Journal Pre-proof

Development of advanced artificial intelligence models for daily

rainfall prediction

Binh Thai Pham, Lu Minh Le, Tien-Thinh Le, Kien-Trinh Thi

Bui, Vuong Minh Le, Hai-Bang Ly, Indra Prakash

PII: S0169-8095(19)31123-8
DOI: https://doi.org/10.1016/j.atmosres.2020.104845
Reference: ATMOS 104845

To appear in: Atmospheric Research

Received date: 28 August 2019

Revised date: 26 November 2019
Accepted date: 3 January 2020

Please cite this article as: B.T. Pham, L.M. Le, T.-T. Le, et al., Development of advanced
artificial intelligence models for daily rainfall prediction, Atmospheric Research(2019),
https://doi.org/10.1016/j.atmosres.2020.104845

This is a PDF file of an article that has undergone enhancements after acceptance, such
as the addition of a cover page and metadata, and formatting for readability, but it is
not yet the definitive version of record. This version will undergo additional copyediting,
typesetting and review before it is published in its final form, but we are providing this
version to give early visibility of the article. Please note that, during the production
process, errors may be discovered which could affect the content, and all legal disclaimers
that apply to the journal pertain.

© 2019 Published by Elsevier.

 Journal Pre-proof
Development of Advanced Artificial Intelligence Models for Daily

Rainfall Prediction

Binh Thai Pham1* ; Lu Minh Le 2 ; Tien-Thinh Le 3* ; Kien-Trinh Thi Bui4 ; Vuong Minh Le 2 ; Hai-

Bang Ly1* ; Indra Prakash5

1
University of Transport Technology, Hanoi 100000, Vietnam

2
Faculty of Engineering, Vietnam National University of Agriculture, Gia Lam, Hanoi 100000,

of
Vietnam

ro
3
Institute of Research and Development, Duy Tan University, Da Nang 550000, Vietnam

4
-p
Geomatics Center, Thuyloi University, Hanoi 100000, Vietnam
re
5
Department of Science & Technology, Bhaskarcharya Institute for Space Applications and Geo-
lP

Informatics (BISAG), Government of Gujarat, Gandhinagar 382007, India

*Corresponding authors: Binh Thai Pham (binhpt@utt.edu.vn); Hai-Bang Ly (banglh@utt.edu.vn);

na

Tien-Thinh Le (letienthinh@duytan.edu.vn)
ur

Abstract:
Jo

In this study, the main objective is to develop and compare several advanced Artificial Intelligent (AI)

models namely Adaptive Network based Fuzzy Inference System optimized with Particle Swarm

Optimization (PSOANFIS), Artificial Neural Networks (ANN) and Support Vector Machines (SVM)

for the prediction of daily rainfall in Hoa Binh province, Vietnam. For this, meteorological variable

parameters such as maximum temperature, minimum temperature, wind speed, relative humidity and

solar radiation were collected and used as input parameters and daily rainfall as an output parameter in

the models. Validation of the developed models was achieved using various quality assessment criteria

such as correlation coefficient (R) and Mean Absolute Error (MAE), Skill Score (SS), Probability of
1
 Journal Pre-proof
Detection (POD), Critical Success Index (CSI), and False Alarm Ratio (FAR). The results showed that

all the AI models provided reasonable predictions of daily rainfall but the SVM was found to be the

best method for predicting rainfall. This method was also found to be the most robust and efficient

prediction model while taking into account of input variability using the Monte Carlo approach. This

AI based study would be helpful in quick and accurate prediction of daily rainfall.

Keywords: Rainfall; Artificial Neural Networks; Robustness analysis; Support Vector Machines;

Adaptive Network based Fuzzy Inference System; Particle Swarm Optimization

of
1. Introduction

ro
Rainfall is one of the most crucial meteorological factor that has a direct influence on many fields

-p
including agriculture (Trinh, 2018), hydroelectric production (Haddad, 2011), and water resources
re
management (Hartmann et al., 2016a; Serinaldi and Kilsby, 2012). In fact, agriculture production of the

developing countries depends significantly on rainfall. About 65% total agriculture production of
lP

developing countries like India and Vietnam depends on monsoon rainfall (Sahai et al., 2000; Trinh,
na

2018; Le et al., 2019b). Hydroelectric generation all over the world is mainly depended on the rainfall

in the catchment area. Adequate rainfall is important for recharging depleting ground water resources
ur

(Hartmann et al., 2016a, 2016b). In addition, many catastrophic natural disasters are also directly
Jo

related to rainfall intensity and duration such as flood (Bezak et al., 2016; Bui et al., 2019; Faccini et

al., 2018; Janizadeh et al., 2019; Khosravi et al., 2019; Tien Bui et al., 2019, 2016), drought (Deo et al.,

2018; Mouatadid et al., 2018), landslide (Abedini et al., 2019; Bezak et al., 2016; Chen et al., 2017;

Dou et al., 2019; Pham et al., 2019b, 2018, 2017), local sea-level rise (Senior et al., 2002). Therefore, it

is extremely important to have an quick and accurate method of predicting rainfall intensity and

duration for proper water resources management of an area (Mislan et al., 2015). Moreover, timely

precise rainfall prediction helps in planning of agricultural activities, even in case of unusual

precipitation (Abbot and Marohasy, 2017; Navone and Ceccatto, 1994). Accurate and timely rainfall
2
 Journal Pre-proof
prediction is also important to prevent and minimize adverse effects of natural disasters such as

landslides, floods, and droughts (Bezak et al., 2016; Hung et al., 2009).

Prediction of rainfall is generally done by using operational numerical weather models in conjunction

with meteorological radars data (He et al., 2013; Lynch, 2008). These models have been widely used in

many works involving multiple regressions and climatology average methods (Abbot and Marohasy,

2014a; He et al., 2013; Lynch, 2008; Mahmud and Ross, 2005; Shao and Li, 2013; Tanessong et al.,

2014), numerical methods (Azadi et al., 2013; Novak et al., 2014), and empirical formulations

of
(Silvestro and Rebora, 2014). These traditional models are based on long-term measurement of rainfall

ro
and its statistical correlation with other meteorological factors such as air temperature, solar radiation,

cloud information, wind speed, sunshine duration, relative humidity (G.Gouda et al., 2019; Mghouchi
-p
et al., 2016; Mousavi et al., 2017; Paoli et al., 2010), and geographical parameters (i.e. latitude,
re
longitude and altitude) (Chegaar and Chibani, 2001; TürkToğrul and Onat, 1999). Various empirical
lP

equations for estimating rainfall intensity have been proposed by various researchers (AlHassoun,

2011), (Al Mamun et al., 2018) and (Awadallah et al., 2017). Niu and Zhang (2015) (Niu and Zhang,
na

2015) have proposed operational numerical weather models for rainfall forecasting over a large range
ur

of location in China to explore the influence of space inputs on the meteorological variables. In another
Jo

study, Rajeevan et al. (2007) (Rajeevan et al., 2007) have used multiple linear regression model for the

prediction of rainfall in India using six relevant predictors based on the experimental data. Linear

regression and statistical methods have also been combined in for forecasting rainfall in the Caribbean

area considering various characteristics of sea (Ashby et al., 2005). Many researchers have attempted to

improve the prediction capability of numerical weather models using different meteorological inputs

obtained from field surveys and radar data (Villarini et al., 2014; Wang et al., 2015) to reduce

possibility of error in rainfall prediction (Black, 1994; Tippett, 1999).

3
 Journal Pre-proof
Recently, Artificial Intelligent (AI) based approaches such as Artificial Neural Networks (ANN) and

Adaptive Neuro-Fuzzy Inference Systems (ANFIS) have been adopted for predicting rainfall intensity

considering climate and weather conditions by many countries such as Australia (Abbot and Marohasy,

2014b; Deo and Şahin, 2015), India (Acharya et al., 2014), Thailand (Hung et al., 2009), Greece

(Nastos et al., 2013) and Italy (Faccini et al., 2018). For this, different time period rainfall data: hourly

(Hung et al., 2009; Wei, 2013), daily (Mehdi Keshtkar et al., 2013; Nastos et al., 2014), weekly (El-

Shafie et al., 2012; Warsito et al., 2016; Le et al., 2020), monthly (Abbot and Marohasy, 2014b; Mislan

of
et al., 2015), seasonally (Garric et al., 2002) and annually (Philip and Joseph, 2003) have been analyzed

in the models.

ro
Dabhi and Chaudhary (2014) (Dabhi and Chaudhary, 2014) employed hybrid wavelet-postfix-GP
-p
model for the prediction of daily rainfall of Anand region, India, using various meteorological variables
re
such as maximum temperature, minimum temperature, evaporation index and relative humidity.
lP

Discrete wavelet transform was applied as a data preprocessing technique for reducing noise in the raw

database to improve the prediction capability. Wu et al. (2010) (Wu et al., 2010) investigated modeling
na

of daily rainfall prediction based on three parametric studies including model inputs, modeling methods
ur

and data preprocessing techniques. AI based methods such as ANN and K-nearest-neighbors, and
Jo

processing technique of principal component analysis were employed for the prediction of rainfall.

Altunkaynak and Nigussie (2015) (Altunkaynak and Nigussie, 2015) integrated Seasonally Adjusted

Series (SAS) technique in a multilayer perceptron model for rainfall prediction based on data collected

from two stations in Turkey. Partal et al. (2015) (Partal et al., 2015) utilized daily mean temperature,

daily max temperature, daily min temperature, daily total specific humidity, daily total evaporation

predicting daily total precipitation using different AI techniques such as feed forward back propagation,

radial basis function and generalized regression neural network. Use of Support Vector Machine

(SVM) in daily rainfall prediction was investigated by Ortiz-García et al.(2014) (Ortiz-García et al.,

4
 Journal Pre-proof
2014) involving meteorological variables such as temperature, wind speed, wind direction, and

humidity. In this study, SVM model exhibited greater prediction performance than other alternative AI

approaches namely multi-layer perceptron, extreme learning machine, decision trees and K-nearest

neighbor models.

In terms of hybrid AI models, which are combination of single AI and meta-heuristic algorithms,

Nasseri et al. (2008) (Nasseri et al., 2008) combined neural network model and genetic algorithm for

rainfall prediction based on data from rain gauges in the Upper Parramatta catchment, Sydney,

of
Australia. Real coded genetic algorithm has also been applied for optimizing weight parameters of

ro
ANN for daily rainfall–runoff prediction in Ourika basin, Morocco (Sedki et al., 2009). Even though

many works using the AI based algorithms have been done on the rainfall prediction, there is still scope
-p
of improving the accuracy and effectiveness of the models. In addition, prediction of rainfall remains a
re
big challenge as meteorological parameters are of stochastic in nature (Nielsen et al., 2014; Villarini et
lP

al., 2014). Rainfall in any area depends on various meteorological parameters such as temperature,

humidity, wind besides time and space (Abbot and Marohasy, 2017; Navone and Ceccatto, 1994;
na

Zellou and Rahali, 2019).

ur

Till now, most of the developed AI rainfall prediction models are of deterministic type which needed
Jo

sensitivity analysis to evaluate their robustness considering variable inputs (Khosravi et al., 2010). If

the model is not robust enough, uncertainties could be produced in the predicted output (Hong et al.,

2006; Kasiviswanathan and Sudheer, 2013). Moreover, sensitivity analysis can provide relevant

information about the complex relationship between various meteorological variables during running of

AI models and can also identify irrelevant parameters which may not be required as input in the model.

The main objective of the present study is to develop and compare several advanced Artificial

Intelligent (AI) models namely Adaptive Network based Fuzzy Inference System optimized with

Particle Swarm Optimization (PSOANFIS), ANN and SVM for the prediction of daily rainfall in Hoa
5
 Journal Pre-proof
Binh province, Vietnam. Significance of the present study is that we have developed and applied first

time the hybrid model: PSOANFIS for accurate prediction of daily rainfall. Moreover, in this study, the

influence of randomness of the database on the prediction capability of the AI models was investigated

and analyzed using Monte Carlo technique as well as statistical analysis. To construct and validate the

predictive models, daily meteorological data (minimum temperature, maximum temperature, wind

speed, relative humidity, solar radiation and rainfall) of the study area was used. Validation of the

models was done using various criteria such as Mean Absolute Error (MAE), Correlation Coefficient

of
(R), Skill Score (SS), Probability of Detection (POD), Critical Success Index (CSI), and False Alarm

Ratio (FAR).

ro
2. Materials and Methods

2.1. Methods used

-p
re
2.1.1. Artificial Neural Networks (ANN)
lP

ANN is a computational model based on the structure and functions of biological neural networks
na

(McCulloch and Pitts, 1943). In fact, the neural network itself is not an algorithm, it is rather a

framework in which many machine learning algorithms can work and process complex input data.
ur

ANN algorithm does not need to define any conditions on the input data. Once the learning process has
Jo

been done, ANN can successfully process new data or predict the outcome of a problem without any

given input data. There are three main layers in a neural network: input layers, hidden layers and output

layer (Fig 1) (Narayanakumar and Raja, 2016). There are many algorithms for training an ANN. In this

study, the scaled conjugate gradient algorithm was applied for training the neural network as it

accelerates the convergence by guiding the search in the conjugate directions (Møller, 1991).

6
 Journal Pre-proof

of
ro
Fig 1. Architecture of ANN used in this study.

-p
2.1.2. Adaptive Network Based Fuzzy Inference System (ANFIS)
re
ANFIS is a combination between the learning rules of adaptive networks and a fuzzy inference system
lP

in order to make precise predictions in many aspects of human knowledge. Out of these, the inference
na

system is based on if-then rules (Takagi and Sugeno, 1983) and the adaptive networks system is based

on the gradient descent and the chain rule introduced (Werbos, 1974). Basically, the ANFIS structure
ur

consists of five main layers (Fig 2), each layer contains node functions of the same function family
Jo

(Jang, 1993). An overview of these layers are presented in many literatures (Dao et al., 2019a; Ly et al.,

2019b, 2019d; Nguyen et al., 2019a).

7
 Journal Pre-proof

of
Fig 2. Illustration of basic ANFIS structure with two input parameters.

ro
2.1.3. Particle Swarm Optimization (PSO)
-p
PSO is an efficient-swarm intelligence technique proposed and developed based on the social behavior
re
of bird flocks for solving complex optimization problems (Kennedy and Eberhart, 1995; Shi and
lP

Eberhart, 1998). PSO defines swarm and particle as the population of the potential solution and each
na

individual in that, respectively. The particles fly in the domain space according to some simple

formulae to find their best positions, which are the optimized solutions of an optimization problem (Bui
ur

et al., 2018; Le et al., 2019a; Tien Bui et al., 2016).

Jo

2.1.4. Support Vector Machines (SVM)

SVM is a supervised learning algorithm that analyzes data for classification and regression, which was

developed using the idea of statistical learning theory (CORTES and VAPNIK, 1995; Vapnik, 1998).

The main objective of SVM algorithm is to find a hyperplane in an N dimensional space that distinctly

separates data points, maximizing the distance between two datasets (Fig 3). In SVM, support vectors

are data points that are close to the hyperplane, which contribute to the determination of the position

and the orientation of the hyperplane. This hyper plane is in a feature space induced by a kernel K,

8
 Journal Pre-proof
which defines a dot product in that space (Evgeniou and Pontil, n.d.; Wahba, 1990). Thus, SVM try to

maximize the distance between the hyperplane and the data points through a cost function for solving

the regression problem.

of
ro
-p
re
Fig 3. Illustration of SVM algorithm.
lP

2.1.5. Validation criteria

na

In the present work, various validation criteria such as Correlation Coefficient (R) and Mean Absolute

Error (MAE), Score Skill (SS), Probability of Detection (POD), Critical Success Index (CSI) and False
ur

Alarm Ratio (FAR) were used to validate the developed AI models. Out of these, R allows us to
Jo

identify the statistical relationship between two actual and predicted values (Pearson Karl and Galton

Francis, 1895). It yields a value between 0 and 1 inclusive, where 0 is no correlation and 1 is total

correlation (Ly et al., 2019c; Dao et al., 2019b; Pham et al., 2020). In the case of MAE, which has the

same units as the quantity being estimated, low value of MAE basically indicates good accuracy of

prediction output using the models (Montavon et al., 2013; Ly et al., 2019e; Nguyen et al., 2019b;

Thanh et al., 2020). In term of SS, it is given by the following formula (Murphy, 1988; Benedetti,

2010; Wilks, 2011):

9
 Journal Pre-proof
A-A ref
SS= ×100, (1)
A perf -A ref

where A is the measure of accuracy (i.e. MAE or R) of the proposed forecast, Aref is the set of reference

forecasts and Aperf is a perfect forecast. In this case, Aperf is taken as what actually happened. SS has a

range of negative infinity to 100. A positive value of SS means the proposed forecast is an

improvement over the reference forecast. On the other hand, a negative value of SS means that the

proposed forecast has smaller performance than the reference forecast. In this study, the reference

forecast was chosen as Global Forecast System (GFS) produced by United States' National Weather

of
Service (Kanamitsu et al., 1991).

ro
With contingency scores, the formulation for calculating POD, CSI and FAR for a given threshold are

-p
expressed as follows (Scharf and Demeure, 1991; Gneiting and Raftery, 2007):
re
Hits
POD= , (2)
lP

Hits+Misses

Hits
(3)
na

CSI= ,
Hits+Misses+FalseAlarms
ur

FalseAlarms
FAR= (4)
Hits+FalseAlarms
Jo

where, Number of Hits, Number of Misses, Number of False Alarms and Number of Correct Negative

are defined in Table 1.

Table 1. Description of contingency table.

Contingency table Observed

for a given threshold Rain No rain

Number of False
Forecast Rain Number of Hits
Alarms

10
 Journal Pre-proof
Number of Number of
No rain
Misses Correct Negative

2.1.6. Monte Carlo Simulation

In this study, Monte Carlo method which is highly efficient for propagating variability of input

parameters through the models (Robert and Casella, 2004; Ly et al., 2019a; Hun et al., 2019) has been

used in order to investigate the robustness of AI three models: ANN, PSOANFIS and SVM. In this

of
method, calculation of output is realized by repeating randomly inputs following the probability density

ro
function of each (Mordechai, 2011). In this way, statistical behavior of each input is fully propagated

-p
on the output through the models. Statistical analysis of the output is then determined in order to
re
explore the influence of input variability and the robustness of applied models (Goffart et al., 2017;

Pham et al., 2019a; Ly et al., 2019d; Yuan et al., 2015).

lP

In Monte Carlo simulation, statistical analysis of output requires usually a convergence of variability
na

propagation, i.e. an optimal number of realizations. In this work, statistical convergence function
ur

CONV of the mean of the output result has been calculated using the following equation (Guilleminot

et al., 2013; Le et al., 2016; Soize et al., 2015):

Jo

n MC
1
CONV(n MC ) 
n MC
 Output ,
i1
i
(5)

where n MC is the number of Monte Carlo realizations. Optimal value of nMC could demonstrate a time-

to-convergence to the stationary solution. Various other statistical parameters such as mean, median,

standard deviation and quantiles were also used in this paper in order to fully quantify behavior of

probability distribution. Q25 , Q 50 and Q 75 denote, respectively, the 25th , 50th and 75th quantiles. For

illustration, Q25 splits off the highest 75% of data values from the lowest 25%.

11
 Journal Pre-proof
2.2. Data used

In this study, rainfall data was collected from meteorological gauge located at Latitude 20.763m and

Longitude 105.312m in the Cao Phong district, Hoa Binh province, Vietnam (Fig 4). Weather data was

obtained from the Global Weather Data for SWAT (Dile and Srinivasan, 2014; Fuka et al., 2014)

(available at: https://globalweather.tamu.edu). Hoa Binh province is a tropical moon region with two

separate seasons: dry season (November to April) and rainy season (May to October). Max.

Temperature is in July (32.4°C) whereas min. Temperature (°C) is in January (12.8°C). Highest

of
monthly rainfall is in September (332mm) (Table 2). This province has the largest hydroelectric dam in

ro
Vietnam with 128 m in height, and 970 m in length provided the required electricity for the most

northern province of Vietnam. The present study is important as the rainfall is the main water source of
-p
the Black River Reservoir where the water is stored for the dam operation.
re
Table 2. Monthly meteorological characteristics of Hoa Binh province (https://en.climate-
lP

data.org/asia/vietnam/hoa-binh-province/hoa-binh-4274/#climate- graph).
na

Months
Parameters
January February March April May June July August September October November December
ur

Avg. 16.5 17.8 20.3 24.2 27.7 29 28.6 28.5 27.3 24.8 21.8 18.4
Jo

Temperature

(°C)

Min. 12.8 14.4 16.7 20.5 23.4 25.1 24.9 24.9 23.8 20.7 17.8 14.4

Temperature

(°C)

Max. 20.3 21.3 23.9 28 32 32.9 32.4 32.1 30.9 29 25.9 22.5

Temperature

(°C)

Rainfall 11 26 26 75 216 228 315 316 332 134 39 17

12
 Journal Pre-proof
(mm)

In modeling, a total of five parameters namely maximum temperature, minimum temperature, wind

speed, relative humidity, and solar radiation were used as input variables and daily rainfall data was

used as an output variable for generating training and testing datasets. In total 3653 data samples were

collected during the period January 01, 2004 to December 31, 2013.

Initial analysis of the data used is presented in Table 3. It can be observed that the daily rainfall varies

from 0 mm (that is no rain) to 59.1507 mm with mean value of 7.4195 mm and standard deviation of

of
8.8599 mm. The maximum temperature ranges from 8.0804°Cto 42.2402°C with mean value of

ro
26.8487°C and standard deviation of 6.4102°C. The minimum temperature varies from -0.8650°C to

-p
27.2890°C with mean value of 17.7903°C and standard deviation of 5.0405°C. The wind speed ranges

from 0.6414 m/s to 2.8707m/s with mean value of 1.3493m/s and standard deviation of 0.3070 m/s.
re
The relative humidity varies from 0.2198 to 0.9822 with mean value of 0.8260 and standard deviation
lP

of 0.1063. The solar radiation ranges from 1.5933 MJ/m2 to 29.5522 MJ/m2 with mean value of

15.3547 MJ/m2 and standard deviation of 5.8087 MJ/m2 .

na
ur
Jo

13
 Journal Pre-proof

of
ro
-p
re
lP

Fig 4. Location of rainfall station in Hoa Binh province, Vietnam

na

Table 3. Initial analysis of data used

ur

Parameters Minimum Maximum Average Standard deviation

Jo

Maximum Temperature (o C) 8.0804 42.2402 26.8487 6.4102

Minimum Temperature (o C) -0.8650 27.2890 17.7903 5.0405

Wind speed (m/s) 0.6414 2.8707 1.3493 0.3070

Relative Humidity (fraction) 0.2198 0.9822 0.8260 0.1063

Solar radiation (MJ/m2 ) 1.5933 29.5522 15.3547 5.8087

14
 Journal Pre-proof

Daily rainfall (mm) 0.0000 59.1507 7.4195 8.8599

3. Methodology

of
ro
-p
re
lP
na
ur
Jo

15
 Journal Pre-proof

of
ro
-p
re
lP
na
ur
Jo

Fig 5. Methodology flow chart of the present study.

16
 Journal Pre-proof
Modeling of prediction of daily rainfall using AI models was carried out in four main steps (Fig 5): (1)

generation of datasets, (2) construction of the models, (3) validation of the models, and (4) robustness

analysis.

Step 1, Generation of datasets: In this step, the data collected was randomly split into two parts. One

part includes 70% of the data, which was then used to create the training dataset. Another part consists

of 30% remaining data, which was used to create the testing dataset.

Step 2, Construction of the models: In this step, the training dataset was used to train and construct the

of
models. With PSOANFIS model, PSO optimization was used with 25 particles, the inertia weight of

ro
0.4 and 1000 iterations to optimize the consequent and antecedent parameters of ANFIS model. Using

-p
the optimal parameters, ANFIS was trained with the Gaussian membership function and the c-means

clustering algorithm. In ANN, the scaled conjugate gradient algorithm was used to train the model, a
re
trial-and-error test was applied to select the best number of hidden layers of 10. To construct SVM,
lP

third degree polynomial kernel function was chosen, the regularization constant (c) of box constraint
na

and epsilon parameter were selected as 0.1112and 0.0111, respectively.

Step 3, Validation of the models: The constructed models were validated using both training and
ur

testing datasets. Various criteria namely R, MAE, SS, CSI, POD, and FAR were used to validate and
Jo

compare the predictive capability of the models.

Step 4, Robustness analysis: Monte Carlo simulation was prepared in order to investigate the

robustness of AI methods. In the Monte Carlo simulations, 1000 runs were performed using each AI

method. A total of 1000 Monte Carlo runs were then executed using all three AI methods in order to

compare statistical information. Statistical and convergence analysis were finally carried out to deduce

efficient feedback.

4. Results and analysis

17
 Journal Pre-proof
4.1. Validation of the AI models

4.1.1. Using R, MAE and error analysis

Performance of three AI prediction models was evaluated and presented in Fig 6 and Table 4. Figs 6a,

b, and c present output results of ANN, PSOANFIS and SVM, associated with the training part (70% of

data) in function of the corresponding targets. Figs 6d, e, and f present output results of ANN,

PSOANFIS and SVM, associated with the testing part and the corresponding testing targets. Linear

equations “Predicted = a*Target +b” were also presented in each case to estimate the correlation of the

of
results. Figs 6a, b and c present a slope a = 0.717, a = 0.758 and a = 0.659 for the linear fit equation,

ro
showing that the angles between the linear fit and horizontal line were 35.64°, 37.16° and 33.38° for

-p
training ANN, PSOANFIS and SVM, respectively. This indicates that PSOANFIS provided the closest

linear fit to the diagonal line compared to ANN and SVM. This was also confirmed by the values of R
re
(Table 4). For the training part, PSOANFIS gave the highest value of R compared to ANN and SVM
lP

(i.e. R=0.844, 0.861, 0.850 for ANN, PSOANFIS and SVM, respectively). On the contrary, SVM
na

provided the best training result as per the MAE criterion (i.e. MAE=3.171, 3.141, 2.853 for ANN,

PSOANFIS and SVM, respectively).

ur
Jo

18
 Journal Pre-proof

of
ro
-p
re
lP
na

Fig 6. Predicted versus target rainfall for the training dataset using: (a) ANN, (b) PSOANFIS, and (c)
ur

SVM. Predicted versus target rainfall for the testing dataset using: (d) ANN, (e) PSOANFIS, and (f)
Jo

SVM.

For the testing dataset, SVM algorithm exhibited the best prediction results with respect to both MAE

and R criteria (MAE = 3.209, 3.281, 2.728 and R = 0.829, 0.844, 0.863 using ANN, PSOANFIS and

SVM, respectively). However, the slope of the linear equation using SVM is not the case of smallest

deviation regarding to the diagonal line. As shown in Fig 6, the slopes are a = 0.683, 0.819 and 0.733

using ANN, PSOANFIS and SVM, respectively. This indicates that the angles between linear equation

and horizontal line are 34.33°, 39.32° and 36.24°, respectively.

19
 Journal Pre-proof
Table 4. Summary information of prediction capability of three AI models.

Part Method MAE (mm) R

Training ANN 3.171 0.844

PSOANFIS 3.141 0.861

SVM 2.853 0.850

of
Testing ANN 3.209 0.829

ro
PSOANFIS 3.281 0.844

SVM -p 2.728 0.863

re
Histogram of errors between predicted outputs and targets for the testing dataset is also plotted in Fig 7
lP

where error e is defined using the equation: e=Predicted-Target .

na
ur
Jo

Fig 7. Histogram of error between predicted result and corresponding target using three AI models for

the testing dataset.

20
 Journal Pre-proof
As shown in Fig 7, SVM method exhibited a highest peak and centered at zero. Statistical information

of histograms of e is also indicated in Table 5. The mean and median values of e are -0.065, 0.227

using ANN, 0.317, 0.542 for PSOANFIS and -0.333, 0.224 for SVM, respectively. Based on these

values, ANN and SVM provided the closest mean and median values with respect to zero error. SVM

model presented lowest variation of error distribution with respect to all Q 25 , Q 75 and standard

deviation (StD) values. In the case of Q 25 , ANN, PSOANFIS and SVM presented values of -1.600, -

1.829, and -1.441, respectively. As regards to Q 75 , ANN, PSOANFIS and SVM produced values of

of
2.236, 2.574, and 1.544, respectively. Considering StD, ANN, PSOANFIS and SVM gave values of

4.935, 4.634, and 4.244, respectively. These observations confirmed the highest concentration of error

ro
distribution around zero value with SVM model. From overall analysis of error distribution, SVM gave

-p
smallest deviation between prediction output and target rainfall.
re
Table 5. Error distribution of the AI techniques.
lP

Method Q25 (mm) Q50 (mm) Q75 (mm) Mean (mm) StD (mm)
na

ANN -1.600 0.227 2.236 -0.065 4.935

ur

PSOANFIS -1.829 0.542 2.574 0.317 4.634

Jo

SVM -1.441 0.224 1.544 -0.333 4.244

4.1.2. Using SS criteria

The accuracy of the proposed AI models is referenced to the accuracy of the GFS forecast. As the size

of GFS data is very important (i.e. approximately 1Gb for one day information under GRB2 format, 1°

grid size), it is not possible to request a subset of the archived database; thus, only GFS data from June

21
 Journal Pre-proof
01, 2010 to August 31, 2010, were retrieved for calculation of SS. Indeed, based on the experimental

data, all 92 days from June 01, 2010 to August 31, 2010 had the rain at the study area (the minimum

rainfall recorded was 1.27mm). In addition, the average measured rainfall was 6.27, 13.92 and 19.18

mm for June, July and August, 2010, respectively. Therefore, we only compared the proposed AI

models and the reference forecast under a real rainy condition. Consequently, there were around 90 Gb

of GRB2 files which have been downloaded from the NOAA database. GRB2 files were visualized

using zyGrib (Version 8.0.1) and Matlab (Version R2018). The cumulative daily rainfall was down

of
scaled from the global GFS forecast map at the location of the study area (i.e. longitude 105.312 and

latitude 20.763). Fig 8 presents an example of visualization of GFS rainfall forecast for Northern region

ro
of Vietnam including study area.

-p
re
lP
na
ur
Jo

Fig 8. GFS rainfall forecast map for Northern region of Vietnam (August 24, 2010, at 12h), the study

area is marked by red square. GFS rainfall forecast 24h ahead is 5.66 mm/h at the study area.

Fig 9 shows the evaluation of daily rainfall from June 1, 2010 to August 31, 2010, including the

measured data and the predicted values by ANN, PSOANFIS, SVM and GFS. Table 6 indicates the

values of SS combining the use of three AI models and two quality assessment criteria such as MAE

and R. Based on MAE, the values of SS are 62.56, 67.17, and 66.16 using ANN, PSOANFIS and

22
 Journal Pre-proof
SVM, respectively. Based on R, the values of SS are 71.59, 80.19, and 71.83 using ANN, PSOANFIS

and SVM, respectively. It is seen that all values of SS given in Table 6 are positive, indicating that the

proposed AI models exhibited an improvement compared to the reference forecast. However, it should

also be noticed that such calculation of SS was performed only for three months in summer of 2010. A

longer range of days data analysis (for instance, by trying to overpass the GFS archived data size

inconvenience) needs to be evaluated for comparison of the results.

of
ro
-p
re
lP
na
ur
Jo

Fig 9. Daily rainfall from June 2010 to August 2018, from measured data, predicted by ANN,

PSOANFIS, SVM and GFS, respectively.

Table 6. Values of SS of the models based on different criteria such as MAE and R.

Value of SS (%) Based on MAE Based on R

ANN 62.56 71.59

PSOANFIS 67.17 80.19

SVM 66.16 71.83

23
 Journal Pre-proof
4.1.3. Using contingency scores

Contingency scores (POD, CSI and FAR) were evaluated to assess performance of the AI models. The

daily rainfall threshold values range from 2 to 30 mm. Fig 10 presents the evaluation of CSI, POD and

FAR in function of daily rainfall threshold whereas all these values of the models are indicated in Table

7. As shown in Figs10a and 10b, the values of CSI and POD decrease on increasing the value of

threshold for all three AI models. In addition, the evaluation result doesn’t follow a linear

approximation. At 2 mm of threshold, the values of CSI are 0.76, 0.75 and 0.78; the values of POD are

of
0.91, 0.94 and 0.89; the values of FAR are 0.18, 0.21 and 0.14, using ANN, PSOANFIS and SVM,

ro
respectively. Whereas at 10 mm of threshold, the values of CSI decrease to 0.69, 0.70 and 0.70; the

values of POD decrease to 0.85, 0.80 and 0.81; the values of FAR are 0.21, 0.15 and 0.15, using ANN,
-p
PSOANFIS and SVM, respectively. Based on the observation of contingency scores at different
re
thresholds, it could be deduced that all three AI models exhibit good performance for threshold ranging
lP

from 2 to 18 mm. It is also interesting noticed that SVM model provides lowest FAR from 2 to 26 mm

of threshold.
na
ur
Jo

24
 Journal Pre-proof

Fig 10. Evaluation of contingency scores such as (a) CSI, (b) POD and (c) FAR in function of

of
threshold.

ro
Table 7. Values of contingency scores of the models at different thresholds.

Threshold -p
re
ANN PSOANFIS SVM
(mm)
lP

CSI POD FAR CSI POD FAR CSI POD FAR

2 0.76 0.91 0.18 0.75 0.94 0.21 0.78 0.89 0.14

na

4 0.71 0.91 0.23 0.70 0.91 0.25 0.75 0.88 0.17

6 0.70 0.89 0.23 0.71 0.86 0.19 0.73 0.86 0.17

ur

8 0.71 0.88 0.21 0.71 0.82 0.16 0.73 0.84 0.16

Jo

10 0.69 0.85 0.21 0.70 0.80 0.15 0.70 0.81 0.15

12 0.66 0.83 0.23 0.68 0.80 0.18 0.68 0.79 0.17

14 0.61 0.78 0.27 0.64 0.77 0.21 0.63 0.74 0.20

16 0.54 0.72 0.31 0.58 0.73 0.26 0.57 0.69 0.24

18 0.49 0.65 0.32 0.53 0.68 0.30 0.51 0.60 0.24

20 0.43 0.56 0.35 0.48 0.68 0.37 0.39 0.49 0.34

22 0.36 0.47 0.37 0.47 0.67 0.40 0.36 0.42 0.30

24 0.31 0.38 0.37 0.40 0.63 0.47 0.28 0.31 0.24

25
 Journal Pre-proof
26 0.20 0.23 0.38 0.38 0.58 0.48 0.16 0.18 0.35

28 0.15 0.16 0.21 0.33 0.47 0.48 0.08 0.08 0.45

30 0.13 0.14 0.12 0.24 0.27 0.35 0.06 0.06 0.22

4.2. Robustness analysis of the AI models

Robustness of the AI models was conducted to evaluate prediction capability of these models including

selection of training dataset. In this study, 70% of data were randomly taken for the development of

ANN, PSOANFIS and SVM algorithms. In order to investigate the robustness of each AI technique

of
under input variability, 1000 Monte Carlo simulations were performed to take into account random

ro
combinations of data index. Therefore, 1000 values of R and MAE criteria were obtained by

calculating the deviation between the testing target and the corresponding outputs for each model.
-p
Statistical analysis of R and MAE distributions was conducted to finally provide significant feedback to
re
deduce the robustness of each AI model with the presence of input variability.
lP

The statistical convergence of the mean of R and MAE was first investigated in order to determine the
na

optimal number of Monte Carlo simulations as shown in Fig 11. A large variation of both R and MAE

was observed while using ANN technique along with a small number of nMC (i.e. nMC< 150). It means,
ur

output obtained by ANN exhibited a high order of fluctuation (i.e. a low degree of robustness in the
Jo

presence of input variability). The stationary solution using ANN was found at nMC=500 and 200 for R

and MAE criteria, respectively. In the cases of PSOANFIS and SVM, the amplitude of variation was

smaller than that of the case using ANN. Consequently, the rate of convergence of these two models

was better than ANN algorithm. The optimal numbers nMC were approximately 400, 200 for R using

PSOANFIS and SVM, respectively, whereas they were 150 and 100 in the case of MAE for

PSOANFIS and SVM, respectively. It is worth noting that for both R and MAE criteria, SVM method

provided rather small variation (high degree of robustness) and quick convergence (time-consuming) to

the stationary result, which could indicate that SVM technique is the most stable prediction algorithm

26
 Journal Pre-proof
while accounting for input uncertainty. Besides, the number of 1000 realizations were proven to be

sufficient in order to obtain reliable results of statistical analysis.

of
ro
Fig 11. Statistical convergence function using the AI models: (a) R and (b) MAE

-p
Normalized histograms of 1000 values of R and MAE criteria are presented in Fig 12 corresponding to
re
1000 Monte Carlo simulations. Quantitative summary of statistical information is also presented in
lP

Table 8. In the case of R, the values of quantile Q 25 were 0.817, 0.836, 0.843, the values of median

were 0.841, 0.846, 0.849, values of quantile Q 75 were 0.854, 0.854, 0.855 and the values of mean were
na

0.829, 0.844, 0.849 for ANN, PSOANFIS and SVM algorithms, respectively. These results indicate
ur

that SVM model could provide better prediction performance compared to others. It is also noticed that

SVM model contributed a lowest value of dispersion of R (StD = 0.009) compared to ANN and
Jo

PSOANFIS algorithms (StD = 0.039 and 0.014, respectively).

27
 Journal Pre-proof
Fig 12. Normalized histograms after 1000 Monte Carlo realizations: (a) R (with a resolution of 0.01)

and (b) MAE (with a resolution of 0.05).

Similar observations were also for the statistical results of MAE. The MAE values of Q 25 , Q 50 , Q 75 and

mean were 3.004, 3.164, 3.490, 3.294 for ANN method, whereas that of PSOANFIS model were 3.069,

3.145, 3.233, 3.155, respectively. The lowest values of Q 25 , Q 50 , Q 75 and mean of MAE were observed

using SVM method (i.e. 2.782, 2.844, 2.911 and 2.846, respectively). Standard deviation of MAE in

the case of SVM was also the smallest value, StD = 0.097, compared with 0.407, 0.129 for ANN and

of
PSOANFIS, respectively.

ro
Table 8. R and MAE distribution of the AI models.

Criteria Method Q25 -p

Q50 Q75 Mean StD
re
R ANN 0.817 0.841 0.854 0.829 0. 039
lP

PSOANFIS 0.836 0.846 0.854 0.844 0.014

na

SVM 0.843 0.849 0.855 0.849 0.009

ur

MAE ANN 3.004 3.164 3.490 3.294 0.407

Jo

(mm)
PSOANFIS 3.069 3.145 3.233 3.155 0.129

SVM 2.782 2.844 2.911 2.846 0.097

5. Discussion

Accurate prediction of rainfall is one of the key problems for proper water resource management, flood

control and agriculture. In this study, three advanced AI techniques namely PSOANFIS, ANN and

SVM were developed and applied models in the Hoa Binh province, Vietnam to compare for daily

28
 Journal Pre-proof
rainfall prediction. Results of these model studies were compared to select best model. Comparative

results of these models along with the results of other model studies are presented in the Table 9.

Analysis of results indicated that performance of these models in comparison to previous model studies

for prediction of rainfall 24h ahead (i.e. RSVM=0.829, MAESVM=2.728 mm, PODSVM(2mm)=0.89,

CSISVM(2mm)=0.78, FARSVM(2mm)=0.14) is very good (Table 9).

Rainfall is a dichotomous variable. Therefore, validation of the ability of the proposed AI models for

rainfall prediction for the transit period from a non-rain to a rainy day is desirable to indicate the

of
strengths and/or weaknesses of the models. In view of this, evaluation of daily rainfall for four years

ro
from 2010 to 2013 was done for measured and predicted data by ANN, PSOANFIS and SVM (Fig 13).

In the study area, the non-rain period (i.e. daily rainfall lower than 10 mm) typically occurs from
-p
November to April, whereas the rain period between May and October. Based on the plotted time series
re
rainfall data, the three proposed AI models, especially SVM, exhibit strong ability to track behavior of
lP

rain process in passing from a non-rain period to a rain period. Meteorological record indicated that the

period January to April in the year 2010 was non-rain period. During this period, ANN, PSOANFIS
na

and SVM models also predicted very small quantity of rainfall (except a high peak of rainfall was
ur

observed in January month as rainfall is a dichotomous variable). For the period April to October in the
Jo

year 2010, and from July to October (wet season), the predictive ability of precipitation of proposed AI

models was very good. For the years 2011, 2012 and 2013, prediction performance of the proposed AI

models when passing from a non-rain period to a rainy day was also very good.

29
 Journal Pre-proof
Table 9. Comparison between this study and previously published works for short-term rainfall prediction.

Reference AI methods Lead Input variables Space scale Nr of Metrics used Values of metrics
used time stations
(Kuligows Back- 6h, 24h Moisture: total column precipitable Medium- 4 R, RMSE, RMSE=2.552 mm
ki and propagation water, column-average relative scale Threat score R=0.549
Barros, neural humidity, pressure levels, (Youghioghe Threat scrore6h max=0.32
1998) network equivalent potential temperature ny basin and Threat scrore24h max=0.47
Vertical lift: vertical velocity, Swatara
zonal/meridional velocity,
horizontal divergence, absolute
vorticity advection, thermal
Creek basin,
USA)

o f
ro
advection, vertical lapse rates
(Sedki et ANN-back- 24h Rainfall and runoff daily data Local 3 RMSE, R2 RMSEGABP =0.162 mm/day
al., 2009) propagation
(BP);
(Ourika,
Morocco)
- p R2 GABP =0.91
RMSEBP =0.199 mm/day

e
ANN-genetic R2 BP =0.87

(Hung et
algorithm-BP

ANN 1h, 2h, Cloudiness, relative humidity, wet

P r
Local 75 R, RMSE, R1h =0.88, R2h =0.77, R3h =0.67,
al., 2009) 3h, 4h,
5h, 6h
bulb temperature, dry bulb

speed
a l
temperature, air pressure, wind
(Bangkok,
Thailand)
Efficiency
Index
R4h =0.64, R5h =0.62, R6h =0.60
RMSE1h =0.87 mm, RMSE2h =1.36
mm, RMSE3h =1.72 mm,

r n RMSE4h =1.85 mm, RMSE5h =1.88

mm, RMSE6h =1.93 mm
(PARTAL
and
Wavelet-ANN 24h

u
Mean temperature, daily max
temperature, min temperature, total
Global
(Turkey)
28 MSE, R2 R2 min =0.701
R2 max=0.882

Jo
CIGIZOG specific humidity, total evaporation MSEmin =2.12 mm2
LU, 2009) MSEmax=9.37 mm2

(Wu and ANN-Moving 24h Max temperature, min temperature Local 1 RMSE, RMSEANN-MA =5.3 mm.
Chau, Average; (Wuxi, Willmott’s RMSEANN-SSA =5.8 mm
2013) ANN-Singular China) index
Spectrum

(Dabhi and Wavelet- 24h Max temperature, min temperature, Local - MAE, MSE, MAEGP =7.02 mm
Chaudhary PostFix- mean temperature evaporation, (Anand, R, Adjusted MSEGP =49.25 mm2
, 2014) Genetic relative humidity India) fitness RGP =0.67
Programming; MAEW-ANN =7.23 mm
Wavelet ANN MSEW-ANN =52.66 mm2
RW-ANN =0.69

30
 Journal Pre-proof
(Sánchez- Hierarchical 24h Water vapor: total precipitable Local 1 Accuracy Accuracy=80.440
Monedero nominal– water, equivalent potential (Santiago de Geometric GM=35.470
et al., ordinal temperature, humidity, pressure Compostela, mean (GM) AMAE=0.710 mm/6h
2014) support vector levels Spain) Average of the
classifier Vertical movements: temperature, Mean
wind speed, wind direction, CAPE, Absolute Error
CIN (AMAE)
(Altunkayn Wavelet - 24h Rainfall Local 2 RMSE, R2 , R2 Develi =0.952, CEDeveli =0.921
ak and Multilayer (Develi and CE, Skill RMSEDeveli =0.907 mm/day
Nigussie, Perceptron - Tomarza, Score R2 Tomarza=0.941, CETomarza=0.919
2015)

(Sehad et
Seasonalgorith
m
Support vector 3h, 24h Optical and microphysical
Turkey)

Medium- 304
o f
R, RMSE,
RMSETomarza=0.968 mm/day

R3h =0.78, RMSE3h =1.27mm,

al., 2017) machine properties of clouds: cloud phase,
optical thickness, effective particles
scale
(Northern
r o MAE, Bias Bias 3h =0.24 mm, MAE3h = 0.64mm
R24h =0.94, RMSE24h =2.23mm,
radius, cloud top temperature, cloud
top height;
Algeria,

-
domain with
p Bias 24h =0.61 mm, MAE24h =1.14mm

Cloud texture: water vapor

channels, IR channels, visible
channels and near infrared channel
r
250 km)
e
a radius of

(Asanjan
et al.,
Recurrent
neural
From
0.5h to
Longwave IR channel of GOES

l
provided by Climatic Prediction
P
Medium-
scale
- R, RMSE,
POD, CSI,
For state of Oregon:
RMSE1h =3.18mm/h, R1h =0.65,

a
2018) network; 6h Center (Oregon, FAR POD1h =0.68, CSI1h =0.5, FAR1h =0.35
Long every Oklahoma, RMSE6h =2.20mm/h, R6h =0.5,
Short‐Term
Memory -
0.5h

r n and Florida,
USA)
POD6h =0.35, CSI6h =0.35,
FAR6h =0.45
PERSIANN

u
Jo
algorithm
(Jing et al., Multi-Level 0.5, 1 Raw CINRAD/SA radar Medium- 5 POD, FAR, POD1h =0.6937, FAR1h =0.2404,
2019) Correlation and observations scale CSI, Heidke CSI1h =0.5810, HSS1h =0.6779
Long Short- 1.5h (Hangzhou, skill score POD1.5h =0.6538, FAR1.5h =0.2818,
Term Memory Nanjing, (HSS) CSI1.5h =0.5348, HSS1.5h =0.6298
Xiamen,
Changsha,
and Fuzhou,
China)
Our study ANN 24h Min temperature, max temperature, Local (Hoa 1 R, MAE, RSVM=0.829
PSOANFIS solar radiation, wind speed, relative Binh, POD, CSI, MAESVM=2.728 mm
SVM humidity Vietnam) FAR, PODSVM(2mm)=0.89
Robustness CSISVM(2mm)=0.78
analysis FARSVM(2mm)=0.14

31
 Journal Pre-proof

o f
r o
- p
r e
l P
na
u r
Jo

32
 Journal Pre-proof

of
ro
-p
re
lP
na
ur
Jo

33
 Journal Pre-proof

of
ro
-p
re
lP

Fig 13. Evaluation of daily rainfall for four years from 2010 to 2013 based on measured data and
na

predicted data of the AI models.

ur

However, it should be noted that the group of input variables employed in this study for training the AI
Jo

models was not the most appropriate for the 24-h physical process of precipitation. It has been pointed

out in many published works that the mechanisms producing precipitation in the next 24 hours are

highly related to information of clouds. Sánchez-Monedero et al. (2014) (Sánchez-Monedero et al.,

2014) have classified information of clouds into three main groups: condensation nuclei, water vapor

and vertical movements. Therefore, in their studies, a robust ensemble of input variables was selected

at different pressure levels involving total precipitable water, equivalent potential temperature,

humidity (group of water vapor); wind speed, wind direction, convective available potential energy

(CAPE), convective inhibition (CIN) (group of movements). Hashim et al. (2016) (Hashim et al., 2016)
34
 Journal Pre-proof
has also confirmed the significant influence of information of clouds when studying rainfall in Patna

city, India, based on a set of variables including cloud cover, vapor pressure, max and min temperature

and wet day frequency. Thus, as rainfall is a complex non-linear atmospheric process largely depending

on local-scale space (Applequist et al., 2002), it is still difficult to identify the most suitable set of

meteorological variables for training the AI models taking into account the physical mechanism of 24-h

rainfall process. As investigated by Kuligowski and Barros (1998) (Kuligowski and Barros, 1998), a

combination of ten small-scale meteorological variables (regrouping moisture and vertical lift

of
properties) at different altitudes has gave a RMSE around 2.5 mm when applying neural network for

short-term precipitation forecast. Considering such efficient prediction results, it is highly

ro
recommended to select meteorological variables taking into account characteristics of clouds as

suggested by -p
many researchers for daily rainfall prediction problems (Kuligowski and Barros, 1998;
re
Hall et al., 1999; Sánchez-Monedero et al., 2014; Ortiz-García et al., 2014).
lP

Despite constrains mentioned above, the present model study suggests that performance of SVM is the

best for daily rainfall prediction in comparison to other two models (PSOANFIS and ANN) as this
na

technique provided highest value of R mean and lowest value of MAE mean over 1000 Monte Carlo
ur

simulations.
Jo

6. Conclusion

In the present study, three advanced Artificial Intelligent (AI) models namely PSOANFIS, ANN and

SVM were applied for the prediction of daily rainfall in Hoa Binh province. Maximum temperature,

minimum temperature, wind speed, relative humidity and solar radiation were used as input parameters

and daily rainfall as an output parameter in the models. Validation of the models was done using R and

MAE, SS, POD, CSI, and FAR. Performance of these models in prediction of daily rainfall is very

good (i.e. RSVM=0.829, MAESVM=2.728 mm, PODSVM(2mm)=0.89, CSISVM(2mm)=0.78,

FARSVM(2mm)=0.14).
35
 Journal Pre-proof
The present model study suggested that performance of SVM is the best for daily rainfall prediction in

comparison to other two models (PSOANFIS and ANN) as this technique provided highest value of R

mean and lowest value of MAE mean over 1000 Monte Carlo simulations. This model also found to be

the most robust and efficient prediction model using the Monte Carlo approach.

The AI based study would be helpful in quick and accurate prediction of daily rainfall. However, as the

rainfall is a dichotomous variable, validation of the proposed AI models for rainfall prediction for the

transit period from a non-rain to a rainy day is desirable to indicate the strengths and/or weaknesses of

of
the models. Analysis indicated that for the study years 2010, 2011, 2012 and 2013, prediction

ro
performance of the proposed AI models was very good. However, as the mechanisms producing

precipitation in the next 24 hours are highly related to information of clouds, integrated study may have
-p
to be adopted in refining AI models for the daily prediction of rainfall.
re
Conflict of Interest: The authors declare that there is no conflict of interest.
lP

Reference
na

Abbot, J., Marohasy, J., 2017. Skilful rainfall forecasts from artificial neural networks with long duration series
and single-month optimization. Atmospheric Research 197, 289–299.
ur

https://doi.org/10.1016/j.atmosres.2017.07.015
Abbot, J., Marohasy, J., 2014a. Input selection and optimisation for monthly rainfall forecasting in Queensland,
Australia, using artificial neural networks. Atmospheric Research 138, 166–178.
Jo

https://doi.org/10.1016/j.atmosres.2013.11.002
Abbot, J., Marohasy, J., 2014b. Input selection and optimisation for monthly rainfall forecasting in Queensland,
Australia, using artificial neural networks. Atmospheric Research 138, 166–178.
https://doi.org/10.1016/j.atmosres.2013.11.002
Abedini, M., Ghasemian, B., Shirzadi, A., Shahabi, H., Chapi, K., Pham, B.T., Ahmad, B.B., Bui, D.T., 2019. A
novel hybrid approach of Bayesian Logistic Regression and its ensembles for landslide susceptibility
assessment. Geocarto International 34, 1427–1457. https://doi.org/10.1080/10106049.2018.1499820
Acharya, N., Shrivastava, N.A., Panigrahi, B.K., Mohanty, U.C., 2014. Development of an artificia l neural
network based multi-model ensemble to estimate the northeast monsoon rainfall over south peninsular
India: an application of extreme learning machine. Climate Dynamics 43, 1303–1310.
Al Mamun, A., bin Salleh, Md.N., Noor, H.M., 2018. Estimation of short-duration rainfall intensity from daily
rainfall values in Klang Valley, Malaysia. Appl Water Sci 8, 203. https://doi.org/10.1007/s13201-018-
0854-z
AlHassoun, S.A., 2011. Developing an empirical formulae to estimate rainfall intensity in Riyadh region. Journal
of King Saud University - Engineering Sciences 23, 81–88. https://doi.org/10.1016/j.jksues.2011.03.003

36
 Journal Pre-proof
Altunkaynak, A., Nigussie, T.A., 2015. Prediction of daily rainfall by a hybrid wavelet-season-neuro technique.
Journal of Hydrology 529, 287–301. https://doi.org/10.1016/j.jhydrol.2015.07.046
Applequist, S., Gahrs, G.E., Pfeffer, R.L., Niu, X.-F., 2002. Comparison of Methodologies for Probabilistic
Quantitative Precipitation Forecasting. Wea. Forecasting 17, 783–799. https://doi.org/10.1175/1520-
0434(2002)017<0783:COMFPQ>2.0.CO;2
Asanjan, A.A., Yang, T., Hsu, K., Sorooshian, S., Lin, J., Peng, Q., 2018. Short-Term Precipitation Forecast
Based on the PERSIANN System and LSTM Recurrent Neural Networks. Journal of Geophysical
Research: Atmospheres 123, 12,543-12,563. https://doi.org/10.1029/2018JD028375
Ashby, S.A., Taylor, M.A., Chen, A.A., 2005. Statistical models for predicting rainfall in the Caribbean. Theor.
Appl. Climatol. 82, 65–80. https://doi.org/10.1007/s00704-004-0118-8
Awadallah, A.G., Magdy, M., Helmy, E., Rashed, E., 2017. Assessment of Rainfall Intensity Equations Enlisted
in the Egyptian Code for Designing Potable Water and Sewage Networks [WWW Document]. Advances
in Meteorology. https://doi.org/10.1155/2017/9496787
Azadi, M., Taghizadeh, E., Memarian, M.H., Dmitrieva-Arrago, L.R., 2013. Comparing the results of
precipitation forecast based on mesoscale models on the territory of Iran during the cold season. Russian
Meteorology and Hydrology.

of
Benedetti, R., 2010. Scoring Rules for Forecast Verification. Mon. Wea. Rev. 138, 203–211.
https://doi.org/10.1175/2009MWR2945.1

ro
Bezak, N., Šraj, M., Mikoš, M., 2016. Copula-based IDF curves and empirical rainfall thresholds for flash floods
and rainfall-induced landslides. Journal of Hydrology, Flash floods, hydro-geomorphic response and risk
management 541, 272–284. https://doi.org/10.1016/j.jhydrol.2016.02.058

-p
Black, T.L., 1994. The New NMC Mesoscale Eta Model: Description and Forecast Examples. Wea. Forecasting
9, 265–278. https://doi.org/10.1175/1520-0434(1994)009<0265:TNNMEM>2.0.CO;2
Bui, D.T., Tsangaratos, P., Ngo, P.-T.T., Pham, T.D., Pham, B.T., 2019. Flash flood susceptibility modeling
re
using an optimized fuzzy rule based feature selection technique and tree based ensemble methods.
Science of The Total Environment 668, 1038–1054. https://doi.org/10.1016/j.scitotenv.2019.02.422
lP

Bui, K.-T.T., Tien Bui, D., Zou, J., Van Doan, C., Revhaug, I., 2018. A novel hybrid artificial intelligent
approach based on neural fuzzy inference model and particle swarm optimization for horizontal
displacement modeling of hydropower dam. Neural Comput & Applic 29, 1495–1506.
https://doi.org/10.1007/s00521-016-2666-0
na

Chegaar, M., Chibani, A., 2001. Global solar radiation estimation in Algeria. Energy Conversion and
Management 42, 967–973. https://doi.org/10.1016/S0196-8904(00)00105-9
Chen, W., Panahi, M., Pourghasemi, H.R., 2017. Performance evaluation of GIS-based new ensemble data
ur

mining techniques of adaptive neuro-fuzzy inference system (ANFIS) with genetic algorithm (GA),
differential evolution (DE), and particle swarm optimization (PSO) for landslide spatial modelling.
CATENA 157, 310–324. https://doi.org/10.1016/j.catena.2017.05.034
Jo

CORTES, C., VAPNIK, V., 1995. Support-Vector Networks. Machine Learning 20, 273–297.
Dabhi, V.K., Chaudhary, S., 2014. Hybrid Wavelet-Postfix-GP Model for Rainfall Prediction of Anand Region
of India [WWW Document]. Advances in Artificial Intelligence. https://doi.org/10.1155/2014/717803
Dao, D.V., Ly, H.-B., Trinh, S.H., Le, T.-T., Pham, B.T., 2019a. Artificial Intelligence Approaches for
Prediction of Compressive Strength of Geopolymer Concrete. Materials 12, 983.
https://doi.org/10.3390/ma12060983
Dao, D.V., Trinh, S.H., Ly, H.-B., Pham, B.T., 2019b. Prediction of Compressive Strength of Geopolymer
Concrete Using Entirely Steel Slag Aggregates: Novel Hybrid Artificial Intelligence Approaches.
Applied Sciences 9, 1113. https://doi.org/10.3390/app9061113
Deo, R.C., Şahin, M., 2015. Application of the Artificial Neural Network model for prediction of monthly
Standardized Precipitation and Evapotranspiration Index using hydrometeorological parameters and
climate indices in eastern Australia. Atmospheric Research 161–162, 65–81.
https://doi.org/10.1016/j.atmosres.2015.03.018
Deo, R.C., Salcedo-Sanz, S., Carro-Calvo, L., Saavedra-Moreno, B., 2018. Chapter 10 - Drought Prediction
With Standardized Precipitation and Evapotranspiration Index and Support Vector Regression Models,
in: Samui, P., Kim, D., Ghosh, C. (Eds.), Integrating Disaster Science and Management. Elsevier, pp.
151–174. https://doi.org/10.1016/B978-0-12-812056-9.00010-5
37
 Journal Pre-proof
Dile, Y.T., Srinivasan, R., 2014. Evaluation of CFSR climate data for hydrologic prediction in data-scarce
watersheds: an application in the Blue Nile River Basin. JAWRA Journal of the American Water
Resources Association 50, 1226–1241. https://doi.org/10.1111/jawr.12182
Dou, J., Yunus, A.P., Tien Bui, D., Merghadi, A., Sahana, M., Zhu, Z., Chen, C.-W., Khosravi, K., Yang, Y.,
Pham, B.T., 2019. Assessment of advanced random forest and decision tree algorithms for modeling
rainfall-induced landslide susceptibility in the Izu-Oshima Volcanic Island, Japan. Science of The Total
Environment 662, 332–346. https://doi.org/10.1016/j.scitotenv.2019.01.221
El-Shafie, A., Noureldin, A., Taha, M., Hussain, A., Mukhlisin, M., 2012. Dynamic versus static neural network
model for rainfall forecasting at Klang River Basin, Malaysia. Hydrology and Earth System Sciences 16,
1151–1169. https://doi.org/10.5194/hess-16-1151-2012
Evgeniou, T., Pontil, M., n.d. WORKSHOP ON SUPPORT VECTOR MACHINES: THEORY AND
APPLICATIONS.
Faccini, F., Luino, F., Paliaga, G., Sacchini, A., Turconi, L., de Jong, C., 2018. Role of rainfall intensity and
urban sprawl in the 2014 flash flood in Genoa City, Bisagno catchment (Liguria, Italy). Applied
Geography 98, 224–241. https://doi.org/10.1016/j.apgeog.2018.07.022
Fuka, D.R., Walter, M.T., MacAlister, C., Degaetano, A.T., Steenhuis, T.S., Easton, Z.M., 2014. Using the

of
Climate Forecast System Reanalysis as weather input data for watershed models. Hydrological
Processes 28, 5613–5623. https://doi.org/10.1002/hyp.10073

ro
Garric, G., Douville, H., Déqué, M., 2002. Prospects for improved seasonal predictions of monsoon precipitation
over Sahel. International Journal of Climatology 22, 331–345. https://doi.org/10.1002/joc.736
G.Gouda, S., ZakiaHussein, ShuaiLuo, QiaoxiaYuan, 2019. Model selection for accurate daily global solar

-p
radiation prediction in China. Journal of Cleaner Production 221, 132–144.
Gneiting, T., Raftery, A.E., 2007. Strictly Proper Scoring Rules, Prediction, and Estimation. Journal of the
American Statistical Association 102, 359–378. https://doi.org/10.1198/016214506000001437
re
Goffart, J., Mara, T., Wurtz, E., 2017. Generation of stochastic weather data for uncertainty and sensitivity
analysis of a low-energy building. Journal of Building Physics 41, 41–57.
lP

https://doi.org/10.1177/1744259116668598
Guilleminot, J., Le, T.T., Soize, C., 2013. Stochastic framework for modeling the linear apparent behavior of
complex materials: Application to random porous materials with interphases. Acta Mechanica Sinica 29,
773–782. https://doi.org/10.1007/s10409-013-0101-7
na

Haddad, M.S., 2011. Capacity choice and water management in hydroelectricity systems. Energy Economics 33,
168–177. https://doi.org/10.1016/j.eneco.2010.05.005
Hall, T., Brooks, H.E., Doswell, C.A., 1999. Precipitation Forecasting Using a Neural Network. Wea.
ur

Forecasting 14, 338–345. https://doi.org/10.1175/1520-0434(1999)014<0338:PFUANN>2.0.CO;2

Hartmann, H., Snow, J.A., Stein, S., Su, B., Zhai, J., Jiang, T., Krysanova, V., Kundzewicz, Z.W., 2016a.
Predictors of precipitation for improved water resources management in the Tarim River basin: Creating
Jo

a seasonal forecast model. Journal of Arid Environments 125, 31–42.

https://doi.org/10.1016/j.jaridenv.2015.09.010
Hartmann, H., Snow, J.A., Su, B., Jiang, T., 2016b. Seasonal predictions of precipitation in the Aksu-Tarim
River basin for improved water resources management. Global and Planetary Change 147, 86–96.
https://doi.org/10.1016/j.gloplacha.2016.10.018
Hashim, R., Roy, C., Motamedi, S., Shamshirband, S., Petković, D., Gocic, M., Lee, S.C., 2016. Selection of
meteorological parameters affecting rainfall estimation using neuro-fuzzy computing methodology.
Atmospheric Research 171, 21–30. https://doi.org/10.1016/j.atmosres.2015.12.002
He, S., Raghavan, S.V., Nguyen, N.S., Liong, S.-Y., 2013. Ensemble rainfall forecasting with numerical weather
prediction and radar-based nowcasting models. Hydrological Processes 27, 1560–1571.
https://doi.org/10.1002/hyp.9254
Hong, Y., Hsu, K., Moradkhani, H., Sorooshian, S., 2006. Uncertainty quantification of satellite precipitation
estimation and Monte Carlo assessment of the error propagation into hydrologic response. Water
Resources Research 42. https://doi.org/10.1029/2005WR004398
Hun, D.-A., Guilleminot, J., Yvonnet, J., Bornert, M., 2019. Stochastic multiscale modeling of crack propagation
in random heterogeneous media. International Journal for Numerical Methods in Engineering 119,
1325–1344. https://doi.org/10.1002/nme.6093
38
 Journal Pre-proof
Hung, N.Q., Babel, M.S., Weesakul, S., Tripathi, N.K., 2009. An artificial neural network model for rainfall
forecasting in Bangkok, Thailand. Hydrology and Earth System Sciences 13, 1413–1425.
Jang, J.-S.R., 1993. ANFIS Adaptive-Network-based Fuzzy Inference System. IEEE Transactions on Systems
Man and Cybernetics. https://doi.org/10.1109/21.256541
Janizadeh, S., Avand, M., Jaafari, A., Phong, T.V., Bayat, M., Ahmadisharaf, E., Prakash, I., Pham, B.T., Lee,
S., 2019. Prediction Success of Machine Learning Methods for Flash Flood Susceptibility Mapping in
the Tafresh Watershed, Iran. Sustainability 11, 5426. https://doi.org/10.3390/su11195426
Jing, J., Li, Q., Peng, X., 2019. MLC-LSTM: Exploiting the Spatiotemporal Correlation between Multi-Level
Weather Radar Echoes for Echo Sequence Extrapolation. Sensors 19, 3988.
https://doi.org/10.3390/s19183988
Kanamitsu, M., Alpert, J. c., Campana, K. a., Caplan, P. m., Deaven, D. g., Iredell, M., Katz, B., Pan, H. -L.,
Sela, J., White, G. h., 1991. Recent Changes Implemented into the Global Forecast System at NMC.
Wea. Forecasting 6, 425–435. https://doi.org/10.1175/1520-0434(1991)006<0425:RCIITG>2.0.CO;2
Kasiviswanathan, K.S., Sudheer, K.P., 2013. Quantification of the predictive uncertainty of artificial neural
network based river flow forecast models. Stoch Environ Res Risk Assess 27, 137–146.
https://doi.org/10.1007/s00477-012-0600-2

of
Kennedy, J., Eberhart, R., 1995. Particle swarm optimization, in: Proceedings of ICNN’95 - International
Conference on Neural Networks. Presented at the Proceedings of ICNN’95 - International Conference

ro
on Neural Networks, pp. 1942–1948 vol.4. https://doi.org/10.1109/ICNN.1995.488968
Khosravi, A., Nahavandi, S., Creighton, D., 2010. A prediction interval-based approach to determine optimal
structures of neural network metamodels. Expert Systems with Applications 37, 2377–2387.
https://doi.org/10.1016/j.eswa.2009.07.059
-p
Khosravi, K., Shahabi, H., Pham, B.T., Adamawoski, J., Shirzadi, A., Pradhan, B., Dou, J., Ly, H. -B., Gróf, G.,
Ho, H.L., 2019. A Comparative Assessment of Flood Susceptibility Modeling Using Multi-Criteria
re
Decision-Making Analysis and Machine Learning Methods. Journal of Hydrology.
Kuligowski, R.J., Barros, A.P., 1998. Localized Precipitation Forecasts from a Numerical Weather Prediction
lP

Model Using Artificial Neural Networks. Wea. Forecasting 13, 1194–1204.

https://doi.org/10.1175/1520-0434(1998)013<1194:LPFFAN>2.0.CO;2
Le, L.M., Ly, H.-B., Pham, B.T., Le, V.M., Pham, T.A., Nguyen, D.-H., Tran, X.-T., Le, T.-T., 2019a. Hybrid
Artificial Intelligence Approaches for Predicting Buckling Damage of Steel Columns Under Axial
na

Compression. Materials 12, 1670. https://doi.org/10.3390/ma12101670

Le, L.M., Nguyen, T.C., Pham, B.T., Ly, H.-B., Le, V.M., Le, T.-T., 2019b. Development and Identification of
Working Parameters for a Lychee Peeling Machine Combining Rollers and a Pressing Belt.
ur

AgriEngineering 1, 550–566. https://doi.org/10.3390/agriengineering1040040

Le, T.T., Guilleminot, J., Soize, C., 2016. Stochastic continuum modeling of random interphases from atomistic
simulations. Application to a polymer nanocomposite. Computer Methods in Applied Mechanics and
Jo

Engineering 303, 430–449. https://doi.org/10.1016/j.cma.2015.10.006

Le, T.-T., Pham, B.T., Ly, H.-B., Shirzadi, A., Le, L.M., 2020. Development of 48-hour Precipitation
Forecasting Model using Nonlinear Autoregressive Neural Network, in: Ha-Minh, C., Dao, D.V.,
Benboudjema, F., Derrible, S., Huynh, D.V.K., Tang, A.M. (Eds.), CIGOS 2019, Innovation for
Sustainable Infrastructure, Lecture Notes in Civil Engineering. Springer Singapore, pp. 1191–1196.
Ly, H.-B., Desceliers, C., Le, L.M., Le, T.-T., Pham, B.T., Nguyen-Ngoc, L., Doan, V.T., Le, M., 2019a.
Quantification of Uncertainties on the Critical Buckling Load of Columns under Axial Compression
with Uncertain Random Materials. Materials 12, 1828. https://doi.org/10.3390/ma12111828
Ly, H.-B., Le, L.M., Duong, H.T., Nguyen, T.C., Pham, T.A., Le, T.-T., Le, V.M., Nguyen-Ngoc, L., Pham,
B.T., 2019b. Hybrid Artificial Intelligence Approaches for Predicting Critical Buckling Load of
Structural Members under Compression Considering the Influence of Initial Geometric Imperfections.
Applied Sciences 9, 2258. https://doi.org/10.3390/app9112258
Ly, H.-B., Le, L.M., Phi, L.V., Phan, V.-H., Tran, V.Q., Pham, B.T., Le, T.-T., Derrible, S., 2019c. Development
of an AI Model to Measure Traffic Air Pollution from Multisensor and Weather Data. Sensors 19, 4941.
https://doi.org/10.3390/s19224941

39
 Journal Pre-proof
Ly, H.-B., Monteiro, E., Le, T.-T., Le, V.M., Dal, M., Regnier, G., Pham, B.T., 2019d. Prediction and Sensitivity
Analysis of Bubble Dissolution Time in 3D Selective Laser Sintering Using Ensemble Decision Trees.
Materials 12, 1544.
Ly, H.-B., Pham, B.T., Dao, D.V., Le, V.M., Le, L.M., Le, T.-T., 2019e. Improvement of ANFIS Model for
Prediction of Compressive Strength of Manufactured Sand Concrete. Applied Sciences 9, 3841.
https://doi.org/10.3390/app9183841
Lynch, P., 2008. The origins of computer weather prediction and climate modeling. Journal of Computational
Physics, Predicting weather, climate and extreme events 227, 3431–3444.
https://doi.org/10.1016/j.jcp.2007.02.034
Mahmud, M., Ross, R.S., 2005. Precipitation assessment of a superensemble forecast over South-East Asia.
Meteorological Applications 12, 177–186.
McCulloch, W.S., Pitts, W., 1943. A logical calculus of the ideas immanent in nervous activity. Bulletin of
Mathematical Biophysics 5, 115–133. https://doi.org/10.1007/BF02478259
Mehdi Keshtkar, M., Merad, L., Sidi Mohammed, M., Miraoui, A., BOUSAHLA, M., Hanaeia, H., Kakaei
Lafdani, E., Moghaddam Nia, A., Pahlavanravi, A., Ahmadi, A., JAJARMIZADEH, M., 2013. Daily
Rainfall-Runoff Prediction and Simulation Using ANN, ANFIS and Conceptual Hydrological

of
MIKE11/NAM Models. International Journal of Engineering & Technology Sciences 1.
Mghouchi, Y.E., T. Ajzoul, A., Bouardi, E., 2016. Prediction of daily solar radiation intensity by day of the year

ro
in twenty-four cities of Morocco. Renewable and Sustainable Energy Reviews 53, 823–831.
Mislan, Haviluddin, Hardwinarto, S., Sumaryono, Aipassa, M., 2015. Rainfall Monthly Prediction Based on
Artificial Neural Network: A Case Study in Tenggarong Station, East Kalimantan - Indonesia. Procedia

-p
Computer Science, International Conference on Computer Science and Computational Intelligence
(ICCSCI 2015) 59, 142–151. https://doi.org/10.1016/j.procs.2015.07.528
Møller, M.F., 1991. A Scaled Conjugate Gradient Algorithm for Fast Supervised Learning. Neural Networks.
re
Montavon, G., Rupp, M., Gobre, V., Vazquez-Mayagoitia, A., Hansen, K., Tkatchenko, A., Müller, K.-R.,
Lilienfeld, O.A. von, 2013. Machine learning of molecular electronic properties in chemical compound
lP

space. New J. Phys. 15, 095003. https://doi.org/10.1088/1367-2630/15/9/095003

Mordechai, S., 2011. Applications of Monte Carlo Method in Science and Engineering.
https://doi.org/10.5772/1954
Mouatadid, S., Raj, N., Deo, R.C., Adamowski, J.F., 2018. Input selection and data-driven model performance
na

optimization to predict the Standardized Precipitation and Evaporation Index in a drought-prone region.
Atmospheric Research 212, 130–149. https://doi.org/10.1016/j.atmosres.2018.05.012
Mousavi, S.M., Mostafavi, E.S., Jiao, P., 2017. Next generation prediction model for daily solar radiation on
ur

horizontal surface using a hybrid neural network and simulated annealing method. Energy Conversion
and Management 153, 671–682. https://doi.org/10.1016/j.enconman.2017.09.040
Murphy, A.H., 1988. Skill Scores Based on the Mean Square Error and Their Relationships to the Correlation
Jo

Coefficient. Mon. Wea. Rev. 116, 2417–2424. https://doi.org/10.1175/1520-

0493(1988)116<2417:SSBOTM>2.0.CO;2
Narayanakumar, S., Raja, K., 2016. A BP Artificial Neural Network Model for Earthquake Magnitude
Prediction in Himalayas, India. Circuits and Systems 7, 3456–3468.
Nasseri, M., Asghari, K., Abedini, M.J., 2008. Optimized scenario for rainfall forecasting using genetic
algorithm coupled with artificial neural network. Expert Systems with Applications 35, 1415–1421.
https://doi.org/10.1016/j.eswa.2007.08.033
Nastos, P.T., Moustris, K.P., Larissi, I.K., Paliatsos, A.G., 2013. Rain intensity forecast using Artificial Neural
Networks in Athens, Greece. Atmospheric Research, ADVANCES IN PRECIPITATION SCIENCE
119, 153–160. https://doi.org/10.1016/j.atmosres.2011.07.020
Nastos, P.T., Paliatsos, A.G., Koukouletsos, K.V., Larissi, I.K., Moustris, K.P., 2014. Artificial neural networks
modeling for forecasting the maximum daily total precipitation at Athens, Greece. Atmospheric
Research, Perspectives of Precipitation Science - Part II 144, 141–150.
https://doi.org/10.1016/j.atmosres.2013.11.013
Navone, H.D., Ceccatto, H.A., 1994. Predicting Indian Monsoon Rainfall - a Neural-Network Approach.
Climate Dynamics 10, 305–312.

40
 Journal Pre-proof
Nguyen, H.-L., Le, T.-H., Pham, C.-T., Le, T.-T., Ho, L.S., Le, V.M., Pham, B.T., Ly, H.-B., 2019a.
Development of Hybrid Artificial Intelligence Approaches and a Support Vector Machine Algorithm for
Predicting the Marshall Parameters of Stone Matrix Asphalt. Applied Sciences 9, 3172.
https://doi.org/10.3390/app9153172
Nguyen, H.-L., Pham, B.T., Son, L.H., Thang, N.T., Ly, H.-B., Le, T.-T., Ho, L.S., Le, T.-H., Tien Bui, D.,
2019b. Adaptive Network Based Fuzzy Inference System with Meta-Heuristic Optimizations for
International Roughness Index Prediction. Applied Sciences 9, 4715.
https://doi.org/10.3390/app9214715
Nielsen, J.E., Thorndahl, S., Rasmussen, M.R., 2014. Improving weather radar precipitation estimates by
combining two types of radars. Atmospheric Research 139, 36–45.
https://doi.org/10.1016/j.atmosres.2013.12.013
Niu, J., Zhang, W., 2015. Comparative analysis of statistical models in rainfall prediction, in: 2015 IEEE
International Conference on Information and Automation. Presented at the 2015 IEEE International
Conference on Information and Automation, pp. 2187–2190.
https://doi.org/10.1109/ICInfA.2015.7279650
Novak, D.R., Bailey, C., Brill, K.F., Burke, P., Hogsett, W.A., Rausch, R., Schichtel, M., 2014. Precipitation and

of
Temperature Forecast Performance at the Weather Prediction Center. Weather and Forecasting 29, 489–
504.

ro
Ortiz-García, E.G., Salcedo-Sanz, S., Casanova-Mateo, C., 2014. Accurate precipitation prediction with support
vector classifiers: A study including novel predictive variables and observational data. Atmospheric
Research 139, 128–136. https://doi.org/10.1016/j.atmosres.2014.01.012

-p
Paoli, C., Voyant, C., Muselli, M., Nivet, M.-L., 2010. Forecasting of preprocessed daily solar radiation time
series using neural networks. Solar Energy 84, 2146–2160. https://doi.org/10.1016/j.solener.2010.08.011
PARTAL, T., CIGIZOGLU, H.K., 2009. Prediction of daily precipitation using wavelet—neural networks.
re
Hydrological Sciences Journal 54, 234–246. https://doi.org/10.1623/hysj.54.2.234
Partal, T., Cigizoglu, H.K., Kahya, E., 2015. Daily precipitation predictions using three different wavelet neural
lP

network algorithms by meteorological data. Stoch Environ Res Risk Assess 29, 1317–1329.
https://doi.org/10.1007/s00477-015-1061-1
Pearson Karl, Galton Francis, 1895. VII. Note on regression and inheritance in the case of two parents.
Proceedings of the Royal Society of London 58, 240–242. https://doi.org/10.1098/rspl.1895.0041
na

Pham, B.T., Nguyen, M.D., Dao, D.V., Prakash, I., Ly, H.-B., Le, T.-T., Ho, L.S., Nguyen, K.T., Ngo, T.Q.,
Hoang, V., Son, L.H., Ngo, H.T.T., Tran, H.T., Do, N.M., Van Le, H., Ho, H.L., Tien Bui, D., 2019a.
Development of artificial intelligence models for the prediction of Compression Coefficient of soil: An
ur

application of Monte Carlo sensitivity analysis. Science of The Total Environment 679, 172–184.
https://doi.org/10.1016/j.scitotenv.2019.05.061
Pham, B.T., Nguyen, M.D., Ly, H.-B., Pham, T.A., Hoang, V., Van Le, H., Le, T.-T., Nguyen, H.Q., Bui, G.L.,
Jo

2020. Development of Artificial Neural Networks for Prediction of Compression Coefficient of Soft
Soil, in: Ha-Minh, C., Dao, D.V., Benboudjema, F., Derrible, S., Huynh, D.V.K., Tang, A.M. (Eds.),
CIGOS 2019, Innovation for Sustainable Infrastructure, Lecture Notes in Civil Engineering. Springer
Singapore, pp. 1167–1172.
Pham, B.T., Prakash, I., Jaafari, A., Bui, D.T., 2018. Spatial Prediction of Rainfall-Induced Landslides Using
Aggregating One-Dependence Estimators Classifier. J Indian Soc Remote Sens 46, 1457–1470.
https://doi.org/10.1007/s12524-018-0791-1
Pham, B.T., Prakash, I., Singh, S.K., Shirzadi, A., Shahabi, H., Tran, T.-T.-T., Bui, D.T., 2019b. Landslide
susceptibility modeling using Reduced Error Pruning Trees and different ensemble techniques: Hybrid
machine learning approaches. CATENA 175, 203–218. https://doi.org/10.1016/j.catena.2018.12.018
Pham, B.T., Tien Bui, D., Pham, H.V., Le, H.Q., Prakash, I., Dholakia, M.B., 2017. Landslide Hazard
Assessment Using Random SubSpace Fuzzy Rules Based Classifier Ensemble and Probability Analysis
of Rainfall Data: A Case Study at Mu Cang Chai District, Yen Bai Province (Viet Nam). J Indian Soc
Remote Sens 45, 673–683. https://doi.org/10.1007/s12524-016-0620-3
Philip, N.S., Joseph, K.B., 2003. A neural network tool for analyzing trends in rainfall. Computers &
Geosciences 29, 215–223. https://doi.org/10.1016/S0098-3004(02)00117-6

41
 Journal Pre-proof
Rajeevan, M., Pai, D.S., Anil Kumar, R., Lal, B., 2007. New statistical models for long-range forecasting of
southwest monsoon rainfall over India. Clim Dyn 28, 813–828. https://doi.org/10.1007/s00382-006-
0197-6
Robert, C., Casella, G., 2004. Monte Carlo Statistical Methods, 2nd ed, Springer Texts in Statistics. Springer-
Verlag, New York.
Sahai, A.K., Soman, M.K., Satyan, V., 2000. All India summer monsoon rainfall prediction using an artificial
neural network. Climate Dynamics 16, 291–302.
Sánchez-Monedero, J., Salcedo-Sanz, S., Gutiérrez, P.A., Casanova-Mateo, C., Hervás-Martínez, C., 2014.
Simultaneous modelling of rainfall occurrence and amount using a hierarchical nominal–ordinal support
vector classifier. Engineering Applications of Artificial Intelligence 34, 199–207.
https://doi.org/10.1016/j.engappai.2014.05.016
Scharf, L.L., Demeure, C., 1991. Statistical Signal Processing: Detection, Estimation, and Time Series Analysis.
Addison-Wesley Publishing Company.
Sedki, A., Ouazar, D., El Mazoudi, E., 2009. Evolving neural network using real coded genetic algorithm for
daily rainfall–runoff forecasting. Expert Systems with Applications 36, 4523–4527.
https://doi.org/10.1016/j.eswa.2008.05.024

of
Sehad, M., Lazri, M., Ameur, S., 2017. Novel SVM-based technique to improve rainfall estimation over the
Mediterranean region (north of Algeria) using the multispectral MSG SEVIRI imagery. Advances in

ro
Space Research 59, 1381–1394. https://doi.org/10.1016/j.asr.2016.11.042
Senior, C.A., Jones, R.G., Lowe, J.A., Durman, C.F., Hudson, D., 2002. Predictions of extreme precipitation and
sea-level rise under climate change. Philos Trans A Math Phys Eng Sci 360, 1301–1311.
https://doi.org/10.1098/rsta.2002.1001

management and impact studies.

-p
Serinaldi, F., Kilsby, C.G., 2012. A modular class of multisite monthly rainfall generators for water resource
Journal of Hydrology 464–465, 528–540.
re
https://doi.org/10.1016/j.jhydrol.2012.07.043
Shao, Q., Li, M., 2013. An improved statistical analogue downscaling procedure for seasonal precipitation
lP

forecast. Stochastic Environmental Research and Risk Assessment 27, 819–830.

Shi, Y., Eberhart, R., 1998. A modified particle swarm optimizer, in: 1998 IEEE International Conference on
Evolutionary Computation Proceedings. IEEE World Congress on Computational Intelligence (Cat.
No.98TH8360). Presented at the 1998 IEEE International Conference on Evolutionary Computation
na

Proceedings. IEEE World Congress on Computational Intelligence (Cat. No.98TH8360), pp. 69–73.
https://doi.org/10.1109/ICEC.1998.699146
Silvestro, F., Rebora, N., 2014. Impact of precipitation forecast uncertainties and initial soil moisture conditions
ur

on a probabilistic flood forecasting chain. Journal of Hydrology 519, 1052–1067.

Soize, C., Desceliers, C., Guilleminot, J., Le, T.T., Nguyen, M.T., Perrin, G., Allain, J.M., Gharbi, H., Duhamel,
D., Funfschilling, C., 2015. Stochastic representations and statistical inverse identification for
Jo

uncertainty quantification in computational mechanics. Presented at the UNCECOMP 2015 - 1st

ECCOMAS Thematic Conference on Uncertainty Quantification in Computational Sciences and
Engineering, pp. 1–26.
Takagi, H., Sugeno, M., 1983. Derivation of fuzzy control rules from human operator’s control actions. Proc. of
the IFAC Symp. on Fuzzy Information, Knowledge Representation and Decision Analysis 55–60.
Tanessong, R.S., Igri, P.M., Vondou, D.A., Tamo, P.H.K., Kamga, F.M., 2014. Evaluation of probabilistic
precipitation forecast determined from WRF forecasted amounts. Theoretical and Applied Climatology
116, 649–659.
Thanh, T.T.M., Ly, H.-B., Pham, B.T., 2020. A Possibility of AI Application on Mode-choice Prediction of
Transport Users in Hanoi, in: Ha-Minh, C., Dao, D.V., Benboudjema, F., Derrible, S., Huynh, D.V.K.,
Tang, A.M. (Eds.), CIGOS 2019, Innovation for Sustainable Infrastructure, Lecture Notes in Civil
Engineering. Springer, Singapore, pp. 1179–1184. https://doi.org/10.1007/978-981-15-0802-8_189
Tien Bui, D., Khosravi, K., Shahabi, H., Daggupati, P., Adamowski, J.F., Melesse, A.M., Thai Pham, B.,
Pourghasemi, H.R., Mahmoudi, M., Bahrami, S., Pradhan, B., Shirzadi, A., Chapi, K., Lee, S., 2019.
Flood Spatial Modeling in Northern Iran Using Remote Sensing and GIS: A Comparison between
Evidential Belief Functions and Its Ensemble with a Multivariate Logistic Regression Model. Remote
Sensing 11, 1589. https://doi.org/10.3390/rs11131589
42
 Journal Pre-proof
Tien Bui, D., Pradhan, B., Nampak, H., Bui, Q.-T., Tran, Q.-A., Nguyen, Q.-P., 2016. Hybrid artificial
intelligence approach based on neural fuzzy inference model and metaheuristic optimization for flood
susceptibilitgy modeling in a high-frequency tropical cyclone area using GIS. Journal of Hydrology 540,
317–330. https://doi.org/10.1016/j.jhydrol.2016.06.027
Tippett, M.K. daSilva, 1999. Model Error Estimation for the CPTEC Eta Model.
Trinh, T.A., 2018. The Impact of Climate Change on Agriculture: Findings from Households in Vietnam.
Environ Resource Econ 71, 897–921. https://doi.org/10.1007/s10640-017-0189-5
Türk Toğrul, I., Onat, E., 1999. A study for estimating solar radiation in Elaziğ using geographical and
meteorological data. Energy Conversion and Management 40, 1577–1584.
https://doi.org/10.1016/S0196-8904(99)00035-7
Vapnik, V., 1998. Statistical Learning Theory. Wiley.
Villarini, G., Seo, B.-C., Serinaldi, F., Krajewski, W.F., 2014. Spatial and temporal modeling of radar rainfall
uncertainties. Atmospheric Research 135–136, 91–101. https://doi.org/10.1016/j.atmosres.2013.09.007
Wahba, G., 1990. An Introduction to Reproducing Kernel Hilbert Spaces and Why They are So Useful.
Wang, L.-P., Ochoa-Rodríguez, S., Van Assel, J., Pina, R.D., Pessemier, M., Kroll, S., Willems, P., Onof, C.,
2015. Enhancement of radar rainfall estimates for urban hydrology through optical flow temporal

of
interpolation and Bayesian gauge-based adjustment. Journal of Hydrology, Hydrologic Applications of
Weather Radar 531, 408–426. https://doi.org/10.1016/j.jhydrol.2015.05.049

ro
Warsito, B., Gernowo, R., Sugiharto, A., 2016. Rainfall Prediction by Using Wavelet General Regression Neural
Network. International Journal of Applied Mathematics and Statistics TM 54, 32–41.
Wei, C.-C., 2013. Soft computing techniques in ensemble precipitation nowcast. Applied Soft Computing 13,

-p
793–805. https://doi.org/10.1016/j.asoc.2012.10.006
Werbos, P., 1974. Beyond regression: New tools for prediction and analysis in the behavioral sciences.
Wilks, D.S., 2011. Statistical Methods in the Atmospheric Sciences. Academic Press.
re
Wu, C.L., Chau, K.W., 2013. Prediction of rainfall time series using modular soft computing methods.
https://doi.org/10.1016/j.engappai.2012.05.023
lP

Wu, C.L., Chau, K.W., Fan, C., 2010. Prediction of rainfall time series using modular artificial neural networks
coupled with data-preprocessing techniques. Journal of Hydrology 389, 146–167.
Yuan, W., Göncü, A., Ökten, G., 2015. Estimating sensitivities of temperature-based weather derivatives.
Applied Economics 47, 1942–1955. https://doi.org/10.1080/00036846.2014.1002888
na

Zellou, B., Rahali, H., 2019. Assessment of the joint impact of extreme rainfall and storm surge on the risk of
flooding in a coastal area. Journal of Hydrology 569, 647–665.
https://doi.org/10.1016/j.jhydrol.2018.12.028
ur
Jo

43
 Journal Pre-proof
Highlights :

 Rainfall prediction was carried out using AI methods such as ANN, PSOANFIS and SVM;
 R, MAE, Skill Score and contingency scores were employed to validate the models;
 Monte Carlo method was applied to analyze the robustness of AI models;
 AI based study would be helpful in quick and accurate prediction of daily rainfall.

of
ro
-p
re
lP
na
ur
Jo

44
 Journal Pre-proof
Author statement:

Binh Thai Pham: Writing- Reviewing and Editing, Project administration

Lu Minh Le: Conceptualization, Supervision

Tien-Thinh Le: Investigation, Writing - Review & Editing

Kien-Trinh Thi Bui: Writing- Original draft preparation

Vuong Minh Le: Data curation, Visualization, Writing- Original draft preparation

Hai-Bang Ly: Software, Validation, Writing- Original draft preparation

Indra Prakash: Methodology, Writing - Review & Editing

of
ro
-p
re
lP
na
ur
Jo

45
 Journal Pre-proof

of
ro
-p
re
lP
na
ur
Jo

46