CHEMICAL ENGINEERING TRANSACTIONS  
 

VOL. 56, 2017 

A publication of 

 
The Italian Association 

of Chemical Engineering 
Online at www.aidic.it/cet 

Guest Editors: Jiří Jaromír Klemeš, Peng Yen Liew, Wai Shin Ho, Jeng Shiun Lim 
Copyright © 2017, AIDIC Servizi S.r.l., 

ISBN 978-88-95608-47-1; ISSN 2283-9216 

Graphical User Interface Application in Matlab Environment 
for Water and Air Quality Process Monitoring 

Nazira Abdul Rahima, Zainal Ahmad*,b 
aRiver Basin Research Centre, National Hydraulic Research Institute of Malaysia, Lot 5377, Jalan Putra Permai, 43300 Seri 
 Kembangan, Selangor, Malaysia 
bSchool of Chemical Engineering, Engineering Campus, Universiti Sains Malaysia, Seri Ampangan, 14300 Nibong Tebal, 
 Penang,Malaysia  
 chzahmad@usm.my 

The principle aim of water and air quality modelling is to describe and to predict the observed effects of a change 
in the river and air water system. They can be used to forecast the characteristics of water and air quality 
conditions in aquatics and air systems in order to ensure the water and air quality objectives will be maintained 
under a wide variety conditions as well as the condition of air. Therefore, friendly user interfaces is introduced 
in which it is likely to have the largest impact on whether or not a particular decision support system is actually 
used by decision makers. Indeed, Graphical User Interface (GUI) has progressively been used in water and air 
quality management information systems for over a past decade purposely for validation and verification of 
expert system in reservoir management and also in air pollution index. Numerous researchers have also 
remarked that appropriate and intelligent use of GUI can help increase the effectiveness of management 
information systems. The framework of the underlying database is also general (and standardise) enough to 
allow users to add new input data without major hassle.  The system also offers a Graphical Interface Editing 
Sub-module, which allows a system administrator to change the water quality data of other rivers as well as the 
air quality data. This study is about exploring more towards the advanced monitoring system of the river water 
and air quality by neural network approach, which later will be presented in GUI for online testing/prediction 
which result in ease of monitoring, diagnostic and control. 

1. Introduction 
The water quality management, which consists of monitoring data, could make some addition to the application 
of model's prediction. In this study, water quality prediction model could be applied when the modelling is feasible 
in the situation where the monitoring model is not available or impractical. In advance, the idea of integrating 
the monitoring and modelling models could provide better information for nearly the same cost. The introduction 
of prediction model also could help in assessing and predicting future water quality situation if there is a need 
for different management strategies (Loucks and van Beek, 2005). Application of this water quality prediction 
model gives a benefit in determining the river water quality status, where it needs the measurement of various 
water quality variables, to ensure that the water is safe to use for any purpose. The water quality index (WQI) 
shows the status of the water quality assessment and has been developed with suitability to the rivers, lakes 
and reservoirs. This WQI approach has applied the concept based on the comparison between the water quality 
variables and their respective regulatory standards (Khan et al., 2003). It combines several of the water quality 
variables that have affected to the overall water quality. The simplest WQI has considered variables like 
dissolved oxygen, total suspended solid, pH and some nutrients (Davis and McCuen, 2005). The monitoring 
data availability and accuracy have also contributed to the development of these models. The developed model 
should have also been opted to have the most meaningful and understandable response to the communities; 
either to inform about the pollutant sources or water quality conditions. Some of the researchers have developed 
water quality prediction models with a different approach and application. Some of the models were based on 
regression analysis (Abaurrea et al., 2011), time series (Ömer Faruk, 2010), fuzzy reasoning (Mahapatra et al., 

                               
 
 

 

 
   

                                                  
DOI: 10.3303/CET1756017

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Please cite this article as: Rahim N.A., Ahmad Z., 2017, Graphical user interface application in matlabtm environment for water and air quality 
process monitoring, Chemical Engineering Transactions, 56, 97-102  DOI:10.3303/CET1756017   

97



2011) and neural network method (Eynard et al., 2011). Their study areas also involved sewage treatment, 
drinking water management and various other fields.  
In the other hand, the air quality monitoring and forecasting tools are necessary to take precautionary measures, 
such as reducing the effect of a predicted pollution peaks on the surrounding population and ecosystem. Air 
quality is monitored continuously and manually to detect any changes in the ambient air quality status that may 
cause harm to human health and the environment. The Malaysian DOE monitors the ambient air quality via a 
network of 51 monitoring stations. These monitoring stations are strategically located in residential, traffic and 
industrial areas to detect any significant changes in the air quality which could be harmful to human health and 
the environment. The efficient methods for the assessment of air quality are needed in order to establish 
mechanisms for managing pollutant concentration and preventing illness in health-sensitive people. With 
increasingly severe air pollution, it is important to predict air quality exactly in order to provide actions and 
controlling strategies so that the adverse effects can be minimised. In response to this concern, several studies 
on air quality prediction using artificial neural network have been done (Zhang et al., 2012). Unlike other 
modelling technique, artificial neural network (ANN) makes no prior assumptions concerning the data 
distribution. ANN is capable of modelling highly nonlinear relationships and can be trained to accurately 
generalise when presented with a new data set (Kurt et al., 2008). Even though there were successful in many 
applications of ANN and having considerably less restrictions on the environmental input data, large training 
data sets are usually required to improve the accuracy and minimise uncertainty in the output data, which up to 
now has been a significant disadvantage of these models. In addition to that, it is needed a user friendly 
interfaces that can disseminate the information about the WQI and API in an efficient manner in alerting the 
health-sensitive people in the surrounding area. 
Therefore in this study, GUI that user friendly interface embedded in MatlabTM environment is introduced in 
which it is likely to have the significant impact on a decision support system that actually used by decision 
makers (MATLAB, 2012). Indeed, GUI has progressively been used in water and air quality management 
information systems for over a past decade purposely for validation and verification of expert system in reservoir 
management and also in air pollution index. Numerous researchers have also remarked that appropriate and 
intelligent use of GUI can help increase the effectiveness of management information systems  

2. Methodology 
In this study, two feedforward artificial neural network (FANN) model predictions for WQI and API were 
developed to be embedded in the GUI for process monitoring. For WQI prediction, two output variables were 
selected which are predicted BOD and COD as an outputs from NET1 and NET2 respectively after the FANN 
model development was carried out using the input sets as shown in Table 1. As for NET3 which is a WQI 
predictor, the network input was based on the output of NET1 and NET2 as shown in Figure 1. The FANN 
structure for NET3 was based on the Figure 1 and the FANN network training parameters were shown in Table 
2. Meanwhile the data distribution for FANN development was shown in Table 3. Similar to WQI prediction, there 
were also two output variables selected in the air pollution index (API) prediction model. PM10 and O3 predicted 
outputs were assigned as NET4 and NET5 respectively after the FANN model development was carried out 
using the input sets in Table 1. For NET6 which is an API predictor, the network input was based on the output 
of NET4 and NET5 and it applied the MISO structure as shown in Figure 1.  In that figure, outputs from NET4 
and NET5 were used as an input for NET6.  

Table 1: Input sets for WQI and API model prediction 

 FANN WQI Prediction FANN API Prediction 
 NET1 and NET2 NET3 NET4 and NET5 NET6 

Output BOD (Net 1) 
COD (Net 2) 

WQI PM10 (Net 4) 
O3 (Net 5) 

API 

Input Suspended solid 
NO3 
Kalium 
NH3-NL 
Total solid 
Zink 
Turbidity 

Biological oxygen 
demand, BOD 
 
Chemical oxygen 
demand, COD 

Wind speed, m/s 
Ambient Temperature, °C 
Humidity, % 

Particulate 
matters, 
PM10 
Ozone, O3 

 

98



 
(a) 

 
(b) 

Figure 1: FANN modelling structure for (a) WQI prediction and (b) API prediction 

Table 2: FANN network training parameters for WQI (Net 3) and API (Net 6) model prediction 

Fix parameters 
Learning rate  0.05 
Epochs 1000 
Target error goal  10-5 
Minimum performance gradient 10-5 
Varied parameters 
Number of hidden neuron 1 to 20 
Transfer function (hidden layer)  Log-sigmoid (logsig) 
Transfer function (output layer)  Linear (purelin) 
Training algorithm Levenberg Marquardt backpropagation (trainlm) 

The FANN-based prediction system for WQI and API was developed as an end-user product for GUI. The need 
of simple and compact system was highly focused during the development of the prediction system. The 
prediction system was built based on MatlabTM platform and presented as Graphical User Interface (GUI). The 
implementation of neural network in the system makes the developed GUI had to be user-friendly in terms of 
loading of data and retraining the networks.  The flow chart in Figure 2(a) shows the outlines or the procedure 
for organising and utilising the GUI data by the callback function in the MatlabTM. In addition, from Figure 2(b), 

99



the GUI page layout has six pages for this study. The developments of the pages follow the basic procedures 
of GUI development as shown in Figure 3(a). However, each of these pages is different from another, based on 
their purpose and function. 

3. Result and discussions 
NET3 for WQI was developed based on the prediction output of NET1 and NET2. The output of these networks; 
BOD and COD estimation values will be used as inputs for NET3 network to predict the values of water quality 
index (WQI). Apart from the utilisation of other network’s output as an input, the development steps for FANN 
model of this NET3 were much the same as NET1 and others. FANN network with single hidden layer and one 
output was applied with the Levenberg-Marquardt training algorithm (trainlm) with sigmoid logsig transfer 
function in the hidden layer and linear purelin transfer function in the output layer. The determination of the 
number of neurons in the hidden layer was carried out using cross validation technique and it shows that 6 
neurons in the hidden layer give good performance in training and testing data while for API is consist of 8 
hidden nodes in the hidden layer respectively. Thus, NET3 and NET6 with 6 and 8 hidden neurons were selected 
as a final architecture or structure of the FANN.  Table 3 shows the summary of the NET3 and NET6 WQI and 
API predictor performance respectively. 
 

 

Figure 2: (a) Flowchart of data flow procedure, (b) Page layout for GUI 

Table 3: Coefficient determination and mean sum squared error for unseen validation data for WQI and API 

prediction 

 WQI  2-6-1 API  2-8-1 
Data set No of 

data 
mean 
squared 
error (MSE) 

Coefficient of 
determination 
(r2) 

No of 
data 

mean 
square error 
(MSE) 

Coefficient of 
determination 
(r2) 

Training 800 0.1530 0.9200 1100 0.1936 0.9070 
Testing 100 0.1326 0.9300 138 0.1770 0.9297 
Unseen/Validation 100 0.1373 0.9000 138 0.1850 0.9031 

 
As shown in Figure 3(a), the GUI main page was displayed once the model of estimation system was executed. 
This environmental qualities predictor main page shows two main options to the user, the prediction of water 
quality and prediction of air quality. Each of this option led the user to a specific page for their specific tasks, as 
shown in Figure 4(a) and 5(b). In Figure 4(b), the GUI layout of water quality predictor main page was displayed. 
Two main buttons were included in this page, RELOAD and ENTER INPUT buttons with their specific functions 
and objectives. The other button, HOME was included in the page to bring the user back to the main page of 
the environmental quality prediction model. Once the ENTER INPUT button was selected, the user was 
presented with a new page as shown in Figure 4(b). In Figure 4(a), seven inputs for the water quality estimation 
needed to be filled. These input values are suspended solids (SS), ammonia (NH3-NL), turbidity (TUR), total 
solid (TS), nitrate (NO3), zinc (Zn) and Kalium (K). The complete input with the DONE button brought the user 
back to the water quality main page, together with the prediction of COD, BOD and water quality index. The 

100



water quality index class also was displayed in the main page. As shown also in Figure 4(b), the RELOAD button 
connected the user to the neural network reload page, as in Figure 3(b). RELOAD button was built to help the 
user with a different set of environmental data to use in this estimation system. The user has the option to retrain 
the estimation model based on their historical data-based using this estimation system. With new input and 
specific data like COD, BOD and WQI database, user can load this data into the system and retrain the networks. 
The training progress for the respecting networks can be evaluated by the performance value shown in MSE 
column.  
In Figure 5(a) and 5(b), the end-user layout shows the air quality prediction main page. The GUI layout is nearly 
the same as the water quality prediction main page, with the same function and objective for the RELOAD, 
HOME and ENTER INPUT buttons. The HOME button directed the user to the GUI main page as in Figure 5(a) 
while ENTER INPUT button brought the user to the input page of air quality estimation network. The input page, 
as shown in Figure 5(b) requires user to enter three input values, wind speed, ambient temperature and 
humidity. The requirement input values have to be in the range stated in the respective column, so that the input 
is in the boundary of the network training. Once the DONE button is pressed, the user gets the air quality 
estimation value on the main page, as in Figure 5(a) together with the results. 
 

 

Figure 3: Environmental qualities predictor model: (a) GUI main page (b) Water quality main page 

 

Figure 4: Water quality estimator: (a) Input page, (b) Quality main page with the estimation results 

101



 

Figure 5: (a) Air quality: Main page with results, (b) Estimator: Input page 

4. Conclusions 
As a conclusion, the developed GUI for the environmental quality prediction system has been executed quite 
well. The developed prediction system based on FANN model with the input from the prediction output of NET1 
and NET2 for WQI and NET4 and NET5 for API has performed very well. A developed GUI had been positively 
affected by the simplicity of the FANN model prediction that has been embedded in the system. A simple 
interface in GUI has been successfully developed with only seven measurable variables needed to predict the 
water quality and three variables for air quality prediction. This is an advantage for end-users, especially the 
ones without much knowledge in environmental quality index calculation to predict one. 

Acknowledgments  

The authors would like to acknowledge the support from the Universiti Sains Malaysia (USM) through Graduate 
on Time (GOT) grant 8023006 and special gratitude to Department of Environment (DOE) Malaysia for providing 
the water and air quality data for this study. 

Reference  

Abaurrea J., Asín J., Cebrián A.C., García-Vera M.A., 2011, Trend analysis of water quality series based on 
regression models with correlated errors, Journal of Hydrology 400 (3-4), 341–352. 

Davis A.P., McCuen R.H., 2005, Stormwater Management for Smart Growth Stormwater Management for Smart 
Growth, 1st edition, Springer Science and Business Media, New York, United States. 

Eynard J., Grieu S., Polit M., 2011, Wavelet-based multi-resolution analysis and artificial neural networks for 
forecasting temperature and thermal power consumption, Engineering Applications of Artificial Intelligence 
24, 501–516. 

Khan F., Husain T., Lumb A., 2003, Water quality evaluation and trend analysis in selected watersheds of the 
Atlantic region of Canada, Environmental Monitoring and Assessment 88 (1-3), 221–42.  

Kurt A., Gulbagci B., Karaca F., Alagha O., 2008, An air pollution forecasting system using neural networks, 
Environment International 34, 592-598. 

Loucks D.P., van Beek E., 2005, An Introduction to Methods, Models and Applications, Water Resources 
Systems Planning and Management, UNESCO Publishing, Paris, France. 

Mahapatra S.S., Nanda S.K., Panigrahy B.K., 2011, A Cascaded Fuzzy Inference System for Indian river water 
quality prediction, Advances in Engineering Software 42 (10), 787–796. 

MATLAB and Statistics Toolbox Release 2012b, The MathWorks, Inc., Natick, Massachusetts, United States. 
Ömer Faruk D., 2010, A hybrid neural network and ARIMA model for water quality time series prediction, 

Engineering Applications of Artificial Intelligence 23 (4), 586–594. 
Zhang Y., Bocquet M., Mallet V., Seigneur C., Baklanov A., 2012, Real-time air quality forecasting, part I: history, 

techniques, and current status, Atmospheric Environment 60, 632–655. 

102