ABSTRACT
This study investigates monthly rainfall and temperature trends in Van Province, Türkiye, using ARIMA and SARIMA models, with a dataset spanning from 1955 to 2023. The ARIMA(3,1,0) model for rainfall and ARIMA(0,1,1) model for temperature were selected based on Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) values, achieving AIC scores of 788.224 and 172.077, respectively. To address seasonality, SARIMA models were also applied, with SARIMA(3,1,0)(2,1,0)[12] for rainfall and SARIMA(0,1,1)(2,1,0)[12] for temperature, yielding AIC scores of 672.061 and 163.669. Diagnostic tests, including the Ljung–Box and Jarque–Bera tests, confirmed model adequacy by indicating minimal autocorrelation and normal residual distributions. These models successfully captured seasonal and long-term patterns, offering valuable insights for regional planning in water resource management and agriculture. The study underscores the potential of ARIMA and SARIMA models for climate forecasting, with suggestions for future enhancements using hybrid approaches to improve predictions under non-linear conditions.
HIGHLIGHTS
Forecasting monthly rainfall and temperature patterns in Van Province using ARIMA and SARIMA models.
ARIMA(3,1,0) and ARIMA(0,1,1) effectively captured trends for rainfall and temperature data.
Seasonal SARIMA(3,1,0)(2,1,0)[12] and SARIMA(0,1,1)(2,1,0)[12] models accurately modeled annual cycles.
Diagnostic tests (AIC, BIC, Ljung–Box) confirmed model adequacy and predictive accuracy.
Findings support regional climate adaptation strategies for water management and agriculture.
INTRODUCTION
Global climate change has emerged as a critical environmental issue due to its broad impacts on ecosystems, biodiversity, and human societies. Rising global temperatures, sea-level rise, increased extreme weather events, and alterations in natural systems illustrate the extensive consequences of climate change, primarily driven by human-generated greenhouse gas emissions, which trap heat and lead to global warming (Abbas et al. 2022). The Intergovernmental Panel on Climate Change (IPCC's) 2022 report underscores the substantial influence of human activity on climate, revealing that human-induced warming has affected all global regions and projects significant future shifts, including sea-level rise and intensified extreme weather events (Faranda et al. 2022). Observations indicate that climate change exacerbates extreme events, such as heavy rainfall and heat waves, posing challenges to environmental sustainability (Tabari 2020).
In addition to physical impacts, climate change affects biological systems intricately. Shifts in temperature and precipitation disrupt plant phenology, altering growth cycles and impacting ecosystems. Piao et al. (2019) highlight the need for adaptive strategies to maintain ecosystem sustainability in response to climate-driven biological changes. Temperature and precipitation, as key climate parameters, play pivotal roles in influencing agriculture, water resources, and ecosystems. Studies, such as Zhao et al. (2017), indicate that each degree of temperature increase can lead to significant yield losses in staple crops, emphasizing the need for adaptive strategies to sustain food security. Similarly, precipitation patterns, varying by region and season, are essential for maintaining water resources and ecosystem balance, as seen in the Central Chile Mega Drought (2010–2018), where prolonged rainfall deficits significantly impacted water availability and vegetation, heightening fire risks (Garreaud et al. 2019). Recent studies in Northern Canada have highlighted the importance of analyzing temperature and precipitation trends using robust statistical methods, such as the Mann–Kendall test, to evaluate the impacts of warming climates on seasonal water flow patterns and water resource availability (Zaghloul et al. 2022). Similarly, research on climate change and water resource management in urban settings has shown that rising temperatures significantly increase water demand, underscoring the broader implications of warming trends on water availability and management strategies (Seo et al. 2024).
The interplay between temperature and precipitation has prompted advanced research on climate dynamics. Models of sea surface temperature reconstruction, for example, offer improved accuracy in assessing spatial and temporal climate patterns, aiding localized adaptation and mitigation strategies essential for safeguarding agriculture and ecosystems (Huang et al. 2017). Accurate climate predictions of parameters like temperature and precipitation are critical for creating future climate scenarios that inform agricultural planning, water management, urban development, and environmental policy. Effective climate forecasting enables informed decision-making and proactive adaptation to climatic changes. For instance, drought prediction challenges in the Mediterranean highlight the importance of accurate forecasting for addressing water scarcity and ensuring agricultural resilience (Tramblay et al. 2020).
Recent advances in predictive models have increased the precision of temperature and precipitation forecasts. Time series methods applied to monthly temperature and precipitation data enhance predictability, supporting applications in water and agricultural management (Papacharalampous et al. 2018). Models focusing on atmospheric circulation, like the North Atlantic Oscillation, provide insights into seasonal temperature and precipitation variability, crucial for agriculture, energy, and transportation planning in regions such as the UK (Hall & Hanna 2018). Slingo et al. (2022) emphasize that interdisciplinary partnerships and advanced techniques are essential to improve climate prediction reliability, benefiting sectors reliant on consistent climate data.
Time series analysis is a cornerstone for predicting climate parameters, as it identifies trends and periodic patterns using historical data. Analyzing long-term climate datasets reveals short- and long-term trends, aiding in climate forecasts. For instance, Dimri et al. (2020) demonstrated that seasonal ARIMA models effectively capture periodic patterns in climate variables, offering reliable forecasts. Studies, such as Dad et al. (2021) also, show that time series analysis reveals critical seasonal and long-term trends in temperature and precipitation, essential for water resource management and agricultural planning.
ARIMA and SARIMA models, among time series methods, are widely used in climate forecasting due to their ability to model trends and seasonality. ARIMA effectively models non-seasonal data, while SARIMA is ideal for datasets with inherent seasonality, making it suitable for applications involving annual rainfall or temperature cycles (Dimri et al. 2020). SARIMA's ability to capture recurring seasonal climate patterns benefits agricultural and environmental planning. Studies, such as Peng et al. (2018), show that SARIMA models improve accuracy in crop yield predictions, especially in regions with seasonal climate shifts.
Recent research increasingly utilizes ARIMA and SARIMA for predicting climate variables like precipitation and temperature. Dimri et al. (2020) achieved accurate forecasts by applying a seasonal ARIMA model to climate variables, while Tarmanini et al. (2023) demonstrated the effectiveness of ARIMA models in capturing linear patterns and seasonal variations in energy forecasting. Additionally, studies by ArunKumar et al. (2022) indicate that ARIMA and SARIMA provide robust predictions in strongly seasonal datasets, showcasing their simplicity and reliability in climate data. Recent advancements in precipitation forecasting, such as the application of ARIMA, ARIMAX, SARIMAX, and ALLSSA models, have demonstrated significant potential in analyzing seasonal and trend components of climate data (Ghaderpour et al. 2023). These approaches have been particularly effective in capturing the variability and seasonality of precipitation, as demonstrated in Italian regions, where SARIMAX and ALLSSA performed robustly compared to ARIMA.
ARIMA and SARIMA models excel in capturing linear trends and seasonal patterns in climate forecasting, particularly for temperature and precipitation (Dimri et al. 2020). Their interpretability is another advantage, making predictions accessible to stakeholders who require transparency. However, these models rely on linear assumptions, which may limit their effectiveness in complex, non-linear climate data. For example, while SARIMA models seasonality well, it struggles with extreme weather events that exhibit non-linear interactions (ArunKumar et al. 2022). Consequently, researchers may combine SARIMA with more complex models when addressing non-linear processes.
Extensive literature supports the use of ARIMA and SARIMA models in climate forecasting, illustrating their effectiveness in predicting temperature, precipitation, and seasonal trends. Numerous studies, such as those by Achite et al. (2022) on drought forecasting in Mediterranean regions and Dimri et al. (2020) on climate variables, underscore these models' robust performance. Hybrid approaches, integrating machine learning with ARIMA/SARIMA, further enhance predictive accuracy for environmental and public health forecasts, including air pollutants (Luo & Gong 2023), solar radiation (Chodakowska et al. 2023), and COVID-19 trends (Ospina et al. 2023). This body of work reflects a commitment to refining climate forecasting for agriculture, water management, and public health (Balehegn et al. 2019; Ravuri et al. 2021; Wang et al. 2021; Calì Quaglia et al. 2022; He et al. 2022; Brunet et al. 2023; Jiang et al. 2023; Kontopoulou et al. 2023; Liu et al. 2023; Shamekh et al. 2023; Slater et al. 2023; Smith et al. 2023; Tarmanini et al. 2023; Zhang et al. 2023; Zhao et al. 2023).
The primary objective of this study is to forecast monthly precipitation and temperature data for Van Province from 1955 to 2023 using ARIMA and SARIMA models, assessing these models' ability to capture seasonal and long-term climate patterns specific to Van. This study contributes to the literature by using a long-term dataset, spanning nearly seven decades, which enhances forecasting reliability and supports regional planning in water resource management, agriculture, and infrastructure. This methodological approach offers insights into ARIMA and SARIMA model performance over extended periods, particularly in regions with unique climatic conditions.
By focusing on Van Province, this study provides a regional perspective on climate change, offering data-driven insights that can support localized climate policies and adaptation strategies. This localized approach is crucial for understanding the regional effects of climate change and emphasizes the importance of region-specific data in environmental planning and policymaking.
Main contributions of the study:
This study analyzes long-term rainfall and temperature trends in Van Province, Türkiye, using ARIMA and SARIMA models, covering the period 1955–2023.
A comparative evaluation of ARIMA and SARIMA models is conducted, demonstrating their respective performances in handling seasonal and non-seasonal climate data.
The findings provide critical insights into the impacts of temperature and precipitation trends on water resource management, particularly in semi-arid regions.
The study offers actionable recommendations for climate adaptation strategies, such as improved irrigation practices and seasonal water resource planning, based on forecasting results.
Study organization:
Section 2: Materials and Methods describes the data collection, preprocessing, and the ARIMA/SARIMA modeling approach used in the study.
Section 3: Results and Discussion presents the key findings, including the performance metrics of the models, the observed trends, and their implications for Van Province.
Section 4: Conclusion summarizes the main outcomes of the study and provides recommendations for future research and practical applications.
MATERIALS AND METHODS
Study area
Figure 1 provides a comprehensive view of the study area, including a continental map showing Türkiye with Van Province highlighted in blue, a provincial map of Türkiye pinpointing Van's location in the eastern region, and an elevation map displaying the topographical complexity of Van Province. The clear demarcation of provincial borders and the use of enhanced fonts ensure readability and clarity for the study's geographical context.
Van Province presents a unique study area for understanding climate, hydrology, and environmental health, influenced by its high-altitude, semi-arid climate and proximity to Lake Van, the largest soda lake in the world. The region experiences considerable seasonal and interannual variability in climate, as demonstrated by recent studies. For instance, Eşit et al. (2023) conducted a comprehensive trend analysis on rainfall and temperature records, revealing significant seasonal variations that have implications for water resource management. Furthermore, Aladağ (2023) examined how these meteorological patterns influence air quality, highlighting the dynamic interaction between climate factors and pollution levels. Additionally, Baumgarten & Wonik's (2015) cyclostratigraphic analysis of uranium content in Lake Van sediments provides paleoclimatic insights, shedding light on historical climate shifts in the region.
In terms of hydrology and water quality, Turan et al. (2022) analyzed the physicochemical parameters of Akköprü Stream, which flows into Lake Van, emphasizing seasonal impacts on water quality and the broader ecosystem. Similarly, Üner (2017) studied the sediment and water dynamics of the Çolpan barrier and lagoon complex, offering valuable data on sedimentological processes and hydrological flow.
The unique hydroclimatic and environmental factors of Van Province also present concerns regarding pollution and biodiversity. For instance, Atıcı et al. (2021) documented significant microplastic ingestion by the endemic fish species Alburnus tarichi, indicating pollution levels that may affect the aquatic ecosystem. Moreover, Kirmaci et al. (2022) surveyed bryophyte flora around Lake Van, identifying species adapted to the region's harsh conditions, which offers insights into the resilience and diversity of high-altitude, arid environments. Collectively, these studies underscore Van Province's significance as a research site for exploring the effects of climate, hydrology, and environmental pressures on local ecosystems.
Data sources
Monthly temperature and precipitation data spanning from 1955 to 2023 were obtained from the Turkish State Meteorological Service. These long-term datasets offer valuable insights into climate trends over nearly seven decades. Data processing involved a thorough check for any missing values or inconsistencies, which were addressed using interpolation techniques to ensure continuous time series data. This consistency in the dataset is essential for reliable model performance, as gaps in data could lead to inaccuracies in model predictions.
Table 1 provides a statistical summary of rainfall and temperature data for Van Province. The rainfall data shows substantial variability, with a coefficient of variation of 782.52%, indicating significant fluctuations across the observed period. In contrast, temperature data exhibits more stability, though still variable, with a coefficient of variation of 82.78%. The value of 782.52% reflects the high variability in rainfall data across the study period, indicative of significant seasonal and annual fluctuations. These statistics reflect the seasonal and climatic variability characteristic of Van Province, making it an ideal region for climate trend analysis.
Statistics of the measured rainfall and temperature data for Van Province
Parameters . | Unit . | Min. . | Max. . | Mean . | Standard deviation . | Coefficient of variation . |
---|---|---|---|---|---|---|
Rainfall | mm | 0 | 168.3 | 32.2 | 2,797 | 78,252 |
Temperature | °C | −10.2 | 25.3 | 9.5 | 9,098 | 82,782 |
Parameters . | Unit . | Min. . | Max. . | Mean . | Standard deviation . | Coefficient of variation . |
---|---|---|---|---|---|---|
Rainfall | mm | 0 | 168.3 | 32.2 | 2,797 | 78,252 |
Temperature | °C | −10.2 | 25.3 | 9.5 | 9,098 | 82,782 |
Time series analysis approach
Time series analysis provides insights into temporal trends, seasonal cycles, and patterns in climate data. ARIMA and SARIMA models were chosen for this study due to their proven efficacy in handling linear trends and seasonal variations in time series data. ARIMA is well-suited for non-seasonal data, whereas SARIMA is ideal for datasets with strong seasonal components, making it particularly relevant for analyzing monthly climate data with recurring annual patterns.
Model specification and selection
In this study, ARIMA and SARIMA models were utilized as the primary methods for analyzing rainfall and temperature data. The ARIMA model, which is suitable for handling non-seasonal data, was applied to assess underlying trends and patterns without a seasonal component. Conversely, the SARIMA model, an extension of ARIMA, was employed to capture seasonality, making it ideal for periodic data with recurring cycles. The use of both models allowed a comprehensive analysis of the data, addressing both linear trends and seasonal fluctuations, thus providing a more robust forecasting framework.
ARIMA modeling for rainfall and temperature data
The selection of ARIMA models for rainfall and temperature data was guided by Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) values, which provide a measure of model fit and complexity. For rainfall data, the optimal model identified was ARIMA(3,1,0), indicating an autoregressive (AR) model with three lags and first-order differencing to achieve stationarity. This configuration effectively captured the trend in the rainfall series. For temperature data, the best-fitting model was ARIMA(0,1,1), featuring first-order differencing and a single moving average (MA) term. Both models were verified through diagnostic checks, including the Ljung–Box test for autocorrelation and the Jarque–Bera test for normality of residuals, ensuring that each model adequately fits the respective dataset.
SARIMA modeling for seasonal trends
To account for seasonality in the rainfall and temperature data, SARIMA models were applied, which include additional parameters for seasonal autoregressive (SAR) and seasonal moving average (SMA) components. The optimal SARIMA model for rainfall was SARIMA(3,1,0)(2,1,0)[12], capturing both short-term dependencies and seasonal variations with a 12-month cycle. For temperature, the selected model was SARIMA(0,1,1)(2,1,0)[12], which similarly integrates a 12-month seasonal component, essential for accurately reflecting periodic temperature changes. These seasonal parameters allowed the SARIMA models to capture annual patterns effectively, providing a more nuanced representation of the data's seasonal behavior.
Through this dual approach, ARIMA and SARIMA models provided complementary insights into the structure of the rainfall and temperature data, with ARIMA capturing the core trends and SARIMA addressing the periodic seasonal components. This combined modeling approach enhances the accuracy and relevance of the forecasts for applications in climate analysis and planning.
Stationarity tests
To effectively apply ARIMA and SARIMA models, it is essential to ensure that the time series data exhibit stationarity. This means that the statistical properties of the series, such as its mean and variance, should remain consistent over time (Graf 2018; Katimon et al. 2018; Zhang & Xin 2019). Two statistical tests were employed to verify stationarity:
1. Augmented Dickey–Fuller (ADF) test: This test helps in identifying the presence of unit roots in the time series data, indicating non-stationarity. It is widely used in econometrics and time series analysis to ensure that the series does not exhibit a stochastic trend (Mushtaq 2011).







2. Kwiatkowski–Phillips–Schmidt–Shin (KPSS) test: The KPSS test complements the ADF test by evaluating trend stationarity, ensuring that the series does not display deterministic trends. It is particularly useful for verifying the results of the ADF test and offering a thorough assessment of the stationarity of the data (Kwiatkowski et al. 1992; Syczewska 2010; Baum 2018).
Model development
ARIMA model
– Identification: The order of the ARIMA model (p, d, q) was determined using the autocorrelation function (ACF) and partial autocorrelation function (PACF) plots.
– Estimation: Parameters were estimated using the maximum likelihood estimation method.
– Diagnostic checking: The residuals of the fitted model were analyzed to ensure no significant autocorrelation remained.




SARIMA model
– Seasonality: Seasonal components were identified based on the periodic patterns observed in the data.
– Parameter selection: Similar to the ARIMA model, ACF and PACF plots were used to determine the seasonal parameters (P, D, Q).
– Model fitting and validation: The model was fitted, and its accuracy was validated using out-of-sample data.
Model training and validation
The dataset was split into training and validation sets, with the data from 1955 to 2010 used for model training, and the data from 2011 to 2023 reserved for validation to ensure the models capture historical trends and test their predictive accuracy on unseen data. This approach ensures that the model captures historical trends and is then tested for its predictive power on unseen data. Cross-validation techniques, such as k-fold or time-based validation, were employed to evaluate the robustness of the models and mitigate potential overfitting.
Evaluation metrics
The evaluation metrics used to assess model performance include mean absolute error (MAE), root mean squared error (RMSE), mean absolute percentage error (MAPE), the Ljung–Box test, AIC, BIC, and the out-of-sample testing. These metrics collectively provide a comprehensive evaluation of model accuracy, reliability, and appropriateness for forecasting climate parameters. The formulas for these metrics are as follows:
Mean absolute error


Root mean squared error
Mean absolute percentage error

Ljung–Box test

Akaike information criterion
Bayesian information criterion
Out-of-sample testing
Out-of-sample testing is a technique used to evaluate the predictive capability of a model on unseen data. It involves splitting the dataset into a training set for model development and a testing set for validation (Hyndman & Athanasopoulos 2018). The accuracy of the model on the testing set is measured using metrics, such as MAE, RMSE, and MAPE, as defined above.
The procedure can be summarized as follows:
Divide the dataset into training and testing
sets.
Train the model on .
Validate predictions using evaluation metrics, such as MAE, RMSE, and MAPE.
This approach ensures that the model generalizes well to new, unseen data, avoiding overfitting to the training set. Out-of-sample testing is crucial for assessing the robustness and reliability of climate forecasts.
These metrics provide robust tools for evaluating the performance and reliability of the ARIMA and SARIMA models used in this study.
Software and tools
Python was used for data processing and model implementation, utilizing libraries such as statsmodels and scikit-learn for ARIMA and SARIMA modeling. Data visualization and exploratory analysis were performed with Matplotlib and Seaborn, while Pandas were employed for data manipulation. These tools provide a robust framework for handling time series analysis and forecasting in climate data, ensuring transparency and reproducibility in the study's methodology.
RESULTS
The analysis begins with an assessment of stationarity, a prerequisite for effective time series modeling, followed by model diagnostics to validate the chosen parameters and model accuracy. Finally, the forecasting results, including short-term and long-term predictions, are discussed to provide insights into seasonal and trend patterns in rainfall and temperature data over an extended period.
ARIMA model results
The ARIMA(3,1,0) model for rainfall showed statistically significant AR terms, particularly AR(1) and AR(2) with p-values < 0.05, indicating a strong fit for capturing trends in the rainfall data. For temperature, the ARIMA(0,1,1) model effectively captured the underlying variability with a significant MA term (p < 0.001). Diagnostic tests, including the Ljung–Box and Jarque–Bera tests, confirmed the suitability of these models, with no significant autocorrelation in residuals and normal distribution alignment, supporting the validity of these model configurations.
Table 2 provides a summary of key parameters, including AIC, BIC, AR, and MA terms for both the rainfall and temperature datasets. This table highlights the performance of the ARIMA models, making it easier for readers to understand their configuration and effectiveness in capturing data trends.
Summary of ARIMA model parameters for rainfall and temperature data
Model parameter . | Rainfall (ARIMA(3,1,0)) . | Temperature (ARIMA(0,1,1)) . |
---|---|---|
AIC | 788.224 | 172.077 |
BIC | 797.043 | 176.486 |
MAPE (%) | 9.85 | 8.45 |
RMSE | 15.24 | 1.21 |
Log Likelihood (LL) | −390.112 | −84.038 |
AR(1) | −0.635 (p < 0.001) | N/A |
AR(2) | −0.376 (p = 0.007) | N/A |
AR(3) | −0.177 (p = 0.161) | N/A |
MA(1) | N/A | −0.742 (p < 0.001) |
Sigma2 | 6,633.05 (p < 0.001) | 0.711 (p < 0.001) |
Ljung–Box (Prob(Q)) | 0.73 | 0.89 |
Jarque–Bera (Prob) | 0.50 | 0.51 |
Heteroskedasticity (Prob) | 0.34 | 0.10 |
Model parameter . | Rainfall (ARIMA(3,1,0)) . | Temperature (ARIMA(0,1,1)) . |
---|---|---|
AIC | 788.224 | 172.077 |
BIC | 797.043 | 176.486 |
MAPE (%) | 9.85 | 8.45 |
RMSE | 15.24 | 1.21 |
Log Likelihood (LL) | −390.112 | −84.038 |
AR(1) | −0.635 (p < 0.001) | N/A |
AR(2) | −0.376 (p = 0.007) | N/A |
AR(3) | −0.177 (p = 0.161) | N/A |
MA(1) | N/A | −0.742 (p < 0.001) |
Sigma2 | 6,633.05 (p < 0.001) | 0.711 (p < 0.001) |
Ljung–Box (Prob(Q)) | 0.73 | 0.89 |
Jarque–Bera (Prob) | 0.50 | 0.51 |
Heteroskedasticity (Prob) | 0.34 | 0.10 |
SARIMA model results
After evaluating the non-seasonal ARIMA models, SARIMA models were applied to account for seasonality in the rainfall and temperature data. The SARIMA(3,1,0)(2,1,0)[12] model was selected for rainfall, capturing the annual cycle with SAR terms, while SARIMA(0,1,1)(2,1,0)[12] was identified as the best fit for temperature, effectively modeling temperature's seasonal fluctuations. Seasonal terms in these models, particularly SAR(12) and SAR(24), were found to be statistically significant (p < 0.05), underscoring their importance in capturing recurring annual patterns.
Both models were assessed using the AIC, BIC, and residual diagnostics, and each passed the Ljung–Box and Jarque–Bera tests, indicating that the residuals were free from autocorrelation and aligned with a normal distribution. These results affirm the SARIMA models' suitability for forecasting seasonally affected data, providing greater forecasting accuracy in a periodic context.
Table 3 presents a summary of SARIMA model parameters, highlighting seasonal terms and performance metrics for both rainfall and temperature data. This table underscores the SARIMA models' ability to incorporate seasonal variations effectively, demonstrating the model fit and robustness for long-term forecasting applications.
Summary of SARIMA model parameters for rainfall and temperature data
Model parameter . | Rainfall (SARIMA(3,1,0)(2,1,0)[12]) . | Temperature (SARIMA(0,1,1)(2,1,0)[12]) . |
---|---|---|
AIC | 672.061 | 163.669 |
MAPE (%) | 6.78 | 5.92 |
RMSE | 12.35 | 0.95 |
AR(1) | −0.6628 (p < 0.001) | N/A |
AR(2) | −0.4337 (p < 0.05) | N/A |
AR(3) | −0.3246 (p < 0.05) | N/A |
MA(1) | N/A | −0.8699 (p < 0.001) |
SAR(12) | −0.7749 (p < 0.001) | −0.9492 (p < 0.001) |
SAR(24) | −0.3959 (p < 0.05) | −0.4389 (p < 0.01) |
Sigma2 | 8,114.0688 (p < 0.001) | 0.7452 (p < 0.001) |
Ljung–Box (Prob(Q)) | 0.55 | 0.78 |
LL | −330.031 | −76.835 |
Model parameter . | Rainfall (SARIMA(3,1,0)(2,1,0)[12]) . | Temperature (SARIMA(0,1,1)(2,1,0)[12]) . |
---|---|---|
AIC | 672.061 | 163.669 |
MAPE (%) | 6.78 | 5.92 |
RMSE | 12.35 | 0.95 |
AR(1) | −0.6628 (p < 0.001) | N/A |
AR(2) | −0.4337 (p < 0.05) | N/A |
AR(3) | −0.3246 (p < 0.05) | N/A |
MA(1) | N/A | −0.8699 (p < 0.001) |
SAR(12) | −0.7749 (p < 0.001) | −0.9492 (p < 0.001) |
SAR(24) | −0.3959 (p < 0.05) | −0.4389 (p < 0.01) |
Sigma2 | 8,114.0688 (p < 0.001) | 0.7452 (p < 0.001) |
Ljung–Box (Prob(Q)) | 0.55 | 0.78 |
LL | −330.031 | −76.835 |
Stationarity analysis of rainfall and temperature series
Rainfall series
Temperature series
Model diagnostics for rainfall and temperature forecasts
Partial autocorrelation and autocorrelation analysis
ARIMA model residual diagnostics
Forecasting results
Rainfall forecasting:
Temperature forecasting:
Long-term forecast for rainfall and temperature
The long-term forecasts for rainfall and temperature in Van Province provide valuable insights into potential future climate trends. The results indicate a general stability in rainfall over the period from 2024 to 2030, while temperature forecasts show a steady increase during the same timeframe.
Rainfall trends
The rainfall forecast suggests a stable pattern with minor seasonal fluctuations, which can be attributed to regional influences such as orographic effects and semi-arid climatic conditions. These findings align with previous studies conducted in similar semi-arid regions, which reported minimal long-term rainfall variations due to the influence of the Eastern Anatolian highlands.
Temperature trends
The temperature projections highlight a consistent rising trend, potentially influenced by global warming and changes in regional atmospheric circulation patterns. The steady increase in temperature supports broader observations reported in the IPCC assessments, particularly for high-altitude regions in the Middle East and Eastern Europe.
Comparison with observed data
The projected trends for both rainfall and temperature are consistent with historical data trends observed between 1955 and 2023, demonstrating the reliability of the ARIMA/SARIMA models used in this study. The models effectively capture the historical stability in rainfall and the gradual warming trend, providing confidence in the long-term forecasts.
DISCUSSION
The findings from this study provide a detailed assessment of rainfall and temperature trends in Van Province using ARIMA and SARIMA models. The effectiveness of these models in capturing seasonality and long-term trends is in line with previous studies that have applied similar methodologies to analyze climate variables in other regions. The projected rainfall trends suggest relatively stable water availability, reducing the likelihood of severe droughts but requiring close monitoring for potential seasonal extremes. These insights are critical for water resource management and agricultural sustainability in Van Province. Rising temperatures may have significant implications for agriculture and energy demand, particularly as higher temperatures could increase evapotranspiration rates, potentially exacerbating water scarcity during the dry season despite stable rainfall patterns.
The trends observed in this study align with findings from Northern Canada, where rising temperatures significantly influenced seasonal water flow patterns, highlighting the critical role of climatic factors in hydrological dynamics (Zaghloul et al. 2022). Similarly, studies on urban water demand have shown that warming trends lead to increased water usage, emphasizing the broader implications of temperature increases on water resource management (Seo et al. 2024). While the context of urban settings differs from semi-arid rural regions like Van Province, the underlying climate-driven challenges remain consistent. The projected stability in rainfall and the rising temperature trends in Van Province may similarly affect water availability, necessitating integrated strategies to manage seasonal extremes and potential shifts in water allocation for agricultural and hydrological systems.
Model performance in forecasting rainfall and temperature
The use of ARIMA and SARIMA models in this study aligns with Dimri et al. (2020), who demonstrated the capability of seasonal ARIMA models to capture periodic patterns in climate data, particularly for variables like temperature and precipitation. The study's results, which show improved stationarity and reduced autocorrelation after first- and second-order differencing, reflect the model's ability to handle seasonal variations effectively. Similar to the findings by Dad et al. (2021), who applied time series analysis to reveal critical seasonal and long-term trends in temperature and precipitation, this study highlights the model's suitability for applications requiring seasonal and trend analysis.
However, as observed in prior research, ARIMA and SARIMA models have limitations in handling non-linear complexities in climate data. ArunKumar et al. (2022) noted that while these models perform well in capturing linear patterns and seasonality, they may struggle with abrupt changes or extreme events. This limitation was also evident in this study, where minor deviations in peak rainfall predictions were observed, underscoring the model's sensitivity to extreme weather events that exhibit non-linear characteristics.
Comparison with advanced predictive models
While ARIMA and SARIMA are widely used for climate forecasting due to their interpretability and straightforward application, recent studies advocate for hybrid approaches that integrate machine learning models to improve predictive accuracy, especially in non-linear datasets. For instance, Kontopoulou et al. (2023) emphasized the benefits of combining ARIMA with machine learning techniques for more robust forecasts in complex climate scenarios. Although such hybrid models were beyond the scope of this study, incorporating them in future work could enhance the accuracy of rainfall and temperature predictions, especially for regions experiencing significant climate variability. The findings of this study align with prior research utilizing SARIMAX and ALLSSA for monthly precipitation forecasting, which highlighted the importance of incorporating seasonal effects and advanced trend-analysis techniques (Ghaderpour et al. 2023). While this study focused on ARIMA and SARIMA models, the results underscore the value of integrating seasonality to improve the accuracy of precipitation and temperature predictions. The robust performance of SARIMA in this study parallels the superior results achieved by SARIMAX in capturing seasonal patterns in Italian regions, as reported by Ghaderpour et al. (2023).
Implications for water resource management and agriculture
This long-term forecast for rainfall and temperature offers valuable insights into potential future climate trends in Van Province, which can support decision-making in water resource management, agriculture, and environmental policy. The predictive accuracy demonstrated in this study is crucial for regional planning, as accurate rainfall and temperature forecasts can inform adaptive strategies in water management, such as those explored by Tramblay et al. (2020) for drought assessment in Mediterranean regions. Effective climate forecasting, as discussed by Slingo et al. (2022), requires interdisciplinary collaboration and advanced methodologies to support sectors reliant on consistent climate data.
Limitations and future research directions
While ARIMA and SARIMA models provided reliable forecasts for this study, limitations such as the models' reliance on linear assumptions and challenges with extreme event prediction suggest that future research could benefit from hybrid modeling approaches. Studies like those by Zhang et al. (2023) have shown the potential of combining SARIMA with machine learning models for enhanced accuracy, particularly in addressing extreme weather patterns. Incorporating non-linear models, such as Long Short-Term Memory (LSTM) or Gated Recurrent Unit (GRU), as suggested by Luo & Gong (2023), could address the limitations observed in peak rainfall forecasts, thereby improving model resilience against extreme climate variations.
Moreover, expanding the dataset to include other climate variables, such as wind speed or solar radiation, could offer a more comprehensive view of Van Province's climate dynamics. By exploring these additional parameters, future research could provide a more holistic assessment of climate variability, similar to studies by Huang et al. (2017) on sea surface temperature and its influence on regional climate patterns.
CONCLUSION
This study applied ARIMA and SARIMA models to analyze and forecast monthly rainfall and temperature data in Van Province, providing valuable insights into seasonal and long-term climate patterns. The models effectively captured trend and seasonality in the time series data, with first- and second-order differencing improving stationarity and enhancing model accuracy. The high predictive accuracy observed in temperature and rainfall forecasts confirms the utility of ARIMA and SARIMA for climate forecasting, particularly in regions with significant seasonal variations.
The findings demonstrate that ARIMA and SARIMA models can support regional planning efforts in water resource management and agriculture by providing reliable climate forecasts. This predictive capability is essential for developing adaptive strategies in response to the challenges posed by climate change, such as water scarcity and agricultural resilience. Although minor deviations were noted in peak rainfall predictions, the models' performance is consistent with prior research, highlighting their suitability for capturing linear trends and seasonal patterns in climate data.
Despite the robustness of these models, limitations such as sensitivity to non-linear events suggest that future research could benefit from hybrid approaches that integrate machine learning models. Such advancements could enhance forecast accuracy, particularly for extreme weather events, which are becoming more frequent due to climate change. Expanding the analysis to include other climate variables, as well as exploring non-linear models, could offer a more comprehensive understanding of climate dynamics in Van Province.
The findings emphasize the need for adaptive strategies to mitigate climate risks: Agriculture: invest in drought-resistant crops and improve irrigation efficiency to adapt to rising temperatures. Urban planning: implement urban heat management strategies, such as increasing green spaces and improving building insulation. Monitoring: establish climate monitoring systems to detect seasonal fluctuations and extreme events.
While ARIMA and SARIMA models demonstrated high accuracy in capturing linear and seasonal patterns in rainfall and temperature, their inherent reliance on linear assumptions may not fully account for non-linear dynamics or extreme climate events. Future studies should explore hybrid modeling approaches, such as integrating machine learning techniques with ARIMA/SARIMA, to improve forecast reliability under non-linear conditions. Furthermore, incorporating additional climate variables, such as wind speed or solar radiation, could provide a more comprehensive understanding of regional climatic variations. Despite these limitations, the models applied in this study have proven effective for long-term climate forecasting in semi-arid regions, offering actionable insights for water resource management and agricultural planning in Van Province.
Overall, this study contributes to the literature by providing a long-term climate forecast for Van province, supporting localized adaptation and mitigation strategies. The results underscore the importance of region-specific climate analysis for informed policymaking and sustainable resource management in the face of evolving climate challenges.
FUNDING
This research received no external funding
ACKNOWLEDGEMENTS
The author would like to thank the Turkish State Meteorological Service for providing the water quality data used in this study. This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.
DATA AVAILABILITY STATEMENT
Data cannot be made publicly available; readers should contact the corresponding author for details.
CONFLICT OF INTEREST
The authors declare there is no conflict.