Listen

Due to global climate change, managing water resources is one of the most critical challenges for most countries in the world, especially in the Middle East. In the Kurdistan Region of Iraq (KRI), there is a good amount of precipitation, surface water, and groundwater, but the main issue is mismanagement of those sources. Rainfall is one of the major sources of water resources in KRI. In order to manage the available water resources and prevent natural disasters such as floods and droughts, there is a need for reliable models for forecasting rainfall. The current study focuses on developing a hybrid model, namely seasonal autoregressive integrated moving average combined with an artificial neural network (SARIMA-ANN) for forecasting monthly rainfall at Sulaymaniyah City for the duration of 1938–2012. For comparison purposes, a conventional machine learning model, namely artificial neural networks (ANN) has been applied on the same data. Two different statistical measurements, namely, root mean square error (RMSE) and coefficient of determination (R2), have been used to check the accuracy of the proposed models. According to the findings, SARIMA-ANN outperformed ANN with RMSE = 11.5, RMSE = 51.002, R2 = 0.98, R2 = 0.43, respectively. The findings of the current study could contribute to Sustainable Development Goal (SDG) 6.

Listen

  • An innovative hybrid model has been developed for rainfall prediction.

  • Accurate rainfall prediction will lead to better management of water resources.

  • Rainfall prediction is an important hydrological tool due to global climate change.

hybrid model, rainfall prediction, SARIMA-ANN, Sulaymaniyah City, Sustainable Development Goal (SDG) 6, water resources management
Listen
SARIMA-ANN

Seasonal autoregressive integrated moving average combined with artificial neural network

ANN

Artificial neural network

RMSE

Root mean square error

R2

Coefficient of determination

SDG

Sustainable development goal

ML

Machine learning

SVM

Support vector machine

Listen

Water is one of the basic needs of humans as 60% of the human body is made of water. Water is utilized for different categories such as washing, cleaning, cooking, and so on. Just as life does not exist without oxygen, so it does not exist without water, and it is the only source that supports the Earth to distinguish itself from other planets and provide opportunities for life. The Earth's surface consists of three parts water and one part land. Finding water anywhere else in the universe is nearly impossible. It is important for all living things; therefore, water must be preserved for future generations. Water resource management is the process of planning, protecting, and improving water resources in terms of quality and quantity. Managing water resources is important in order to be able to supply water for domestic, irrigation, hydropower production, etc. Water resources management also helps to reduce the risks of extreme events such as floods (Akram & Hamid 2013; Vieira et al. 2020; Yang & Liu 2020; Fereshtehpour et al. 2021).

Nowadays, machine learning (ML) models have been used in the field of water management, especially to predict rainfall (Latif et al. 2023). ML is able to predict rainfall based on historical rainfall data. Nowadays, accurate prediction of rainfall is a big challenge due to climate change (Basha et al. 2020). ML models are widely used to predict future rainfall for the short and long term. The ML model relies on hydrological variables and data on previous rainfall and accurately processes it to predict future rainfall. Researchers around the world rely on ML due to the in-depth and accurate predictions it makes for the data (Parmar et al. 2017; Ahmed et al. 2020).

Nowadays, the Kurdistan Region of Iraq (KRI) faces water scarcity. The water level has decreased in both Dokan and Darbandikhan dams which are the main sources of water for Sulaymaniyah City in terms of supplying water for society. This water scarcity will lead to other issues in the future if the authority is not able to resolve it. The biggest problem in the KRI is the lack of a good water management plan. Developing a reliable model for predicting rainfall is considered one of the most important tools for managing water resources (Tinti 2017; Mohammed et al. 2018).

Having a wider population would increase water demand. To ensure there is enough water for humans, agriculture, and industry, effective water management is needed. Water resources are significantly affected by climate change, including shifts in precipitation patterns, more severe droughts, and more intense rainfall events (Boretti & Rosa 2019; Zubaidi et al. 2020; Barker et al. 2021). In today's world, starting to predict water resources with ML techniques is increasingly considered an important approach due to their many advantages over conventional modeling techniques (Latif & Ahmed 2023). ML algorithms can process a lot of data and find patterns that manual analysis might miss. Predicting the availability, demand, and quality of water can be made more accurate with ML to recognize complex relationships and patterns in data. The speed with which ML algorithms can process a large amount of data makes it ideal for real-time applications like flood forecasting and water management. ML algorithms are useful for predicting water resources under changing climate conditions and other environmental factors since they can adapt to changing conditions and learn from the new data. By determining the best methods for allocating and managing water to satisfy the requirements of various stakeholders, ML algorithms can enhance water management strategies. By automating data processing and analysis, ML methods can lower the cost of water resource management (Shen 2018).

It can be concluded that ML methods are important for predicting water resource parameters since they improve accuracy, speed, adaptability, optimization, and cost-effectiveness for effective management of water resources in today's rapidly changing world (Li & Sansalone 2022). Many studies have been proposed for forecasting monthly rainfall in different regions. For example, a study conducted by Ali & Shahbaz (2020) proposes an effective method for runoff forecasting by modeling the relationship between precipitation and runoff to identify the optimal rainfall pattern for forecasting daytime streamflow. They have identified different sets of rainfall antecedent components and developed an artificial neural network (ANN) model. Their findings show the potential of ANN based approaches as an effective alternative for solving hydrological problems. Furthermore, another study proposed by Adede et al. (2015) aims to describe an ML experiment that uses ANN to predict rainfall. The data set was divided into three subsets of training, validation, and testing. The test was run 100 times with different random distribution of data. For each fold, ANN was trained 100 times resulting in 10,000 prediction ensembles. The goal of repeatedly training an ANN was to build a more robust model that could handle different data variations. Moreover, another study conducted by Pham et al. (2020) uses meteorological variables such as maximum and minimum temperature, wind speed, relative humidity, and solar radiation at different altitudes as input parameters to predict daily rainfall in Hoa Province in Vietnam. The utilized models were adaptive network-based fuzzy inference systems (ANFIS) optimized with particle swarm optimization (PSO), ANN, and support vector machine (SVM). The results showed that all AI models provided adequate forecasts, but SVM proved to be the best method for forecasting rainfall. In another study, Malki et al. (2020) aimed to investigate the relationship between weather variables, particularly temperature and humidity, and the spread of COVID-19. Various ML models were proposed and employed to extract this relationship using data on the number of confirmed cases and weather variables in certain regions. The study found that weather variables, particularly temperature, are more relevant in predicting the mortality rate of COVID-19 compared to census variables. This suggests that temperature and humidity are important features for predicting the mortality rate of COVID-19. Furthermore, their study indicates that higher temperatures are associated with lower numbers of infection cases, suggesting a potential relationship between temperature and the spread of COVID-19.

The aim of the current study is to develop a hybrid ML model, namely SARIMA-ANN, for forecasting monthly rainfall in Sulaymaniyah City, located in the north of Iraq. In order to compare the accuracy of the proposed hybrid SARIMA-ANN model, ANN has been also applied to the same dataset.

Listen

Study area

Listen
Sulaymaniyah is a city in the KRI located approximately 370 km northeast of Iraq's capital, Baghdad. The city is located in a low-lying territory that is around 2,400 km2 in size, with Goizha Mountain to the north and Qaradagh Mountain to the south. The study area is characterized by a distinct continental interior climate of Mediterranean type, with an average annual precipitation ranging from 500 to 700 mm. There were around 856,990 people living in the Sulaymaniyah governorate in 2016 (Alkaradaghi et al. 2019). The reason for selecting Sulaymaniyah as the study area is due to several important points. For instance, Sulaymaniyah is an important center in the region due to its fertile soil for agriculture and its geographical location between Iraq and Iran. Also, it has two important dams, namely Darbandikhan and Dokan, along with several other small dams that work to store water and reduce the risk of flooding. Furthermore, Sulaymaniyah is also rich in natural resources such as gas and oil, and is considered an important center for energy. Its mountains and valleys exhibit a beautiful landscape and a geological structure. The climate of Sulaymaniyah is relatively dry, cold in winter, and hot in summer. Finally, the position of Sulaymaniyah is important because of its geographical and strategic location. This has made it a valuable city in the region. Figure 1 represents the location of the study area.
Figure 1
Location of the study area (Source: Google Map, 2024).
View largeDownload slide

Location of the study area (Source: Google Map, 2024).

Figure 1
Location of the study area (Source: Google Map, 2024).
View largeDownload slide

Location of the study area (Source: Google Map, 2024).

Close modal

Data

Listen
The weather in Sulaymaniyah is constantly changing throughout the year, the weather in summer is generally dry and hot, the highest temperature in some areas reaches 50 °C, winter is generally cold and rainy, and the temperature reaches 5 °C or even less. In spring and autumn, the weather is generally cool and occasionally rainy. The climate of Sulaymaniyah is generally Mediterranean with moderate rainfall. In this research, rainfall data of Sulaymaniyah City has been collected in a monthly manner recorded in the rainfall station of Sulaymaniyah in the years 1938–2012. The statistics showed that the amount of rainfall in January is the highest compared to all other months in Sulaymaniyah since the sky is cloudy and the temperature is cool. The statistical data were determined in order to understand the environmental condition. Figure 2 represents the monthly rainfall from 1938 to 2012. Figure 3 shows the flowchart of the proposed SARIMA-ANN and single ANN ML models.
Figure 2
Monthly rainfall in Sulaymaniyah City from January 1938 to October 2012.
View largeDownload slide

Monthly rainfall in Sulaymaniyah City from January 1938 to October 2012.

Figure 2
Monthly rainfall in Sulaymaniyah City from January 1938 to October 2012.
View largeDownload slide

Monthly rainfall in Sulaymaniyah City from January 1938 to October 2012.

Close modal
Figure 3
The framework of the current study.
View largeDownload slide

The framework of the current study.

Figure 3
The framework of the current study.
View largeDownload slide

The framework of the current study.

Close modal

Artificial neural network

Listen
ANN is one of the most commonly used ML techniques for prediction purposes. Layers of interconnected artificial neurons, or nodes, make up ANN models, which use a number of mathematical operations to analyze data. Each neuron in the network takes inputs from other neurons or external data sources, processes those inputs to produce an output signal and then transmits that signal to other neurons. Once an ANN model has been trained on a substantial amount of data, it can be applied to forecast the data (Latif et al. 2020). ANN models can be used in conjunction with other models and data sources to increase the accuracy of rainfall forecasts. The structure of the current study's ANN model is shown in Figure 4.
Figure 4
The structure of the current study's ANN model.
View largeDownload slide

The structure of the current study's ANN model.

Figure 4
The structure of the current study's ANN model.
View largeDownload slide

The structure of the current study's ANN model.

Close modal

SARIMA-ANN

Listen
For the purpose of forecasting time series, the SARIMA-ANN hybrid model combines the strengths of two separate models, the SARIMA and ANN. The hybrid model's SARIMA component focuses on identifying linear and seasonal trends in the data. SARIMA models, which include moving average, integrated, and autoregressive components, are frequently employed in time series analysis. By doing so, the historical dependencies and trends in the data are better captured. The SARIMA model could have trouble capturing subtle patterns and complex nonlinear interactions in the data. ANNs can capture nonlinearities and complex interactions and are very adaptable. The residuals (the discrepancies between the actual values and SARIMA predictions) are used to train an ANN, which allows the hybrid model to capture any patterns that the SARIMA model could have missed. By simulating the linear and seasonal components of the time series, the SARIMA component of the hybrid model offers a baseline forecast. Additional nonlinear patterns and dependencies are captured by the ANN component. The final forecast is created by combining the results of the two models. The data must be divided into linear and nonlinear components before being used in a hybrid model:
(1)
where is the actual rainfall, is the linear component of actual rainfall, and is the nonlinear component of actual rainfall (Moeeni & Bonakdari 2017). Iteratively fitting the SARIMA model to the data comes first, then an examination of the residuals in the SARIMA-ANN hybrid model. The ANN is then trained using the residuals as the variable, and the SARIMA model is then integrated for the outcome prediction. Using the right metrics, the hybrid model's performance is assessed, and changes can be made as needed to make the model better. Figure 5 shows the structure of the developed SARIMA-ANN model.
Figure 5
The structure of the developed SARIMA-ANN model (Moeeni & Bonakdari 2017).
View largeDownload slide

The structure of the developed SARIMA-ANN model (Moeeni & Bonakdari 2017).

Figure 5
The structure of the developed SARIMA-ANN model (Moeeni & Bonakdari 2017).
View largeDownload slide

The structure of the developed SARIMA-ANN model (Moeeni & Bonakdari 2017).

Close modal

Input selections

Listen

Auto-correlation function (ACF) is a statistical technique used to determine how closely a time series of data and its lagged values correlate. In other words, it evaluates how closely a data point resembles its previous values. The correlation coefficient between time series data and a lagged version of that data at various time lags is used to calculate the ACF. The outcome is a collection of correlation coefficients, each of which is associated with a different time lag. The ideal lag value for a time series model can be chosen using ACF to spot patterns and trends in time series data. In this research, ACF was used to find the most correlated lagged data. All the training data in the ANN system were built on 4-lag-time as recommended by ACF.

Evaluation metrics

Listen

Metrics used to assess an ML model's performance are called evaluation metrics. With the support of these measures, it is possible to evaluate the model's effectiveness in terms of accuracy, precision, recall, and other crucial factors. This research depends on two types of evaluation metrics. The first type is RMSE, and the second type is R2.

Root mean squared error

Listen
Root mean squared error (RMSE) is a widely used metric for assessing how well a ML model performs, particularly when doing regression tasks. It calculates the difference between the dataset's actual values and the predicted values.
(2)
where and are observed and predicted rainfall values.

R2

Listen
An evaluation of a regression model's ability to fit the data is done statistically using the R2 metric. It displays the proportion of the dependent variable's variance that the model's independent variables are responsible for explaining.
(3)
Listen

Predicting the amount of rainfall in Sulaymaniyah after testing the data through the ANN model could not achieve accurate results. In the months when the amount of rainfall in the city was high, the level of prediction was very poor in testing the data through the ANN model. In the next step, SARIMA-ANN was developed, and the best model obtained an accurate rainfall forecasting result.

ANN results

Listen
Initially, the ANN model was used to forecast the amount of rainfall in Sulaymaniyah based on data from previous years. As can be seen in the first testing set (Figure 6), in the first model, it is clear that the prediction is generally positive, and there is no negative prediction. However, the first model could not provide accurate and reliable predictions. The model failed to predict extreme events. The R2 for the first model is 0.335, which is a high error rate, with RMSE of 55.05 which is also a high error rate. Obviously, the rainfall level in Sulaymaniyah is not stable, so this model could not predict the low and high levels of rainfall, especially in months such as January when usually heavy rains and sometimes floods occur in some parts of the city.
Figure 6
Rainfall forecast results obtained for the training and testing periods in Model 1.
View largeDownload slide

Rainfall forecast results obtained for the training and testing periods in Model 1.

Figure 6
Rainfall forecast results obtained for the training and testing periods in Model 1.
View largeDownload slide

Rainfall forecast results obtained for the training and testing periods in Model 1.

Close modal
In the second experiment with the ANN model, which is model 2 as shown in Figure 7, it was found that the positive and negative ratios do not represent a good model for predicted rainfall and there is a significant error compared to the first model. In the ANN model, the Rt−1, and Rt−2 were used for the second model as input combination selection. However, the model failed to provide an accurate rainfall forecast. The error rate of the second model is higher than the first model. The R2 for the second model is 0.362, which is a high error rate and causes the instability of the prediction. The RMSE is 54.312 which is obviously a high error rate compared to the first model. The instability of rainfall and the complexity of the seasons in Sulaymaniyah resulted in high errors in prediction accuracy.
Figure 7
Rainfall forecast results obtained for the training and testing periods in Model 2.
View largeDownload slide

Rainfall forecast results obtained for the training and testing periods in Model 2.

Figure 7
Rainfall forecast results obtained for the training and testing periods in Model 2.
View largeDownload slide

Rainfall forecast results obtained for the training and testing periods in Model 2.

Close modal
As shown in Figure 8, the error rate for the third model is higher than for the previous models. Rt−1, Rt−2, and Rt−3 are used as input combinations for the third model. The ability to predict high levels of rainfall is weak, as shown in Figure 8. There are positive and negative parts. This shows the weaknesses and errors of the current model compared to the previous models. The R2 for the third model is 0.386 which indicates a high error rate and causes the instability of the prediction. Also, RMSE is 53.292 for the first model which is a high error rate compared to the first and second models.
Figure 8
Rainfall forecast results obtained for the training and testing periods in Model 3.
View largeDownload slide

Rainfall forecast results obtained for the training and testing periods in Model 3.

Figure 8
Rainfall forecast results obtained for the training and testing periods in Model 3.
View largeDownload slide

Rainfall forecast results obtained for the training and testing periods in Model 3.

Close modal
As shown in Figure 9, the lowest error rate has been obtained compared to the other previous models. The input combination selection for the fourth model was Rt−1, Rt−2, Rt−3, and Rt−4. The R2 for the fourth model is 0.431, which shows a low error rate compared to the other five models. RMSE is 51.002, which shows a similarly low error rate. The availability of appropriate models with the least error will help predict low levels of rainfall in the future in conjunction with high levels of rainfall, which has concluded with the fourth model. The temperature difference in the city was an important factor in the error rate, especially in the first and second months of the year which had an unstable amount of rainfall. In the next steps, Model 4 was developed which was an appropriate model compared to the others. The fourth model was successfully developed with the SARIMA-ANN.
Figure 9
Rainfall forecast results obtained for the training and testing periods in Model 4.
View largeDownload slide

Rainfall forecast results obtained for the training and testing periods in Model 4.

Figure 9
Rainfall forecast results obtained for the training and testing periods in Model 4.
View largeDownload slide

Rainfall forecast results obtained for the training and testing periods in Model 4.

Close modal
It was found that Model 5 has a higher error rate than Model 4. Based on both RMSE and R2, it showed that the fourth model is the most appropriate among all models. The R2 for the fifth model is 0.372, which shows a high error rate compared to the other models, especially the fourth. The RMSE for Model 5 is 55.501, which showed another low coefficient compared to the fourth model as shown in Figure 10. According to the results given in Table 1, there are five different models with five different components, and Model 4 shows the better accuracy among the five. Finally, the input combination for Model 4 is selected to be developed utilizing the SARIMA-ANN model.
Table 1

Represents the results of ANN models in terms of RMSE and R2

ANN modelsInput selectionR2RMSE
Model 1 Rt−1 0.335 55.05 
Model 2 Rt−1, Rt−2 0.362 54.312 
Model 3 Rt−1, Rt−2, Rt−3 0.386 53.292 
Model 4 Rt−1, Rt−2, Rt−3, Rt−4 0.431 51.002 
Model 5 Rt−1, Rt−2, Rt−3, Rt−4, Rt−5 0.372 55.501 
ANN modelsInput selectionR2RMSE
Model 1 Rt−1 0.335 55.05 
Model 2 Rt−1, Rt−2 0.362 54.312 
Model 3 Rt−1, Rt−2, Rt−3 0.386 53.292 
Model 4 Rt−1, Rt−2, Rt−3, Rt−4 0.431 51.002 
Model 5 Rt−1, Rt−2, Rt−3, Rt−4, Rt−5 0.372 55.501 
View Large
Figure 10
Rainfall forecast results obtained for the training and testing periods in Model 5.
View largeDownload slide

Rainfall forecast results obtained for the training and testing periods in Model 5.

Figure 10
Rainfall forecast results obtained for the training and testing periods in Model 5.
View largeDownload slide

Rainfall forecast results obtained for the training and testing periods in Model 5.

Close modal

SARIMA-ANN results

Listen

Since Model 4 outperforms all models based on both RMSE and R2 metrics, SARIMA-ANN has been developed based on the input combination scenario of Model 4. Table 2 shows the results of the developed SARIMA-ANN model for predicting rainfall. It shows that RMSE is 11.5 and R2 is 0.98 which outperformed the most accurate model of ANN (Model 4) with a significant difference of RMSE = 51.002 and R2 = 0.431. It shows the successful results of using the SARIMA-ANN model in this study. The SARIMA-ANN model successfully identified the data and accurately worked on it to forecast monthly rainfall in Sulaymaniyah. The model forecasted low and high levels of rainfall in Sulaymaniyah accurately, which helps sustainable water management.

Table 2

Represents SARIMA-ANN results with RMSE, R2.

SARIMA-ANNInput selectionR2RMSE
Developed model Rt−1, Rt−2, Rt−3, Rt−4 0.98 11.5 
SARIMA-ANNInput selectionR2RMSE
Developed model Rt−1, Rt−2, Rt−3, Rt−4 0.98 11.5 
View Large

The main problem in Sulaymaniyah City is that its summers are very hot, and its winters are very cold, and this vast change makes the prediction process harder and contains more errors. SARIMA-ANN can anticipate severe weather, such as heavy floods and drought. Therefore, through this model, weather disasters can be predicted before they occur. Also, this model was able to predict low levels of rain, unlike previous models. Finally, it can be mentioned that the developed SARIMA-ANN model is considered accurate and can be relied upon to predict the rainfall levels of Sulaymaniyah City accurately. Figure 11 represents the performance of the proposed hybrid SARIMA-ANN model.
Figure 11
Rainfall forecast results with SARIMA-ANN model.
View largeDownload slide

Rainfall forecast results with SARIMA-ANN model.

Figure 11
Rainfall forecast results with SARIMA-ANN model.
View largeDownload slide

Rainfall forecast results with SARIMA-ANN model.

Close modal

Based on the findings of the current study, SARIMA-ANN could be very useful for Sulaymaniyah City since it could accurately predict the monthly rainfall data. On the other hand, studies for rainfall prediction in Sulaymaniyah using hybrid models are rarely found in the literature. Therefore, the findings of the current study could fill this gap in the literature.

Listen

The water level of Sulaymaniyah City rises and falls due to climate change. ANN failed to perform well in predicting monthly rainfall in Sulaymaniyah. The main obstacle to the low prediction results was the low quantity of data. The fourth model was selected as the highest accuracy in the ANN model. Later, the input selection scenario of the fourth model was developed using SARIMA-ANN and obtained a high accuracy of RMSE = 12.209 and R2 = 0.979. This developed model of the current research could be considered as a suitable tool for managing rainfall in Sulaymaniyah City. It is recommended for future studies to use daily rainfall data since it will help the ML models train better and have more accurate predicted values.

Listen

Data will be made available on reasonable request.

Listen

Data cannot be made publicly available; readers should contact the corresponding author for details.

Listen

The authors declare there is no conflict.

Adede
C.
,
Oboko
R.
,
Wagacha
P.
 &
Atzberger
C.
2015
A mixed model approach to drought prediction using artificial neural networks: Case of an operational drought monitoring environment
.
MDPI Remote Sens
11
(
9
),
1099
.
Google Scholar
Crossref
Search ADS
 
Ahmed
A. N.
,
Latif
S. D.
 &
El-Shafie
A.
2020
Research Series in Civil Engineering: Soft Computing Applications in Hydro-Environment
.
UNITEN Press
,
Selangor
.
Google Scholar
 
Akram
M.
 &
Hamid
A.
2013
A comprehensive review on water balance
.
Biomed. Prev. Nutr.
3
,
193
–
195
.
https://doi.org/10.1016/j.bionut.2012.10.003
.
Google Scholar
Crossref
Search ADS
 
Ali
S.
 &
Shahbaz
M.
2020
Streamflow forecasting by modeling the rainfall–streamflow relationship using artificial neural networks
.
Model. Earth Syst. Environ.
6
,
1645
–
1656
.
https://doi.org/10.1007/s40808-020-00780-3
.
Google Scholar
Crossref
Search ADS
 
Alkaradaghi
K.
,
Ali
S. S.
,
Al-ansari
N.
 &
Laue
J.
2019
Landfill site selection using MCDM methods and GIS in the sulaymaniyahyah governorate, Iraq
.
Sustainability
.
https://doi.org/10.3390/su11174530
.
Google Scholar
 
Barker
A.
,
Pitman
A.
,
Evans
J. P.
,
Spaninks
F.
 &
Uthayakumaran
L.
2021
Drivers of future water demand in Sydney, Australia: Examining the contribution from population and climate change
.
J. Water Clim. Chang
12
,
1168
–
1183
.
https://doi.org/10.2166/wcc.2020.230
.
Google Scholar
Crossref
Search ADS
 
Basha
C. Z.
,
Bhavana
N.
,
Bhavya
P.
 &
Sowmya
V.
2020
Rainfall prediction using machine learning deep learning techniques
. In:
Proceedings of the International Conference on Electronics and Sustainable Communication Systems, ICESC 2020
.
https://doi.org/10.1109/ICESC48915.2020.9155896
.
Google Scholar
Crossref
Search ADS
 
Boretti
A.
 &
Rosa
L.
2019
Reassessing the projections of the world water development report
.
npj Clean Water
2
.
https://doi.org/10.1038/s41545-019-0039-9
.
Google Scholar
 
Fereshtehpour
M.
,
Javidi Sabbaghian
R.
,
Farrokhi
A.
,
Jovein
E. B.
 &
Ebrahimi Sarindizaj
E.
2021
Evaluation of factors governing the use of floating solar system: A study on Iran's important water infrastructures
.
Renewable Energy
171
,
1171
–
1187
.
https://doi.org/10.1016/j.renene.2020.12.005
.
Google Scholar
Crossref
Search ADS
 
Latif
S. D.
 &
Ahmed
A. N.
2023
A review of deep learning and machine learning techniques for hydrological inflow forecasting
.
Environ. Dev. Sustainability
.
https://doi.org/10.1007/s10668-023-03131-1
.
Google Scholar
 
Latif
S. D.
,
Azmi
M. S. B. N.
,
Ahmed
A. N.
,
Fai
C. M.
 &
El-Shafie
A.
2020
Application of artificial neural network for forecasting nitrate concentration as a water quality parameter: a case study of Feitsui Reservoir, Taiwan
.
Int. J. Des. Nat. Ecodyn.
.
https://doi.org/10.18280/ijdne.150505
.
Google Scholar
 
Latif
S. D.
,
Alyaa Binti Hazrin
N.
,
Hoon Koo
C.
,
Lin Ng
J.
,
Chaplot
B.
,
Feng Huang
Y.
,
El-Shafie
A.
 &
Najah Ahmed
A.
2023
Assessing rainfall prediction models: Exploring the advantages of machine learning and remote sensing approaches
.
Alexandria Eng. J.
82
,
16
–
25
.
https://doi.org/10.1016/j.aej.2023.09.060
.
Google Scholar
Crossref
Search ADS
 
Li
H.
 &
Sansalone
J.
2022
Implementing machine learning to optimize the cost-benefit of urban water clarifier geometrics
.
Water Res.
220
,
118685
.
https://doi.org/10.1016/j.watres.2022.118685
.
Google Scholar
Crossref
Search ADS
PubMed
 
Malki
Z.
,
Atlam
E. S.
,
Hassanien
A. E.
,
Dagnew
G.
,
Elhosseini
M. A.
 &
Gad
I.
2020
Association between weather data and COVID-19 pandemic predicting mortality rate: Machine learning approaches
.
Chaos, Solitons Fractals
138
,
110137
.
https://doi.org/10.1016/j.chaos.2020.110137
.
Google Scholar
Crossref
Search ADS
PubMed
 
Moeeni
H.
 &
Bonakdari
H.
2017
Forecasting monthly inflow with extreme seasonal variation using the hybrid SARIMA-ANN model
.
Stochastic Environ. Res. Risk Assess.
31
,
1997
–
2010
.
https://doi.org/10.1007/s00477-016-1273-z
.
Google Scholar
Crossref
Search ADS
 
Mohammed
M. N.
,
Daham
B. F. A.
 &
Mohammed
K. S.
2018
Darbandikhan reservoir operation optimization using ant colony optimization algorithm
.
Zanco J. Pure Appl. Sci.
30
,
135
–
141
.
https://doi.org/10.21271/zjpas.30.s1.15
.
Google Scholar
 
Parmar
A.
,
Mistree
K.
 &
Sompura
M.
2017
Machine learning techniques for rainfall prediction: A review
.
Int. Conf. Innov. Inf. Embed. Commun. Syst.
,
Coimbatore, India, 17-18 March 2017. Curran Associates, Red Hook, NY
.
Google Scholar
 
Pham
B. T.
,
Le
L. M.
,
Le
T. T.
,
Bui
K. T. T.
,
Le
V. M.
,
Ly
H. B.
 &
Prakash
I.
2020
Development of advanced artificial intelligence models for daily rainfall prediction
.
Atmos. Res.
.
https://doi.org/10.1016/j.atmosres.2020.104845
.
Google Scholar
 
Shen
C.
2018
A transdisciplinary review of deep learning research and its relevance for water resources scientists
.
Water Resour. Res.
54
,
8558
–
8593
.
https://doi.org/10.1029/2018WR022643
.
Google Scholar
Crossref
Search ADS
 
Tinti
A.
2017
Water Resources Management in the Kurdistan Region of Iraq, The Institute of Regional and International Studies
.
Vieira
A. d. N.
,
Kleinermanns
K.
,
Martin
W. F.
 &
Preiner
M.
2020
The ambivalent role of water at the origins of life
.
FEBS Lett.
594
,
2717
–
2733
.
https://doi.org/10.1002/1873-3468.13815
.
Google Scholar
Crossref
Search ADS
PubMed
 
Yang
T. H.
 &
Liu
W. C.
2020
A general overview of the risk-reduction strategies for floods and droughts
.
Sustainability
12
,
1
–
20
.
https://doi.org/10.3390/su12072687
.
Google Scholar
PubMed
 
Zubaidi
S. L.
,
Ortega-Martorell
S.
,
Al-Bugharbee
H.
,
Olier
I.
,
Hashim
K. S.
,
Gharghan
S. K.
,
Kot
P.
 &
Al-Khaddar
R.
2020
Urban water demand prediction for a city that suffers from climate change and population growth: Gauteng province case study
.
Water (Switzerland)
12
,
1
–
17
.
https://doi.org/10.3390/W12071885
.
Google Scholar
 
© 2024 The Authors
This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY-NC-ND 4.0), which permits copying and redistribution for non-commercial purposes with no derivatives, provided the original work is properly cited (http://creativecommons.org/licenses/by-nc-nd/4.0/).