ABSTRACT
The present research examines the impact of climate change on meteorological parameters using global climate models (GCMs) and artificial intelligence, with a case study in Fars Province, Iran. In this study, the meteorological parameters of minimum temperature, maximum temperature, precipitation, and solar radiation, as well as 12 GCMs from the Fifth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC), are utilized. A statistical downscaling model (multilayer perceptron neural network) is employed to extract climate change predictions under two scenarios, representative concentration pathway (RCP)4.5 and RCP8.5, for four synoptic stations (Shiraz, Abade, Fasa, and Lar), each representing different climatic regions. Correlation analysis is used to identify the most influential predictor variables for each meteorological parameter. The results indicate a projected increase in maximum temperature by up to 2.67 °C, which could significantly impact agricultural productivity in Fars Province. This finding is accompanied by minimum temperature ranges from 0.23 to 2.71 °C, solar radiation increasing by up to 1.91 MJ/m², and precipitation fluctuation between a decrease of 7% and an increase of 36.5%. These findings suggest that the region may face increased agricultural stress due to higher temperatures and variable precipitation patterns, necessitating adaptive strategies for sustainable water resource management.
HIGHLIGHTS
Evaluating climate change impact in Fars Province by 12 global climate models (GCMs).
Downscaling climate variables by using artificial neural networks (ANNs).
CanESM2, HadGEM2-CC, ACCESS1.3, and HadGEM2-ES were the best-performance models.
Increase in temperature and solar radiation, decrease precipitation in the future.
INTRODUCTION
The increasing reliance on fossil fuels, pollutant production, and greenhouse gas emissions serve as a primary driver of global warming and climate change. Scientists worldwide agree on the impact of climate change in the past and present, and its continuation in the future, although the extent and direction of change are highly uncertain (Libanda & Nkolola 2019). Climate change, manifested in the warming of the Earth's temperature, has become one of the most destructive environmental crises and a barrier to global sustainable development, including social and economic activities (IPCC 2018).
Numerous studies have been conducted on the effects of climate change on weather parameters in different parts of the world, including Iran, highlighting the importance and need for continued study of this phenomenon (Costa et al. 2023; Gholami et al. 2023; Xiao et al. 2024). It is evident that climate change has significant impacts on regions with more vulnerable climates, such as southern Iran where rainfall is predominant in the cold season (Bahrami & Mahmoudi 2022). Consequently, identifying and predicting these impacts can help planners prepare for upcoming challenges and assist them in strategic planning. Climatologists employ various methodologies to simulate and predict climate variables (Chim et al. 2021). Among these, general circulation models have gained prominence due to their ability to represent atmospheric processes at large scales. However, GCMs often operate at spatial and temporal resolutions that may not be suitable for smaller regions (Keller et al. 2022), necessitating additional techniques such as dynamical downscaling (e.g., RegCM4 (Valcheva et al. 2024) and WRF (Maichandee et al. 2024)) and statistical downscaling methods (e.g., LARS-WG (Ostad-Ali-Askari et al. 2020; Kavwenje et al. 2022) and Statistical DownScaling Model (SDSM) (Rakhimova et al. 2020)). Recently, artificial intelligence methods have emerged as innovative tools that enhance predictions related to climate change, with multilayer perceptron artificial neural networks (MLP ANNs) demonstrating high predictive capabilities in environmental modeling (Ansari et al. 2015; Bahraminejad et al. 2018; Golkar Hamzee yazd et al. 2019; Bahrami & Mahmoudi 2020).
Research indicates that ANNs can yield more accurate long-term predictions compared with traditional models by effectively capturing complex nonlinear relationships (Liu et al. 2010). For example, Babel et al. (2017) reported a decline in precipitation from 2011 to 2040 in a Thai watershed using ANN methodologies. Similarly, Samuels et al. (2018) utilized GCMs to forecast a significant reduction in total precipitation in the eastern Mediterranean region. These findings highlight the necessity of employing advanced modeling techniques to understand climate dynamics better. Zhu et al. (2019) anticipated climate change impacts on temperature and precipitation in two Chinese wetlands by 2100 using GCMs and found that temperature and precipitation variations will be greater in the future. He et al. (2019) utilized seven GCMs from CMIP5 to project climate conditions in various regions of China until 2100. Findings demonstrated variations in the GCMs' ability to simulate climate conditions. For the United Arab Emirates, an increase in future temperature of up to 2.5 °C for 2021–2050 and up to 4.19 °C for 2051–2080 is anticipated if ANN is used based on the representative concentration pathway (RCP)4.5 and RCP8.5 scenarios (Ashour et al. 2022). Additionally, under RCP4.5, RCP8.5, and RCP2.6 scenarios in Cambodia, Chim et al. (2021) forecasted that changes in temperature and precipitation would result in future droughts. In Iraq, Mohammed & Hassan (2022) analyzed climate change effects on temperature and precipitation, projecting a temperature increase of 5.5–5.91 °C under the RCP8.5 scenario and 1.4–1.5 °C under the RCP4.5 scenario for the years 2021–2100.
In this study, we focus on Fars Province in southern Iran – a region particularly vulnerable to climate change due to its dry and semi-arid conditions. If unaddressed, these climatic shifts could severely impact water resources and ecosystems within the province. Therefore, this research innovatively integrates GCMs with ANN-based methods to improve the accuracy of weather parameter predictions and climate change impact assessments through enhanced downscaling. It also introduces customized algorithms for climatic data and focuses on practical applications in sectors such as water resource management and agriculture. By utilizing state-of-the-art GCMs alongside ANN-based methods, this study aims to assess the impacts of climate change on these vital meteorological parameters in Fars Province. Expected results will not only enhance scientific knowledge but also provide practical solutions for addressing the adverse effects of climate change in vulnerable regions like southern Iran. By focusing on both theoretical advancements and practical applications, this study highlights its relevance in light of increasing concerns about global climate variability and its impact on sustainable development.
MATERIALS AND METHODS
Study area
Geographical location of the studied stations: Abade (31°12′N, 52°37′E), Shiraz (29°33′N, 52°36′E), Fasa (28°54′N, 53°43′E), and Lar (27°40′N, 54°22′E).
Geographical location of the studied stations: Abade (31°12′N, 52°37′E), Shiraz (29°33′N, 52°36′E), Fasa (28°54′N, 53°43′E), and Lar (27°40′N, 54°22′E).
Data collection
Data on precipitation and surface temperature from Fars Provincial Meteorological Organization for 1986–2005 were used for this study. Additionally, monthly solar radiation data for the same period were downloaded from the NASA website (https://www.nasa.gov). To achieve the study's objective, 12 different GCMs published in the Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report by the Earth System Grid Federation at Lawrence Livermore National Laboratory (https://esgf-node.llnl.gov/search/cmip5) were used. GCMs provided minimum and maximum temperature, precipitation, and solar radiation in NETCDF format (Table 1).
Characteristics of applied GCMs in current research
Model . | Spatial resolution (Geog. Deg.) . |
---|---|
CanESM2 | 2.81 × 2.79 |
Miroce-ESM-CHEM | 2.8 × 2.8 |
Miroce-ESM | 2.8 × 2.8 |
HaddGem2-ES | 1.9 × 1.25 |
NorESM1-ME | 1.87 × 2.5 |
GIIS-E2-H-CC | 2.5 × 2 |
ACCESS1.3 | 1.25 × 1.9 |
HadGem2-CC | 1.25 × 1.9 |
CNRM-CM5 | 1.41 × 1.4 |
GISS-E2-H | 2.5 × 2 |
FGOALS-S2 | 2.81 × 2.79 |
GISS-E2-R-CC | 2.5 × 2 |
Model . | Spatial resolution (Geog. Deg.) . |
---|---|
CanESM2 | 2.81 × 2.79 |
Miroce-ESM-CHEM | 2.8 × 2.8 |
Miroce-ESM | 2.8 × 2.8 |
HaddGem2-ES | 1.9 × 1.25 |
NorESM1-ME | 1.87 × 2.5 |
GIIS-E2-H-CC | 2.5 × 2 |
ACCESS1.3 | 1.25 × 1.9 |
HadGem2-CC | 1.25 × 1.9 |
CNRM-CM5 | 1.41 × 1.4 |
GISS-E2-H | 2.5 × 2 |
FGOALS-S2 | 2.81 × 2.79 |
GISS-E2-R-CC | 2.5 × 2 |
This study investigated the most significant climatic variables as predictors for each of the IPCC models. The researchers have explored various situations using different prediction scenarios for the state of the economy, society, technology, and environment. RCP scenarios describe four possible climatic states for future climate.
RCP scenarios and GCMs
RCPs are trajectory scenarios that were developed in 2010 by a scientific committee working under the direction of the Inter-State Climate Change Commission. The primary goal of these scenarios was to provide a set of data that could be used to identify the main causes of climate change and to improve climate models (Sun et al. 2022). The outcomes of these scenarios are used by climate models to display greenhouse gas concentration and emission, pollution levels, and changes in land use. The RCP family includes four scenarios: 8.5, 4.5, 2.6, and 6 (Table 2). These scenarios are based on various factors such as the state of technology, economy, society, and policies that may be in place in the future. Depending on the circumstances, these scenarios may result in different levels of greenhouse gas emissions and climate change.
Four scenarios in the RCP family: 8.5, 4.5, 2.6, and 6
Scenario . | Description . | References . |
---|---|---|
RCP2.6 | The IMAGE modeling team at the Netherlands Environmental Assessment Agency created this hypothetical situation. This scenario predicts that by the middle of the century, radiative forcing will have increased to around 1.3 kW/m2, and that it will finally reach 2.6 kW/m2 in 2100. | |
RCP4.5 | According to the RCP4.5 scenario, greenhouse gas emissions will have stabilized by the middle of the twenty-first century, increasing solar radiation intensity by 4.5 kW/m2 by the year 2100. | |
RCP6 | The AIM modeling team of Japan's National Institute for Environmental Studies created this scenario. This scenario assumes that the introduction of new technologies and methods for reducing greenhouse gas emissions causes post-2020 radiative force to remain constant. | |
RCP8.5 | According to solar radiation intensity scenario RCP8.5, this level will climb to 8.5 kW/m2/m2 by 2100 along with a rise in CO2 concentration to 1,370 ppm. It is important to note that the MESSAGE modeling group created this scenario. | Hausfather & Peters (2020), Schwalm et al. (2020) |
Scenario . | Description . | References . |
---|---|---|
RCP2.6 | The IMAGE modeling team at the Netherlands Environmental Assessment Agency created this hypothetical situation. This scenario predicts that by the middle of the century, radiative forcing will have increased to around 1.3 kW/m2, and that it will finally reach 2.6 kW/m2 in 2100. | |
RCP4.5 | According to the RCP4.5 scenario, greenhouse gas emissions will have stabilized by the middle of the twenty-first century, increasing solar radiation intensity by 4.5 kW/m2 by the year 2100. | |
RCP6 | The AIM modeling team of Japan's National Institute for Environmental Studies created this scenario. This scenario assumes that the introduction of new technologies and methods for reducing greenhouse gas emissions causes post-2020 radiative force to remain constant. | |
RCP8.5 | According to solar radiation intensity scenario RCP8.5, this level will climb to 8.5 kW/m2/m2 by 2100 along with a rise in CO2 concentration to 1,370 ppm. It is important to note that the MESSAGE modeling group created this scenario. | Hausfather & Peters (2020), Schwalm et al. (2020) |
In these equations, is the observed values,
is the predicted data,
is the average of the predicted data,
is the average of observed data, and n is the number of data used (Babel et al. 2017; Piraei et al. 2023).
Artificial neural network
ANNs have become increasingly popular for nonlinear modeling in recent years. The ANN model processes information in parallel and does not rely on complex mathematical formulas to establish the relationship between input and output data. It is a data processing system that imitates the human brain and is made up of input, hidden, and output layers. The input layer generates and transmits data, while the output layer provides predicted values of the model and defines the model's output. The hidden layer includes processing nodes where data processing occurs. When training a neural network, the number of neurons in hidden and input layers is determined by trial and error. During this process, the best-performing network is one with a high r-value and low values of MSE, RMSE, and MAE (Mulyadi et al. 2024; Olasehinde et al. 2024). To achieve the desired accuracy, the training of the network should be stopped after a few iterations. The number of iterations is optional, but it should be chosen to reduce network error. When modeling time series data, there are two categories of data: calibration and validation. To achieve better results, it is recommended to normalize data.
Regional simulation study
This section outlines the methodology employed to assess the impacts of climate change on meteorological variables in Fars Province, utilizing both GCMs and ANNs. The integration of these approaches allows for a more nuanced understanding of local climate dynamics and enhances the predictive accuracy of climate models.
Selection of GCMs
The study utilized 12 GCMs from the IPCC Fifth Assessment Report, selected based on their performance in simulating historical climate data and their relevance to the regional context.
Downscaling methodology
To refine projections from GCMs, which operate at coarse spatial resolutions, statistical downscaling using ANNs is employed. This study first utilized statistical downscaling based on ANNs to downscale data from GCMs (Reboita et al. 2022). The MLP method was implemented in the MATLAB software environment to select the most effective GCMs. A training dataset comprising 70% of historical data was used, while the remaining 30% was reserved for testing (Qing et al. 2021). The number of hidden layers and neurons within these layers were determined through trial and error, ensuring sufficient training by maintaining a minimum error slope for weights at e − 100. After downscaling with ANN, two models were selected as best for predicting minimum temperature, maximum temperature, precipitation, and solar radiation based on common evaluation criteria. These criteria included a high correlation coefficient (r-value) and low MSE, RMSE, and MAE, assessed over a total of 500 iteration sets.
Predictor variable identification
To enhance predictive capabilities, a correlation analysis was conducted to identify significant predictor variables affecting precipitation, maximum temperature, minimum temperature, and solar radiation across all four stations from 1986 to 2005. Using Statistical Package for Social Science (SPSS), we calculated correlation values between observational data and GCM outputs. Variables with the highest correlation were selected as influential inputs for ANN models (MLP model). This analysis ensures that ANN models are informed by relevant atmospheric conditions influencing local weather patterns. The performance of ANN models is evaluated using statistical metrics such as RMSE, MAE, and R-squared values (R²). These metrics provide a comprehensive assessment of how well the models replicate observed historical data and their ability to predict future climate scenarios accurately.
Future climate projections
RESULTS AND DISCUSSION
Evaluation of optimal GCMs and their predictive variables
Evaluation statistics calculations between downscaled GCMs and observational data of synoptic stations to select the best model are presented in Table 3 (for Shiraz) and Supplementary Tables S1–S3 (for other stations). Based on the evaluation results of the indicators, CanESM2 and HadGem2-CC models have proven superior for the Shiraz station. Similarly, HadGem2-CC and HaddGem2-ES models are better for the Fasa station. For the Abade and Lar stations, ACCESS1.3 and HaddGem2-ES models have shown better results.
Performance results of GCMs in the Shiraz synoptic station
Model . | Statistics criteria . | Solar radiation . | Precipitation . | Minimum temperature . | Maximum temperature . |
---|---|---|---|---|---|
CanESM2 | MSE (TEST) | 0.02 | 0.005 | 0.003 | 0.003 |
RMSE | 0.16 | 0.07 | 0.05 | 0.06 | |
MAE (TEST) | 0.12 | 0.05 | 0.04 | 0.04 | |
r | 0.14 | 0.97 | 0.98 | 0.97 | |
Miroce-ESM-CHEM | MSE (TEST) | 0.02 | 0.008 | 0.005 | 0.006 |
RMSE | 0.15 | 0.09 | 0.07 | 0.08 | |
MAE (TEST) | 0.11 | 0/06 | 0.05 | 0.06 | |
r | 0.47 | 0.95 | 0.96 | 0.97 | |
Miroce-ESM | MSE (TEST) | 0.03 | 0.007 | 0.004 | 0.006 |
RMSE | 0.17 | 0.08 | 0.06 | 0.07 | |
MAE (TEST) | 0.14 | 0.06 | 0.04 | 0.05 | |
r | 0.35 | 0.91 | 0.98 | 0.96 | |
HaddGem2-ES | MSE (TEST) | 0.01 | 0.006 | 0.004 | 0.003 |
RMSE | 0.13 | 0.07 | 0.06 | 0.06 | |
MAE (TEST) | 0.09 | 0.05 | 0.05 | 0.05 | |
r | 0.68 | 0.95 | 0.97 | 0.98 | |
NorESM1-ME | MSE (TEST) | 0.02 | 0.01 | 0.006 | 0.005 |
RMSE | 0.15 | 0.12 | 0.07 | 0.07 | |
MAE (TEST) | 0.11 | 0.09 | 0.06 | 0.05 | |
r | 0.55 | 0.95 | 0.95 | 0.96 | |
GIIS-E2-H-CC | MSE (TEST) | 0.02 | 0.01 | 0.006 | 0.04 |
RMSE | 0.16 | 0.12 | 0.07 | 0.22 | |
MAE (TEST) | 0.08 | 0.09 | 0.06 | 0.17 | |
r | 0.48 | 0.91 | 0.93 | 0.7 | |
ACCESS1.3 | MSE (TEST) | 0.02 | 0.006 | 0.004 | 0.004 |
RMSE | 0.16 | 0.08 | 0.06 | 0.06 | |
MAE (TEST) | 0.11 | 0.06 | 0.048 | 0.05 | |
r | 0.5 | 0.96 | 0.98 | 0.98 | |
HadGem2-CC | MSE (TEST) | 0.01 | 0.00 | 0.004 | 0.004 |
RMSE | 0.13 | 0.07 | 0.06 | 0.06 | |
MAE (TEST) | 0.09 | 0.05 | 0.04 | 0.05 | |
r | 0.62 | 0.97 | 0.97 | 0.97 | |
CNRM-CM5 | MSE (TEST) | 0.01 | 0.01 | 0.004 | 0.008 |
RMSE | 0.14 | 0.10 | 0.06 | 0.09 | |
MAE (TEST) | 0.09 | 0.08 | 0/05 | 0.06 | |
r | 0.82 | 0.96 | 0/95 | 0.97 | |
GISS-E2-H | MSE (TEST) | 0.02 | 0.08 | 0.005 | 0.005 |
RMSE | 0.15 | 0.29 | 0.07 | 0.07 | |
MAE (TEST) | 0.09 | 0.08 | 0.05 | 0.05 | |
r | 0.43 | 0.93 | 0.93 | 0.97 | |
FGOALS-S2 | MSE (TEST) | 0.04 | 0.01 | 0.003 | 0.003 |
RMSE | 0.22 | 0.12 | 0.05 | 0.06 | |
MAE (TEST) | 0.13 | 0.07 | 0.04 | 0.04 | |
r | 0.45 | 0.96 | 0.98 | 0.964 | |
GISS-E2-R-CC | MSE (TEST) | 0.02 | 0.01 | 0.006 | 0.006 |
RMSE | 0.14 | 0.12 | 0.07 | 0.07 | |
MAE (TEST) | 0.11 | 0.09 | 0.06 | 0.06 | |
r | 0.58 | 0.92 | 0.96 | 0.96 |
Model . | Statistics criteria . | Solar radiation . | Precipitation . | Minimum temperature . | Maximum temperature . |
---|---|---|---|---|---|
CanESM2 | MSE (TEST) | 0.02 | 0.005 | 0.003 | 0.003 |
RMSE | 0.16 | 0.07 | 0.05 | 0.06 | |
MAE (TEST) | 0.12 | 0.05 | 0.04 | 0.04 | |
r | 0.14 | 0.97 | 0.98 | 0.97 | |
Miroce-ESM-CHEM | MSE (TEST) | 0.02 | 0.008 | 0.005 | 0.006 |
RMSE | 0.15 | 0.09 | 0.07 | 0.08 | |
MAE (TEST) | 0.11 | 0/06 | 0.05 | 0.06 | |
r | 0.47 | 0.95 | 0.96 | 0.97 | |
Miroce-ESM | MSE (TEST) | 0.03 | 0.007 | 0.004 | 0.006 |
RMSE | 0.17 | 0.08 | 0.06 | 0.07 | |
MAE (TEST) | 0.14 | 0.06 | 0.04 | 0.05 | |
r | 0.35 | 0.91 | 0.98 | 0.96 | |
HaddGem2-ES | MSE (TEST) | 0.01 | 0.006 | 0.004 | 0.003 |
RMSE | 0.13 | 0.07 | 0.06 | 0.06 | |
MAE (TEST) | 0.09 | 0.05 | 0.05 | 0.05 | |
r | 0.68 | 0.95 | 0.97 | 0.98 | |
NorESM1-ME | MSE (TEST) | 0.02 | 0.01 | 0.006 | 0.005 |
RMSE | 0.15 | 0.12 | 0.07 | 0.07 | |
MAE (TEST) | 0.11 | 0.09 | 0.06 | 0.05 | |
r | 0.55 | 0.95 | 0.95 | 0.96 | |
GIIS-E2-H-CC | MSE (TEST) | 0.02 | 0.01 | 0.006 | 0.04 |
RMSE | 0.16 | 0.12 | 0.07 | 0.22 | |
MAE (TEST) | 0.08 | 0.09 | 0.06 | 0.17 | |
r | 0.48 | 0.91 | 0.93 | 0.7 | |
ACCESS1.3 | MSE (TEST) | 0.02 | 0.006 | 0.004 | 0.004 |
RMSE | 0.16 | 0.08 | 0.06 | 0.06 | |
MAE (TEST) | 0.11 | 0.06 | 0.048 | 0.05 | |
r | 0.5 | 0.96 | 0.98 | 0.98 | |
HadGem2-CC | MSE (TEST) | 0.01 | 0.00 | 0.004 | 0.004 |
RMSE | 0.13 | 0.07 | 0.06 | 0.06 | |
MAE (TEST) | 0.09 | 0.05 | 0.04 | 0.05 | |
r | 0.62 | 0.97 | 0.97 | 0.97 | |
CNRM-CM5 | MSE (TEST) | 0.01 | 0.01 | 0.004 | 0.008 |
RMSE | 0.14 | 0.10 | 0.06 | 0.09 | |
MAE (TEST) | 0.09 | 0.08 | 0/05 | 0.06 | |
r | 0.82 | 0.96 | 0/95 | 0.97 | |
GISS-E2-H | MSE (TEST) | 0.02 | 0.08 | 0.005 | 0.005 |
RMSE | 0.15 | 0.29 | 0.07 | 0.07 | |
MAE (TEST) | 0.09 | 0.08 | 0.05 | 0.05 | |
r | 0.43 | 0.93 | 0.93 | 0.97 | |
FGOALS-S2 | MSE (TEST) | 0.04 | 0.01 | 0.003 | 0.003 |
RMSE | 0.22 | 0.12 | 0.05 | 0.06 | |
MAE (TEST) | 0.13 | 0.07 | 0.04 | 0.04 | |
r | 0.45 | 0.96 | 0.98 | 0.964 | |
GISS-E2-R-CC | MSE (TEST) | 0.02 | 0.01 | 0.006 | 0.006 |
RMSE | 0.14 | 0.12 | 0.07 | 0.07 | |
MAE (TEST) | 0.11 | 0.09 | 0.06 | 0.06 | |
r | 0.58 | 0.92 | 0.96 | 0.96 |
Tables containing the correlation test results are presented to determine the most effective weather variables on precipitation, solar radiation, minimum temperature, and maximum temperature for the best GCMs obtained. These tables include Tables 4 and 5 for the Shiraz station, and Supplementary Tables S4–S9 for other studied stations. Significant correlation values between weather variables and studied parameters at the 5% level are highlighted in the tables. For instance, based on results obtained from the Shiraz synoptic station for the HadGEM2-CC model, certain variables were found to significantly affect the maximum temperature parameter. These variables include surface upwelling longwave radiation (sulr), TOA incident shortwave radiation (tisr), TOA outgoing shortwave radiation (tosr), near-surface wind speed (sfcwind), surface downward eastward wind stress (sdew), tmax, tmin, and surface downwelling shortwave radiation (sdsr) were the most positive significant parameters. On the other hand, the variable near-surface relative humidity (hurs) showed a significant negative influence.
Correlation coefficient between predictor variables and minimum and maximum temperature, solar radiation, and precipitation for the HadGEM2-CC model of the Shiraz station
Predictor . | Maximum temperature . | Minimum temperature . | Solar radiation . | Precipitation . |
---|---|---|---|---|
evspsbl | −0.20 | −0.18 | 0.26 | 0.16 |
hurs | −0.70 | −0.66 | −0.16 | 0.49 |
huss | −0.11 | −0.14 | −0.47 | 0.06 |
ps | −0.62 | −0.57 | −0.08 | 0.47 |
psl | −0.61 | −0.56 | −0.024 | 0.45 |
rlds | 0.49 | 0.45 | −0.11 | −0.34 |
sulr | 0.63 | 0.59 | 0.01 | −0.43 |
tisr | 0.86 | 0.84 | 0.37 | −0.58 |
rlut | 0.49 | 0.45 | −0.07 | −0.39 |
tosr | 0.86 | 0.84 | 0.50 | −0.56 |
sfcwind | 0.78 | 0.78 | 0.53 | −0.48 |
sdew | 0.71 | 0.69 | 0.39 | −0.54 |
uas | 0.69 | 0.66 | 0.28 | −0.23 |
pr | −0.46 | −0.47 | −0.54 | 0.303 |
tas | 0.6 | 0.56 | −0.02 | −0.41 |
tmax | 0.96 | 0.96 | 0.84 | −0.62 |
tmin | 0.97 | 0.97 | 0.79 | −0.63 |
sdsr | 0.82 | 0.84 | 0.96 | −0.51 |
Predictor . | Maximum temperature . | Minimum temperature . | Solar radiation . | Precipitation . |
---|---|---|---|---|
evspsbl | −0.20 | −0.18 | 0.26 | 0.16 |
hurs | −0.70 | −0.66 | −0.16 | 0.49 |
huss | −0.11 | −0.14 | −0.47 | 0.06 |
ps | −0.62 | −0.57 | −0.08 | 0.47 |
psl | −0.61 | −0.56 | −0.024 | 0.45 |
rlds | 0.49 | 0.45 | −0.11 | −0.34 |
sulr | 0.63 | 0.59 | 0.01 | −0.43 |
tisr | 0.86 | 0.84 | 0.37 | −0.58 |
rlut | 0.49 | 0.45 | −0.07 | −0.39 |
tosr | 0.86 | 0.84 | 0.50 | −0.56 |
sfcwind | 0.78 | 0.78 | 0.53 | −0.48 |
sdew | 0.71 | 0.69 | 0.39 | −0.54 |
uas | 0.69 | 0.66 | 0.28 | −0.23 |
pr | −0.46 | −0.47 | −0.54 | 0.303 |
tas | 0.6 | 0.56 | −0.02 | −0.41 |
tmax | 0.96 | 0.96 | 0.84 | −0.62 |
tmin | 0.97 | 0.97 | 0.79 | −0.63 |
sdsr | 0.82 | 0.84 | 0.96 | −0.51 |
Correlation coefficient between predictor variables and minimum and maximum temperature, solar radiation, and precipitation for the CanESM2 model of the Shiraz station
Predictor . | Maximum temperature . | Minimum temperature . | Solar radiation . | Precipitation . |
---|---|---|---|---|
clt | −0.48 | −0.48 | −0.21 | 0.16 |
evespsbl | −0.29 | −0.28 | −0.10 | 0.09 |
hfls | −0.29 | −0.28 | −0.10 | 0.08 |
hur | −0.64 | −0.66 | −0.78 | 0.33 |
hurs | −0.64 | −0.66 | −0.76 | 0.35 |
prc | −0.09 | −0.09 | −0.13 | −0.01 |
prsn | −0.12 | −0.11 | −0.13 | 0.05 |
wpr | 0.56 | 0.57 | 0.68 | −0.41 |
psl | −0.04 | −0.03 | −0.02 | 0.05 |
sulr | 0.74 | 0.77 | 0.91 | −0.43 |
rsdscs | 0.39 | 0.44 | 0.85 | −0.22 |
tosr | 0.22 | 0.25 | 0.71 | −0.16 |
sci | −0.33 | −0.35 | −0.50 | 0.15 |
ta | 0.42 | 0.45 | 0.52 | −0.22 |
tas | 0.42 | 0.45 | 0.53 | −0.22 |
sdew | −0.37 | −0. 38 | −0.11 | 0.21 |
tauv | −0.52 | −0.53 | −0.50 | 0.22 |
ts | 0.43 | 0.46 | 0.55 | −0.23 |
nsw | −0.49 | −0.48 | −0.45 | 0.33 |
gh | −0.73 | −0.76 | −0.92 | 0.43 |
pr | −0.12 | −0.11 | 0.08 | 0.02 |
tmax | 0.97 | 0.97 | 0.78 | −0.60 |
tmin | 0.97 | 0.97 | 0.79 | −0.6 |
sdsr | 0.81 | 0.83 | 0.94 | −0.50 |
Predictor . | Maximum temperature . | Minimum temperature . | Solar radiation . | Precipitation . |
---|---|---|---|---|
clt | −0.48 | −0.48 | −0.21 | 0.16 |
evespsbl | −0.29 | −0.28 | −0.10 | 0.09 |
hfls | −0.29 | −0.28 | −0.10 | 0.08 |
hur | −0.64 | −0.66 | −0.78 | 0.33 |
hurs | −0.64 | −0.66 | −0.76 | 0.35 |
prc | −0.09 | −0.09 | −0.13 | −0.01 |
prsn | −0.12 | −0.11 | −0.13 | 0.05 |
wpr | 0.56 | 0.57 | 0.68 | −0.41 |
psl | −0.04 | −0.03 | −0.02 | 0.05 |
sulr | 0.74 | 0.77 | 0.91 | −0.43 |
rsdscs | 0.39 | 0.44 | 0.85 | −0.22 |
tosr | 0.22 | 0.25 | 0.71 | −0.16 |
sci | −0.33 | −0.35 | −0.50 | 0.15 |
ta | 0.42 | 0.45 | 0.52 | −0.22 |
tas | 0.42 | 0.45 | 0.53 | −0.22 |
sdew | −0.37 | −0. 38 | −0.11 | 0.21 |
tauv | −0.52 | −0.53 | −0.50 | 0.22 |
ts | 0.43 | 0.46 | 0.55 | −0.23 |
nsw | −0.49 | −0.48 | −0.45 | 0.33 |
gh | −0.73 | −0.76 | −0.92 | 0.43 |
pr | −0.12 | −0.11 | 0.08 | 0.02 |
tmax | 0.97 | 0.97 | 0.78 | −0.60 |
tmin | 0.97 | 0.97 | 0.79 | −0.6 |
sdsr | 0.81 | 0.83 | 0.94 | −0.50 |
The most significant variables on minimum temperature were TOA incident shortwave radiation (tisr), TOA outgoing shortwave radiation (tosr), near-surface wind speed (sfcwind), surface downward eastward wind stress (sdew), tmax, tmin, and surface downwelling shortwave radiation (sdsr) with a positive impact. The most significant positive variables on solar radiation were near-surface wind speed (sfcwind), tmax, tmin, and surface downwelling shortwave radiation (sdsr), while precipitation (pr) was the only most significant negative variable. In contrast to other parameters, precipitation was influenced negatively by TOA incident shortwave radiation (tisr), near-surface wind speed (sfcwind), surface downward eastward wind stress (sdew), tmax, tmin, and surface downwelling shortwave radiation (sdsr), while near-surface relative humidity (hurs) was the only most significant positive variable.
Increased wind speed can enhance the dispersion of clouds and pollutants, leading to clearer skies and higher solar radiation levels. Wind can also help in reducing local temperature inversions that trap heat and moisture, which may otherwise obstruct sunlight (Vidal Bezerra et al. 2023). Higher temperatures generally correlate with increased solar radiation. This relationship can be attributed to the fact that warmer air can hold more moisture, which can lead to clearer skies. Additionally, maximum temperatures often coincide with clear days, enhancing solar exposure. Conversely, minimum temperatures can indicate the presence of atmospheric conditions conducive to clearer skies during the day (Malik et al. 2022). The surface downwelling shortwave radiation variable directly measures the amount of solar energy received at the surface. A positive correlation indicates that as sdsr increases, so does available solar radiation for surface absorption, which is critical for applications like solar energy generation (Wehrli et al. 2013). On the other hand, increased rainfall typically leads to cloud formation, which blocks sunlight and reduces solar radiation reaching the surface. The negative correlation suggests that higher precipitation levels are associated with lower solar radiation, primarily due to cloud cover and increased atmospheric moisture that scatters and absorbs solar energy (Medvigy & Beaulieu 2012).
To predict variables such as precipitation and temperature, certain atmospheric parameters have a greater impact on the process than others. For example, factors such as relative humidity, minimum temperature, maximum temperature, and wind speed are more significant in predicting precipitation. Consequently, in precipitation forecasting using ANNs, only those factors with a greater influence are utilized as inputs to the network, rather than all parameters available in GCM output. This approach ensures that the results are more accurate while optimizing volume, time, and systematic error in calculations.
Also, correlation coefficients for the CanESM2 model of the Shiraz station indicate the most significant negative and positive variables that affect maximum temperature, minimum temperature, solar radiation, and precipitation. For maximum temperature, the most significant negative variables are total cloud cover percentage (clt), relative humidity (hur), near-surface relative humidity (hurs), northward near-surface wind (nsw), and geopotential height (gh), while water vapor path (wpr), surface upwelling longwave radiation (sulr), tmax, tmin, and surface downwelling shortwave radiation (rsds) have the most significant positive variables.
For minimum temperature, the most significant negative variables are relative humidity (hur), near-surface relative humidity (hurs), and geopotential height (gh), while water vapor path (wpr), surface upwelling longwave radiation (sulr), tmax, tmin, and surface downwelling shortwave radiation (sdsr) have most significant positive variables.
For solar radiation, the most significant negative variables are relative humidity (hur), near-surface relative humidity (hurs), and geopotential height (gh), while water vapor path (wpr), surface upwelling longwave radiation (sulr), surface downwelling clear-sky shortwave radiation (sdcsr), TOA outgoing shortwave radiation (tosr), tmax, tmin, and surface downwelling shortwave radiation (sdsr) have the most significant positive effects. For precipitation, water vapor path (wpr), surface upwelling longwave radiation (sulr), tmax, tmin, and surface downwelling shortwave radiation (sdsr) have the most significant negative effects, while northward near-surface wind (nsw) and geopotential height (gh) have the most significant positive variables.
CanESM2 model forecast values for maximum and minimum temperature, precipitation, and solar radiation of Shiraz for 2026–2045 (a, c, e, and g) and 2046–2065 (b, d, f, and h).
CanESM2 model forecast values for maximum and minimum temperature, precipitation, and solar radiation of Shiraz for 2026–2045 (a, c, e, and g) and 2046–2065 (b, d, f, and h).
HadGEM2-CC model forecast values for maximum and minimum temperature, precipitation, and solar radiation of Shiraz for 2026–2045 (a, c, e, and g) and 2046–2065 (b, d, f, and h).
HadGEM2-CC model forecast values for maximum and minimum temperature, precipitation, and solar radiation of Shiraz for 2026–2045 (a, c, e, and g) and 2046–2065 (b, d, f, and h).
The comparison between RCP4.5 and RCP8.5 scenarios suggests that both provide similar precipitation predictions for the Shiraz station in future periods. According to simulated results of HadGEM2-CC and CanESM2 models for this area, precipitation will decline in the future. Under RCP4.5 and RCP8.5 scenarios, the HadGEM2-CC model predicts a decrease of 11.26 and 19.75% in precipitation during 2026–2045, and a decrease of 21.45 and 36.51% during 2046–2065. On the other hand, the CanESM2 model predicts a reduction of 7.44 and 16.86% in precipitation under RCP4.5 and RCP8.5 scenarios, respectively, between 2026 and 2045, and a decrease of 30.03 and 35.18%, respectively, during the period of 2046–2065. Based on research findings for Shiraz, winter precipitation will decrease in both scenarios. However, there will be small and almost constant changes in summer precipitation. This is in line with Levin & Cotton's (2008) perspective, which suggests that the rise in greenhouse gas emissions leads to an increase in pollution, causing a decrease in winter rainfall.
According to the results of the HadGEM2-ES model simulation for Abade, precipitation is expected to increase by 9.21% under the RCP4.5 scenario and decrease by 7.65% under the RCP8.5 scenario between 2026 and 2045. Between 2046 and 2065, precipitation is predicted to increase by 3.8% under the RCP4.5 scenario and decrease by 10.85% under the RCP8.5 scenario. However, the ACCESS1.3 model simulation for Abade suggests that precipitation will decrease by 6.11 and 11.04% under RCP4.5 and RCP8.5 scenarios, respectively, between 2026 and 2045. Between 2046 and 2065, precipitation is estimated to decrease by 14.62 and 18.8%, respectively, under RCP4.5 and RCP8.5 scenarios. Based on simulation results of the HadGem2-ES model for the station of Fasa, it has been observed that there is a decrease in precipitation by 7.84 and 14.63% in the period of 2026–2045 under RCP4.5 and RCP8.5 scenarios, respectively. From 2046 through 2065, the decrease in precipitation is 10.91 and 18.7%. Similarly, for the HadGem2-CC model, it has been observed that precipitation decreases by 8.62 and 11.15% in the period 2026–2045 under scenarios RCP4.5 and RCP8.5, respectively, and by 13.13 and 20.36% in the period 2046–2065. According to the simulation results of the HadGem2-ES model, precipitation at the Lar station is expected to decrease by 8.34 and 12.59% during 2026–2045 under the RCP4.5 and RCP8.5 scenarios. Similarly, a descent of 15.42 and 21.64% is anticipated from 2046 to 2065. On the other hand, the ACCESS1.3 model suggests a decline in precipitation of 7 and 10.28% during 2026–2045 under RCP4.5 and RCP8.5 scenarios, respectively. Moreover, a reduction of 13.16 and 20.48% is expected between 2046 and 2065. The IPCC reported in 2018 that while precipitation levels may decrease in some regions, extreme weather events are likely to rise (IPCC 2018). This means that many areas may experience more frequent and severe weather conditions, such as droughts or floods, due to changes in precipitation patterns.
The HadGEM2-CC model predicts that maximum temperature in Shiraz will increase by 0.48 and 1.21 °C between 2026 and 2045 under RCP4.5 and RCP8.5 scenarios and by 0.92 and 1.84 °C between 2046 and 2065. The CanESM2 model shows maximum temperature increases of 0.6 and 1.19 °C between 2026 and 2045 under RCP4.5 and RCP8.5 scenarios, and 1.64 and 2.67 °C between 2046 and 2065, respectively. The HadGem2-ES model projects that maximum temperature in Abade will increase by 0.27 and 0.57 °C between 2026 and 2045 under RCP4.5 and RCP8.5 scenarios, and by 1.24 and 2.16 °C between 2046 and 2065. In comparison, the ACCESS1.3 model forecasts an increase of 0.56 and 0.91 °C for the period 2026–2045 under RCP4.5 and RCP8.5 scenarios and an increase of 1.35 and 1.98 °C between 2046 and 2065. Results from the HaddGem2-ES model for the Fasa station indicate that maximum temperature is projected to increase by 0.43 and 1.1 °C under RCP4.5 and RCP8.5 scenarios, respectively, during 2026–2045. In 2046–2065, maximum temperature is expected to increase by 0.73 and 1.87 °C under the same scenarios. For the HadGem2-CC model in Fasa, maximum temperature is forecasted to rise by 1.37 and 1.39 °C during 2026–2045 under RCP4.5 and RCP8.5 scenarios, and by 0.2 and 2.21 °C during 2046–2065. The HaddGem2-ES model predicts an increase in maximum temperature of 0.77 and 1.25 °C for the Lar station under RCP4.5 and RCP8.5 scenarios in 2026–2045, and of 1.41 and 1.69 °C for the same scenarios. For the ACCESS1.3 model, the maximum temperature is expected to rise by 0.32 and 0.68 °C under RCP4.5 and RCP8.5 scenarios in 2026–2045, and by 1 and 1.43 °C in the period 2046–2065, respectively.
The minimum temperature parameter for the HadGEM2-CC model in Shiraz is projected to increase by 0.44 and 1.02 °C under RCP4.5 and RCP8.5 scenarios, respectively, in 2026–2045. For 2046–2065, the increase is projected to be 0.72 and 2.01 °C under the same scenarios. For the CanESM2 model in Shiraz, the minimum temperature is expected to rise by 0.32 and 0.92 °C under RCP4.5 and RCP8.5 scenarios, respectively, in 2026–2045. For 2046–2065, the increase is projected to be 0.77 and 1.71 °C under the same scenarios. For the Abade station, the minimum temperature parameter under RCP4.5 and RCP8.5 scenarios will increase by 0.35 and 0.83 °C, respectively, in 2026–2045. It will also increase by 1.46 and 2.11 °C, respectively, in 2046–2065. As for the ACCESS1.3 model, minimum temperature under RCP4.5 and RCP8.5 scenarios will rise by 0.24 and 0.9 °C in 2026–2045 and by 1.16 and 1.62 °C in 2046–2065. The minimum temperature for the HaddGem2-ES model in the Fasa station is projected to increase by 0.31 and 0.9 °C during the period 2026–2045 under RCP4.5 and RCP8.5 scenarios and by 0.68 and 1.44 °C during the period 2046–2065. For the HadGem2-CC model, the temperature is expected to increase by 1 and 1.37 °C during the period 2026–2045 under scenarios RCP4.5 and RCP8.5 and by 2.48 and 2.71 °C in the period 2046–2065. The HaddGem2-ES model predicts an increase in the minimum temperature parameter for the Lar station of 0.45 and 0.81 °C during the 2026–2045 period under RCP4.5 and RCP8.5 scenarios and a further increase of 1.18 and 1.58 °C during the 2046–2065 period. Similarly, the ACCESS1.3 model forecasts an increase of 0.23 and 0.44 °C for the 2026–2045 period under RCP4.5 and RCP8.5 scenarios, with a subsequent rise of 0.77 and 1.14 °C during the 2046–2065 period for the Lar station. The maximum and minimum temperature forecasts align with similar conclusions from studies conducted in Iran regarding the projected temperature rise due to climate change (Doulabian et al. 2021).
For the Shiraz station, the solar radiation parameter for the HadGEM2-CC model is projected to increase by 0.35 and 0.71 kW/m2/day under RCP4.5 and RCP8.5 scenarios, respectively, during the period 2026–2045. For 2046–2065, the increase is projected to be 0.66 and 1.17 kW/m2/day under the same scenarios. For the CanESM2 model, the solar radiation is projected to increase by 0.32 and 0.83 kW/m2/day from 2026 to 2045 and by 0.79 and 1.36 kW/m2/day from 2046 to 2065. The solar radiation parameter for the HaddGem2-ES model in Abade will increase by 0.22 and 0.42 kW/m2/day in 2026–2045 under RCP4.5 and RCP8.5 scenarios and by 0.63 and 0.72 kW/m2/day in 2046–2065. For the ACCESS1.3 model in this station, the parameter will increase by 0.23 and 0.5 in 2026–2045 under RCP4.5 and RCP8.5 scenarios and by 0.72 and 0.95 kW/m2/day in 2046–2065. The solar radiation parameter for the HadGem2-ES model in Fasa is projected to increase by 0.25 and 0.72 kW/m2/day during 2026–2045 under RCP4.5 and RCP8.5 scenarios. It is also expected to increase by 0.43 and 1 kW/m2/day during 2046–2065. For the HadGem2-CC model, the parameter will increase by 0.56 and 1 kW/m2/day during the period 2026–2045 under scenarios RCP4.5 and RCP8.5 and by 1.58 and 1.91 kW/m2/day in 2046–2065. The solar radiation parameter for the HaddGem2-ES model at the Lar station is projected to increase by 0.23 and 0.44 kW/m2/day during the 2026–2045 period under RCP4.5 and RCP8.5 scenarios. It is expected to further increase by 0.6 and 0.85 kW/m2/day during the 2046–2065 period. For the ACCESS1.3 model, the parameter will increase by 0.3 and 0.48 kW/m2/day during the 2026–2045 period under RCP4.5 and RCP8.5 scenarios and by 0.72 and 0.91 kW/m2/day during the 2046–2065 period.
The study shows that there will be significant increases in both minimum and maximum temperatures, as well as solar radiation in Fars Province. Projected temperature increases range from 0.23 to 2.71 °C for minimum temperatures, 0.27 to 2.67 °C for maximum temperatures, and 0.22 to 1.91 for solar radiation. These findings are consistent with previous studies conducted in similar arid regions, such as work by Mohammed & Hassan (2022), which projected a temperature increase of up to 5.91 °C under the RCP8.5 scenario in Iraq. This consistency highlights the urgent need for localized climate adaptation strategies. Anticipated changes in precipitation range from a decrease of 7% to an increase of 36.5%. Policymakers must consider these variations when developing water resource management plans. For example, increased variability in precipitation could worsen drought conditions, affecting agricultural productivity and water availability. Therefore, it is essential to incorporate these findings into strategic planning to mitigate adverse effects on local communities and ecosystems. The impact of monsoonal troughs on climate change in Iran is also an important factor to consider. During summer, temperatures in the interior regions of Iran rise significantly, creating areas of low pressure. This low pressure draws moist air from the Indian Ocean toward Iran. As a result, monsoonal trough typically forms in June and July, leading to monsoon rains in parts of Iran, particularly in the south and southeast. The development of these monsoonal troughs significantly impacts climate and precipitation patterns across various regions of Iran, owing to the country's unique geographical position and proximity to open waters (Ghassabi et al. 2023; Mahoutchi et al. 2023). Fars Province, especially the stations studied, is among the areas affected by these troughs. The formation of monsoonal troughs during specific seasons can lead to intense and sudden rainfall in these regions. These rains typically occur in warmer months and can affect positively and negatively on agriculture, water resources, and the environment. In the Shiraz station, rainfall associated with monsoonal troughs can help meet water needs for agriculture and gardening. Similarly, at the Fasa and Lar stations, these rains can increase soil moisture and improve the quality of agricultural products. Additionally, in Abade, this phenomenon can help generate new water resources and strengthen local ecosystems.
While this study uses robust methodologies combining GCMs and ANNs, it is important to acknowledge potential limitations, such as uncertainties inherent in GCM outputs and assumptions made during model calibration. Future research should focus on refining these models by integrating more localized datasets and exploring additional climate scenarios to provide a comprehensive understanding of climate impacts in Fars Province.
CONCLUSION
The findings of this study indicate a significant projected increase in maximum temperatures in Fars Province, southern Iran, with potential rises of up to 2.67 °C under climate change scenarios RCP4.5 and RCP8.5. This temperature increase poses serious challenges to agricultural productivity, highlighting the urgent need for adaptive strategies in water resource management and agricultural practices. Additionally, the variability in precipitation – ranging from a decrease of 7% to an increase of 36.5% – underscores the unpredictability of future climate conditions in the region. Given these results, policymakers and stakeholders must implement proactive measures that address the anticipated climatic shifts. This includes investing in sustainable agricultural practices, enhancing water conservation techniques, and developing robust climate adaptation frameworks tailored to the unique climatic conditions of Fars Province. The integration of advanced modeling techniques, such as ANNs with GCM outputs, not only improves predictive accuracy but also provides a valuable tool for understanding and mitigating the impacts of climate change on vulnerable regions. In conclusion, this research contributes to a deeper understanding of climate dynamics in southern Iran and serves as a call to action for comprehensive strategies that can safeguard both environmental and agricultural resilience against the backdrop of ongoing climate change.
FUNDING
The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.
AUTHOR CONTRIBUTIONS
All authors contributed to the study conception and design. Material preparation, data collection, and analysis were performed by N.R., A.M.A.-A., and M.A.M. The first draft of the manuscript was written by N.R. and M.B. and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.
DATA AVAILABILITY STATEMENT
All relevant data are included in the paper or its Supplementary Information.
CONFLICT OF INTEREST
The authors declare there is no conflict.