computationally more efficient and often better aligned with local observational data, but they do not directly model physical processes and may have limited validity under future climate conditions [14].

To produce reliable regional climate projections, the Coordinated Regional Climate Downscaling Experiment (CORDEX) initiative was established, aiming to generate high-resolution climate data through dynamical downscaling methods [15]. CORDEX projects are typically based on outputs from global models developed under the Coupled Model Intercomparison Project Phase 5 (CMIP5), such as HadGEM2-ES, MPI-ESM-LR, and EC-Earth. In recent years, the next generation of GCMs developed under the Coupled Model Intercomparison Project Phase 6 (CMIP6)—which incorporates updated climate scenarios (SSPs: Shared Socioeconomic Pathways) and enhanced physical representations—has started to be utilized [16]. Nevertheless, dynamically downscaled products based on CMIP6-driven RCMs have not yet been completed for all regions, and data availability remains limited.

EC-Earth2 is one of the prominent global climate models developed under the CMIP5 framework and is frequently utilized in Euro-CORDEX studies [17,18]. RCMs derived from EC-Earth2, such as RACMO22E, CCLM4-8-17, and RCA4, are capable of successfully reproducing the seasonal and spatial patterns of key climate variables—including precipitation and temperature—across Europe, owing to their high spatial resolution of approximately 12 km (0.11°) [19,20]. Despite the advantages offered by these high-resolution simulations, they may contain systematic biases when compared with observational data. These biases arise from uncertainties inherent in both the global and regional model structures [21]. Therefore, before their application in environmental impact assessments, hydrological modeling, or decision support systems, these datasets must be processed using statistical bias correction methods such as Distribution Mapping, Linear Scaling, etc. Bias correction methods exhibit different performances depending on the type of data. Linear Scaling and Variance Scaling are methods suitable for precipitation and temperature data, respectively, and they correct for mean and variance. Power Transformation and Local Intensity Scaling (LOCI) are effective for daily precipitation data with skewed distributions, and LOCI offers the advantage of separately correcting precipitation frequency and intensity [22]. Distribution Mapping provides the most comprehensive correction by targeting the entire distribution of climate variables such as temperature and precipitation; however, due to its high computational cost, it should be applied with caution in large datasets. To sum up, downscaling and bias correction methods are usually applied together to improve the accuracy and usability of climate model outputs at a local level. Downscaling increases the spatial resolution of climate models to enable the analysis of local effects, while error correction eliminates systematic errors in model outputs to make them more consistent with observations.

Climate change studies are frequently conducted for the Mediterranean climate region in Türkiye and around the world [23,24,25,26]. Although climate hydrology studies in Türkiye commonly focus on large basins within the Mediterranean climate zone, detailed modeling and climate scenario integration for smaller strategically significant sub-basins—such as the Mudurnu River Basin—remain limited in the literature [25,26]. This study aims to comprehensively analyze the hydrological impacts of climate change in the Mudurnu River Basin. For this purpose, the following points have been followed:

• Simulations from the EC-Earth2 global climate model, dynamically downscaled using the RCA4 regional climate model and obtained from the CMIP5 Euro-CORDEX data archive, were used.
• Future streamflow regimes of the basin were assessed in detail using the SWAT hydrological model, which forecasts the dynamics of actual water resource systems by considering various parameters, including precipitation, temperature, soil properties, etc., under the Representative Concentration Pathways (RCP4.5 and RCP8.5) climate scenarios.
• Moreover, five different statistical bias correction methods were evaluated comparatively, and the most suitable method was integrated into the hydro-meteorological modeling process.
• Thus, this study also aims to fill this gap by contributing to a better understanding of the vulnerability of small watersheds to climate change and the potential adaptation strategies they may need.

2. Materials and Methods

2.1. Study Area

Mudurnu Stream Basin is located between 30°24′–31°26′ east longitude and 40°22′–40°54′ north latitude. The basin area is located within the borders of Sakarya, Bolu, and Düzce provinces (Figure 1). Mudurnu Stream, one of the tributaries of the Sakarya River, is located in the Lower Sakarya Sub-basin and is 131 km long. The other major tributary in the basin is the 41 km long Dinsiz Stream. These two streams join the Sakarya River by merging 6.5 km upstream of the confluence point. Mudurnu Stream Basin has a drainage area of 2.211 km². Ballikaya Dam is being constructed on the Mudurnu Stream to meet a significant portion of the drinking and potable water needs of Kocaeli and Sakarya provinces, which are important industrial centers of Türkiye with a population of more than 3 million people. The dam, which is of Roller-Compacted Concrete (RCC) dam type with a height of 98 m from the foundation, supplies a total of 189 hm³ (6 m³/s) of water annually to Kocaeli and Sakarya provinces [27]. Due to this important dam, the Mudurnu Stream Basin was selected as the study area, since the establishment of a hydrological model of the basin can help in the assessment of the effects of climate change in future projections, the determination of basin management strategies, and dam operation studies.

The Mudurnu Stream Basin sits at the junction of three tectono-stratigraphic domains: the Istanbul Zone in the north, the Armutlu–Almacık Zone in the center, and the Sakarya Zone to the south, creating a geologically intricate setting [28,29]. The northern part includes Precambrian gneiss–amphibolite of the Yedigoller Complex and is unconformably overlain by fluvial-to-shallow marine Paleozoic successions (Kurtkoy and Eregli formations), deeper-water Yılanlı turbidites, and, finally, Maastrichtian Akveren limestones. The central belt includes Metamorphics of the Iznik and Akçay groups enveloping slices of Almacık ophiolite, while the Abant Mélange preserves a chaotic mix of oceanic and continental fragments. The southern part of it comprises Jurassic–Early Cretaceous volcanic–clastic rocks of the Mudurnu Formation pass upward into shelf carbonates of the Bilecik and Sogukcam limestones, which are, in turn, overlain by Cenomanian–Paleocene Yenipazar flysch. Eocene Yıgılca volcanics, Eocene Caycuma turbidites, and scattered Pliocene Özlü basalts mantle the older units, whereas Quaternary alluvium fills the valley floors [30].

2.2. Applied Methodology

In this study, in order to examine the effects of climate change on the Mudurnu Stream Basin, the hydrological model of the Mudurnu Basin established on the SWAT2012 rev.685 program was run with Sakarya and Mudurnu MOS climate data representing the basin for the period 1978–1995, the first 3 years of which were the warming period, and the flow results given by the model were calibrated and validated with the flow measurements of Dokurcun, Gebes, and Yagbasan SGS in the basin using the SWAT-CUP 2019 v5.3 (SWAT Calibration and Uncertainty Programs).

In order to apply the future climate projection to the basin, the EC-Earth2 climate model was selected by analyzing the historical data of EC-Earth2, HadGEM2-ES, and MPI-ESM-LR models presented within the scope of the EURO-CORDEX project with observation data. The data of the RCP4.5 and RCP8.5 scenarios for the future 2021–2100 period, which were downscaled with the RCA4 regional climate model within the scope of the EURO-CORDEX project, were subjected to a correction process to remove biases. Four bias correction methods were used for precipitation, and three methods were used for temperature. By using the historical precipitation and temperature data of the climate model for the period 1970–1995 and the observation precipitation and temperature data of each MGI station for the period 1970–1995, bias-corrected climate data were obtained for each station according to these methods.

The obtained climate model data of future projection RCP4.5 and RCP8.5 scenarios for the 2026–2100 period were analyzed in the same period in the monthly time period, and in the 2026–2060 and 2061–2100 periods in the seasonal and annual time periods. Based on this, the climate model data of the adjusted future projection RCP4.5 and RCP8.5 scenarios were entered into the hydrological model established in the SWAT2012 rev. 685 program as climate data, and the hydrological model was run for the period 2021–2100, with the first 5 years being the warming period, and future simulation flow data were obtained. The flowchart of the applied methodology is shown in Figure 2.

2.3. SWAT Hydrological Model

The SWAT, designed by the USDA, plays a crucial role in the assessment of water resources and non-point source pollution in watersheds and large river basins. Operating at the watershed scale, this model is conceptual, physically based, and continuous time in nature. SWAT integrates a range of key components to comprehensively model watershed processes. These components include weather patterns, hydrology, soil temperature, soil properties, plant growth dynamics, nutrient cycling, pesticide transport, and bacteria and pathogens fate and transport, as well as the influence of land management practices. By considering these various factors, SWAT provides a holistic approach to simulating the complex interactions within a watershed, making it a valuable tool for understanding and managing water resources and pollution at a regional scale [31,32,33].

The SWAT model employs a hierarchical approach in which the watershed is initially subdivided into multiple sub-watersheds. These sub-watersheds are further divided into Hydrological Response Units (HRUs), with each HRU characterized by uniform land use, soil properties, and slope. Subsequently, the model conducts simulations for the hydrological balance within each HRU. This involves intricate considerations of various processes, including but not limited to rainfall retention by vegetation, rainfall partitioning, snowmelt water dynamics, irrigation water influence, surface runoff patterns, infiltration processes, water redistribution within the soil profile, evapotranspiration phenomena, lateral subsurface flow through the soil, and the return flow from shallow aquifers. Through this comprehensive modeling approach at the HRU level, the SWAT model provides a nuanced understanding of the hydrological interactions and processes occurring within the watershed, contributing to effective water resource management and pollution assessment [33].

2.4. Model Input Data and Setup

To develop a SWAT model, four essential datasets are necessary: a digital elevation model (DEM), a land cover and use map (LULC), a soil property map, and climate data. Furthermore, during the calibration phase, flow observation data are essential to refine and adjust the flow outcomes produced by the constructed model.

Table 1. Description and resolution of the data used in this study.

Data Type	Source	Data Resolution
Digital Elevation Model (DEM)	ALOS-PALSAR	12.5 m
Land Use and Cover Map	ESA WorldCover 2020	10 m
Soil Map	UNESCO—FAO Global Soil Map	5 km
Climate Data	General Directorate of Meteorology	2 Stations
Streamflow Observation Data	General Directorate of State Hydraulic Works	3 Stations

provides a summary of the data utilized in both the model setup and calibration processes.

The digital elevation model was obtained from the open-source data [34] of the ALOS-PALSAR (Advanced Land Observing Satellite) satellite developed by the Japan Aerospace Exploration Agency (JAXA, Tokyo, Japan) (Figure 3a). The basin area was calculated as 2211.3 km², the basin was divided into 135 sub-basins, and a total of 3459 hydrological processing units (HRUs) were formed (Figure 3b).

The LULC map utilized in this study was acquired from the WorldCover 2020 project [35], generously provided by the European Space Agency (ESA) (Figure 3c). Moreover, the map depicting soil properties was sourced from comprehensive global data compiled by the Food and Agriculture Organization (FAO), an organization affiliated with UNESCO, accessible through the FAO data portal (Figure 3d).

The climate data were selected by analyzing data from 12 meteorological observation stations (MOSs) installed by the Directorate General of Meteorology around the Mudurnu Stream Basin. Data from Sakarya MOS and Mudurnu MOS were used to reflect the precipitation and temperature characteristics of the basin, and also to run the SWAT model promptly to provide accurate results (Figure 4). SWAT uses a Thiessen polygon approach to determine which stations represent which sub-basins. As a result, the relationship between the Mudurnu Basin and the meteorological stations is shown in Figure 4.

Considering the time intervals of the MOS data in

Table 2. MOS data information used in the study.

Station Code	Station Name	Elevation (m)	Data Time Range	Selected Model Working Time
17069	Sakarya	28	1960–2022	1978–1995
18158	Mudurnu	858	1965–1995

, the model runtime was set to 18 years to cover both the dry and rainy seasons. Due to the closure of Mudurnu MOS after 1995, as a limiting factor, the end of the working period of the model had to be chosen as 1995. The flow results provided by the model were calibrated using measured data from three Stream Gauging (Observation) Stations (SGSs) in the basin. The information about these SGSs is shown in

Table 3. SGS data information used in the study.

Station Code	Station Name	Stream	Cath. Area (km2)	Observation Time Range
E12A037	Dokurcun	Mudurnu	1073.4	1956–2011
D12A136	Gebes	Mudurnu	1521.0	1972–1983
E12A019	Yagbasan	Dinsiz	410.8	1953–2011

. There are missing years in the measurement data of Gebes SGS in the study years of the model. For this reason, the data of Gebes SGS was extended to 1995, the end year of the study, by using the measurements of Dokurcun SGS, which is located 41 km upstream of Gebes SGS on the Mudurnu Stream mainstem. The coefficient of determination (R²) between Gebes SGS and Dokurcun SGS was 0.823, indicating that the correlation between the two SGSs is quite high.

Accurate calculation of potential evapotranspiration (PET) is crucial for the SWAT model to produce realistic, balanced, and reliable results. A large number of methods have been developed to estimate PET. Three of these methods—Penman–Monteith [36], Priestley–Taylor [37], and Hargreaves [38]—are included in SWAT. The three PET methods included in the model vary depending on the specific inputs required. The Penman–Monteith method requires solar radiation, air temperature, relative humidity, and wind speed data, while the Priestley–Taylor method requires solar radiation, air temperature, and relative humidity data. The Hargreaves method, on the other hand, only requires air temperature. The Hargreaves method was chosen for this study based on an examination of the basic model results from all three PET methods, as it showed better agreement with the observed streamflow data.

2.5. Model Calibration and Validation

The calibration process was conducted to enhance the alignment between the model results and observed flow data, aiming to improve the overall performance of the model. Subsequently, validation was carried out to validate the effectiveness of this calibration process.

For the calibration and validation procedures, SWAT-CUP 2019 v5.3 was employed, utilizing SUFI-2 (Sequential Uncertainty Fitting) as the optimization algorithm. These tools facilitated the refinement of model parameters to achieve a better fit with observed data and ensured the reliability of the model’s predictive capabilities. Various statistical measures were employed to assess the agreement between the model results and observed data, essentially indicating the success of the model. The evaluation criteria included the coefficient of determination (correlation) denoted as R², Nash–Sutcliffe efficiency (NSE), percent bias (PBIAS), and the ratio of root mean square error (RMSE) to the standard deviation of the observation data (RSR). The model performance in terms of hydrological simulations over a monthly time period is categorized and rated based on established benchmarks, as outlined in

Table 4. Overall performance ratings of hydrological models for the proposed statistics in the monthly time frame [39].

Performance Rating	R2	NSE	RSR	PBIAS (%)
Very Good	0.85 < R2 ≤ 1.00	0.80 < NSE ≤ 1.00	0.00 ≤ RSR ≤ 0.50	PBIAS < ±5
Good	0.75 < R2 ≤ 0.85	0.70 < NSE ≤ 0.80	0.50 < RSR ≤ 0.60	±5 ≤ PBIAS < ±10
Satisfactory	0.60 < R2 ≤ 0.75	0.50 < NSE ≤ 0.70	0.60 < RSR ≤ 0.70	±10 ≤ PBIAS < ±15
Unsatisfactory	0.60 ≤ R2	0.50 ≤ NSE	0.70 > RSR	PBIAS ≥ ±15

[39].

2.6. Climate Change Models and Scenarios

EURO-CORDEX regional climate model (RCM) simulations, which involve generating ensemble climate simulations from various dynamical and empirical–statistical downscaling models, are conducted based on a broad array of global climate models within the framework of the Coupled Model Intercomparison Project Phase 5 (CMIP5). These simulations are carried out at two spatial resolutions: the standard CORDEX resolution of 0.44 degrees (EUR-44, approximately 50 km) and a more refined resolution of 0.11 degrees (EUR-11, approximately 12.5 km) [40].

In this research, historical data from the EC-Earth2, HadGEM2-ES, and MPI-ESM-LR models, developed under the CMIP5 framework, are compared with observational data obtained from the MOS. The comparison results are presented in

Table 5. Comparison of GCM historical and observational data for precipitation.

GCM	MOS	Sakarya		Mudurnu
	Scenario	RCP4.5	RCP8.5	RCP4.5	RCP8.5
EC-Earth2	Observed	824.1		544.2
	Historical Raw	424.7		523.4
	Historical Bias	815.6		544.0
	Scenario Raw	477.7	463.1	575.8	562.4
	Scenario Bias	896.2	847.1	580.8	565.1
HadGEM2-ES	Observed	824.1		544.2
	Historical Raw	475.8		551.1
	Historical Bias	759.0		518.6
	Scenario Raw	498.3	518.0	573.4	589.4
	Scenario Bias	848.8	897.8	410.3	449.5
MPI-ESM-LR	Observed	824.1		544.2
	Historical Raw	487.1		615.7
	Historical Bias	808.6		540.3
	Scenario Raw	447.3	453.5	592.9	587.9
	Scenario Bias	729.5	742.2	495.3	485.1

and

Table 6. Comparison of GCM historical and observational data for temperature.

GCM	MOS	Sakarya				Mudurnu
	Scenario	RCP4.5		RCP8.5		RCP4.5		RCP8.5
		Max	Min	Max	Min	Max	Min	Max	Min
EC-Earth2	Observed	19.31	9.70	19.31	9.70	15.78	2.34	15.78	2.34
	Historical Raw	19.60	8.55	19.60	8.55	12.12	1.29	12.12	1.29
	Historical Bias	19.31	9.70	19.31	9.70	15.78	2.34	15.78	2.34
	Scenario Raw	21.22	10.21	22.16	11.04	14.38	3.57	15.69	4.76
	Scenario Bias	21.18	11.36	22.22	12.17	18.38	4.79	19.86	6.07
HadGEM2-ES	Observed	19.31	9.70	19.31	9.70	15.78	2.34	15.78	2.34
	Historical Raw	20.75	9.71	20.75	9.71	13.87	3.01	13.87	3.01
	Historical Bias	19.39	9.75	19.39	9.75	15.88	2.42	15.88	2.42
	Scenario Raw	23.25	12.19	24.19	13.19	17.04	5.97	18.33	7.25
	Scenario Bias	20.44	10.91	21.17	11.66	17.59	3.39	18.50	4.43
MPI-ESM-LR	Observed	19.31	9.70	19.31	9.70	15.78	2.34	15.78	2.34
	Historical Raw	20.85	10.05	20.85	10.05	13.89	3.10	13.89	3.10
	Historical Bias	19.31	9.70	19.31	9.70	15.78	2.34	15.78	2.34
	Scenario Raw	22.79	11.72	23.70	12.73	16.56	5.42	17.68	6.52
	Scenario Bias	21.43	11.33	22.45	12.31	18.74	4.81	19.99	5.97

, revealing generally similar performances among the three models. Specifically, the EC-Earth (European community Earth System Model) version 2 was chosen for this study. The EC-Earth2 model data, dynamically downscaled using the Rossby Center Regional Atmospheric Model v4 (RCA4) regional climate model (RCM) developed by the Swedish Meteorological and Hydrological Institute (SMHI) as part of the CORDEX project, were acquired from the Copernicus Climate Change Service, the Earth observation component of the European Union space program. Details about the climate models can be found in

Table 7. Climate model information used in the study.

Database	GCM	RCM	Scenarios	Resolution	Historical Data Time Period	Projection Data Time Period
CORDEX	EC- Earth2	SMHI RCA4	RCP4.5 RCP8.5	0.11°–12.5 km	1951–2005	2006–2100

.

Among the RCPs that depict various greenhouse gas (GHG) emission scenarios and potential land use changes as outlined in the IPCC Fifth Assessment Report (AR5), RCP4.5 and RCP8.5 have been specifically chosen due to their significant distinctions. Broadly, the RCP4.5 pathway suggests gradual reductions in emissions, leading to CO₂ concentrations stabilizing around 540 ppm by 2100. In contrast, the RCP8.5 trajectory, representing the upper limit among the RCPs, forecasts that CO₂ concentrations will stabilize at approximately 940 ppm by 2100. This is considered a pessimistic scenario due to the substantial increase in greenhouse gas emissions.

For the climate model data stations in this study, the closest ones to the Sakarya and Mudurnu MOSs, serving as references in the bias correction process, were chosen from the data station network of the EC-Earth2 climate model, encompassing the basin boundaries. Raw data from these selected stations underwent bias correction. Figure 5 displays the network of data stations from the EC-Earth2 climate model and highlights the selected data stations.

2.7. Bias Correction Methods

Systematic errors, also known as biases, exist in the output of both GCMs and RCMs. These errors stem from factors such as limited spatial resolution (large grid sizes), simplified thermodynamic processes and physics, or an incomplete understanding of the global climate system. Consequently, utilizing unadjusted outputs in impact models or climate impact assessments can often lead to unrealistic results. Various bias correction methods have been devised to address significant biases in climate models. It is crucial to recognize that the quality of observational datasets plays a pivotal role in determining the efficacy of bias correction methods. A high-quality observation dataset is essential for conducting effective bias correction [41].

In this study, the CMhyd (Climate Model data for hydrologic modeling) program [42] was employed. This program is designed to extract and bias-correct data from global and regional climate models. A concise overview of the methods applied is provided in

Table 8. Bias correction methods used in the study.

Bias Correction Method	Variable	Description
LS	Precipitation Temperature	The adjusted regional climate model (RCM) employs monthly correction values to guarantee a strong alignment between the monthly mean values of the simulations and the observed data, as outlined by Lenderink et al. [43].
LOCI	Precipitation	It corrects the mean values using monthly correction values, wet day frequencies, and wet day rainfall intensities individually, employing a three-step process as described by Schmidli et al. [44].
PT	Precipitation	It modifies both the variance and mean of monthly data through a non-linear correction in exponential form, as demonstrated by Leander and Buishand [45] and Leander et al. [46].
VS	Temperature	It aligns both the mean and variance of the temperature time series in a manner akin to power transformation, as detailed by Chen et al. [47,48].
DM	Precipitation Temperature	By utilizing the Gamma distribution for precipitation data and the Gaussian distribution for temperature data, the regional climate model (RCM) adjusts the distribution function of simulated climate values to align with the observed distribution function, as discussed by Sennikovs and Bethers [49] and Teutschbein and Seibert [50].

Notes: LS: Linear Scaling; LOCI: Local Intensity Scaling; PT: Power Trans.; VS: Variance Scaling; DM: Distr. Mapping.

.

3. Results

After the hydrological model was created by using the input data of the basin, the model was executed by entering the climate data into the model. A period of 18 years between 1978 and 1995 was determined as the model study period. In order to create the initial conditions of the model and to increase the accuracy of the outputs, 3 years were entered as the “warm-up period”. A calibration period of 10 years between 1981 and 1990 and a validation period of 5 years between 1991 and 1995, including wet and dry periods, were determined.

After the values of the preliminary parameters (TLAPS, PLAPS, SFTMP, SMTMP, SMFMX, SMFMN, and TIMP) were determined and fixed, the calibration of the parameters representing the other hydrological processes started. CN2, GW_DELAY, GWQMN, SOL_Z, SOL_BD, and RCHRG_DP were found to be the most sensitive parameters, respectively. The parameter values of the calibration processes are shown in

Table 9. Calibration parameter values.

Parameter	Description	Mathematical Operation	Limit Values		Initial Values		Optimal Value
			Min	Max	Min	Max	Dokurcun	Gebes	Yagbasan
PLAPS	Precip. laps rate (mm/km)	v	−1000	1000	0	500	200	400	500
TLAPS	Temp. laps rate (°C/km)	v	−10	10	0	−10	−10	−10	−10
SFTMP	Snowfall temp.(°C)	v	−20	20	−5	5	−2.07
SMTMP	Snow melt base temp. (°C)	v	−20	20	−5	5	−2.09
SMFMX	Snow melt factor on 21.07 (mm/°C-day)	v	0	20	0	10	9.03
SMFMN	Snow melt factor on 21.12 (mm/°C-day)	v	0	20	0	10	0.41
TIMP	Snow pack temp. lag factor	v	0	1	0	1	0.06
CN2	Initial SCS runoff curve number for moisture condition II	r	35	98	−0.1	0.1	−0.094	−0.093	0.092
GW_DELAY	Groundwater delay (days)	v	0	500	0	100	15.25	9.25	0.25
GWQMN	Threshold depth of water in the shallow aquifer for return flow to occur (mm)	v	0	5000	0	2000	1045	315	505
SOL_Z	Depth to bottom of soil layer (mm)	r	0	3500	−0.1	0.1	0.013	−0.035	−0.089
SOL_BD	Moist bulk density of soil layer (mg/m3 or g/cm3)	r	0.9	2.5	−0.1	0.1	−0.044	0.069	0.036
RCHRG_DP	Deep aquifer percolation fraction	v	0	1	0	0.1	0.076	0.095	0.005

Notes: v: replace value, r: multiply by.

.

Figure 6 shows the observation data for Dokurcun, Gebes, and Yagbasan SGSs, where calibration and validation procedures were performed, and the graphs show the results of the calibrated and validated models.

The statistical performance metrics and model evaluation grades for each SGS are shown in

Table 10. Model calibration performance results and evaluation ratings.

SGS	Calibration (1981–1990)				Validation (1991–1995)
	R2	NSE	PBIAS	RSR	R2	NSE	PBIAS	RSR
Dokurcun	0.87 (VG)	0.87 (VG)	3.70 (VG)	0.37 (VG)	0.80 (G)	0.78 (G)	7.80 (G)	0.47 (VG)
Yagbasan	0.80 (G)	0.80 (G)	2.50 (VG)	0.45 (VG)	0.72 (S)	0.71 (G)	11.80 (S)	0.54 (G)
Gebes	0.91 (VG)	0.91 (VG)	4.00 (VG)	0.30 (VG)	0.87 (VG)	0.87 (VG)	3.60 (VG)	0.36 (VG)

Notes: VG: very good, G: good, S: satisfactory.

, which measure the agreement between calibrated and validated model flow data and observational data.

When the performance values of the model were analyzed, it was determined that the model was successful in reflecting the hydrological processes of the basin and simulating the stream flows observed in the real situation by receiving “Very Good” and “Good” ratings.

3.1. EC-Earth2 Climate Model Data and Bias Correction Results

For the RCP4.5 and RCP8.5 scenarios of the EC-Earth2 model, bias corrections were applied to the model data from Sakarya and Mudurnu meteorological stations separately. For precipitation, four methods were used: LS, LOCI, PT, and DM. For temperature, three methods were used: LS, VS, and DM. Daily precipitation and minimum and maximum temperature data from 1970 to 1995 for Sakarya and Mudurnu MOSs were used as reference historical observation data. The results obtained from these methods were analyzed with the climate indices in

Table 11. Bias correction methods and evaluation indices.

Precipitation	Temperature
Monthly Mean (μ)	Monthly Mean (μ)
90th Percentiles (X90)	10th and 90th Percentiles (X10, X90)
Standard Deviations (σ)	Standard Deviations (σ)
Coefficient of Variation (CV)
Probability of Wet Days (Prwet)
Intensity of Wet Days (iwet)

, which are frequently used in the literature [50,51,52,53,54]. For the EC-Earth2 model under the RCP4.5 and RCP8.5 scenarios, bias corrections were applied to the model data using data from Sakarya and Mudurnu meteorological stations separately. Precipitation underwent correction through four methods: LS, LOCI, PT, and DM. Temperature corrections, on the other hand, employed three methods: LS, VS, and DM.

Reference historical observation data, spanning from 1970 to 1995, for daily precipitation and minimum and maximum temperatures were obtained from Sakarya and Mudurnu meteorological observation stations.

The outcomes of these correction methods were scrutinized using climate indices listed in Table 11. These indices, widely utilized in the literature [50,51,52,53,54], provided a comprehensive analysis of the corrected model results against established climatic benchmarks. While all five methods reduced systematic bias to varying degrees, small residual errors remained. These residual errors are consistent with the known limitations of statistical correction methods in the context of non-stationary climate conditions.

The precipitation graphs of the climate indices calculated for precipitation and temperature parameters as a result of bias correction and evaluation processes for Sakarya and Mudurnu MOSs are shown in Figure 7 and temperature graphs are shown in Figure 8.

As a result of this analysis, the bias correction method that gave the best results for both precipitation and temperature parameters according to both stations was DM. As a result, the final climate model data were obtained by applying the DM correction method to the EC-Earth2 model historical and future climate data. The raw climate model data and the climate model data after bias correction are shown in

Table 12. EC-Earth2 model climate data raw and bias-corrected annual total precipitation results.

Meteorology Observation Station	Sakarya		Mudurnu
Bias Correction Method	DM		DM
Scenarios	RCP4.5	RCP8.5	RCP4.5	RCP8.5
Observation (1970–1995)	824.1		544.2
Historical Raw (1970–1995)	424.7		523.4
Historical Corrected (1970–1995)	815.6	815.6	544.0	544.0
Future Raw (2026–2100)	477.7	463.1	575.8	562.4
Future Corrected (2026–2100)	896.2	847.1	580.8	565.1

for annual total precipitation and

Table 13. EC-Earth2 model climate data raw and bias-corrected monthly average maximum and minimum temperature results.

Meteorology Observation Station	Sakarya				Mudurnu
Bias Correction Method	DM				DM
Scenarios	RCP4.5		RCP8.5		RCP4.5		RCP8.5
	Max	Min	Max	Min	Max	Min	Max	Min
Observation (1970–1995)	19.31	9.70	19.31	9.70	15.78	2.34	15.78	2.34
Historical Raw (1970–1995)	19.60	8.55	19.60	8.55	12.12	1.29	12.12	1.29
Historical Corrected (1970–1995)	19.31	9.70	19.31	9.70	15.78	2.34	15.78	2.34
Future Raw (2026–2100)	21.22	10.21	22.16	11.04	14.38	3.57	15.69	4.76
Future Corrected (2026–2100)	21.18	11.36	22.22	12.17	18.38	4.79	19.86	6.07

for monthly average maximum and minimum temperature.

3.2. Meteorological Results

Using the DM of the EC-Earth2 model, bias-corrected precipitation and temperature data according to the RCP4.5 and RCP8.5 scenarios were analyzed for the future projection 2026–2100. Monthly data are analyzed for the period 2026–2100, while seasonal and annual data are analyzed for the periods 2026–2060 and 2061–2100. For comparison purposes, observation data of Sakarya and Mudurnu MOSs for the period 1970–1995 are also included in the graphs (Figure 9, Figure 10 and Figure 11).

According to the results (

Table 14. Sakarya MOS and Mudurnu MOS average monthly total precipitation and changes.

Sakarya MOS						Mudurnu MOS
Months	Observation (1970−1995) (mm)	RCP4.5 (2026−2100) (mm)	RCP8.5 (2026−2100) (mm)	Rate of Change (RCP4.5)	Rate of Change (RCP8.5)	Observation (1970−1995) (mm)	RCP4.5 (2026−2100) (mm)	RCP8.5 (2026−2100) (mm)	Rate of Change (RCP4.5)	Rate of Change (RCP8.5)
Jan	85.82	95.42	99.26	(↑) 11%	(↑) 16%	72.60	82.01	86.76	(↑) 13%	(↑) 20%
Feb	68.87	84.93	98.77	(↑) 23%	(↑) 43%	51.45	68.67	81.61	(↑) 33%	(↑) 59%
Mar	65.35	74.20	78.03	(↑) 14%	(↑) 19%	51.71	67.69	67.59	(↑) 31%	(↑) 31%
Apr	57.47	66.86	62.57	(↑) 16%	(↑) 9%	47.64	56.82	55.82	(↑) 19%	(↑) 17%
May	48.12	49.88	44.27	(↑) 4%	(↓) 8%	49.98	42.09	39.00	(↓) 16%	(↓) 22%
Jun	76.38	84.56	83.28	(↑) 11%	(↑) 9%	40.01	29.64	28.47	(↓) 26%	(↓) 29%
Jul	49.78	34.69	19.92	(↓) 30%	(↓) 60%	21.54	9.51	10.24	(↓) 56%	(↓) 52%
Aug	46.23	35.16	21.84	(↓) 24%	(↓) 53%	22.35	22.63	5.25	(↑) 1%	(↓) 77%
Sep	44.20	49.18	37.49	(↑) 11%	(↓) 15%	16.03	20.05	12.12	(↑) 25%	(↓) 24%
Oct	89.52	91.35	87.51	(↑) 2%	(↓) 2%	39.65	39.61	38.00	(↓) 0%	(↓) 4%
Nov	89.96	88.55	89.00	(↓) 2%	(↓) 1%	53.93	58.38	52.51	(↑) 8%	(↓) 3%
Dec	102.41	141.47	125.16	(↑) 38%	(↑) 22%	77.31	83.71	87.69	(↑) 8%	(↑) 13%

Note: ↑: Increase, ↓: Decrease.

and

Table 15. Sakarya MOS and Mudurnu MOS average seasonal and annual total precipitation and its variations.

Sakarya MOS						Mudurnu MOS
Seasons	Observation (1970−1995) (mm)	RCP4.5 (2026−2100) (mm)	RCP8.5 (2026−2100) (mm)	Rate of Change (RCP4.5)	Rate of Change (RCP8.5)	Observation (1970−1995) (mm)	RCP4.5 (2026−2100) (mm)	RCP8.5 (2026−2100) (mm)	Rate of Change (RCP4.5)	Rate of Change (RCP8.5)
Annual	824.12	896.25	847.09	(↑) 9%	(↑) 3%	544.21	580.81	565.05	(↑) 7%	(↑) 4%
Spring	170.95	190.94	184.87	(↑) 12%	(↑) 8%	149.34	166.61	162.41	(↑) 12%	(↑) 9%
Summer	172.40	154.41	125.03	(↓) 10%	(↓) 27%	83.89	61.78	43.95	(↓) 26%	(↓) 48%
Autumn	223.68	229.08	214.00	(↑) 2%	(↓) 4%	109.62	118.04	102.64	(↑) 8%	(↓) 6%
Winter	257.09	321.82	323.19	(↑) 25%	(↑) 26%	201.36	234.38	256.05	(↑) 16%	(↑) 27%

Note: ↑: Increase, ↓: Decrease.

), dry season summer precipitation in Sakarya MOS region will decrease by 10% and this decrease will be 27% in the RCP8.5 scenario. In the wet season, it was determined that precipitation would increase by 25% in both scenarios. Total annual precipitation increases by 9% and 3% for RCP4.5 and RCP8.5 scenarios, respectively (Table 15).

In the Mudurnu MOS region, there is a greater decrease in summer precipitation in the dry season by 26% and 48% for RCP4.5 and RCP8.5, respectively, while there is an increase of 16% and 27% in the wet season, respectively. Total annual precipitation increases by 7% and 4% for RCP4.5 and RCP8.5 scenarios, respectively (Table 15).

For both MOS regions, it is observed that the temperatures increase in all periods (

Table 16. Sakarya MOS and Mudurnu MOS average monthly max. and min. temperatures and their changes.

MOS	Months	Observation (1970−1995) (°C)		RCP4.5 (2026−2100) (°C)		RCP8.5 (2026−2100) (°C)		Rate of Change (RCP4.5)		Rate of Change (RCP8.5)
		Max	Min	Max	Min	Max	Min	Max	Min	Max	Min
Sakarya	Jan	9.30	2.76	12.22	4.81	12.98	5.38	(↑) 31%	(↑) 74%	(↑) 40%	(↑) 95%
	Feb	10.31	2.93	12.98	5.01	13.84	5.56	(↑) 26%	(↑) 71%	(↑) 34%	(↑) 90%
	Mar	13.36	4.39	16.04	6.19	16.83	6.81	(↑) 20%	(↑) 41%	(↑) 26%	(↑) 55%
	Apr	18.64	8.06	19.98	9.17	20.83	9.66	(↑) 7%	(↑) 14%	(↑) 12%	(↑) 20%
	May	22.81	11.86	24.71	13.32	25.97	14.13	(↑) 8%	(↑) 12%	(↑) 14%	(↑) 19%
	Jun	27.18	15.40	29.35	17.00	30.17	17.65	(↑) 8%	(↑) 10%	(↑) 11%	(↑) 15%
	Jul	28.36	17.38	29.88	19.30	31.05	20.17	(↑) 5%	(↑) 11%	(↑) 9%	(↑) 16%
	Aug	28.33	17.28	29.54	19.00	31.12	20.26	(↑) 4%	(↑) 10%	(↑) 10%	(↑) 17%
	Sep	25.83	14.01	27.19	15.52	28.48	16.79	(↑) 5%	(↑) 11%	(↑) 10%	(↑) 20%
	Oct	20.82	10.79	22.38	12.44	23.52	13.36	(↑) 7%	(↑) 15%	(↑) 13%	(↑) 24%
	Nov	15.34	6.80	17.20	8.54	17.85	9.32	(↑) 12%	(↑) 26%	(↑) 16%	(↑) 37%
	Dec	10.94	4.32	12.29	5.63	13.57	6.54	(↑) 12%	(↑) 30%	(↑) 24%	(↑) 51%
Mudurnu	Jan	3.99	−6.01	7.47	−1.92	8.70	−0.76	(↑) 87%	(↑) 68%	(↑) 118%	(↑) 87%
	Feb	5.89	−5.04	9.94	−1.41	11.23	−0.39	(↑) 69%	(↑) 72%	(↑) 91%	(↑) 92%
	Mar	10.34	−2.35	15.06	0.54	17.03	1.64	(↑) 46%	(↑) 123%	(↑) 65%	(↑) 170%
	Apr	15.41	1.90	18.29	4.10	19.54	5.21	(↑) 19%	(↑) 116%	(↑) 27%	(↑) 174%
	May	19.88	5.25	22.94	8.23	24.66	9.84	(↑) 15%	(↑) 57%	(↑) 24%	(↑) 87%
	Jun	23.67	8.15	26.61	10.72	28.10	12.15	(↑) 12%	(↑) 32%	(↑) 19%	(↑) 49%
	Jul	26.14	10.26	28.05	12.43	29.67	14.02	(↑) 7%	(↑) 21%	(↑) 14%	(↑) 37%
	Aug	26.47	10.05	27.55	11.62	29.27	13.08	(↑) 4%	(↑) 16%	(↑) 11%	(↑) 30%
	Sep	23.54	6.64	25.19	8.65	26.57	9.98	(↑) 7%	(↑) 30%	(↑) 13%	(↑) 50%
	Oct	17.75	3.45	19.80	5.53	21.32	6.90	(↑) 12%	(↑) 60%	(↑) 20%	(↑) 100%
	Nov	10.53	−0.84	12.53	0.98	13.52	1.98	(↑) 19%	(↑) 217%	(↑) 28%	(↑) 336%
	Dec	5.19	−3.84	6.69	−2.42	8.17	−1.20	(↑) 29%	(↑) 37%	(↑) 57%	(↑) 69%

Note: ↑: Increase.

). The highest temperature increase in the Sakarya MOS region is observed in the winter minimum temperature values by 54% and 75% for RCP4.5 and RCP8.5 scenarios, respectively (

Table 17. Sakarya MOS and Mudurnu MOS average seasonal and annual max.and min. temp. and their changes.

MOS	Seasons	Observation (1970−1995) (°C)		RCP4.5 (2026−2100) (°C)		RCP8.5 (2026−2100) (°C)		Rate of Change (RCP4.5)		Rate of Change (RCP8.5)
		Max	Min	Max	Min	Max	Min	Max	Min	Max	Min
Sakarya	Annual	19.31	9.70	21.18	11.36	22.22	12.17	(↑) 10%	(↑) 17%	(↑) 15%	(↑) 25%
	Spring	18.27	8.10	20.25	9.56	21.21	10.20	(↑) 11%	(↑) 18%	(↑) 16%	(↑) 26%
	Summer	27.96	16.69	29.59	18.43	30.78	19.36	(↑) 6%	(↑) 10%	(↑) 10%	(↑) 16%
	Autumn	20.66	10.53	22.26	12.16	23.28	13.16	(↑) 8%	(↑) 15%	(↑) 13%	(↑) 25%
	Winter	10.19	3.34	12.50	5.15	13.46	5.83	(↑) 23%	(↑) 54%	(↑) 32%	(↑) 75%
Mudurnu	Annual	15.78	2.34	18.38	4.79	19.86	6.07	(↑) 16%	(↑) 105%	(↑) 26%	(↑) 159%
	Spring	15.21	1.60	18.76	4.29	20.41	5.56	(↑) 23%	(↑) 168%	(↑) 34%	(↑) 248%
	Summer	25.43	9.48	27.40	11.59	29.01	13.08	(↑) 8%	(↑) 22%	(↑) 14%	(↑) 38%
	Autumn	17.27	3.09	19.17	5.05	20.47	6.29	(↑) 11%	(↑) 63%	(↑) 19%	(↑) 104%
	Winter	5.02	−4.96	8.03	−1.92	9.37	−0.78	(↑) 60%	(↑) 61%	(↑) 87%	(↑) 84%

Note: ↑: Increase.

). Annual average maximum and minimum temperatures increase by 15–25%.

In the Mudurnu MOS region, the increase in temperatures is much more dramatic. The highest temperature increases are observed in spring minimum temperatures of 168% and 248% for RCP4.5 and RCP8.5, respectively (Table 17). Annual average minimum temperatures increased by 105% and 159%, respectively.

3.3. Streamflow Results

The final climate data, which were bias-corrected using the DM of the EC-Earth2 model, were entered into the calibrated and validated SWAT2012 rev. 685 hydrological model as climate data and the hydrological model was run for the period 2021–2100, with a warming period of 2021–2025. As a result, flow results were obtained for the points where Dokurcun, Gebes, and Yagbasan SGSs are located in the basin according to RCP4.5 and RCP8.5 scenarios. The results obtained were analyzed for monthly flows for the period 2026–2100, and seasonal and annual flows for the periods 2026–2060 and 2061–2100. For comparison purposes, observation data of Dokurcun, Gebes, and Yagbasan SGSs for the period 1956–2010 are also included in the graphs. Average monthly and annual flows and seasonal average flows for Dokurcun, Gebes, and Yagbasan SGSs are shown in Figure 12.

Changes in average monthly flows for RCP4.5 and RCP8.5 scenarios according to observation data are shown in

Table 18. Average monthly flows and changes in Dokurcun SGS, Gebes SGS, and Yagbasan SGS.

SGS	Months	Observation (1956−2010) (m3/s)	RCP4.5 (2026−2100) (m3/s)	RCP8.5 (2026−2100) (m3/s)	Rate of Change (RCP4.5)	Rate of Change (RCP8.5)
Dokurcun	Jan	7.45	20.39	20.76	(↑) 174%	(↑) 179%
	Feb	11.25	23.32	26.87	(↑) 107%	(↑) 139%
	Mar	16.49	21.78	22.59	(↑) 32%	(↑) 37%
	Apr	17.36	16.18	15.55	(↓) 7%	(↓) 10%
	May	11.79	11.06	10.28	(↓) 6%	(↓) 13%
	Jun	6.74	6.33	6.94	(↓) 6%	(↑) 3%
	Jul	3.63	3.65	3.63	(↑) 1%	(↑) 0%
	Aug	2.62	4.58	2.50	(↑) 75%	(↓) 5%
	Sep	2.51	3.43	2.44	(↑) 37%	(↓) 3%
	Oct	2.83	3.34	3.28	(↑) 18%	(↑) 16%
	Nov	3.27	5.48	4.70	(↑) 68%	(↑) 44%
	Dec	5.27	12.54	13.86	(↑) 138%	(↑) 163%
Gebes	Jan	20.04	36.38	37.38	(↑) 82%	(↑) 87%
	Feb	28.15	42.54	48.45	(↑) 51%	(↑) 72%
	Mar	37.82	43.95	44.98	(↑) 16%	(↑) 19%
	Apr	38.33	36.47	34.22	(↓) 5%	(↓) 11%
	May	26.38	24.24	22.24	(↓) 8%	(↓) 16%
	Jun	16.82	18.79	19.22	(↑) 12%	(↑) 14%
	Jul	9.18	10.62	9.02	(↑) 16%	(↓) 2%
	Aug	6.83	10.34	6.42	(↑) 51%	(↓) 6%
	Sep	6.67	8.04	6.09	(↑) 21%	(↓) 9%
	Oct	7.76	10.51	9.88	(↑) 36%	(↑) 27%
	Nov	9.45	14.42	13.24	(↑) 52%	(↑) 40%
	Dec	15.50	26.97	28.18	(↑) 74%	(↑) 82%
Yagbasan	Jan	12.22	13.93	13.48	(↑) 14%	(↑) 10%
	Feb	13.73	14.90	16.82	(↑) 8%	(↑) 22%
	Mar	14.61	14.56	14.80	(↓) 0%	(↑) 1%
	Apr	9.17	11.58	10.95	(↑) 26%	(↑) 19%
	May	5.77	7.25	6.81	(↑) 26%	(↑) 18%
	Jun	3.92	8.39	8.41	(↑) 114%	(↑) 115%
	Jul	2.65	4.20	2.90	(↑) 58%	(↑) 9%
	Aug	2.26	3.44	1.93	(↑) 52%	(↓) 15%
	Sep	2.18	2.23	1.75	(↑) 2%	(↓) 20%
	Oct	3.54	4.00	3.81	(↑) 13%	(↑) 8%
	Nov	5.82	4.94	4.54	(↓) 15%	(↓) 22%
	Dec	10.71	12.17	10.37	(↑) 14%	(↓) 3%

Note: ↑: Increase, ↓: Decrease.

for Dokurcun, Gebes, and Yagbasan SGSs.

For Dokurcun and Gebes SGSs on the Mudurnu Stream mainstem, significant increases in flows are generally observed during the wet season. For RCP4.5 and RCP8.5, this increase is 140% and 160% for Dokurcun SGS and 70% and 80% for Gebes SGS, respectively. For the Yagbasan SGS located on the Dinsiz stream, there is a significant increase in flows in the spring and summer months with an increase of 55% and 32% for RCP4.5 and RCP8.5, respectively.

4. Evaluation and Recommendations

The analysis of the results reveals notable trends for the Sakarya region of the basin. In general, there is an observed increase in precipitation during the wet season and a decrease during the dry season. The magnitude of these changes is more pronounced in the RCP8.5 scenario compared to the RCP4.5 scenario. Total annual precipitation shows an increase of approximately 70 mm in the RCP4.5 scenario, while remaining relatively constant in the RCP8.5 scenario.

Concerning temperatures, both maximum and minimum temperatures exhibit a consistent and significant increase throughout the year in the Sakarya region. The average annual maximum temperature rises by approximately 2 °C in the RCP4.5 scenario and by about 3 °C in the RCP8.5 scenario. The annual average minimum temperature experiences an increase of 1.5 °C in the RCP4.5 scenario and about 2.5 °C in the RCP8.5 scenario.

In the Mudurnu region of the basin, similar patterns are observed in terms of precipitation and temperature changes. There is an increase in precipitation during the wet period and a decrease during the dry period, with the RCP8.5 scenario showing more substantial variations. Total annual precipitation in the RCP4.5 scenario increases by around 40 mm, while, in the RCP8.5 scenario, there is an initial increase of 50 mm in the first half of the future period, followed by proximity to the observation average in the second half.

Temperature-wise, the Mudurnu region experiences a more pronounced warming compared to the Sakarya region in the future projection. The average annual maximum temperature increases by approximately 2.5 °C in the RCP4.5 scenario and about 4 °C in the RCP8.5 scenario. The annual average minimum temperature rises by 2.5 °C in the RCP4.5 scenario and about 3.5 °C in the RCP8.5 scenario.

In summary, both regions are expected to undergo warming trends, but the Mudurnu region is projected to experience a more significant increase in temperatures compared to the Sakarya region. Moreover, while both regions show an increase in precipitation during the wet season and a decrease during the dry season, the Sakarya region has the potential to have a higher overall increase in precipitation compared to the Mudurnu region.

When the results of EC-Earth2 climate model future simulation in terms of stream flows are analyzed, it is seen that the flows of Dokurcun and Gebes SGS on the main tributary of Mudurnu Stream increased significantly by 113% (10 m³/s)–130% (11 m³/s) in Dokurcun SGS and 56% (12 m³/s)–65% (15 m³/s) in Gebes SGS during the wet period between December and March compared to RCP4.5 and RCP8.5 scenarios, respectively. This increase in flows reaches very high levels in January in the RCP4.5 and RCP8.5 scenarios, ranging from 174% (13 m³/h) to 179% (13.5 m³/h) for Dokurcun SGS and 82% (16.5 m³/h) to 87% (17.5 m³/h) for Gebes SGS. At Yagbasan SGS, flows are 31% (2 m³/s) higher than the observation data in both scenarios between January and June.

This significant increase in the amount of precipitation and, consequently, in the flow results in the future projection may pose risks in terms of flooding in the rivers in the basin. Urban areas in the region are also at potential risk of urban flooding. For this reason, in order to see the effects of climate change on both river floods and urban floods and the results of future climate projections, studies can be carried out to create intensity–duration–frequency (IDF) curves and calculate instantaneous peak flood flows according to climate models and scenarios. Studies to be carried out in this direction can improve the adaptation of urban drainage systems and river flood protection structures to possible climatic changes in the future projection and prevent possible flood events and their negative effects.

In other words, this hydrological model can help the operation studies of Ballikaya Dam, which will be an important water source for Sakarya and Kocaeli provinces, to determine the current and future management strategies of the basin and make decisions to mitigate the effects of climate change.

5. Limitations and Future Work

• The CMIP5 dataset and RCP scenarios are based on older projections than the more recent CMIP6 and SSP scenarios.
• The SSP scenarios provide a more comprehensive projection of the future, considering socioeconomic developments and emissions.
• Using only a single climate model (EC-Earth2) in the study resulted in model-induced uncertainties not being adequately reflected.
• Future studies using multiple GCM outputs and CMIP6-SSP scenarios can provide more reliable and up-to-date projections.
• Despite bias corrections, residual biases may remain in climate model outputs. Therefore, more advanced techniques, such as quantile delta mapping or bias correction methods based on machine learning, are recommended.
• Integrating the SWAT model with more detailed groundwater models, such as MODFLOW, can be beneficial where the long-term interaction between groundwater and surface water is important.
• The SWAT model is limited in its ability to simulate short-term extreme precipitation events because it operates on a daily time step. To better model such events, hourly precipitation data or remote sensing-based precipitation forecasts could be integrated, and the model could be calibrated to focus on periods of extreme precipitation.