Civil Engineering and Architecture

ISSN 2332-1091

Volume 7 Number 6A 2019

Civil Engineering and Architecture http://www.hrpub.org

Horizon Research Publishing, USA

http://www.hrpub.org

Civil Engineering and Architecture Civil Engineering and Architecture is an international peer-reviewed journal that publishes original and high-quality research papers in all areas of civil engineering and architecture. As an important academic exchange platform, scientists and researchers can know the most up-to-date academic trends and seek valuable primary sources for reference. The subject areas include, but are not limited to the following fields: Architectural design competition, Architectural drawing, Architectural style, Architectural theory, Atmospheric sciences, Biomechanics, Building materials, Coastal engineering, Computational science and engineering, Construction engineering, Control engineering, Earthquake engineering, Environmental engineering, Forensic engineering, Geodesy, Geophysics, Geotechnical engineering, Materials engineering, Municipal or urban engineering, Organic architecture, Sociology of architecture, Structural engineering. General Inquires Publish with HRPUB, learn about our policies, submission guidelines etc. Email: [email protected] Tel: +1-626-626-7940 Subscriptions Journal Title: Civil Engineering and Architecture Journal’s Homepage: http://www.hrpub.org/journals/jour_info.php?id=48 Publisher: Horizon Research Publishing Co.,Ltd Address: 2880 ZANKER RD STE 203 SAN JOSE, CA 95134 USA Publication Frequency: bimonthly Electronic Version: freely online available at http://www.hrpub.org/journals/jour_info.php?id=48 Online Submission Manuscripts should be submitted by Online Manuscript Tracking System (http://www.hrpub.org/submission.php). If you are experiencing difficulties during the submission process, please feel free to contact the editor at [email protected]. Copyright Authors retains all copyright interest or it is retained by other copyright holder, as appropriate and agrees that the manuscript remains permanently open access in HRPUB 's site under the terms of the Creative Commons Attribution International License (CC BY). HRPUB shall have the right to use and archive the content for the purpose of creating a record and may reformat or paraphrase to benefit the display of the record. Creative Commons Attribution License (CC-BY) All articles published by HRPUB will be distributed under the terms and conditions of the Creative Commons Attribution License(CC-BY). So anyone is allowed to copy, distribute, and transmit the article on condition that the original article and source is correctly cited. Open Access Open access is the practice of providing unrestricted access to peer-reviewed academic journal articles via the internet. It is also increasingly being provided to scholarly monographs and book chapters. All original research papers published by HRPUB are available freely and permanently accessible online immediately after publication. Readers are free to copy and distribute the contribution under creative commons attribution-non commercial licence. Authors can benefit from the open access publication model a lot from the following aspects: • High Availability and High Visibility-free and unlimited accessibility of the publication over the internet without any restrictions; • Rigorous peer review of research papers----Fast, high-quality double blind peer review; • Faster publication with less cost----Papers published on the internet without any subscription charge; • Higher Citation----open access publications are more frequently cited. Civil Engineering and Architecture

Editor-in-Chief Elizabeth Martin-Malikian Southern Polytechnic State University, USA

Members of Editorial Board

Chia-hui Lin The University of Melbourne, Australia

Janendranath Mandal Indian Institute of Technology, India

Engin Eroglu Duzce University Faculty of Forestry, Turkey

Hasan Tahsin Ozturk Karadeniz Technical University, Turkey

Antonio Brencich University of Genoa, Italy

Santina Di Salvo University of Palermo, Italy

Hongyuan Liu University of Tasmania, Australia

Aldo Renato Daniele Accardi Architectural Department, Italy

Halil Zafer Alibaba Eastern Mediterranean University, North Cyprus

Adel Francis École de technologie supérieure (ÉTS), Canada

Amira Elnokaly University of Lincoln, UK

Damian Beben Opole University of Technology, Poland

Mohamed Mostafa Hassan Mostafa Central University of Technology, South Africa

Stephen Eromobor Central University of Technology, South Africa

Ying Wang Deakin University, Australia

H.M.A. Mahzuz Shahjalal University of Science & Technology Sylhet, Bangladesh

Tomaz Tollazzi University of Maribor, Slovenia

Jonas Jakaitis Vilnius Gediminas Technical University, Lithuania

Daniela Staffenova University of Žilina, Slovenia

Petros Christou Frederick University, Cyprus

George Vasdravellis Heriot-Watt University, UK

Lee Pugalis Northumbria University, UK

Maurizio Francesco Errigo Delft University of Technology, Netherlands

Elizabeth Grant University of Adelaide, Australia

Luisa Maria Calabrese Delft University of Technology, Netherlands

Horizon Research Publishing http://www.hrpub.org ISSN 2332-1091 Table of Contents Civil Engineering and Architecture

Volume 7 Number 6A 2019

Monitoring Slope Condition Using UAV Technology (https://www.doi.org/10.13189/cea.2019.071401) Norhayati Ngadiman, Ibrahim Adham Badrulhissham, Mazlan Mohamad, Nurazira Azhari, Masiri Kaamin, Nor Baizura Hamid ...... 1

Sub-surface Profiling Using Electrical Resistivity Tomography (ERT) with Complement from Peat Sampler (https://www.doi.org/10.13189/cea.2019.071402) Kasbi Basri, Norhaliza Wahab, Mohd Khaidir Abu Talib, Adnan Zainorabidin ...... 7

Empirical Mode Decomposition Couple with Artificial Neural Network for Water Level Prediction (https://www.doi.org/10.13189/cea.2019.071403) Eng Chuen Loh, Shuhaida Binti Ismail, Azme Khamis ...... 19

Prediction of Future Climate Change for Rainfall in the Upper Kurau River Basin, Perak Using Statistical Downscaling Model (SDSM) (https://www.doi.org/10.13189/cea.2019.071404) Nuramidah Hamidon, Sobri Harun, Norshuhaila Mohamed Sunar, Nor Hazren A.Hamid, Mimi Suliza Muhamad, Hasnida Harun, Roslinda Ali, Mariah Awang, Mohamad Ashraf Abdul Rahman, Faridahanim Ahmad, Kamaruzaman Musa, Fatimah Mohamed Yusof, Mohd Syafiq Syazwan Mustafa ...... 33

User Perception on Urban Light Rail Transit (https://www.doi.org/10.13189/cea.2019.071405) Seuk Yen Phoong, Seuk Wai Phoong, Sedigheh Moghavvemi, Kok Hau Phoong ...... 43

Mathematical Modeling for Flood Mitigation: Effect of Bifurcation Angles in River Flowrates (https://www.doi.org/10.13189/cea.2019.071406) Iskandar Shah Mohd Zawawi, Nur Lina Abdullah, Hazleen Aris, Badrul Amin Jaafar, Nur Arif Husaini Norwaza, Muhammad Haris Fadzillah Mohd Yunos ...... 50

Predicting the Capability of Carboxylated Cellulose Nanowhiskers for the Remediation of Copper from Wastewater Effluent Using Statistical Approach (https://www.doi.org/10.13189/cea.2019.071407) Hazren A. Hamid, Hasnida Harun, Norshuhaila Mohamed Sunar, Faridahanim Ahmad, Nuramidah Hamidon, Mimi Suliza Muhamad, Latifah Jasmani, Norhidayah Suleiman ...... 58

Determining the Chaotic Dynamics of Hydrological Data in Flood-Prone Area (https://www.doi.org/10.13189/cea.2019.071408) Adib Mashuri, Nur Hamiza Adenan, Nor Zila Abd Hamid ...... 71

Comparison between Multiple Gradient and Pole Dipole Array Protocols for Groundwater Exploration in Quaternary Formation (https://www.doi.org/10.13189/cea.2019.071409) A. K. Abd Malik, A. Madun, M. F. Md Dan, M. K. Abu Talib, F. Pakir, S. A. Ahmad Tajudin, M. N. H. Zahari, M. E. Z. Mat Radzi ...... 77

Civil Engineering and Architecture 7(6A): 1-6, 2019 http://www.hrpub.org DOI: 10.13189/cea.2019.071401

Monitoring Slope Condition Using UAV Technology

Norhayati Ngadiman*, Ibrahim Adham Badrulhissham, Mazlan Mohamad, Nurazira Azhari, Masiri Kaamin, Nor Baizura Hamid

Department of Civil Engineering, Centre for Diploma Studies (CeDS), Universiti Tun Hussein Onn Malaysia (UTHM), Malaysia

Received July 30, 2019; Revised September 30, 2019; Accepted December 10, 2019

Abstract Slope failure is a serious geologic hazard in there is a wide range in their predictability, rapidity of many countries in the world including Malaysia. In order to occurrence and movement, and ground area affected, all of prevent slope failure, the hazardous symptoms can be which relate directly to the consequences of failure [3]. detected early in slope monitoring process. Nowadays, There are several types of slope failures that can occur slope failure symptoms monitoring has been done by when the shear resistance along the slip plane is exceeded. human by on site observation at the slope spot and it is Slope falls, slope topples, landslides, flows and spreads of dangerous for the human safety. Furthermore, it takes slope are the types of slope failure. These can be caused by longer time to complete the investigation and some of the excessive load imposed at the slope crest or compromised data collected are inaccurate because human view is stability of the slope, and disturbed dimensions of the limited. Therefore, this project is to evaluate the slope. performances of Unmanned Aerial Vehicle (UAV) to Monitoring of slopes is a crucial method to predict early monitor the slope condition. A slope at Taman Harmoni sign of slope failure and prevent slope from collapse. Vista at Pagoh, Johor has been selected as a study area. The Conducting a quantitative study on the unstable slope area measurement of slope studied was 150 meters in length and will be able to detect any movement on the slope. 20 meters in height respectively. The aim of this project is Therefore, detailed orthophotos are a valuable resource for to evaluate the performances of Unmanned Aerial Vehicle mapping and classifying morphological phenomena that (UAV) to monitor the risk slope. Visual image of 3D occur at that point [4]. Initial information obtained will modelling was obtained from Pix4D Mapper and Global provide preliminary information on the slope and be able to Mapper software was used to analyze the slope failure detect slope stability level prior to the occurrence of slope symptoms. There are defects that have been identified from failure [5]. Monitoring is an important element of hazard 3D modelling of the slope which is tension crack on slope identification, hazard assessment and hazard information and surface crack on the drainage. The use of UAV is an which can be supported by Geographical Information alternative method to obtain quality picture and 3D model Systems (GIS) and related databases. Nowadays, slope for inspection of the slope. This method expected to analysis is done by inspector that inspecting and facilitate the work of inspection slopes in addition to saving monitoring will climb up to the top of the slope for the time, energy and cost as well asreducing the risk of harm. purpose of taking photographs for inclusion in the work report [6]. This method needed more time to complete the Keywords Slope Failure, Slope Monitoring, observation and the data was collected not too accurately Unmanned Aerial Vehicle (UAV) because the human errors sometimes fail to read the data correctly. This study was conducted to determine the extent to which the use of Unmanned Aerial Vehicle (UAV) can 1. Introduction assist in the slope monitoring process. The main objectives of this study are to produce a 3-dimensional model of the Landslides are usually found in the form of slope failure slope to determine defects that occur on the slopes. The in a man-made slope, especially the slopes involved with UAV can be controlled by a safe distance from a risky cut and fill activities which often occur along highway area, slope. UAV can be manually or autonomously controlled residential area and urban area [1]. Slope falls, slope [7]. Basically, the monitoring process can be done safely, topples, landslides, flows and spreads of slope are types of quickly and more energy work can be saved by using this slope failure [2]. Slope failures occur in many forms and technology. The purpose of this paper was to identify the 2 Monitoring Slope Condition Using UAV Technology

effectiveness of an UAV in monitoring high risk slope and method to obtain the slope visual information by using analyze the probability of slope failure symptoms based on Unmanned Aerial Vehicle (UAV). the 3D slope modelling using Pix4D Mapper and Global Mapper software. This study will help researchers and developers to maintain the slopes safety effectively.

1.1. Slope Monitoring Slope is any ground in which surface forms an angle with the horizontal plane [8]. Rainfall or an earthquake can weaken slope structure thus causing slope collapses abruptly. Sudden collapse of slope may cause a great natural disaster that may result loss of life and property. The effectiveness of monitoring depends on the extent of the slope, and giving adequate warning before it failed depends also on the ability of the monitoring system to detect the warning [9]. Visual inspection of the slope often needs to be done to check the stones that are loose and potentially dangerous. Besides identifying the deterioration slope caused by weathering, erosion, cutting Figure 1. Flowchart of the procedure of the method to obtain the slope and blasting damage. Instead of probing failure that will visual information occur, the emergence of crack can be an early sign of failure and it is important to monitor the cracks occurring. 2.1. Site Survey This can be done by recording the number and width of cracks at regular time intervals. This is suitable for low In this study, the chosen location was at Taman hazard potential failure [10]. Harmoni Vista, at Jalan Panchor, Pagoh. Slope within 150 meters in length and 20 meters in height located at 1.2. Unmanned Aerial Vehicle (UAV) 2.15141808° latitude, 102.72160858° longitude and 63.318 meters altitude as a case study location as shown An unmanned aerial vehicle (UAV) is a type of aircraft in Figure 2. From the information given by Sime Darby thatis operated without a human pilot on board. This Officer, a school has planned to build on the high ground current technology is widely applied in various fields to of the slope. As a school it will be built there, the slope facilitate work. In the construction industry, UAV use for alongside must be monitored. monitoring the construction site, construction inspection, thermography, infrared, photogrammetry, transport applications and marketing activities [11]. The most obvious application of UAV camera to produce aerial photography for the project using an airplane or helicopter conventionally has led to high expenses. However, replacing the conventional method with the UAVs will be able to provide high-quality video images at lower cost [12]. Image of aerial photographs taken in accordance with the characteristics of the flight path is straight and parallel. Images are taken at a certain time of intervals so Figure 2. Study Area that the area covered by each image along the flight path contains most of the photo image area covered by the The slope area is also near to the main road as shown in previous image. map in figure 3. Thus, the slope is the best location to run this study because hazardous symptoms on the slope failure can be detected earlier and prevention steps can be taken in 2. Methodology order to avoid any slope failure to happen and enhance the safety of the location. Figure 1 shows the flowchart of the procedure of the

Civil Engineering and Architecture 7(6A): 1-6, 2019 3

Figure 3. Slope’s location on map

Figure 4. Parameters the flight plan and size mission

4 Monitoring Slope Condition Using UAV Technology

2.2. Data Collection and Processing slope.

In the planning process of this project, pilot test has been Table 2. Comparison of slope studied done for data accuracy and to ensure the flight safety. JKR slope maintenance Studied slope analysis Checking weather condition is a must to get the better standard guide visual images and very critical not to fly drone in bad Perieod 1 Perieod 2 Perieod 3 weather conditions especially during rainy days. Next, the Huge Rock(Boulder) Absent Absent Absent drone’s details planning using Pix4D Capture must be set Land movement Absent Absent Absent as return home location and need to set correctly to ensure Tension cracks Absent Absent Absent the safety return of UAV. Moreover, the parameters of Surface crack on Absent Absent Absent flight plan and size mission must be set before launching drainage the drone as shown in Figure 4. In this project, flight plan and size mission were using double grid to build a 3D 3.1. Boulder model and 32 meters for the flight drone altitude. This pilot test is very important in order to avoid any problem Based on the image analysis, boulder was absent in occurring during the flight mission for actual data period 1, 2 and 3 because the slope is vegetated surface collection. covered as shown in Figure 5. There is sufficient sunlight There were two applications that had been used for the to support vegetation growth in the study area. Many data collection which were Dji Go and Pix4Dcapture. The failures in rock slopes involve minor rock falls. The flight plan for the study area was set within the Pix4D presence of unstable upslope boulders which could impact Capture with frontal overlapping of 75%, side overlapping on the boulder, particularly for those that are already ο. of 65% and the angle of camera of 70 [13]. The flight was overhanging or resting on other boulders where the carried out at three different periods of time as shown in contact is open or soil-filled or dipping out of the slope Table 1. [14].

Table 1. Dates of flight

Period Flight date 1 18 October 2018 2 18 November 2018 3 7 December 2018

All the visual images collected from the UAV were saved in the UAV’s storage. The pictures were transferred into the computer and were generated using Pix4D Mapper and Global Mapper software to build slope’s 3D modelling and get the visual image of the slope. Then, the 3D model was analyzed to identify the possibility of slope failure symptoms.

3. Results and Discussion Figure 5. Vegetated surface covered

In the study, three flight missions with a UAV were 3.2. Land Movement carried out to monitor the slope. An average of 90 images was taken for each flight to generate the 3D models using The contour analysis result from Global Mapper pix4D Mapper software. From the 3D models, the slope software did not show any deflection as shown in Figure 6. studied can be analyzed to find the signs or symptoms of This indicates no sign of land movement during the failure based on Jabatan Kerja Raya (JKR) slope observation period. The slope movement sign will show as maintenance standard guide. The slope is categorized as an elevation difference in the contour analysis within the vegetated surface cover [14]. Table 2 shows the JKR slope observation time. maintenance standard guide is compared with the studied

Civil Engineering and Architecture 7(6A): 1-6, 2019 5

Figure 6. Contour in Global Mapper software 3.3. Tension Cracks 3.4. Surface Cracks on Drainage Tension cracks of slope occurred in period 1, 2 and 3 in The finding indicates surface cracks on drainage only displaying calibrated camera in initial position (m) occur in period 3. Figure 8 shows lateral crack on surface 246612.86, 237946.76, 48.1. The length of the tension drainage occurs in period 3. Longitudinal cracks and lateral crack was about 1.5 meter as shown in Figure 7. Tension cracks originate at a certain point. Based on the analysis, crack tends to form near the top of the slope as the the lateral crack was found in displaying calibrated camera condition of limiting equilibrium and failure to develop. A in initial position (m) that is 246602.07, 237969.16, 49.2. tension crack may develop in a slope when the inclination Maintenance work is required to repair minor crack in angle of the slip surface is steep and when the sliding mass order to prevent further crack. This minor crack can be is sitting on a weak foundation material. The typical fixed using cement mortar or flexible sealing compound. maintenance work required is to repair the cracks or Furthermore, clearance works are also needed outside of spalling. Cracked impermeable surface cover should be site boundaries to prevent debris from blocking the repaired by cutting a chase along the line of the crack, drainage system. which is to be filled with similar slope cover material or flexible sealant.

Figure 7. Tension crack of the slope Figure 8. Lateral crack on surface drainage

6 Monitoring Slope Condition Using UAV Technology

Unmanned Aerial Vehicles: A Survey," International 4. Conclusions Journal of Control, Automation, and Systems, 36-44. Currently, the UAV usage is in line with the changes of [8] Ali, P. D. N. (2018). Slope Monitoring and Engineering the nowadays technologies to make the inspection and Concept: Slope Recovery Course, Skudai: Universiti monitoring work easier. This study introduces UAV as an Teknologi Malaysia (UTM). alternative way to obtain data in the form of accurate [9] Cawood, F. T., & Stacey, T. R. (2006). Survey and images, 3D modelling of the slope to facilitate the slope Geotechnical Slope Monitoring Considerations. Journal of monitoring process. Inspection used by Unmanned Aerial the Southern African Institute of Mining and Metallurgy, Vehicle (UAV) is able to obtain clearer and more accurate 106(7), 495–501. visual images. Every slope whether high risk slope or [10] SIG-EHS-GU013. (2013). Slope stability guidelines normal slope should be monitored regularly to identify any sig-ehs-gu013, 36. initial defects or damages. Therefore, if any defects or problems occur, any action may be taken and remedied [11] Rok Cajezk, Gic Dradnje D.O.O, D. U. K. (2016). An Unmanned Aerial Vehicle for Multi-Purpose Task in detected earlier based on visual images and slope 3D Construction Industry, 11, 314–327. modelling. Application of UAVs is a more flexible, fast and effective method for the acquisition of taking the visual [12] Opfer, N. D., & P.E, D. R. S. (2014). Unmanned Aerial data Vehicle Applications and Issues for Construction. [13] Madawalagama S., (2019). Introduction to Photogrammetry. [Art]. Geoi formatics Center Asian Institute of Technology Acknowledgements Thailand. https://www.itu.int/en/ITU-D/Regional-Presence /AsiaPacific/SiteAssets/Pages/Events/2018/Drones-in-agri The authors would like to express our deepest gratitude culture/asptraining/B_introduction_to_photogrammetry1.p df to all parties who have contributed to this research, especially to Tier 1 Grant (Vote H123), Pejabat Pendaftar [14] Jabatan Kerja Raya (JKR), Guidelines on Slope UTHM, Office for Re-search, Innovation, Maintenance in Malaysia, Kuala Lumpur: Slope Commercialization and Consultancy Man-agement, Engineering Branch Public Works Department Malaysia, 2006. (ORICC), Center for Diploma Studies (CeDS), Sime Darby Property, Bandar Universiti Pagoh.

REFERENCES [1] Komoo, I., Aziz, S., & Sian, L. C. (2011). Incorporating the Hyogo Framework for Action into landslide disaster risk reduction in Malaysia. Bulletin of the Geological Society of Malaysia, 57(57), 7–11. [2] Braathen A., Blikra L. H., Berg S. S. and Karslen F. (2004) "Rock-slope failures in Norway; type, geometry, deformation mechanisms and stability," Norwegian Journal of Geology, 67-88. [3] Hunt R. E. (2005). Geotechnical Engineering Investigation Handbook, United States of America: CRC Press. [4] Giordan, D., Manconi, A., Tannant, D. D., & Allasia, P. (2015). UAV: Low-cost remote sensing for high-resolution investigation of landslides. International Geoscience and Remote Sensing Symposium (IGARSS), 5344–5347. [5] Omar, M., Pichan, S., Rosli, M. (2015). A Detailed Investigation on Slope Failures at Federal Road (FT185) - The Impending Improvement to JKR's Existing Guidelines For Road in Hilly Terrain. International Conference on Slopes. Kuala Lumpur, Malaysia: Slope Engineering Branch, JKR, 23-37. [6] Utusan Online. (2016, January). Mengawasi cerun 24 jam. Utusan Online, p. 1. Yong Peng. [7] Chao, H., Cao, Y. and Chen, Y. (2010) "Autopilots for Small

Civil Engineering and Architecture 7(6A): 7-18, 2019 http://www.hrpub.org DOI: 10.13189/cea.2019.071402

Sub-surface Profiling Using Electrical Resistivity Tomography (ERT) with Complement from Peat Sampler

Kasbi Basri1,2,*, Norhaliza Wahab1,2, Mohd Khaidir Abu Talib1,2, Adnan Zainorabidin1,2

1Faculty of Civil Engineering and Environmental, Universiti Tun Hussein Onn Malaysia, Malaysia 2Research Centre for Soft Soil (RECESS), Universiti Tun Hussein Onn Malaysia, Malaysia

Received July 30, 2019; Revised September 23, 2019; Accepted December 10, 2019

Abstract The pre-construction works include the to accumulate sufficient data regarding the soil profile. geotechnical investigation which comprises the surface Conventional method such as boring is proven to be and subsurface exploration. Sub-surface exploration often efficient to provide sufficient data needed. However, the causes several difficulties such as high cost, time method is expensive, time consuming and intrusive. consuming, localized investigation and intrusive. Large-scale project such as highway and railway usually Furthermore, investigation on soft soil such as peat often required high number of boreholes to provide sufficient raises several additional problems such as high risk of data for detail interpretation. Causing the construction time sample disturbance, difficult access for heavy equipment and cost to increase drastically. and inconsistent data due to the heterogeneity of peat. The In recent years, geophysical methods have gained much advancement of geophysical method such as Electrical attention as it is a non-intrusive investigation, and involves Resistivity Tomography (ERT) allows the determination of larger volume of investigation and rapid data interpretation. soil profile in time-efficient manner, economic, larger The advancement of the geophysical method such as volume of investigation and non-intrusive. This study Electrical Resistivity Tomography (ERT) allowed the focused on the determination of soil profile, particularly mapping of the electrical resistivity distribution in the peat layer using ERT method with complement from peat Earth, thus, allowing the estimation of the subsurface sampler data. The study was conducted at Parit Nipah, heterogeneity [1]. The measurement of the ground surface Johor. The results revealed a high accuracy profile (layers of materials with different individual resistivity) delineated by the ERT method with only less than 8% when current is injected into the ground through two percentage of error as compared to peat sampler profile. current electrodes allows the determination of the The comparison between Schlumberger and Wenner array subsurface resistivity distribution [2, 3]. The critical part in showed that; the Schlumberger showed superior depth of ERT is computing the resistivity pseudosection. Everett [4] penetration, with almost 3 times deeper penetration relative mentioned that, ERT imaging is performed by matching to Wenner. The Schlumberger array is also able to the measured apparent resistivity pseudosection to a delineate lateral variation within the peat layer. Finally, the computed pseudosection. The ERT measurement can be resistivity value of peat obtained ranged from 100.8 to interpreted into 1-D, 2-D and 3-D high resolution 139.5 ohm.m with both arrays having consistent results. resistivity images. Compared to other electrical methods, ERT is considered as superior, because quantitative results Keywords ERT, Peat, Peat Sampler are obtained by using controlled source of specific dimensions [5]. Also, the electrical properties are greatly affected by the geological parameters such as mineral, fluid content, porosity and degree of water saturation [3, 6, 1. Introduction 7]. Therefore, it is possible to separate different soil layer by mapping the soil electrical resistivity distribution. The determination of soil profile is very critical in Peat refers to a highly organic soil with an organic geotechnical investigation. To adequately design the content more than 75%. Commonly in Malaysia, and the geotechnical structure for any construction, it is necessary color of peat is dark reddish brown to black [8, 9]. Peat is 8 Sub-surface Profiling Using Electrical Resistivity Tomography (ERT) with Complement from Peat Sampler

also well known to have high natural water content, high studies reported involving the geophysical methods with compressibility, high organic content, low shear strength the conventional method especially on soft soil, the level of and low bearing capacity [8]. According to Huat [10], uncertainty regarding the data obtained using the organic content has significant effect on the resistivity geophysical methods is still high. Therefore, extensive value. The critical aspect of peat was the ability to store study to relate geophysical data with conventional method very high-water content. The ability of holding is critical. Integration between geophysical and considerable amount of water was governed by the soil conventional method would provide rapid, larger volume, structure characterized by organic coarse particles (fibers) and more economic method of investigation. which is characterized as very loose and hollow [11]. The high amount of water stored within peat allows the current flow smoothly which allows subsurface resistivity 2. Site Description mapping. Very often, underneath the peat layer is sensitive soil such as soft clay. The presence of clay minerals The study area was located at Parit Nipah, Johor, increases the conductivity through the ion exchange Malaysia. The location was situated in the quaternary process [6]. Therefore, high contrast of resistivity region, which consists of marine and continental deposits distribution is expected to distinguish between peat and such as clay, silt, sand, peat with minor gravel as shown in soft clay layer as peat has higher resistivity value compared Figure 1. The surrounding area was an agricultural area to soft clay. planted with palm trees and pineapple. The peat sampler This study focused on delineating the soil stratigraphy delineated that the top 4 meters of soil in the study area was by using geophysical method known as ERT method. This mainly peat. Therefore, the location was chosen as the peat method is expected to replace or complement the depth was among the deepest known to date in the area. conventional borehole method for rapid and sustainable The location also was reserved for research purposes, thus, soil profile investigation. However, due to only few prior ease access into the area.

Civil Engineering and Architecture 7(6A): 7-18, 2019 9

Figure 1. Geological map of Parit Raja [12] 3. Field Investigation 3.1. Electrical Resistivity Tomography (ERT) The field study comprises geo-electrical investigations The Electrical Resistivity Tomography (ERT) was using Electrical Resistivity Tomography (ERT) and peat carried out using Schlumberger and Wenner array. The profile using peat sampler. The study focused on the tests were conducted during the rainy season somewhere in determination of the depth of peat by means of June. The weather caused by the Southwest Monsoon (late non-destructive test (NDT) and was verified by the May to September) causes the soil to be medially wet due conventional intrusive test. to average rainfall. Thus, the saturated soil allows the distribution of electrical current easily. Figure 2 shows the general arrangement for Schlumberger and Wenner array. 10 Sub-surface Profiling Using Electrical Resistivity Tomography (ERT) with Complement from Peat Sampler

Figure 2. Array arrangement and pseudosection example for Schlumberger and Wenner arrays [3] Both arrays were chosen as the study only focused on electrodes for two Lund imaging cable) were used as only the shallow profile and involved heterogeneous material. the short setup as shown in Figure 4 that was used. Two According to Morris et al. [13], Schlumberger array is different electrodes spacing which were 1 and 1.5 m superior in detecting lateral resistivity inhomogeneity. electrode spacing was adopted. The short spacings were While, Wenner array is capable in detecting vertical used to maximize the number of measurements, thus, changes [14]. The presence of clay layer also provides obtaining high resolution results and minimising the high cation exchange capacity (CEC) to the soil particles noise which can cause uncertainty to the results. The which suggests that low resistivity values are expected number of measurements, depth of investigation, which makes Schlumberger and Wenner array superior sensitivity to noise, horizontal and vertical resolution are compared to other types of array. As mentioned by governed by the electrode spacing [3, 16]. The data Moreira et al. [15], Schlumberger and Wenner arrays are processing and inversion were performed using recommended for low resistivity with emphasis on RES2DINV software. The RES2DINV inversion code schlumberger array. Therefore, Schlumberger and Wenner allowed the determination of 2-D inverse model resistivity arrays were chosen for this study as the method had the measurement of the subsurface using the measured advantages in term of the suitability of the method on the apparent resistivity [17]. A total of 4 array lines for each target materials and the simplicity of the configurations. array setup were investigated and analyzed. Two array The main equipment involved in the test includes lines were provided for each electrode spacing. The two ABEM Terrameter SAS 4000, Lund imaging cable, array lines were configured to intersect at the midpoint to jumper, electrode selector and steel electrodes. Figure 3 allow comparison. The first array line was fixed as x-axis shows some of the equipment involved in ERT test. A line (West-East) and the second array line was at y-axis total of 41 steel electrodes (maximum number of (North-South).

Civil Engineering and Architecture 7(6A): 7-18, 2019 11

Figure 3. ERT equipment

Figure 4. ERT field arrangement 3.2. Peat Sampler were as shown Figure 5 and Figure 6. The image obtained shows good contrast between different soil layers The peat sampler was used to obtain the semi-disturbed allowing determination of peat thickness. The profiles samples. The procedures follow the guideline provided by obtained by Schlumberger array show superiority in depth Eijkelkamp Agrisearch Equipment [18]. The peat sampler of investigation compared to Wenner array with similar tests were conducted to verify the soil stratigraphy electrode spacing. The finding is in good agreement with obtained by the ERT method. Peat sampler was used as an Apostolopoulos [19] that mentioned Schlumberger array alternative to expensive and complicated borehole. Three provided great sensitivity to depth and good penetration investigation points located along the array lines were of depth through a conductive surface layer. The profile investigated. The profiles were determined for every 0.5 depth obtained using Schlumberger array using 1 meter meter until the depth of 5 meters. and 1.5 meters electrode spacing was 7.8 and 11.8 meters respectively. While, using Wenner array the profile depth was only 2.5 and 3.7 meters. 4. Results Discussion The soil stratigraphy obtained using Schlumberger array shown in Figure 5 shows that the depth of peat layer was 4.1. Soil Resistivity Profile determined at the top 3.9 and 3.7 meters for 1 meter and 1.5 The results for the 2-D stratigraphy profiles obtained meters electrode spacing respectively. The resistivity value

12 Sub-surface Profiling Using Electrical Resistivity Tomography (ERT) with Complement from Peat Sampler

obtained for peat ranged from 100.8 to 139.5 ohm.m. The transition layer between peat and soft clay. The resistivity resistivity value started to drop drastically at the depth value obtained for soft clay ranged from 5.4 to 22.3 between 3.9 to 4.8 meters and 3.7 to 4.7 meters for 1 meter ohm.m. and 1.5 meters electrode spacing respectively showing the

(a)

(b)

(c)

(d)

Figure 5. 2-D soil stratigraphy using Schlumberger array; (a) 1.0 m at x-axis (West-East), (b) 1.5 m at x-axis (West-East) (c) 1.0 m at y-axis (North-South) and (d) 1.5 m at y-axis (North South)

Civil Engineering and Architecture 7(6A): 7-18, 2019 13

(a)

(b)

(c)

(d)

Figure 6. 2-D soil stratigraphy using Wenner array; (a) 1.0 m at x-axis (West-East), (b) 1.5 m at x-axis (West-East) (c) 1.0 m at y-axis (North-South) and (d) 1.5 m at y-axis (North-South)

14 Sub-surface Profiling Using Electrical Resistivity Tomography (ERT) with Complement from Peat Sampler

The soil stratigraphy obtained using Wenner array was occurred mainly on the top 3 meters suggesting the effect as shown in Figure 6. As mentioned earlier, the depth of of lateral heterogeneity of peat and the different electrode penetration using Wenner array was less as compared to spacing. A shorter spacing generates denser data Schlumberger array. Therefore, the results obtained compared to a longer spacing. According to Okpoli [24], covered only the profile for peat layer. The resistivity value the sensitivity of geo-electrical resistivity increases with obtained ranged from 110.9 to 130.6 ohm.m. The results the increase in data density and decreases with the were in good agreement with the Schlumberger array. increase in depth of investigation (longer electrode The high resistivity value of peat was governed by the spacing). For the depth below 3 meters, the results with low degree of humification on the study area. According different configurations agreed well with minimum to Mansor and Zainorabidin [20], the Parit Nipah peat was variations. From Figure 8, the resistivity value of peat hemic peat with 88% of organic content. High organic generally decreased slightly with depth. According to content and low decomposition rate cause the resistivity Ling et al. [23], the organic content of peat at Parit Nipah to increase [10]. The electrical conductivity response of decreases with depth. The resistivity value decreases as peat also depends mainly on the conductivity of the fluid the organic content decreases [10]. saturating the sample [21]. This suggested that, the Further, El-galladi et al. [25] mentioned that the conductivity of fluid saturating the peat on the study area conductivity of peat pore water increases with depth. This was low. While, the drastic drop in resistivity value in soft suggests the conductivity of peat pore water at Parit Nipah clay layer was due to high cation exchange capacity (CEC) increases with depth. As mentioned by Kim et al. [26], the provided by the clay fraction [10, 22]. According to Ling conductivity of pore fluid affects the resistivity of soil. At et al. [23], the CEC value for Parit Nipah peat was only in the transition layer between peat and soft clay the the range of 6.79 to 8.5 meq/100 g. Thus, lower CEC and resistivity value decreased drastically due to the presence high organic content contributed to high resistivity value of clay fraction. As mentioned previously, the presence of of peat. clay fraction provides high CEC, thus, causing a lower The investigation using Schlumberger array had also resistivity value [10]. Also, at the soft clay layer, the revealed a significant variation of resistivity value resistivity value was very low. This was governed by high laterally within the peat layer. Apostolopoulos [19] CEC and presence of saturated clay [10, 22]. mentioned that, Schlumberger array provide high resolution of horizontal variation. This suggests that, 4.2. Soil Profile Schlumberger array could map the resistivity variation due to heterogeneity of peat. As for the soft clay layer, the The soil profile obtained was as shown in Table 1. The resistivity value was consistent as the soil was results showed that, the peat layer was determined until 4 homogeneous. However, the variations of resistivity value meters depth, followed by the transition layer of peat and were not obtained using Wenner array configurations soft clay and finally the soft clay layer. From the table, the suggesting Schlumberger array superiority in delineating image obtained shows visible fibre (root, etc.) showing the lateral variations. The finding agreed well with Morris et heterogeneity of peat. The sample obtained also shows al. [13], which stated that the Schlumberger array is slight changes in peat colours from dark reddish brown to superior compared to Wenner array in detecting lateral reddish brown. This suggests that the degree of resistivity inhomogeneity. Loke [14] also mentioned that decomposition of peat might differ with depth. The degree Wenner array is incapable in detecting horizontal changes. of decomposition is related to the moisture, organic and From the 2-D stratigraphy, 1-D resistivity value was fibre content. This suggests that there is potential extracted from the center of the array line to obtain the correlation between the resistivity value and the changes in resistivity value of soil. Figure 7 shows the resistivity degree of decomposition. However, the purpose of the peat value of soil with depth for different configurations and sampler was only to obtain the soil profile for comparisons Figure 8 shows the resistivity value obtained for the top 5 with the ERT method. Thus, further description regarding meters. From Figure 7, the graph shows a slight variation the effect of degree of decomposition will not be discussed.

Civil Engineering and Architecture 7(6A): 7-18, 2019 15

Figure 7. Peat resistivity value with different configuration

Figure 8. Soil resistivity value

16 Sub-surface Profiling Using Electrical Resistivity Tomography (ERT) with Complement from Peat Sampler

Table 1. Soil profile obtained using peat sampler

Soil Profile Depth (m) Remarks

0 – 0.5 Peat

0.5 – 1.0 Peat

1.0 – 1.5 Peat

1.5 – 2.0 Peat

2.0 – 2.5 Peat

2.5 – 3.0 Peat

3.0 – 3.5 Peat

3.5 – 4.0 Peat

Peat + soft 4.0 – 4.5 clay

4.5 – 5.0 Soft clay

5. Comparison between ERT Method respectively when compared to the peat sampler data. and Peat Sampler While, for the transition layer from peat to soft clay layer, the percentage of error for 1 meter and 1.5 meters The soil stratigraphy obtained using the electrical electrode spacing was 6.7 % and 4.4 % respectively. resistivity method and the soil profile from the peat Overall, the results show that the shorter electrode spacing sampler was compared. The percentage of error for the provides higher accuracy and better sensitivity compared observed and exact value was calculated using the to longer electrode spacing. As mentioned by Okpoli [24], standard equation for percent of error [(observed value – lesser depth of penetration (shorter electrode spacing) exact value)/exact value x 100]. Table 2 summarizes the causes the sensitivity of electrical resistivity of the comparison between the soil profiles obtained using ERT subsurface to increase. The results concluded that, the method and peat sampler. The results obtained using electrical resistivity method was able to determine the Wenner array were not included due to insufficient depth changes of soil strata with high accuracy. The percentage of penetration. The percentage of error for peat layer of error between the peat profiles determines using the determined using Schlumberger array with 1 meter and peat sampler and the electrical resistivity method was 1.5 meters electrode spacing was 2.5 % and 7.5 % only less than 8%.

Civil Engineering and Architecture 7(6A): 7-18, 2019 17

Table 2. Summary of comparison between ERT and peat sampler method Schlumberger Schlumberger Peat sampler Percentage of error for Percentage of error for Soil type 1.5m (m) 1m (m) (m) Schlumberger 1.5 m (%) Schlumberger 1.0 m (%) Peat 0 – 3.7 0 – 3.9 0 – 4 7.5 2.5 Peat + soft clay 3.7 – 4.7 3.9 – 4.8 4 – 4.5 4.4 6.7

6. Conclusions [5] Telford, W.M., et al., Applied geophysics. Vol. 1. 1990: Cambridge university press. The application of ERT method on the determination of [6] Aizebeokhai, A.P., K.D. Oyeyemi, and O.T. Kayode. soil profile shows promising results. The soil profile Multiple-gradient array for near-surface electrical obtained using the ERT method agreed well with the peat resistivity tomography. in Near-Surface Asia Pacific sampler data. The overall percentage of difference was Conference, Waikoloa, Hawaii, 7-10 July 2015. 2015. only less than 8%. Comparison between Schlumberger Society of Exploration Geophysicists, Australian Society of Exploration. and Wenner array shows superiority of Schlumberger in terms of depth of penetration and lateral variation [7] Miele, M., et al. Rectangular Schlumberger resistivity detection. As the Schlumberger array was able to arrays for delineating vadose zone clay-lined fractures in delineate the lateral variations caused by the heterogeneity shallow tuff. in Symposium on the Application of Geophysics to Engineering and Environmental Problems of peat. Overall, the average resistivity value of peat 1996. 1996. Society of Exploration Geophysicists. obtained ranged from 100.8 to 139.5 ohm.m. The high value of resistivity was governed by high organic content [8] Zainorabidin, A. and D.C. Wijeyesekera, Geotechnical and low CEC value. The results concluded that the ERT challenges with Malaysian peat. Advances in Computing and Technology, 2007: p. 252-261. method was able to delineate the soil profile with high accuracy in timely efficient manner, lower cost, larger [9] Zainorabidin, A. and H.M. Mohamad, Engineering volume of investigation and non-intrusive. To further Properties of Integrated Tropical Peat Soil in Malaysia. improve the application of ERT method, the author Electronic Journal of Geotechnical Engineering, 2017. 22(02): p. 457-466. recommends that special attention on peat organic content, moisture content, CEC and degree of humification should [10] Huat, B.B., et al., Geotechnics of organic soils and peat. be needed to better understand the peat resistivity value 2014: CRC Press. changes. [11] Kazemian, S., Organic Soils and Peats, in Encyclopedia of Engineering Geology. 2017, Springer. p. 1-5. [12] Construction Research Institute of Malaysia, Guidelines for Acknowledgements Construction on Peat and Organic Soils in Malaysia. 2015. The authors would like to thank the University of Tun [13] Morris, M., J.S. Rønning, and O.B. Lile, Detecting lateral Hussein Onn Malaysia, and Ministry of Education resistivity inhomogeneities with the Schlumberger array. Malaysia for their generous grant of this research, TIER 1 Geophysical Prospecting, 1997. 45(3): p. 435-448. research grant, GPPS grant vot number H009 and H011. [14] Loke, M., Tutorial: 2-D and 3-D electrical imaging surveys. The authors also would like to extend their gratitude to 2004. Research Centre for Soft Soil (RECESS) for allowing the use of research equipment and facilities. [15] Moreira, C.A., M. Montenegro Lapola, and A. Carrara, Comparative analyzes among electrical resistivity tomography arrays in the characterization of flow structure in free aquifer. Geofísica internacional, 2016. 55(2): p. 119-129. REFERENCES [16] Baines, D., et al., Electrical resistivity ground imaging (ERGI): a new tool for mapping the lithology and geometry [1] Slob, E., Optimal acquisition and synthetic electrode arrays, of channel‐belts and valley‐fills. Sedimentology, 2002. in SEG Technical Program Expanded Abstracts 2004. 2004, 49(3): p. 441-449. Society of Exploration Geophysicists. p. 1389-1392. [17] Griffiths, D. and R. Barker, Two-dimensional resistivity [2] Herman, R., An introduction to electrical resistivity in imaging and modelling in areas of complex geology. geophysics. American Journal of Physics, 2001. 69(9): p. Journal of applied Geophysics, 1993. 29(3-4): p. 211-226. 943-952. [18] Equipment, E.A., Peat Sampler Operating Instruction. [3] Loke, M., Electrical imaging surveys for environmental and Netherlands: Eijkelkamp Agrisearch Equipment, 2014. engineering studies. A practical guide to, 1999. 2. [19] Apostolopoulos, G., Combined Schlumberger and [4] Everett, M.E., Near-surface applied geophysics. 2013: dipole-dipole array for hydrogeologic applications. Cambridge University Press. Geophysics, 2008. 73(5): p. F189-F195.

18 Sub-surface Profiling Using Electrical Resistivity Tomography (ERT) with Complement from Peat Sampler

[20] Mansor, S.H.B. and A.B. Zainorabidin, Stress-Strain Behavior of Parit Nipah Peat, in InCIEC 2014. 2015, Springer. p. 515-523. [21] Ponziani, M., et al., Influence of water content on the electrical conductivity of peat. Int Water Technol J (IWTJ), 2011. 1(1): p. 14-21. [22] Jakalia, I., et al., Implications Of Soil Resistivity Measurements Using The Electrical Resistivity Method: A Case Study Of A Maize Farm Under Different Soil Preparation Modes At KNUST Agricultural Research Station, Kumasi. 2015. [23] Ling, F.N., et al. Geochemistry properties of southern Malaysian organic soil. in Applied Mechanics and Materials. 2013. Trans Tech Publ. [24] Okpoli, C.C., Sensitivity and resolution capacity of electrode configurations. International Journal of Geophysics, 2013. [25] El-Galladi, A., et al., Mapping Peat Layer Using Surface Geoelectrical Methods At Mansoura Environs, Nile Delta, Egypt. Mansoura Journal of Geology and Geophysics, 2007. 34(1): p. 59-78. [26] Kim, M.-I., et al., Surface geophysical investigations of landslide at the Wiri area in southeastern Korea. Environmental Earth Sciences, 2011. 63(5): p. 999-1009.

Civil Engineering and Architecture 7(6A): 19-32, 2019 http://www.hrpub.org DOI: 10.13189/cea.2019.071403

Empirical Mode Decomposition Couple with Artificial Neural Network for Water Level Prediction

Eng Chuen Loh*, Shuhaida Binti Ismail, Azme Khamis

Faculty of Applied Sciences and Technology, Universiti Tun Hussein Onn Malaysia, Malaysia

Received August 4, 2019; Revised October 10, 2019; Accepted December 16, 2019

Abstract Natural disaster brings massive destruction commercial [2]. towards properties and human being and flood is one of There are several impactful factors that affect them. In order for the government to take earlier action to inconsistent flood occurrence. For example, temperature, reduce the damages, an accurate flood prediction is humidity, dew point temperature, wind speed, streamflow necessary. In Malaysia, Kelantan is categorized as a high volume, and rainfall volume. The streamflow volume flood risk area, thus this study focuses on Kelantan flood indicates how much the volume of water the river can hold prediction. This study is to investigate the effect of to sustain the rainfall volume. Higher temperature and decomposition for water level prediction by applying wind speed result in faster water particles’ moves thus Artificial Neural Network (ANN) forecasting model. In easier to evaporate into the atmosphere. Humidity also this study, Empirical Mode Decomposition (EMD) is used affects the water particle in the air to be condensed out of as the decomposition method. The best Intrinsic Mode the atmosphere. Function (IMF) for each input variable is selected using Kelantan is chosen as the study area for water level correlation-based selection method. The results showed prediction. According to reports, the occurrence of flood in that the performance of hybrid EMD and ANN is superior Kelantan is twice as much compared to other states in compared to other models, especially classic ANN model. Peninsular Malaysia. This is because the geological The reason for this outcome is that through decomposition location of Kelantan with located near to the Northeast Sea methods, ANN is able to capture more in-depth which is affected by the Monsoon Season every November information of the Kelantan hydrological time series data. to March. Besides, Malaysia is located near the Equator The resulting model provides new insights for which has higher average temperature compared to other government and hydrologist in Kelantan to have better countries. prediction towards flood occurrence. Water level data are rather complex and consistent. Thus, classical forecasting methods which require the fulfillment Keywords Artificial Neural Network, Empirical Mode of several assumptions such as stationary and linearity are Decomposition, Intrinsic Mode Function, Flood Prediction, not suitable. A more reliability and adaptability approach is Water Level essential to achieve a more accurate prediction. In the past decades, the number of studies applying machine learning approaches in hydrological time series prediction has increased. This is because machine learning approaches are more reliable and accurate compared to the classical 1. Introduction methods which attract the attention of researchers [3]. Flood is the most commonly happened natural disaster Artificial Neural Network (ANN) is one of the in the world. It also causes tremendous damages to commonly used methods in machine learning approaches. economics, properties, besides threatening human life and This method imitates the processes of a human brain which safety. In order to reduce such damages, an early issued receives, processes and produces information through a flood warning is essential. Thus, water level forecasting is series of nodes and connections. This enables ANN to essential to predict future flood occurrence. Prevention, provide higher flexibility and adaptability towards protection and preparation plan can be made by the complex and chaotic data handling. Besides, these methods government while evacuating the affected citizens [1]. can also be reconstructed into different architecture in Water level prediction also benefits other sectors such as order to deal with different data such as Long Short-Term agriculture, plants, domestics and industrial and Memory (LSTM), Convolutional Neural Network (CNN) 20 Empirical Mode Decomposition Couple with Artificial Neural Network for Water Level Prediction

and even hybrid with other methods. 2. Methodology Due to the inconsistency property of the hydrological time series data, data preprocessing methods are needed to EMD with the application of ANN in forecasting reduce the noise of the data [4]. Decomposition methods Kelantan water level for 30 days ahead forecast is can be used to reduce the noises in the signal or time series compared with the single ANN model. The best model is thus increasing the accuracy of the prediction. There are determined by comparing the performance accuracy of several different decomposition methods existing such as each model. Empirical Mode Decomposition (EMD), Singular Value Decomposition (SVD), Principle Component Analysis 2.1. Data (PCA), Discrete Wavelet Transformation (DWT) and more. In this study, Kelantan is chosen as the study area for Previous results show that the application of ANN in water level prediction. The hydrological data of Kelantan dealing with hydrological time series is significant. ANN are retrieved from the Department of Irrigation and has also been used in various fields, for example, Drainage Malaysia (DID). The rainfall data ranging from groundwater [5, 6], flood magnitude [7], water level [8, 9], 2005 to 2014 are taken from 33 stations whereas streamflow volume [10, 11], water quality [12, 13] and streamflow and water level data are taken from 3 stations. more. However, there are spaces for improvement and On the other hand, the temperature, humidity, dew point remodel However, there are still spaces for improvement temperature, wind speed, and pressure are collected at and remodel. Besides, different location would have Sultan Ismail Petra Airport which is located at Kota Bharu different outcome due to the different distribution of data for the same period of study. and variables. That this case study focused on these specific locations This study aimed to predict Kelantan water level using is because the locations are in high land and due to the ANN model. Empirical Mode Decomposition (EMD) is completeness of the data. Moreover, these areas are near to used to decompose the Kelantan hydrological data. The farms and paddy fields. This case study will also benefit to model is evaluated using four performance accuracies the agriculture section as well as the government and which are Mean Absolute Error (MAE), Root Mean Square hydrologists. Error (RMSE), Mean Square Error (MSE) and Mean Figure 1, Figure 2, and Figure 3 show the water level Arctangent Absolute Percentage Error (MAAPE). This time series for 3 different rivers in Kelantan which are study enables the government and hydrologist to have Sungai Lanas, Sungai Kelantan, and Sungai Golok better insight in dealing with hydrological time series respectively. forecasting.

Figure 1. Water Level Time Series of Sungai Lanas

Civil Engineering and Architecture 7(6A): 19-32, 2019 21

Figure 2. Water Level Time Series of Sungai Kelantan

Figure 3. Water Level Time Series of Sungai Golok

2.2. Data Preprocessing is the minimum actual value, and xmax is the maximum actual value. The tendency and accuracy of the ANN models would deteriorate if the input values’ range is inconsistent or 2.3. Artificial Neural Network (ANN) exceptionally large [14]. In order to overcome this issue, normalization method should be carried out to transform Artificial Neural Network is the commonly used method the input value into a predefined range. In most cases, the in machine learning approaches. It was introduced by range is set within 0 and 1 [15]. In this study, min-max McCulloch & Pitts in 1943 that was inspired by the normalization method is used to normalize the Kelantan’s architecture and the processing method of a human brain hydrological time series data. The equation of min-max [16]. This method is able to learn the information through normalization is shown as follows: the input layer and processes it using one or several hidden layers then processes it through the output layer. Due to its xxi  min X i  (1) xx superiority, it can be applied in various fields such as max min image data, signal data, time series data and more [17]. where Xi is the normalized value, xi is the actual value, xmin The basic architecture of ANN consists of one input

22 Empirical Mode Decomposition Couple with Artificial Neural Network for Water Level Prediction

layer, hidden layer and output layer each. Each of the time t, wi is the weight connection between input and layers is interconnected with a series of weighted hidden nodes, wj is the weight connection between hidden connections [18]. Each of the weighted connections is and output nodes, θ is the bias constant, α is the number of constantly changing in order to optimize the result by using input nodes, β is the number of hidden nodes, f(x) and g(x) a back-propagation training algorithm. The error is are the respective activation functions showed in (4) and propagated back in order to adjust the weighted connection (5). until the best performance is obtained. The weight of the fxpurelinxx    (4) connections is adjusted based on (2). ˆ ˆ 2 wwyyyttiii1     (2) fxtansigx     1 (5) 1 e2 x where wt+1 is the weight at time t, yi is the target output, ŷi is the predicted output and η is the learning rate. where x is the input/hidden node values. In this study, a three-layer ANN with Levenberg -Marquardt backpropagation training algorithm is used. 2.4. Hybrid Decomposition with ANN The Levenberg-Marquardt backpropagation training algorithm is modified from the Gauss-Newton method [19]. In this study, hybrid models with various decomposition This architecture consists of α numbers of input nodes, β methods are proposed to predict the water level at 3 rivers numbers of hidden nodes and 7 output nodes. The in Kelantan which are Sungai Lanas, Sungai Kelantan, and predicted outputs are obtained based on (3). Sungai Golok The input data undergo decomposition and then are  yfwgw xww  (3) connected to the hidden layer, simultaneously being tjiiij ji11   processed with the raw input data. The architecture of the where yt is the output value at time t, xt is the input value at hybrid model is shown in Figure 4.

Figure 4. Hybrid ANN Architecture

Civil Engineering and Architecture 7(6A): 19-32, 2019 23

2.5. Empirical Mode Decomposition (EMD) There are several selection methods for reasonable IMF such as frequency-based method, Kurtosis-based method, EMD is the fundamental part of the Hilbert-Huang energy-based method, correlation-based method and more transform (HHT) that decomposes the signal into the terms [22]. In this study, the correlation-based method is used to of Intrinsic Mode Function (IMF) [20]. There are two main determine the suitable IMF. IMF with higher correlation steps needed to perform EMD which are firstly obtaining coefficient indicates that the IMF is significant. The the IMF through the EMD algorithm then obtaining the formula of the correlation coefficient is as shown in (7). instantaneous frequency spectrum of the initial sequence using HHT. 1 ooyy In EMD algorithm, local maxima and minima are r   (7) nSS1  determined using the smooth envelope in order to produce oy upper and lower envelopes by connecting all local maxima where n is the number of observation, o is the actual value, and minima using cubic spline lines. The IMFs are ō is the mean actual value, So is the standard deviation of determined through a series of subtraction with the local actual value, y is the IMF value, ȳ is the mean IMF value, mean value. For every extraction of the IMF, a new set of and Sy is the standard deviation of IMF value. maxima and minima are produced. These sifting processes stopped when the residual became a monotonic function 2.6. Training Parameters where IMF extraction is no longer available [21]. Figure 5 illustrates the flow chart of the EMD algorithm. The The parameters shown in Table 1 were used for all the original signal can be reconstructed as in (6). neural network models. All models are set with the same

n parameter in order to obtain fair results. Meanwhile, the SFR (6) i1 i n architecture of each neural network model is shown in Table 2. where Fi is the IMF, Rn is the final residue, and n is the number of IMF identified. Table 1. Training Parameters of Neural Network

Parameter Value Training Algorithm Levenberg-Marquardt Data Partition 70:15:15 Transfer function tan-sigmoid + linear Maximum fail 6 Maximum epochs 500 Learning rate, α 0.01 Performance goal 0 Minimum gradient 1.00 × 10-6 μ 1.00 × 10-3 Maximum μ 1.00 × 1010

Table 2. Architecture for each Neural Network

Method Model Architecture

M1 ANN 44-89-3

M2 EMD-ANN 55-106-3

There are different approaches in deciding the number of hidden neurons in the neural network [23, 24]. In this study, the number of hidden neurons is set as 2n+1 where n is the number of input neurons [25].

2.7. Performance Accuracy In this study, four types of performance accuracy are applied to evaluate the accuracy of the predicted output for the forecasting models. The measurements that will be used in this study are MAE, MSE, RMSE, and MAAPE, Figure 5. Flow of IMF Search using EMD Algorithm whereby, the best model will be selected based on the

24 Empirical Mode Decomposition Couple with Artificial Neural Network for Water Level Prediction

smallest values for all measurements. 3. Result and Discussions MAAPE is used instead of Mean Absolute Percentage Error (MAPE) because MAPE will have difficulty dealing Figure 6 to Figure 16 show that the best IMF for each with actual value approaches zero. MAAPE is more robust input variables using correlation-based method selection. and less biased compared to MAPE due to the bounded Based on results obtained, it is concluded that hybrid of influence of outliers [26]. The equations for each of the EMD with ANN is able to predict the Kelantan’s water performance accuracy are shown as follows: level accurately. Moreover, the model is able to predict Sungai Lanas with the lowest error compared to other 1 n MAEyyˆ i1 ii (8) location where it obtained MAE of 0.3140, MAAPE of n 0.0122, MSE of 0.2396, and RMSE of 0.4895. The results also indicate that single ANN yielded the 1 n 2 MSEyyˆ i1  ii (9) worst performance accuracy. This is due to the fact that n single ANN is unable to perform well because of the incapability in dealing with an extremely large set of input 1 n 2 RMSEyyˆ (10) i1  ii variables and relatively complex data. Moreover, Table 6 n shows the processing speed for training the ANN architecture. It shows that the single ANN’s speed is as 1 n yy ˆ MAAPE  arctan ii i1 (11) triple as lower compared to other models, however, its ny i accuracy is insignificant. where yi is the actual value, ŷi is the predicted output, and n is the number of observation.

Figure 6. Best IMF of Streamflow Volume of Sungai Lanas

Civil Engineering and Architecture 7(6A): 19-32, 2019 25

Figure 7. Best IMF of Streamflow Volume of Sungai Kelantan

Figure 8. Best IMF of Streamflow Volume of Sungai Golok

26 Empirical Mode Decomposition Couple with Artificial Neural Network for Water Level Prediction

Figure 9. Best IMF of Temperature

Figure 10. Best IMF of Dew Point Temperature

Civil Engineering and Architecture 7(6A): 19-32, 2019 27

Figure 11. Best IMF of Humidity

Figure 12. Best IMF of Wind Speed

28 Empirical Mode Decomposition Couple with Artificial Neural Network for Water Level Prediction

Figure 13. Best IMF of Atmospheric Pressure

Figure 14. Best IMF of Water Level at Sungai Lanas

Civil Engineering and Architecture 7(6A): 19-32, 2019 29

Figure 15. Best IMF of Water Level at Sungai Kelantan

Figure 16. Best IMF of Water Level at Sungai Golok

30 Empirical Mode Decomposition Couple with Artificial Neural Network for Water Level Prediction

Table 3. Performance Accuracy for Sungai Lanas model failed to predict most of the peaks during period th th th th Model MAE MAAPE MSE RMSE 500 day to 1500 day, 2500 day to 3000 day, and 3500th day to 3700th day as shown in purple rounded M1 0.3911 0.0151 0.3526 0.5938 rectangle. Also, the model predicted a dried out situation M2 0.3140* 0.0122* 0.2396* 0.4895* where the water level is much lower compared to the * Indicate the best result among all of the models average water level. This situation occurred during the th th Table 4. Performance Accuracy for Sungai Kelantan period of 2000 day to 2500 day as shown in green rounded rectangle. Model MAE MAAPE MSE RMSE

M1 0.7945 0.0784 1.6873 1.2990

M2 0.6207* 0.0623* 1.0053* 1.0027* * Indicate the best result among all of the models

Table 5. Performance Accuracy for Sungai Golok

Model MAE MAAPE MSE RMSE

M1 0.5094 0.0253 0.5121 0.7156

M2 0.3745* 0.0187* 0.2837* 0.5327* * Indicate the best result among all of the models

Table 6. Processing Speed for each Model

Model Iteration Time Taken (s)

M1 13* 224*

M2 15 765

* Indicate the best result among all of the models Figure 18. Actual vs Predicted Value of Kelantan From the visual inspection of Figure 17, the model is able to capture the peaks within the period of 1st day to From the visual inspection of Figure 19, the model 500th day, 1500th day to 2000th day, and 3000th to 3621st successfully captured peaks during the period of 1000th day, day as shown in red rounded rectangle. However, the 2000th day to 2500th day, and 3000th day to 3700th day as model has failed to predict several peaks during period of shown in red rounded rectangle. However, the model failed 500th day to 1500th day, and 2000th day to 3000th day as to capture a few of the peaks within the period of 1000th shown in purple rounded rectangle. day to 2000th day, and 2500th day to 3000th day as shown in purple rounded rectangle. In general, these models successfully capture the seasonal trend of water level data.

Figure 17. Actual vs Predicted Value of Lanas Figure 19. Actual vs Predicted Value of Sungai Golok From the visual inspection of Figure 18, the model is only able to capture the peaks within period of 1st day to According to the figures of actual versus predicted value, 500th day, 1500th day to 2000th day, and 3000th day to they showed that the predicted values are relatively close to 3500th day as shown in red rounded rectangle. But, the the actual value. In addition, the figures always proved that

Civil Engineering and Architecture 7(6A): 19-32, 2019 31

the hybrid of EMD and ANN has the ability to capture the seasonal trend of the hydrological time series significantly. From the figures, the model is also able to capture the peak of each location. This enables the government to foresight the future flood occurrence more accurately and how severe the flood will be. However, the predicted values contain more noise compared to the actual value. This is due to the noise carried forward from the 33 stations for rainfall volume data collection. Yet, the robustness of the model remains. According to the visual inspection of Figure 20 to Fig 22, it is concluded that the model is significant in predicting Kelantan water level data. Figure 20 and Figure 22 showed that the model is able the capture the uptrend of the water level in Sungai Lanas and Sungai Golok during the time period from 17th day onwards. However, Figure 21 shows that the model unsuccessfully captures the uptrend in Figure 22. A Month Ahead Forecast of Sungai Golok Sungai Kelantan. 4. Conclusions In conclusion, the hybrid of EMD and ANN model is superior compared to the other five models. Moreover, this model is able to predict Sungai Lanas with the lowest error among all three rivers. However, there are some limitations in forecasting Kelantan water level data using the current model. The network structure of the model contributes significant influence toward its accuracy. For example, the number of the hidden layers, the number of the hidden neurons, the type of family member value, the level of decomposition, the training algorithm and more. Thus, in future studies, determining the suitable network structure is essential through trial and error method or optimization methods such as Genetic Algorithm (GA), Particle Swarm Figure 20. A Month Ahead Forecast of Sungai Lanas Optimization (PSO), Ant Colony Optimization (ACO) and more.

Acknowledgements This study is supported by Ministry of Education and Universiti Tun Hussein Onn Malaysia (UTHM) via the Fundamental Research Grant Scheme (FRGS) Vot K082.

REFERENCES [1] Mosavi A, Ozturk O & Chau KW (2018), Flood prediction using machine learning models: Literature review, Water (Switzerland), Vol. 10, No. 11, pp. 1-40.

Figure 21. A Month Ahead Forecast of Sungai Kelantan

32 Empirical Mode Decomposition Couple with Artificial Neural Network for Water Level Prediction

[2] Yadav V & Eliza K (2017), A hybrid wavelet-support within machine learning-based classification system for vector machine model for prediction of lake water level early warnings related to geotechnical problems, fluctuations using hydro-meteorological data, Automation in Construction, Vol. 68, pp. 65-80. Measurement: Journal of the International Measurement Confederation, Vol. 103, pp. 2655-2675. [16] McCulloch WS & Pitts W (1943), A logical calculus of the ideas immanent in nervous activity, Buletting of [3] Alexander AA, Thampi SG & Chithra NR (2018), Mathematical Biophysics, Vol. 5, pp. 115-116. Development of hybrid wavelet-ANN model for hourly flood stage forecasting, ISH Journal of Hydraulic [17] Cocianu C-L & Grigoryan H (2015), An artificial neural Engineering, Vol. 24, No. 2, pp. 266-274 network for data forecasting purposes. Informatica Economica, Vol. 20, No. 2, pp. 34-45. [4] Oh S-K, Kim W-D & Pedrycz W (2016), Design of radial basis function neural network classifier realized with the [18] Rosenblatt F (1961), Principle of Neurodynamics. aid of data preprocessing techniques: Design and analysis, Perceptrons and the Theory of Brain Mechanism, Cornell International Journal of General Systems, Vol. 45, No. 4, Aeroneutical Lab Inc. Buffalo, New York. pp. 434-454. [19] Reynaldi A, Lukas S & Margaretha H (2012). [5] Gong Y, Zhang Y, Lan S & Wang H (2016), A comparative Backpropagation and levenberg-marquardt algorithm for study of Artificial Neural Networks, Support Vector training finite element neural network, Proceedings – Machines and Adaptive Neuro Fuzzy Inference System for UKSim – AMSS 6th European Modelling Symposium, EMS forecasting groundwater levels near Lake Okeechobee, 2012, Vol. 2, pp. 89-94. Florida, Water Resource Management, Vol. 30, No. 1, pp. 375-391. [20] Damaševičius R, Napoli C, Sidekerskienė T & Woźniak M (2017), IMF mode demixing in EMD for jitter analysis, [6] Mohanty S, Jha M, Raul S, Panda RK & Sudheer KP (2015), Journal of Computational Science, Vol. 22, pp. 240-252. Using Artificial Neural Network approach for simultaneous forecasting of weekly groundwater levels at multiple sites, [21] Malik H & Mishra S (2016), Artificial neural network and Water Resources Management, Vol. 29, No. 15, pp. empirical mode decomposition based imbalance fault 5521-5532. diagnosis of wind turbine using TurbSim, FAST and Simulink, IET Renewable Power Generation, Vol. 11, No. [7] Hitokoto M & Sakuraba M (2018), Application of the deep 6, pp. 889-902 learning flood forecast model against the inexperienced magnitude of flood, EPiC Series in Engineering, Vol. 3, pp. [22] Isham MF, Leong MS, Hee LM & Ahmad ZAB (2017), 893-901 Empirical mode decomposition: A review on mode selection method for rotating machinery diagnosis, [8] Khan M & Coulibaly P (2006), Application of Support Internation Journal of Mechanical Engineering and Vector Machine in lake water level prediction, Journal of Technnology (IJMET), Vol. 8, No. 6, pp. 16-26. Hydrological Engineering, Vol. 11, No. 3, pp. 199-205. [23] Madhiarasan M & Deepa SN (2015), A novel criterion to [9] Ang HTN, Dat NQ, Van NT, Doanh NN & An NL (2018), select hidden neuron numbers in improved back Wavelet-Artificial Neural Network model for water level propagation networks for wind speed forecasting, Applied forecasting, Proceedings of 2018 3rd IEEE International Intelligence, Vol. 44, No. 4, pp. 878-893 Conference on Research in Intelligent and Computing in Engineering, RICE 2018, pp 1-6. [24] Mostafa F, Dillon TS & Chang E (2018), Computational Intelligence Applications to Option Pricing, Volatility [10] Deo RC & Şahin M (2016), An extreme learning machine Forecasting and Value at Risk. Springer. model for the simulation of monthly mean streamflow water level in eastern Queensland, Environmental [25] Lippmann RP (1987), An introduction to computing with Monitoring and Assessment, Vol. 188, No. 2, pp. 1-24. neural network, IEEE Assp Magazine, Vol. 4, No. 4, pp. 4-22. [11] Adhikary S, Muttil N & Yilmaz A (2017), Improving streamflow forecast using optimal rain gauge [26] Kim S & Kim H (2016), A new metric of absolute network-based input to Artificial Neural Network models, percentage error for intermittent demand forecasts, Hydrology Research, pp. 1-20. International Journal of Forecasting, Vol. 32, pp. 669-679. [12] Alizadeh MJ & Kavianpour MR (2015), Development of wavelet-ANN models to predict water quality parameters in Hilo Bay, Pacific Ocean, Marine Pollution Bulletin, Vol. 98, No. 1, pp. 171- 178. [13] Sarkar A & Pandey P (2015), River water quality modelling using Artificial Neural Network technique, Aquatic Procedia, Vol. 4, pp. 1070-1077. [14] Zhang GP (2012), Neural network for time-series forecasting, eds Rozenberg G, Bäck T & Kok JN, in Handbook of Natural Computing, Berlin, Heidelberg: Springer, pp. 461-477 [15] Chou JS & Thedja JPP (2016), Metaheuristic optimization

Civil Engineering and Architecture 7(6A): 33-42, 2019 http://www.hrpub.org DOI: 10.13189/cea.2019.071404

Prediction of Future Climate Change for Rainfall in the Upper Kurau River Basin, Perak Using Statistical Downscaling Model (SDSM)

Nuramidah Hamidon1,*, Sobri Harun2, Norshuhaila Mohamed Sunar1, Nor Hazren A.Hamid1, Mimi Suliza Muhamad1, Hasnida Harun1, Roslinda Ali1, Mariah Awang1, Mohamad Ashraf Abdul Rahman1, Faridahanim Ahmad1, Kamaruzaman Musa1, Fatimah Mohamed Yusof1, Mohd Syafiq Syazwan Mustafa1

1Department of Civil Engineering Technology, Faculty of Engineering Technology, Universiti Tun Hussein Onn Malaysia, Malaysia 2Department of Civil Engineering, Faculty of Engineering, Universiti Teknologi Malaysia, Johor

Received July 26, 2019; Revised October 9, 2019; Accepted December 10, 2019

Abstract Climate change is considered to be one of the biggest threats faced by nature and humanity today. The goal of this study is to predict future climate change for rainfall in the Upper Kurau Basin. In this research, the 1. Introduction applicability of statistical downscaling model (SDSM) in Global warming will have a significant impact on local downscaling rainfall in the Upper Kurau River basin, Perak, and regional precipitation and hydrological regimes, which Malaysia was investigated. The investigation includes in turn will affect ecological, social and economic systems calibration of the SDSM model by using large-scale of human, such as health of ecosystems and fish resource atmospheric variables encompassing the National Centers management, industrial and agricultural water supply, for Environmental Prediction (NCEP) reanalysis data. resident living water supply, water energy exploitation, Rainfall data were derived for three 30-year time slices, human health, etc. These potential changes will affect some 2020s, 2050s and 2080s, with A2 and B2 scenarios. A2 is qualitative and quantitative estimation on the impact of considered among the “worst” case scenarios, projecting climate change upon regional water resources [1]. The high emissions for the future. Unlikely, B2 projected a direct impact of climate change can be variation and lower emission for the future and it is considered as changing pattern of water resources availability and “environmental” case scenarios. Results from simulation hydrological extreme events such as floods and droughts, showed that during the calibration and validation stage, the with many indirect effects on agriculture and water supply SDSM model was well acceptable in regards to its [2]. performance in downscaling of daily and annual rainfalls. The Global Climate Models (GCMs) are the optimal Under both scenarios A2 and B2, during the prediction tools to estimate future global climate changes resulting period of 2010–2099, changes of annual mean rainfall in from the continuous increase of greenhouse gas the Upper Kurau River basin would present a trend of concentration in the atmospheres [3]. There are currently increased rainfall in 2020s; insignificant changes in the two major popular downscaling approaches, namely 2050s; and a surplus of rainfall in the 2080s, as compared statistical downscaling (SD) and dynamic downscaling to the mean values of the base period. Annual mean rainfall (DD). “Statistical downscaling” adopts statistical would increase by about 33.7% under scenario A2 and relationships between the regional climates and carefully increase by 27.9% under scenario B2 in the 2080s. Most of selected large-scale parameters [4]. Dynamical the areas of the Upper Kurau River Basin were dominated downscaling methods, on the other hand, are extremely by increasing trend of rainfall and will become wetter in computationally intensive and have data requirements the future. which may not be easily available [5]. Compared to other Keywords Climate Change, Malaysia, Rainfall, downscaling methods, the statistical method is relatively Statistical Downscaling Model feasible to be used as it provides station-scale climate 34 Prediction of Future Climate Change for Rainfall in the Upper Kurau River Basin, Perak Using Statistical Downscaling Model (SDSM) information from GCM-scale output [6]. Statistical method main drainage for paddy fields and the source of drinking has comparable accuracy to that of dynamical downscaling water. The study area used to experience extreme flood and [7,8,9]. Many studies have shown that this model is simple drought that seriously affected the paddy cultivation (staple to handle and operate, and its large and superior capability food), ecosystem and human health. Therefore, good makes it have been widely applied [10,11]. These studies knowledge of future rainfall scenarios in the Upper Kurau indicated that there would be an increase in future rainfall basin will be of great importance in better evaluating the simulation using SDSM model applications. Huang et al. risk of floods and droughts. (2011) showed that the annual mean precipitation in most parts of the Yangtze River basin would be dominated by an 1.1. Study Area increasing trend under both scenarios A2 and B2. A2 is considered among the “worst” case scenarios, projecting The Upper Kurau River basin, Perak, as shown in Figure high emissions for the future. Unlikely, B2 projected a 1, lies between latitude 40 51’ (N) and 50 10’ (N), lower emission for the future and it is considered as longitude 100 38’ (E) and 101 01’ (E). The catchment area “environmental” case scenarios. Hassan et al. (2013) stated is approximately 359.2 km2, and is drained by the Kurau that the southern and central of Malaysia will face higher River and the Ara River. The rivers meet at Pondok raining compared to the northern Malaysia under both Tanjung town, Kurau. The river originates partly in the scenarios of A2 and B2. Bintang Range and partly in the Main Range where the To date, downscaling algorithm of SDSM has been terrain in the upper reaches is steep and mountainous. Mid applied to a host of meteorological, hydrological and valleys of the river are characterized by low to undulating environmental assessments, as well as a range of terrain, which gives way to broad and flat floodplains. geographical contexts, including the Europe, North Ground elevations at the river headwaters are moderately America and Southeast Asia [12,13]. The objective of this high. study is to predict future climate change for rainfall in the From 1961 to 1990, the average annual rainfall was 215 Upper Kurau Basin in Perak using Statistical Downscaling mm. For the same period, the minimum temperature was Model (SDSM) for the year 2010 until 2099. The main around 23oC and the maximum temperature was 34oC. The reason for the selection of the Upper Kurau River basin in relative humidity fluctuated between 54% and 98% and the this research is because the Bukit Merah reservoir is wind speed was in the range of 0 to 12 knots during normal located at the downstream of the basin which acts as the weather. Civil Engineering and Architecture 7(6A): 33-42, 2019 35

Figure 1. Upper Kurau River Sub-basin in Perak

2. Materials and methods data; (4) generation of climate change scenarios; (5) diagnostic testing and statistical analyses. SDSM is well documented and has been successfully tested in numerous 2.1. Statistical Downscaling Model (SDSM) studies [15]. Statistical Downscaling Model (SDSM) is a decision support tool which facilitates the rapid development of 2.2. Data multiple, low-cost, single-site scenarios of daily surface weather variables under current and future regional climate The location of rainfall stations in the study area is [14]. It also assesses the regional impacts of global shown in Figure 2. The observed daily rainfall data from 10 warming by allowing the process of spatial scale reduction stations were used to predict rainfall change in Upper of data provided by large-scale GCMs. Statistical Kurau River basin as listed in Table 1. The missing data of downscaling methods rely on empirical relationships one day or two days were replaced by the average between local-scale predictions and regional-scale precipitation values of the neighboring stations. If predictors to downscale GCM scenarios. The SDSM 4.2 consecutive days had the missing data, the missing values reduces the task of statistically downscaling daily weather were replaced with long term averages of the same days. series into five discrete processes (1) screening of predictor The data was carefully checked and calibrated for avoiding variables; (2) model calibration; (3) synthesis of observed unexpected errors (mainly by human errors).

36 Prediction of Future Climate Change for Rainfall in the Upper Kurau River Basin, Perak Using Statistical Downscaling Model (SDSM)

Figure 2. The location of rainfall gauging stations in Perak

Table 1. Rainfall stations in Upper Kurau River Basin, Perak

No Station Id Station Name Latitude (N) Longitude (E) 1 4807016 Bkt. Larut, Taiping 040 51’ 45” 1000 47 35 2 4907017 Ldg. Windsor, Ulu Sepetang 040 56’ 35” 1000 43’ 55” 3 4907019 Ldg. Norseman 040 57’ 55” 1000 45’ 50” 4 4908013 Ibu Bekalan Sempeneh, Batu Kurau 040 56’05” 1000 49’ 40” 5 4908018 Pusat Kesihatan Kecil,Batu Kurau 040 58’ 45” 1000 48’ 15” 6 5006021 Kolam Air Bkt. Merah 050 02’ 00” 1000 39” 10” 7 5007020 Ldg. Pondoland, Pondok Tanjung 050 00’ 35” 1000 43’ 50” 8 5007029 Ibu Bekalan Jelai 050 01’ 20” 1000 48’ 00” 9 5107006 Ldg. Stoughton, Batu Kurau 050 06’ 25” 1000 46’ 20” 10 5108005 Ibu Bekalan Ulu Ijok 050 07’ 20” 1000 48’ 20” (Sources: Department of Irrigation and Drainage, DID)

2.3. Selection of Predictors (Parameter) HadCM3 is used for predictors. HadCM3 was chosen because the model is widely used in many climate-change Predictor variables are available from the Canadian impact studies [16]. Furthermore, HadCM3 provides daily Institution of website for model output. The predictor predictor variables, which can be used for the SDSM model. variables are supplied on a grid box by grid box basis. On In addition, HadCM3 has the ability to simulate for a entering the location of study area, as Figure 3, the correct period of a thousand years, showing little drift in its surface grid box will be calculated and a zip file will be made climate. Its predictions for temperature change are average, available for downloading. When unzipping this file, there and for the precipitation, the predictions’ increases are are three directories, which are NCEP_1961-2001, below average [17]. The decision process to select suitable H3A2a_1961-2099, H3B2a_1961-2099.Table 2 listed 26 predictors is also complicated due to the fact that the predictor variables from the NCEP reanalysis and explanatory power of individual predictor variables varies HadCM3 simulation output that are used as potential inputs spatially and temporally [18]. to the multiple linear regression model. In this study,

Civil Engineering and Architecture 7(6A): 33-42, 2019 37

Figure 3. Asia continent window with 2.5°latitude x 3.75°longitude grid size

Table 2. Large-scale atmospheric variables from the NCEP reanalysis and HadCM3 simulation output that are used as potential inputs to the multiple linear regression model Predictor Predictor No Predictor Description No Predictor Description Variables Variables 1 mslpas mean sea level pressure 14 p5zhas 500 hpa divergence 2 p_fas surface air flow strength 15 p8_fas 850 hpa airflow strength 3 p_uas surface zonal velocity 16 p8_uas 850 hpa zonal velocity 4 p_vas surface meridional velocity 17 p8_vas 850 hpa meridional velocity 5 p_zas surface vorticity 18 p8_zas 850 hpa vorticity 6 p_thas surface wind direction 19 p850as 850 hpa geopotential height 7 p_zhas Surface divergence 20 p8thas 850 hpa wind direction 8 p5_fas 500 hpa airflow strength 21 p8zhas 850 hpa divergence 9 p5_uas 500 hpa zonal velocity 22 p500as Relative humidity at 500 hpa 10 p5_vas 500 hpa meridional velocity 23 p850as Relative humidity at 850 hpa 11 p5_zas 500 hpa vorticity 24 rhumas Near surface relative humidity 12 p500as 500 hpa geopotential height 25 shumas Surface specific humidity 13 p5thas 500 hpa wind direction 26 tempas Mean temperature at 2 m (Source:http://www.cccsn.ca/Help_and_Contact/Predictors_Help-e.html)

38 Prediction of Future Climate Change for Rainfall in the Upper Kurau River Basin, Perak Using Statistical Downscaling Model (SDSM)

2.4. Calibration and Validation correlation between the predictor variables and each predictor is not high in the case of daily precipitation.The The model was calibrated using output from NCEP steps to identify predictor variables that were used in this reanalysis data which predictor variable(s) (parameter for climate models) have been screening for 30 years data and study being recommended by several researchers [20, 21] were divided into two period times, which were 15 years are all predictors that are chosen and the explained for rainfall calibration (1961 to 1975) and another 15 years variance is run on a group of eight or ten of predictors at a for model rainfall validation (1976 to 1990). The choices of time and of each group, high explained variance of 1961-1990 and 1976-1999 as the calibration and validation predictor(s) is chosen. Then, partial correlation analysis is periods were made based on the availability of the rainfall done for selected predictors based on correction of each data. The selected parameters for all the stations are predictor. There could be a predictor with a high explained precipitation (prcp), surface specific humidity (Shum) and variance, but it might be very highly correlated with wind velocity at 500 hPa. The predictor prcp is the another predictor. This means that it is difficult to tell that dominant predictor in all the station so it may be said that this predictor will add information to the process, and prcp is the super predictor for this area [19]. therefore, it will be dropped from the list. Finally, the During the calibration process, some of the SDSM setup scatter-plot is used to show the relationship between parameters for bias correction and variance inflation were potential predictors. The predictor variables identified for adjusted to obtain a good statistical agreement between the downscaling rainfall used in this study were shown in observed and simulated climate variables. In general, the Table 3 and 4 below.

Table 3. Summary of GCM predictor for the downscaling rainfall analysis

Station Name Predictors Bkt. Larut, Taiping ncepp__fas.dat, ncepp5_uas.dat, ncepp8_fas.dat and ncepshumas.dat Ldg. Windsor, Ulu Sepetang ncepp850as, ncepp8_uas and ncepshumas Ldg. Norseman ncepp5thas.dat, ncepshumas.dat Ibu Bekalan Sempeneh, Batu Kurau ncepp_thas.dat, ncepp850as.dat and ncepshumas.dat Pusat Kesihatan Kecil,Batu Kurau ncepp_fas, ncepp5_fas and ncepshumas Kolam Air Bkt. Merah ncepp__uas.dat, ncepshumas.dat Ldg. Pondoland, Pondok Tanjung ncepp__fas.dat, ncepshumas.dat Ibu Bekalan Jelai ncepp__fas.dat, ncepp_850as and nceppshumas Ldg. Stoughton, Batu Kurau ncepp5_uas.dat, ncepshumas.dat Ibu Bekalan Ulu Ijok ncepp__uas.dat, ncepshumas.dat Ldg. Norseman ncepp5thas.dat, ncepshumas.dat

Table 4. Types of predictors

Variables Descriptions temp Mean temperature at 2m mslp Mean sea level pressure p500 500 hpa geopotential height p850 850 hpa geopotentail height rhum Near surface relative humidity r500 Relative humidity at 500 hpa height r850 Relative humidity at 850 hpa height shum Near surface specific humidity s500 Specific humidity at 500 hpa height s850 Specific humidity at 850 hps height **_f Geostrophic air flow velocity **_z Vorticity **_u Zonal velocity component **_v Meridional velocity component **zh Divergence **thas Wind direction ** refers to different atmospheric levels: the surface (p_), 850 hPaheight (p8), and 500 hPa height (p5).

Civil Engineering and Architecture 7(6A): 33-42, 2019 39

Meanwhile, for the validation process, Weather about HadCM3A2 and B2 are listed in Table 5. Generator is used to produce synthetic current daily weather data based on inputs of the observed time series Table 5. Group of Special Report on Emissions Scenarios (SRES) SRES data and the multiple linear regression parameters Description produced. During this process, the input file is obtained Scenario Lower trade flows, relatively slow capital stock from the calibration process and the predictor direction is turnover, and slower technological change. three sets of atmospheric data, NCEP and HadCM3 A2 Self-reliance in terms of resources and less emphasis Scenario A2 and B2. The output from Weather Generator is on economic, social, and cultural interactions between the synthesized artificial weather time series data which regions are characteristic for this future. represent actual weather. 100 simulations of synthetic daily Increased concern for environmental and social sustainability presents a particularly favorable climate weather are performed. 100 simulations mean 100 numbers B2 for community initiative and social innovation, of assembly sizes of SDSM interface. The result of especially in view of the high educational levels. validation may be different from calibration model and each ensemble of validation due to the relative significance of the relative significance of the deterministic and 3. Results and discussion stochastic components of the regression models. Performance of SDSM model is measured using 3.1. Calibration and Validation of SDSM coefficient of determination (R2) and Root Mean Square Error (RMSE). The Root Mean Square Error (RMSE) is The downscale daily rainfall simulated by SDSM (using frequently used to measure the difference between values the NCEP variables) at Stations 5, 6, 7 and 8 as tabulated in 2 predicted by a model and the values actually observed from Table 5, gives a higher value for R compared to other the environment that is being modelled. These coefficients stations during calibration with 0.24, 0.30, 0.20 and 0.23 are calculated according to the following equations: respectively. For validation part, station 5 gives higher among all stations with value R2 being 0.20. It can be seen 푛 2 푖=1 푄(obs)i – 푄obsave) (Q(sim)i – Qsimave that the SDSM model is unable to predict well for daily 푅 2 = 푛 2 푛 2 0.5 (1) 2 푖=1 푄(obs)i- Qobsave 푖=1 Q(sim)i – Qsimave rainfall when R <0.3 during calibration and validation. Meanwhile Root Mean Square Error (RMSE) at Station 4 푛 2 (Q(obs)i − Q(sim)i ) gives the highest RMSE value during calibration with 푖=1 (2) RMSE = 6.61mm/day and validation with 5.54mm/day. The result 푛 shows that the daily rainfall series simulated from NCEP Definition 1: In above equations 1 and 2, Qobs is the with the mean R2 values is less than 0.3, which is observed value at time, Qsim is the simulated value at time, comparable with literature values [22]. This is because the n is the sum number of observations, and Qobsave and Qsimave amount of rainfall is stochastic processes, the downscaling are the average of observed and predicted values, of daily rainfall is always a difficult subject, and the respectively. stimulation results of daily rainfall in the most of similar researches were worse than those of monthly [23]. From 2.5. Downscaling Precipitation under Future Emission the results obtained, it shows a higher value of RMSE and Scenarios the small value of R2, which indicates poor performance in downscaled rainfall time series. The long term future climate is divided into 30–year In general, the study showed that the SDSM model was period, 2010 to 2039 (2020s), 2040 to 2069 (2050s), and poor in predicting on a daily rainfall between observed and 2070 to 2099 (2080s). For this study, the model output of simulated rainfall. The results were similar with other HadCM3 GCM was used for the A2 (medium-high) and B2 studies such as [24]. Hence, results from this study are (medium-low) emission scenarios. 100 ensembles of considered fully justified according to early research works. synthetic daily time series are produced for HadCM3 A2 Furthermore, daily rainfall is the most difficult variables and B2, 139 years (1961 to 2099). The HadCM3A2 and B2 for prediction and it is a condition process which involves are the emission scenario from GCM output files. Details an inter-connected with many factors/variables.

40 Prediction of Future Climate Change for Rainfall in the Upper Kurau River Basin, Perak Using Statistical Downscaling Model (SDSM)

Table 5. The R2 and RMSE between observed and simulated rainfall results for each station for the SDSM model

Calibration Validation Calibration Validation Station Name R2 RMSE Bkt. Larut, Taiping (1) 0.01 0.11 4.27 4.56 Ldg. Windsor, Ulu Sepetang (2) 0.01 0.11 3.47 4.12 Ldg. Norseman (3) 0.08 0.07 2.94 3.12 Ibu Bekalan Sempeneh, Batu Kurau (4) 0.03 0.01 6.61 5.54 Pusat Kesihatan Kecil,Batu Kurau (5) 0.24 0.20 2.00 1.71 Kolam Air Bkt. Merah (6) 0.30 0.04 2.24 3.13 Ldg. Pondoland, Pondok Tanjung (7) 0.20 0.01 2.63 3.11 Ibu Bekalan Jelai (8) 0.23 0.15 3.27 3.87 Ldg. Stoughton, Batu Kurau (9) 0.01 0.07 2.49 2.15 Ibu Bekalan Ulu Ijok (10) 0.03 0.10 3.60 3.49

3.2. Downscaling for Future Emission stations is according to the result of highest correlation R2 among all the 10 stations. Taking the simulation results of The period of 1961–1990 was taken as the base period as SDSM in the modeling rainfall of current period (1961– it was used in most impact studies worldwide, and the 1990) into account, the change of annual mean rainfall of future period was divided into 2020s (2010–2039), 2050s Kurau River basin under scenarios A2 and B2 was (2040–2069), 2080s (2070–2099). Future annual mean discussed in this paper. rainfall for 10 stations rainfall was depicted in Figure 4 to Figure 5 shows changes of annual mean rainfall for 10 show the pattern between current and future periods under stations in Upper Kurau River Basin. Stoughton (station 5) scenario A2 and B2. has shown the increase in future 2080, followed by station Norseman (station 3) and Sepetang (station 2). It is seen that under scenario A2, the changes of annual mean rainfall of future periods (2020s, 2050s and 2080s) for the whole Upper Kurau River Basin would be 13.87, 25.09 and 39.52% respectively. Under scenario B2, the changes of annual mean rainfall of future periods (2020s, 2050s and 2080s) in the Upper Kurau river basin would be 14.11, 22.41 and 32.38% respectively.

Figure 5. The changes of percentage annual mean rainfall (compared to base period) in 2020s, 2050s, and 2080s under scenarios A2 and B2. Figure 4. Future annual mean rainfall for 10 stations under scenarios A2 The simulation results of future rainfall when compared and B2 to the base period, the annual mean rainfall of the Upper Patterns of change about future rainfall scenarios Kurau river basin of three future periods would show an compared to the base period were then analysed at stations increase of rainfall in the future. As for A2 scenario, in the 5, 6, 7 and 8, which represented the Upper Kurau River 2020s, the change would present a situation of increase of Basin using H3A2 and H3B2 data. The selection of these 14%; as for the 2050s, the change increase of 25%; when it

Civil Engineering and Architecture 7(6A): 33-42, 2019 41

comes to 2080s, the change would present a situation of Report, Universiti Teknologi Malaysia, Skudai. 1-91pp. increase being larger than 35%. Scenario B2 would also [6] Wilby RL, Dawson CW, Barrow EM (2002) SDSM – A show an increase in future rainfall but lower than A2. Decision Support Tool for The Assessment Of Regional Increment of annual rainfall showed that Upper Kurau Climate Change Impacts. Env Mod Soft 17: 147–159 River basin will receive an increase of rainfall in the future. [7] Kidson JW, Thompson CS (1998) A comparison of Results obtained from this study are similar with Huang et statistical and model-based downscaling techniques for al. (2011) and Hassan and Sobri (2012). estimating local climate variations. J Clim 11: 735–753

[8] Solman SA, Nu~nez MN (1999) Local Estimates Of Global Climate Change: A Statistical Downscaling Approach. Int J 4. Conclusions Climatology 19: 835–861

SDSM was applied in this study to predict future rainfall [9] Schoof JT, Pryor SC (2001) Downscaling temperature and in Upper Kurau River Basin. The results obtained from precipitation: a comparison of regression-based methods calibration and validation of SDSM model show that the and artificial neural networks. Int J Climatology 21: 773– model was average in predicting the daily rainfall but 790 successful in simulating annual rainfall. The discrepancies [10] Wilby RL, Wigley TML, Conway D, Jones PD, Hewitson between observed and generated data could be driven from BC, Main J, Wilks DS (1998) Statistical downscaling of uncertainty of input data, errors in the forcing of GCM general circulation model output: a comparison of methods. scenarios and incomplete representation of key processes Water Resource Research 34(11):2995–3008 for the downscaling model. Results for future rainfall [11] Wilby RL, Harris I (2006). A framework for Assessing indicate an increasing trend in all future time horizons for Uncertainties in Climate Change Impacts: Low-Flow both A2 and B2 emission scenarios. It proves that the Scenarios. Water Resource Research 42:W02419. SDSM model is able to predict the future rainfall at the doi:10.1029/2005WR004065 Upper Kurau River Basin. These results would provide [12] Hassan, H., Aramaki, T., Hanaki, K., Matsuo, T., Wilby, important scientific base and practical information for R.L., (1998). Lake stratiﬁcation and temperature proﬁles water resources planning and management in the basin. simulated using downscaled GCM output. Journal of Water Science and Technology 38, 217–226

[13] Hay, L.E., Wilby, R.L., Leavesley, G.H., (2000). A Acknowledgements comparison of delta change and downscaled GCM scenarios for three mountainous basins in the United States. This study was funded by Fundamental Research Grant Journal of the American Water Resources Association 36, (FRGS), under vote 4F325, UTM, Mybrain15 MOHE, 387–397 Malaysia and Grant Tier 1 (H201) UTHM. [14] Wilby, R.L., Hay, L.E., Leavesley, G.H., (1999). A Comparison of Downscaled and Raw GCM output: Implications for Climate Change Scenarios in The San Juan River Basin, Colorado. Journal of Hydrology 225, 67–91. REFERENCES [15] Hassan, Z., Shamsudin, S. and Harun, S., (2013) Application of SDSM and LARS-WG for Simulating and Downscaling [1] Huang J., Zhang J., Zhang Z., Xu C. Y., Wang B., Yao J., of Rainfall and Temperature. Theor Appl Climatol Volume (2011). Estimation of future precipitation change in the 116, Issue 1-2 , pp 243-257. doi: DOI 10.1007/s00704-013 Yangtze River basin by using statistical downscaling -0951-8 method, Stoch Environ Res Risk Assess (2011) 25:781–792 DOI 10.1007/s00477-010-0441-9 [16] Simonovic. S.P (2010) A New Methodology for the Assessment of Climate Change Impacts on a Watershed [2] Fenta Mekonnen, D., & Disse, M. (2018). Analyzing the Scale. Current Science, Vol. 98, No. 8, 25 April 2010 future climate change of Upper Blue Nile River basin using statistical downscaling techniques. Hydrology and Earth [17] McCarthy, J., Canziani, O., Leary, N., Dokken, D., and System Sciences, 22(4), 2391-2408. White, K. (2001) Climate Change 2001: Impacts, Adaptation, and Vulnerability. Cambridge University Press, [3] Busuioc, A., Chen, D., Hellstrom, C. (2001) Performance of New York. 105-110pp Statistical downscaling Models in GCM validation and regional climate change estimates: application for Swedish [18] Hessami, M., Gachon, P., ourda, T. B. M. J., and St-Hilaire, precipitation. International Journal of Climatology. A. (2007) Automated regression based statistical downscaling tool. Environmental Modeling and Software , [4] Wilby, R.L. & Dawson, C.W. (2004) Using SDSM Version Science Direct. 3.1 – A Decision Support Tool for the Assessment of Regional Climate Change Impacts. User Manual. 67pp [19] Tahir, T., Hashim, A. M., & Yusof, K. W. (2018, April). Statistical downscaling of rainfall under transitional climate [5] Harun, S., Hanapi, M. N., Shamsuddin, S., Mohd Amin, M. in Limbang River Basin by using SDSM. In IOP Z., and Ismail, N. A. (2008) Regional Climate Scenarios Conference Series: Earth and Environmental Science (Vol. Using a Statistical Downscaling Approach. Technical 140, No. 1, p. 012037). IOP Publishing.

42 Prediction of Future Climate Change for Rainfall in the Upper Kurau River Basin, Perak Using Statistical Downscaling Model (SDSM)

[20] Hassan, Z. and Harun, S., (2012) Application of Statistical Downscaling Model for Long Lead Rainfall prediction in Kurau River Catchment in Malaysia, Malaysian Journal of Civil Engineering 24(1):1-12 (2012) [21] Wilby, R. L., and Dawson, C. W. (2007) SDSM 4.2 — A Decision Support Tool For The Assessment Of Regional Climate Change Impacts, User Manual [22] Dibike, Y. B. and Coulibaly, P. (2005) Hydrologic impact of climate change in the Saguenay watershed: Comparison of downscaling methods and hydrologic models. Journal of Hydrology, 307(1-4): 145-163. [23] Khan MS, Coulibaly P, Dibike Y (2006) Uncertainty analysis of statistical downscaling methods. J Hydrology 319:357–382 [24] Karamouz, M., Fallahi, M., Nazif, S., and Rahimi Farahani, M. (2009) Long lead rainfall prediction using Statistical Downscaling and Artificial Neural Network modeling. Scientia Iranica, 16(2): 165-172.

Civil Engineering and Architecture 7(6A): 43-49, 2019 http://www.hrpub.org DOI: 10.13189/cea.2019.071405

User Perception on Urban Light Rail Transit

Seuk Yen Phoong1,*, Seuk Wai Phoong2, Sedigheh Moghavvemi2, Kok Hau Phoong3

1Department of Mathematics, Universiti Pendidikan Sultan Idirs, Malaysia 2Department of Operations and Management Information Systems, University of Malaya, Malaysia 3Faculty of Management and Information Technology, Sultan Azlan Shah University, Malaysia

Received July 28, 2019; Revised October 8, 2019; Accepted December 15, 2019

Abstract Public transport is a shared passenger traffic congestion in the city. transport service available for public use. Increased The problems pertaining to the public transportation population is accompanied by the increased demand for system in Kuala Lumpur are the lack of integration and private vehicles. The exponential growth in the number of focus. The government does not actively promote the use private vehicles will result in negative impacts such as air of public transportation, which indirectly influences daily pollution, excessive noise, and traffic congestion. life. According to the statistics in 2017, the average daily Additionally, customers’ perceptions on different aspects ridership of the public transport in 2017 was 638,608 in include safety, operation or time, comfortableness and Kuala Lumpur. The average time a person spent every day cleanness of public transportation that are also essential in in traffic congestion is ~53 minutes in Kuala Lumpur. affecting their mode of choice when travelling. The public ~83% of the respondents in Kuala Lumpur wish to own a transport that is discussed in this paper is light rail transit. car in the next half decade, which will serve to only This paper intends to investigate the main purpose of using exacerbate the current congestion problem. light rail transit and elucidate the public perspective of the To pursue green growth in Malaysia, the government light rail transit via factor analysis and correlation analysis. invested public transport in its effort to lower carbon A questionnaire with five-point Likert scale was designed, mobility. As per the Eleventh Malaysia Plan [1], a single and data were randomly selected from 200 light rail transit occupant car ranks first of the ground public transport users in Kuala Lumpur, Malaysia. The results revealed that which emits carbon dioxide, followed by buses and trains. the majority of the customers use light rail transit to school The use of public transport is expected to reduce or university. Moreover, most of the users satisfy with the congestion and minimize environmental pollution. Also, safety, operation, cost, comfortableness and cleanness of clean and comfortable facilities, increased train the light rail transit. This can be concluded that light rail frequencies and punctuality would also play a role in transit provides mobility and choice for everyone in the encouraging the public to use public transportation in their context of efficiency, health and safety, affordability, daily lives. The public transport system is required to be accessibility, and environmental friendliness. more market-oriented and competitive, which can only be realized when the public’s needs, expectations, and Keywords Light Rail Transit, Public Transport, behavior are identified and addressed appropriately. Safety, Comfortable, Affordability Public transportation can be defined as a shared passenger transport service which is available for use by the general public, as distinct from modes including buses, taxi, Monorail and transit trains. In Malaysia, the modes of 1. Introduction public transportation are buses, taxis, KL Monorail, commuter, Rapid KL, light rail transit (LRT) and mass Kuala Lumpur is the economic capital, as well as the rapid transit (MRT). The LRT and MRT are known as cultural and financial centre of Malaysia. Due to its metro rail transit which is an emerging transit system multirole nature, its municipality experiences many introduced by Malaysian government. These system are problems, such as traffic congestion, emission of carbon available to the public which may require fares, and run at dioxide, and excessive noise pollution. The ever increasing scheduled times. The purpose of introducing or expanding amount of vehicles annually can be problematic in the public transport is to address traffic jam problem in a future in the context of transportation governance, and the densely populated area like Kuala Lumpur. The benefit of introduction of public transport is expected to alleviate this emerging transit system is to minimize the waiting 44 User Perception on Urban Light Rail Transit

time, to envisage improving the poor and inadequate publicly or privately owned and make available the public transportation coverage at Kuala Lumpur. services to the general public. This study focuses on one of the most frequently used Public transport is regarded as a need in many parts of public transportation in Kuala Lumpur, which is light rail the world. A transportation system is regarded to be dismal transit (LRT). This is because LRT is the most frequently if its service is poor [4]. Public transportation plays a used transportation as compared to others. It can easily substantial role in our lives due to the fact that it improves convey users from suburbs far off cities to the heart of our quality of life by expediting traffic, which reduces costs Kuala Lumpur. For instance, the Kelana Jaya LRT station while also creating jobs. It provides accessibility to those is one of the most frequently used stops for residents who who cannot drive, or those who cannot afford to do so. work in the centre of Kuala Lumpur. While for the Ampang Public transport service includes multiple tenancy vehicles LRT line, it’s also frequently used by the office workers services designed to transport passengers on local and who need to convey from other suburban areas to the heart regional routes and their corresponding sub-systems. of Kuala Lumpur. Malaysia has many public transportation systems that are Light rail transit, also known as LRT, is an urban rail in operation, such as buses, taxis, trishaws, and trains. transport service that has two major routes, including However, most people are now highly dependent on car Kelana Jaya and Ampang LRT line in Malaysia. This rail travel [5]. Metropolitan centers are overwhelmed with transit services serve as a large part of the national capital unmanageable trends in the transportation sector due to and largest city, Kuala Lumpur in Malaysia. The current escalation in energy use, pollution, and traffic congestion. daily ridership of LRT is over 464,000 users per day which These problems are even more prominent in the case of has the highest passenger loads as compared to other public developing cities because the vehicle growth rate is far transports. superior to the growth rate of transport infrastructure. The However, there are a few cases about the delay of LRT. progressively active car use in the cities contributed to In 2018, the signaling problem caused over two hours of increased accessibility problems, due to traffic congestion delay at peak hour which resulted in many passengers and parking problems. Traffic congestion is a significant being stranded at several stations. The incident led some of problem faced by many urban areas in Malaysia, which can the passengers to vent out their disappointment over the be mitigated by the usage of public transport. The use of rail transit service through social media [2]. Other issues public transport can reduce traffic accidents, congestion about the weaknesses of LRT were poor customer service and parking problems. Although car use is the most popular quality, overcharging, cleanness and etc. Thus, this study visitor transport mode, congestion, pollution, traffic intends to elucidate public perspective on the environment, problems, and demands for maintainable transport service, and system adopted by the urban LRT. It is practices led to efforts expanded towards improving public expected that Rapid Rail Sdn. Bhd. and Prasarana Malaysia transportation. Public transportation has played an Berhad take into account these factors when evaluating important role in transporting passengers to work and residents’ usage of LRT. The objectives of this study elsewhere, which has the positive effect of reducing traffic include: congestion [6]. i. Determining the main usage of LRT users in Kuala Besides road congestions, private cars also cause serious Lumpur. problems such as CO2 pollution, global warming, and noise. ii. Elucidating public perception on the urban LRT. Jain and Khare [7] said that the increasing alteration in iii. Determining the highest correlation between the motorized mode of transport in urban cities of the items on the usage of LRT using factor analysis. developing countries increased the amount of air pollutants. The transport sector is responsible for the emission of more than a quarter of carbon dioxide (CO2) globally, as well as 2. Literature Review sizeable shares of methane (CH4) and nitrous oxide (N2O) emissions, making it one of the largest single contributors Public transport is known to be environmentally friendly to global greenhouse gas (GHG) emissions. According to when relative to private modes of transportation. Public Parry et al. [8], unrestrained emissions of GHG to the transportation includes buses, coaches, rapid transit, trains, atmosphere have warmed the planet to levels never seen ferries, and airlines, most of which are outright owned or before in history. The production of GHG and the global financed by the state or private individuals. According to warming not only affect the environment and economy, Bachok et al. [3], public transportation helps maintain they also noticeably affect human health. The climate accessibility while also expanding economic opportunities, changes caused by global warming have unpredictable reduce fuel consumption, and mitigate environmental impacts on the environment, society, and economy. concerns. He also pointed out that the public transportation According to the IEA [9], the global CO2 emissions system consists of all multiple occupancy vehicle services reached 32.4 Gt-CO2 in 2014, which is an increase of 0.8% which were designed to transport passengers, such as vans, over that of levels reported in 2013. The transportation buses, taxis, or rails or other transportation, which are sector is one of the major contributors to these numbers, at

Civil Engineering and Architecture 7(6A): 43-49, 2019 45

a share of 23% [9], since most vehicles on the road remain especially travelling with annoying passenger’s attitude dependent on hydrocarbon fuels. Also, the number of such as talking loudly over the phone. This situation could automobiles on the road has been increasing in tandem get much worst during peak hours with students getting to with the global population. Public transport has been touted or from classes. Ismail et al. [21] reported that comfortable as a solution to this problem. travel experience appeared to have a strong relationship According to Santos et al. [10], the rate of vehicle with the overall satisfaction. This be can supported by growth is far greater than the rate of growth of transport Cantwell et al. [22] states that the personal space invasion infrastructure. This means that traffic congestion will be and crowding is one of the main reasons for user rife in urban centers. Similarly, Khalid et al. [11] also dissatisfaction. reported that traffic congestion is one of the significant In addition, satisfaction is the gap between customers’ problems faced by many urban areas in Malaysia. The perception and expectations towards products or services current growth trends in Malaysia are unsustainable, as and their own personal experience on those products or they are directed towards increased use of private vehicles. services [23]. It can bemeasured only when consumers The level of investment in road construction increased have self-experience with the product. Satisfaction is significantly over the past few years, but the level of essential for an organization either in public or private investment in public transport does not comport with that sectors as it has been related with customer loyalty and will of road construction. affect organizations’ sustainability [24]. Without the The Malaysian government is eager to improve public continuous support from customers, a company could not transport in urban areas of the country. This is in line with survive in industry for long term. In public transport its objective of stimulating economic growth and relieving industry, it is necessary to evaluate user public transport traffic congestion. There are many forms of public satisfaction to ensure service provider has a clear picture on transportation in Malaysia, an example of which being the customers’ travel needs and presumption. Sumaedi et al. light rail transit (LRT). The LRT is one of the most [25] mentioned that citizens tend to own a private vehicle frequently used public transport services in Kuala Lumpur, when they are dissatisfied with the public transport services being utilized by those going to work, schools, and going and it will cause traffic congestion. If the standard of about their daily lives. It is constantly being developed and service provided has met with customer perception, they improved over the years. However, it should be pointed out will be satisfied and indirectly the public transport that increased supply does not mean increased demand or company will get a good reputation. When customers are satisfaction in the case of the LRT [12]. In order to ensure satisfied with the services provided, they will have the that the LRT meets its objectives, factors such as frequency, initiative to use public transport as their first choice of travel time, comfort and cleanliness that were prioritized. travelling transport. Network coverage and safety issues [13] were also According to Mouwen [26], public transport users mentioned as crucial factors by customers when evaluating emphasized on the service attributes such as on-time the quality of public transport services. performance, travel speed, and service frequency as the According to Roberts [14], factors affecting the most important aspect, followed by the driver behavior and transportation mode for the people include cost, service, vehicle tidiness. Then, a generic policy is established by product characteristics, capacity, security, and Mouwen [26] to achieve the service attributes such as environment. Taking into account the multiple factors that punctuality, frequency, driver behavior and cleanliness that play into the decision and carefully analyzing them will can increase the level of customers' satisfaction. Clarifying allow us to improve public transport. Botzoris et al. [15] the factors affecting the level of satisfaction is the most pointed out that the facilities and equipment of public important variables to the authorities to design policies for transports are the main factors influencing passengers' encouraging both the actual and potential public perception of service quality. In addition, Leem and Yoon transportation users. [16] also highlighted the fact that customer satisfaction is an assessment of the services and products. Meanwhile, Aworemi et al. [17] suggested that socioeconomic factor is 3. Research Methodology also important, while Pucher et al. [18] and Zheng and Wu [19] concluded that improving service quality would This project is going to be carried out using quantitative ultimately be helpful for reputation and profit margins of approach. Reviews of secondary sources materials, public transportation. including article journal, professional magazines and Comfort and cleanliness are one of the factors affecting reports will be done to identify the current situation, issues customer satisfaction. This can be supported by referring to and challenges of the usage of public transport in Kuala Irtema et al. [20] states that the level of comfort before and Lumpur. The review of literature will enhance the nature of during the journey will affect an user's travelling mood. In the study and provide focus of the study. addition, overcrowding can be a major setback for public Next, questionnaire is used in this project to evaluate the transport users. Overcrowding itself is uncomfortable, main usage of the LRT users in Kuala Lumpur; and the

46 User Perception on Urban Light Rail Transit

service quality and customer satisfaction on the usage of LRT. This questionnaire is divided into two sections which are (2) demographic profile and constructs items. Section A is asking about respondents’ personal details and section B is the item about the satisfaction on safety, operation or time, where i  EXi  denotes the population with means of comfortableness and cleanness of LRT. A five-point Likert variable i. scale, ranging from 1 (strongly disagree) to 5 (strongly Let p be the unobservable common factors f , f ,...,f . agree) was used in the questionnaire for rating the items in 1 2 p section B. Based on the data collection, the service quality Then, the common factors can be collected into a vector, µ. and customer’s satisfaction that emerge from the questionnaire will be identified and codified according to the themes. Thereafter, data analysis using factor analysis (3) is used to measure and analyze the gathering data. Moreover, reliability test is also included in this study to measure the reliability and significance of the study (see The regression coefficients l for multiple regressions Table 1). ij are called factor loadings. Factor loadings are coefficients Table 1. Reliability Coefficients of Instrument used to explain the correlations between observed variables Cronbach's Alpha Value of Reliability in a factor pattern matrix. The rule of thumb for factor loading with 0.7 or above represents that the factor extracts < 0.67 Poor sufficient variance from that variable [28]. A factor loading 0.67-0.80 Fair matrix is a matrix of weight or coefficients for the variable. 0.81-0.90 Good Thus, the factor loading will be collected into a matrix 0.91-0.94 Very Good shown below: > 0.94 Excellent Next, factor analysis is an exploratory tool that can be used to describe the variability obtained and measured (4) from the independent latent variables. This process reduces the information in a model by reducing the dimensions of the observations. Factor analysis can be used in many areas, Thereafter, the errors, ε are called specific factors with and is of particular significance in education, sociology, variable i. Finally, the matrix notation for factor analysis is: and psychology. The implementation of factor analysis in such areas is mainly for identifying how manifest X = µ + Lf + ε (5) behaviour can be interpreted in the context of underlying From the obtained factor loading, we can determine if patterns and structures. there are any correlation(s) between items using correlation This study used factor analysis to analyze the data analysis. Correlation analysis is used to explain the impact obtained from the survey. This method is a useful tool for of changes of independent variables on dependent examining the relationship between complex concepts such variables [29]. as socioeconomic status and psychological scales. This can Y = β +β x + ... + β x + ε (6) be supported by referring to Sharfuddin [27] that factor i 0 1 1 k k analysis is used to show produced clusters of uncorrelated where Y = Dependent variable factors for the public transit ridership. β = Coefficient of Independent Variable Supposing we have a set of n random variables, x = Independent Variable x1, x2,...,xn , with means of 1,2,...,n . Let vector X ε = Random Error denote the vector of traits collecting all random variables, This study attempts to establish a performance x . i evaluation mechanism for consulting residents’ perception on the usage of light rail transit in Kuala Lumpur, Malaysia. The sampling technique of the study is random sampling. (1) There are 200 LRT users being selected as respondents and the questionnaire was distributed to LRT users in both Kelana Jaya and Ampang LRT line in Malaysia. The data Then, assuming that the vector of traits, X, is sampled were analyzed using descriptive statistics, reliability test, from a population with a population mean vector of: validity and factor analysis in SPSS.

Civil Engineering and Architecture 7(6A): 43-49, 2019 47

4. Results and Discussion cost are discussed in the Table 3. From the result of analysis, the internal consistency of Table 3. Perception on the usage of LRT % % % % the items in customers' perception on LRT was examined Item % A using the reliability test. The Cronbach’s Alpha is 0.915, SD D NDA SA The ticket price is which exceeds the acceptable limit of 0.7. This can be 0 7 22.5 49.5 21 reasonable. referred to Tavakol and Dennick [30] that the acceptable The schedule and route range of value for the Cronbach’s Alpha is 0.70 – 0.95 but a information provided at the 3 7 20 51.5 18 lower alpha could still be reliable if the questionnaire was station is sufficient and easy comprised of lesser questions overall. to get. I feel safe while waiting at Then, a descriptive statistics table is used to describe the 1.5 5 17.5 54.5 21.5 the station. information from the respondents. Table 2 shows that there Easiness and convenience on 1.5 7 19 50.5 22 are 90 male and 110 female respondents. In addition, there booking or buying ticket. are 97 respondents of LRT users claiming to own a vehicle, The environment at the 6 9 24.5 42 18.5 while 103 do not. Also, 31% of the respondents use LRT station is clean. I am satisfied with the route for commuting to school/college/university, 18.5% to work, 10 15 23.5 34 17.5 waiting time. 26% shopping, 14% for recreation, and 10% for visiting I am satisfied with the route 5.5 11 20 45.5 18 friends/relatives. travel time. Table 2. Demographic Analysis Note that SD = Strongly Disagree; D = Disagree; NDA = Neither Disagree nor Agree; A = Agree; SA = Strongly Agree Frequency Percentage Variable (n = 200) (%) Table 3 shows that 70.5% of the respondents agree that Gender the ticket price of the LRT is reasonable; 69.5% agree that Male 90 60 the information pertaining to schedules and routes is 40 sufficient; 76% feel safe while waiting at stations; 72.5% Female 110 feel easy and convenient to book or buy ticket; 60.5% agree Have Own Vehicle that the station is clean; 51.5% are satisfied with the route Yes 97 48.5 waiting time; 63.5% are satisfied with the route travel time. It can be surmised that the majority of LRT users are No 103 51.5 satisfied with its current environment and system. Next, the exploratory factor analysis with principal Usage of LRT component analysis and direct oblimin rotation was used to Commuting to 62 31 examine the data set. The results of Bartlett’s test of school/college/university sphericity statistic is 0.000 (p < 0.05), which supports the Work 38 18.5 factorability of the correlation matrix. Moreover, the Shopping 52 26 Kaiser-Meyer-Olkin measure of sampling adequacy (KMO) Recreation 28 14 test value is 0.891, which exceeds the recommended minimum value of 0.6 for a good factor analysis [31]. The Visiting friends/relatives 20 10 results of the exploratory factor analysis using the principal The results of users' perception on LRT based on the component analysis and correlation between the items are comfortableness , cleanness, safety, operation or time and shown in Table 4.

Table 4. Principal component analysis and correlation between items

Correlations Item Factor loading 1 2 3 4 5 6 The ticket price is reasonable. 0.706 The schedule and route information provided at the station is sufficient and easy 0.772 .531** to get. I feel safe while waiting at the station. 0.779 .468** .545** Easiness and convenience on booking or 0.808 .573** .667** .626** buying ticket. The environment at the station is clean. 0.887 .510** .596** .602** .634** I am satisfied with the route waiting time. 0.881 .523** .553** .566** .550** .830** I am satisfied with the route travel time. 0.863 .507** .549** .654** .586** .829** .843** Note: **Correlation is significant at the 0.01 level (2-tailed).

48 User Perception on Urban Light Rail Transit

Table 4 shows that all the items with factor loading helped fund the research. which are greater than 0.40. Thus, there are no items that will be deleted. Then, the correlation matrix that is obtained from the Table 4 shows statistically significant and moderate to high correlations among the observed indicators used in the analysis, especially the route waiting REFERENCES time and travel time that has the strongest positive [1] Eleventh Malaysia Plan. Eleventh Malaysia Plan 2016-2020: correlation which is 0.843 compared to others. Whiler the Anchoring Growth on People, Malaysia, 2015. relationship between routes waiting time and cleanness of station is also strong positively correlated which is 0.830. [2] The Sun Daily (September 25, 2018), Signalling Fault Caused Peak Hour Delay of Kelana Jaya LRT line, Online Meanwhile, the relationship between other items such as available from https://www.thesundaily.my/archive/signall cost, route information, safety, and ticket selling are ing-fault-caused-peak-hour-delay-kelana-jaya-lrt-line-FUA weakly positively correlated. RCH580492 All of the items used in this study confirmed that it can [3] S. Bachok, M. M. Osman, Z. Ponrahono. Passenger’s be one of the determinants to encourage LRT users to Aspiration towards Sustainable Public Transportation choose their own mode choice of transportation. The System: Kerian District, Perak, Malaysia, Procedia Social results are supported by Roberts [14], both of whom and Behavioral Sciences, Vol. 153, 553–565, 2014. posited that the determinants influencing transportation [4] O. Phoebe. Factors Influencing Customer Satisfaction in modes for the citizen include frequency, travel time, cost, Public Transport Sector: A Case of Matatus in Central service, comfort, and cleanliness. Business District Nairobi-Kenya, Thesis, University of Nairobi, Kenya, 2017. [5] M. Hynes. At a Crossroads: Investigating Automobility and 5. Conclusions Its Implications for Local Urban Transport Policy Design, Urban Science, Vol. 1, No. 14, 1-16, 2017. This study investigated user perception(s) on the usage of LRT and the correlation between multiple items. The [6] R. Kamaruddin, O. Ismah, C. P. Che Anizaliana. Customer results indicated that the highest usage of LRT is to Expectations and its Relationship Towards Public Transport in Klang Valley, Journal of Asian Behavioural Studies, Vol. school/college/university. Additionally, most of the LRT 2, No. 5, 29-38, 2017. users agree with the items on LRT, such as reasonable ticket price, the sufficiency of information provided, [7] S. Jain, M. Khare. Urban Air Quality Modelling: A feeling of safety, the cleanliness of the stations, the user Neuro-Fuzzy Approach, VDM Verlag Müller, Germany, 2010. friendliness of the payment system, the available routes, and waiting/travel times are all acceptable. There are 76% [8] M. L. Parry, O. F. Canziani, J. P. Palutikof, P. J. van der of the LRT users feeling safe while waiting at the station. Linden, C. E. Hanson. Climate Change 2007: Impacts, However, the highest dissatisfied rate is waiting time. Adapation and Vulnerability, Cambridge Univ. Press, United Kingdom, 2007. There are about 25% of the users who stated that there are not satisfaction with the route waiting time. The [9] IEA. Energy Technology Perspectives 2016: Towards stakeholders should reschedule the railway timetable Sustainable Urban Energy Systems, International Energy Agency, France, 2016. especially during peak hours because the majority of the respondents use LRT for commuting to [10] G. Santos, H. Behrendt, L. Maconi, T. Shirvani, A. school/college/university and work. User perception on Teytelboym. Part I: Externalities and economic policies in using the LRT allowed us to pinpoint the issues and road transport, Research in Transportation Economics, Vol. 28, No. 1, 2-45, 2010. challenges that could be used to evaluate the quality of the public transport and suggest approaches that would result [11] U. A. Khalid, S. Bachok, M. M. Osman, M. Ibrahim. User in its improvement. The measurement of the determinant Perceptions of Rail Public Transport Services in Kuala attribute of residents’ choices when traveling would result Lumpur, Procedia Social and Behavioral Sciences, Vol. 153, No. 16, 566-573, 2014. in information that unveils crucial information on the needs of customers in the context of improving the overall [12] S. Fujii, R. Kitamura. What does a one-month free bus ticket performance of public transportation systems. do to habitual drivers? An experimental analysis of habit and attitude change, Transportation, Vol. 30, 81-95, 2003. [13] M. Fellesson, M. Friman. Perceived satisfaction with public Acknowledgements transport services in nine European cities, The Journal of Transportation Research Forum, Vol. 47, 93-103, 2008. The authors would like to extend their gratitude to the [14] K. Roberts. Key Factors and Trends in Transportation Mode Research Management and Innovation Centre (RMIC), and Carrier Selection, The Journal of Undergraduate Sultan Idris Education University (UPSI) for the Research at the University of Tennessee, Vol. 4, No. 1, University Research Grants (code: 2018-0082-106-01) that 41-52, 2012.

Civil Engineering and Architecture 7(6A): 43-49, 2019 49

[15] G. Botzoris, A. Galanis, V. Profillidis, N. Eliou. Commuters [30] M. Tavakol, R. Dennick. Making Sense of Cronbach’s Perspective on Urban Public Transport System Service Alpha, International Journal of Medical Education, Vol. 2, Quality, WSEAS Transactions on Environment and 53-55, 2011. Development, Vol. 11, 182-192, 2015. [31] H. F. Kaiser. An index of factorial simplicity, Psychometrika, [16] C. S. Leem, Y. Yoon. A maturity model and an evaluation Vol. 39, 31–36, 1974. system of software customer satisfaction: The case of software companies in Korea, Industrial Management & Data Systems, Vol. 104, No. 4, 347–354, 2004. [17] J. R. Aworemi, A. O. Salami, J. O. Adewoye, M. O. Ilori. Impact of socioeconomic characteristics on formal and informal public transport demands in Kwara state, Nigeria, African Journal of Business Management, Vol. 2, No. 4, 72-76, 2008. [18] J. Pucher, N. Korattyswaropam, N. Mittal, N. Ittyerah. Urban Transport Crisis in India, Transport Policy, Vol. 12, No. 3, 185-198, 2005. [19] L. Zheng, J. Wu. Summary of the Application Effect of Bus Rapid Transit at Beijing South-Centre Corridor of China, Journal of Transportation Systems Engineering and Information Technology, Vol.7, No.4, 137-142, 2007. [20] H. I. M. Irtema, A. Ismail, S. I. Albrka, M. A. Ladin, H. A. Yahia. Evaluating the performance of traffic flow in four intersections and two roundabouts in Petaling Jaya and Kuala Lumpur using Sidra 4.0 software, Jurnal Teknologi, Vol. 72, No. 4, 1-5, 2015. [21] R. Ismail, M. H. Hafezi, R. M. Nor, K. Ambak. Passengers Preference and Satisfaction of Public Transport in Malaysia, Australian Journal of Basic and Applied Sciences, Vol. 6, No. 8, 410–416, 2012. [22] M. Cantwell, B. Caulfield, M. O’Mahony. Examining the Factors that Impact Public Transport Commuting Satisfaction, Journal of Public Transportation, Vol. 12, No. 2, 1-21, 2009. [23] T. Mmutle, L. Shonhe. Customers' perception of Service Quality and its impact on reputation in the Hospitality Industry, African Journal of Hospitality, Tourism and Leisure, Vol. 6, No. 3, 1-25, 2017. [24] G. Sweis, R. Imam, G. Kassab, R. Sweis. Customer Satisfaction in Apartment Buildings: The Case of Jordan, Life Science Journal, Vol. 10, No. 12s, 237–245, 2013. [25] S. Sumaedi, I. G. M. Y. Bakti, T. Rakhmawati, N. J. Astrini, T. Widianti, M. Yarmen. Factors influencing public transport passengers’ satisfaction: a new model, Management of Environmental Quality: An International Journal, Vol. 27, No. 5, 585-597, 2016. [26] A. Mouwen. Drivers of customer satisfaction with public transport services, Transportation Research Part A: Policy and Practice, Vol. 78, 1-20, 2015. [27] J. S. Sharfuddin. Factor Analysis for the Study of Determinants of Public Transit Ridership, Journal of Public Transportation, Vol. 3, No. 3, 1-17, 2000. [28] J. F. Hair Jr, W. C. Black, B. J. Babin, R. E. Anderson, R. L. Tatham. Multivariate data analysis (7th Ed.), Prentice-Hall, Upper Saddle River, NJ, 2010. [29] A. G. Bluman. Elementary Statistics: A Step By Step Approach 10th Edition, McGraw-Hill Education, New York, 2017.

Civil Engineering and Architecture 7(6A): 50-57, 2019 http://www.hrpub.org DOI: 10.13189/cea.2019.071406

Mathematical Modeling for Flood Mitigation: Effect of Bifurcation Angles in River Flowrates

Iskandar Shah Mohd Zawawi1,*, Nur Lina Abdullah2, Hazle e n Aris 1, Badrul Amin Jaafar2, Nur Arif Husaini Norwaza2, Muhammad Haris Fadzillah Mohd Yunos 2

1Institute of Informatics and Computing in Energy, Universiti Tenaga Nasional, M alaysia 2Faculty of Computer and Mathematical Sciences, Universiti Teknologi MARA Negeri Sembilan, M alaysia

Received August 4, 2019; Revised October 12, 2019; Accepted December 15, 2019

Abstract This paper investigates the river flowrate at set at a right-angled midway along the straight main two branches of bifurcated river. The mathematical model channel. The estimation of the flowrate ratio in terms of the fro m the literature is formulated based on momentum Froude number and the depth ratio had been obtained using principle and mass continuity to cope with river flowrate at theoretical model in [2]. The authors provided the different bifurcation angles. The hydraulic variables, experimental data for the validity of their proposed model. geometric properties of trapezoidal cross-sectional river Based on experimental observations, the work of [3] and other physical characteristics of bifurcated river are carried out a study on depth discharge relationship and provided, which may be assumed to be given beforehand energy-loss coefficient for a subcritical, equal-width, for practical applications. An example of river bifurcation right-angled dividing subcritical flow over a horizontal bed problem is given by UTM Centre for Industrial and in a narrow aspect ratio channel. The theoretical model for Applied Mathematics (UTM-CIAM), Universiti Teknologi subcritical flows in dividing open channel junction is Malaysia. Maple software is used to implement the derived in [4] with the aid of the overall mass conservation proposed model equation and generate the results. The together with the momentum principle in the streamwise amount of bifurcated river flowrate with different direction to two control volumes through the junction. bifurcation angles is determined, resulting in a reasonable Further, a physical model with meandering features is discussion. It is shown that for specific bifurcation angles, constructed in [5] to investigate the effect of off-take the river flowrates after the bifurcated junction are less than angles on the flow distribution at a concave channel the critical flowrate. Finally, the results of applied problem bifurcation. indicate that the right-angled river bifurcation would be A theoretical model for predicting depth of water with preferable to mitigate flood. certain dividing angles has been proposed by [6]. The authors developed the model equations for both combining Ke ywo rds Bifurcat ion, Flo wrate, Momentum and dividing types of subcritical flows at channel junctions Principle using the principal of momentum balance. The width of all the channels both in case of combining and dividing has been kept differently. An unsteady mathematical model for predicting flow 1. Introduction divisions at a right-angled open-channel junction [7] and hydrodynamic model [8, 9] for bifurcating stream was a lso River bifurcation is the process that determines the done. More recently, the findings of nearly 10 years of distribution of flow, sediments and contaminants along the researches into modeling bifurcation system w ith downstream river branches. This process is important in numerous simulation techniques have been reviewed [10]. order to mitigate flood due to climate change. There have However, none of the above work analyzes the effect of been several approaches in investigating the river different bifurcation angles in river flowrate. In fact, the bifurcation or bifurcated open-channel flow. For instance, majority of the existing models are designed for a [1] used both analytical and experimental ways to study the right-angled junction. Therefore, the aim of this paper is to bifurcated open-channel flow. The channels used are of investigate the behavior of river flowrates influenced by rectangular cross-sectional and the branch channel being different bifurcation angles using mathematical model Civil Engineering and Architecture 7(6A): 50-57, 2019 51

approach. The following section will describe the characteristics of the bifurcated channel and its geometric properties. Section 3 deals with the formulation of the mathe matical model. An example of river bifurcation problem is given in Section 4. In section 5, the results are analyzed and discussed. Finally, some conclusions are made as well as the recommendation for future study.

2. Methodology Fi gure 2. Geometric details of the typical trapezoidal cross-sect io nal: α = angle of the slope side, b = bottom widt h, y = depth of flow, λ = This section provides the detailed description of the wetted length measured along the slope side, T = top width, z = channel channel and formulation of the model for the present study. side slope According to [11], flow hydraulics and momentum 2.1. Description of the Channel exchange in straight channels are significantly influenced by geometric and hydraulic variables. The cross-sectional The characteristics of the bifurcated open-channel and its cross-sectional properties have to be considered for the area, A is given by A= by + zy2 , in wh ich b is the equation of mathematical model. The schematic layout of width of the channel bottom and y is the depth of flow. the bifurcated channel is illustrated in Figure 1. A main The side slope is usually specified as horizontal : vertical, channel is connected with two branch channels. The angles, z :1. Additional parameters for open channel flow are the θ and θ at the bifurcated junction are called 1 2 wetted perimeter, P , the hydraulic radius, R and the bifurcation angles. For the application of momentum w H conservation law, we consider the boundaries of control hydraulic depth, D . The wetted perimeter, Pbw = + 2λ volume as shown by the dotted line. The section has been is the length of the line of contact between the water and positioned at the distance of two times the width of the the channel where the wetted length measured along the channel at upstream and three times the width of the slope side is given by λ =y2 + ( yz)2 . channel at downstream of the bifurcation. The hydraulic radius, RH is the area divided by the A wetted perimeter, that is, RH = . The hydraulic depth Pw A is the area divided by the top width, D = where T T= b + 2 zy . Flow area is the cross-sectional area of the flow taken perpendicular to the flow direction. Even though the river cross-sectional areas of the main channel and channel 1 are assumed to be similar, the capability of the channel to convey water can vary due to bifurcation angle. If the flowrate is unknown, a uniform velocity, V that applies to an entire cross-sectional can be determined using Manning’s equation [12, 13] as shown below: Fi gure 1. Schemat ic layout of t he bifurcat ed channel: Q = flowrat e, b = bottom width of channel, θ = bifurcat ion angles, 0 = main channel, 1 1 0.66 0.5 = channel 1, 2 = channel 2 VR= H S n , The channels are assumed to be uniform cross section. where n is the roughness coefficient and S is the Channel cross sections can be considered to be either average slope of channel. The dimensionless ratio of the regular or irregular. A regular section is one whose s hape inertial forces to gravitational forces acting on the flow is does not vary along the length of the channel, whereas an represented by Froude number, F is defined as irregular section will have changes in its geometry. The most common irregular section of open channel is a VQ F = = trapezoidal shape. The typical trapezoidal cross-sectional gD A Ag is shown in Figure 2.  T ,

52 Mathematical Modeling for Flood Mitigation: Effect of Bifurcation Angles in River Flowrates

where V is the velocity, D is the hydraulic depth and PP02−cosθθ 21 − P cos 1 − UU 2 − 1 −∆ P = the gravitational acceleration, g is 9.80665m/s2 . The γ (3) (Q22211100 Vcosθθ+− QV cos Q V ) . Froude number plays a significant role in open channel g flow analysis. The hydraulic behavior of channel flow varies significantly depending on whether the flow is The ter ms of momentumtransfering from the main critica l ( F = 1 ), subcritical ( F < 1 ) o r supercrit ical channel to the branch channels are given in the following ( F > 1). forms [7]: The division of flowrate at bifurcated channel can be U1= ρ QVC 10sin θρ 1 , U 2= QVC 20sin θ 2 , determined using the aid of momentum principle and mass continuity with the following assumptions: where (a) Main channel is straight prismatic channel, to which 2 two branches of bifurcated junction are connected. 5 Fk0012+ k 0 C =−− . All channels are trapezoidal cross-sectional. 6 40 12q 2 r (1+ k0 ) (b) The flow is from main channel into channels 1 and 2. (c) The velocities and water surface elevations are 2 3 constant across the channels at the inflow and outflow by02 zy2 Noting that ∆=PPγθ + −22cos wh ile sections of the control volumes. 23 (d) The pressure distribution is hydrostatic at all sections the density of water, ρ is re lated to γ and g wh ich of control volume. γ (e) The geometrical properties such as channels width, can be determined as ρ = . By moving the terms, U1 channels depth, control volume lengths and slope of g channel are known. and U2 to the right hand side (RHS), (3) can be written (f) The depth of flow in the main channel, channels 1 and as follows: 2 are equal. γ (g) The shear stresses on the flow surface due to wind, the P02211− Pcosθθ − P cos −∆ P = [ QV 222cos θ g effects of vertical acceleration and the wall friction (4) force as compared to other forces are neglected. +QV11cosθ 1 −+ Q 00 V ρ Q 20 V Csin θρ 2 + QV 10 Csin θ 1] . By taking the left hand side (LHS) of (4), we have the 2.2. Formulation of the Model following equation: This section describes the detail formulation of PP02−cosθθ 21 − P cos 1 −∆ P. (5) mathematical model based on momentum principle and mass continuity [7]. The basic continuity equation is taken Based on (2), we simplify (5) as follows: as starting point for the formulation, 1 2 22 γθ(by0 0−− by 0 2 by 11cos 1) Q= QQ + , (1) 2 012 (6) z 3 33  where Q0= A 0 V 0, Q 1 = AV 11 and Q2= AV 22. The +( yyy021 −−cosθ 1) . 3  terms QQ01, and Q2 are flowrates, AA01, and A2 Now taking the RHS of (4), are trapezoidal cross-sectional areas while VV01, and V are velocities in main channel, channel 1 and channel γ 2 [Q22211100 Vcosθθ+− QV cos Q V 2 respectively. The hydrostatic force on the horizontal g (7) strip of A will be PA= γ where A= by + zy2 and ++ρQVC20sin θρ 2 QVC 10sin θ 1] . γ are the specific weight of water. Therefore, the total Based on (1), we produce horizontal force can be determined as follows: γ Q Q QQ QQ 2 2θθ+− 11 00 y 23 y 23   cos21 cos 2 by zy by zy gA 2 A 10 A P=+=+=+γγ by zy dA .(2) (8) ∫    230 23 Q20 Q QQ 10  0 ++CCsinθθ21sin . AA00 By applying the continuity equation (1) and momentum principle in the flow direction of the main channel, we By using algebraic manipulation in (8), yields obtain

Civil Engineering and Architecture 7(6A): 50-57, 2019 53

2  22 2 γ Q0 11QQ21 kby001 =  cosθθ21+ cos by11+ gA0QQ22 A 20// A AA10 A1 y0  00 (9) = . (15) A 2 QQ 0 kby000 21 by00+ −+1C  sinθθ21+ sin . y QQ00 0 In our case, it has to be noted that the depths of all Q1 Let the flowrate ratio, qr = . Based on (1) and channels are equal, y012= yy = . Thus, the depth ration Q 0 yy =12 = Q10= Qqr , we get Q0= Q 20 + Qqr and subsequently is given by yr . Equation (15) is written as yy00 Q2 produce =1. − qr Therefore, the following equation follows: Q0 A by+ kb yy is obtained: 1 = 11 001r . (16) A0 by 00+ kby 000 γ 2  − 2 2 Q0 (1 qr ) qr By multiplying the numerator and denominator of (14) cosθθ21+ cos gA0 A 20// A A 10 A . (10) 1 and (16) with , we obtain −+1Cq(( 1 −rr) sinθθ21 + q sin) . by00 Knowing that A (Br10+ k yrr) y 1 = , (17) 22 222 2 Ak001+ Q0000000 Q T A QTA2 A 0 = ×× = =F0 . gA gA T 23TT A (Br+ k y) y 0 00A00 gA 0 0 2 = 20rr. (18) Ak001+ Subsequently, yielding We substitute (17) and (18) into (10), yielding b22 y++2 b zy 3 z 24 y 2 0 0 00 0 (11) F0 . + 2  − 2 b+ 2 zy 22(1 k0 )  (1 qr ) 00 γθF0 by 00 cos 2 (1++ 2k0) ( Br 20 k yrr) y/1( + k 0) kb00 Let z = and factorize, we have 2 y qr 0 +−cosθ1 1 (19) (Br10++ k yrr) y/1( k 0) 22 2 2 by00(12++ k 0 k 0) +−Cq((1) sinθθ + q sin) . F0 . (12) rr21 bk00(12+ ) Simplifying (19), we get 2 + 2 2   2 Q0 22(1 k0 ) A2 (1+ k ) 1+ k (1− q ) Hence, = F0 by 00 . The term γθF22 by 0  0  r cos gA (12+ k ) A 0 00 ++  2 000 (12k0)  yrr ( Br 20 k y ) (20) can be written as follows: 2   qr  + cosθ1 −+ 1Cq(( 1 −rr) sin θθ 21 + q sin) . A b y+ zy 2 (Br+ k y )   2= 22 2 . (13) 10r   A 2 0 b00 y+ zy 0 Finally, LHS of (6) is equal to RHS of (20) that becomes y2 b2 b1 Let = yr , = Br2 and = Br1, we get y b b (12+ k0 ) 1 2 22 0 0 0 (by0 0−− by 0 2 by 11cosθ 1) 2 2 A by+ kbyy by00 2 = 22 002r . (14) A by+ kby z 3 33 2 2 0 00 000 +( yyy021 −−cosθ 1) = F 0( 1 + k 0) 3  (21) A 1   2 Similarly, the term is given by 1+ k (1− q ) q 2  A0  0  r cosθθ+ r cos   21  yrr( Br20++ k y ) (Br 10 k y r) −+1Cq(( 1 −rr) sinθθ21 + q sin) .

54 Mathematical Modeling for Flood Mitigation: Effect of Bifurcation Angles in River Flowrates

After simp lification of (21), the general equation of bifurcated channel used in the proposed model are bifurcated flow is obtained in the following form: presented in Table 1. 1 22 Ta b l e 1 . Geometric and hydraulic properties (GHP) of bifurcat ed (1+ 2k0) ( 1 −− yrr Br 11 y cosθ ) channel 2 GHP Main channel Channel 1 Channel 2 kk0033 2 2 1+ + −−θ = + 0 0 0 (1yyrr cos10) F( 1 k 0) α 60 60 60 3   yr (22) y 3.5m 3.5m 3.5m 2 (1− q ) q 2 r θθ+ r z 4.081632657m 4.081632657m 0.859291084m cos 21cos (Br20++ k yrr) (Br10 k y ) T 300m 300m 60.15037594m 271.285714m 271.285714m 54.13533835m −+1Cq(( 1 −rr) sinθθ21 + q sin) . b A 1000m2 1000m2 200m2 λ 14.70821651m 14.70821651m 4.614668923m 3. River Bifurcation Problem Pw 300.8450044m 300.8450044m 63.3646762m R 3.323970767m 3.323970767m 3.156332708m In this section, we give special attention to river H bifurcation problem in Sungai Nenggiri, Gua Musang, D 3.33m 3.33m 3.325m Kelantan. Sungai Nenggiri that is geographically located in the north eastern part of Peninsular Malaysia within latitude 4.97024° to 4.96951° North and 101.77144° to 4. Results 101.77207° East. This river is considered in this study due to serious floods’ occurrence during Monsoon season in To analyze the results, the model (22) is performed using the past few years. Extensive flooding throughout the Maple software. The bifurcation angles, θ1 and θ2 catchment occurs during heavy and prolonged rainfall considered in this study are resulting in high river flow. The river flow will overspill 0000000 , 15 , 30 , 45 , 60 , 75 and 900 . The values of the banks of Sungai Nenggiri, disrupting road network flowrate ratios, , flowrates in channel 1, and and human life. qr Q1 The main mitigation action that can be taken is by flowrates in channel 2, Q2 are tabulated in Tables 2-8. It diverting some of Sungai Nenggiri’s flow during peak has to be mentioned that qr is the ratio o f flo wrate in flow to a new river, namely Sungai Anak Nenggiri. The channel 1 to the flowrate in main channel. For simplicity, amounts of river flow from Sungai Nenggiri (main the graphical representations of flowrate ratios and channel) going through Sungai Nenggiri after the bifurcation angles are shown in Figure 3. bifurcation junction (channel 1) and Sungai Anak 0 Nenggiri (channel 2) are depending on the bifurcation Table 2. Flowrates in channels 1 and 2 when θ1 = 0 angles, θ and θ . The flowrate in the main channel is θ θ q Q 3 Q 3 1 2 1 2 r 1 , m /s 2 , m /s 3 assumed to be Q0 = 1000m /s while Q1 and Q2 are 0 0 0.8067940552 806.7940552 193.2059448 the flowrates in channel 1 and channel 2 respectively. The 0 15 0.8227092780 822.7092780 177.2907220 critical flowrate in channel 1 is expected to be 0 30 0.8284124759 828.4124759 171.5875241 3 Q1 = 800m /s . We theorize that if the flowrate exceeds 0 45 0.8214011067 821.4011067 178.5988933 this value, flood will occur in channel 1. Therefore, (22) 0 60 0.7928048911 792.8048911 207.1951089 can be applied to determine the amount of flowrates in 0 75 0.7125371460 712.5371460 287.4628540 channel 1 and channel 2 with different bifurcation angles. This general problem is given by UTM Centre for 0 90 0.4161533812 416.1533812 583.8466188 Industrial and Applied Mathematics (UTM-CIAM), 0 Ta b l e 3 . Flowrates in channels 1 and 2 when θ1 =15 Universiti Teknologi Malaysia. However, the real θ θ q Q 3 Q 3 experimental data of Sungai Nenggiri is unavailable at this 1 2 r 1 , m /s 2 , m /s time for error analysis. For application purpos e, we 15 0 0.7911862451 791.1862451 208.8137549 assume that the channels are normal, clean, straight, full s tage, with no rifts or deep pools. Thus, n = 0.03 is 15 15 0.8067650940 806.7650940 193.2349060 selected as manning’s coefficient while the slope of the 15 30 0.8110586289 811.0586289 188.9413711 main channel is S = 0.0001814260235m . Since the 15 45 0.8010471671 801.0471671 198.9528329 Froude number for the main channel is 15 60 0.7663678231 766.3678231 233.6321769 = , it can be said that the flow is F0 0.174902437 15 75 0.6717559187 671.7559187 328.2440813 subcritica l. The geometric and hydraulic properties of 15 90 0.3192905923 319.2905923 680.7094077

Civil Engineering and Architecture 7(6A): 50-57, 2019 55

Table 4. Flowrates in channels 1 and 2 when θ = 300 3 1 The amount of Q1 is less than 800m /s when θ θ q Q 3 Q 3 0 00 0 0 1 2 r 1 , m /s 2 , m /s θ1 = 0 and θ2 = 60 , 75 or 90 . If θ1 = 30 , 30 0 0.7869497049 786.9497049 213.0502951 3 Q is less than 800m /s except when θ = 150 or 30 15 0.8027210236 802.7210236 197.2789764 1 2 0 0 0 3 30 30 0.8067320739 806.7320739 193.2679261 30 . For θ1 = 15 or 45 , Q1 is less than 800m /s 30 45 0.7957057968 795.7057968 204.2942032 θ = 000 0 30 60 0.7582902466 758.2902466 241.7097534 when 2 0 , 60 , 75 or 90 . It also can be 30 75 0.6546484797 654.6484797 345.3515203 3 0 observed that Q1 is less than 800m /s when θ1 = 60 30 90 0.2401904453 240.1904453 759.8095547 and θ = 0 or 0 . Fro m Tables 2-7, it is observed 0 2 75 90 Table 5. Flowrates in channels 1 and 2 when θ1 = 45 that Q decreases significantly when θ = 900 . The θ θ Q Q 1 2 1 2 qr 1 , m3 /s 2 , m3 /s 0 0 45 0 0.7949051046 794.9051046 205.0948954 lowest value of Q1 is when θ1 = 75 and θ2 = 90 45 15 0.8114393561 811.4393561 188.5606439 as presented in Table 7. However, Q1 becomes greater 45 30 0.8164352045 816.4352045 183.5647955 than the critical flowrate when θθ= = 900 . The values 45 45 0.8066871811 806.6871811 193.3128189 12 of bifurcation angle when Q is less than the critical 45 60 0.7705429480 770.5429480 229.4570520 1 45 75 0.6645994258 664.5994258 335.4005742 flowrate that is summarized in Table 9. 45 90 0.1722446834 172.2446834 827.7553166 Table 9. Bifurcation angles when qr < 0.8 0 θ θ Table 6. Flowrates in channels 1 and 2 when θ1 = 60 1 2 00 600, 750, 900 θ θ q Q 3 Q 3 1 2 r 1 , m /s 2 , m /s 150 00, 600, 750, 900 60 0 0.8159655445 815.9655445 184.0344555 300 00, 450, 600, 750, 900 60 15 0.8339066727 833.9066727 166.0933273 0 0 0 0 0 60 30 0.8414225983 841.4225983 158.5774017 45 0 , 60 , 75 , 90 0 0 0 60 45 0.8358869677 835.8869677 164.1130323 60 75 , 90 60 60 0.8066091461 806.6091461 193.3908539 750 900 60 75 0.7097821934 709.7821934 290.2178066 900 00, 150, 300, 450, 600, 750 60 90 0.1112785518 111.2785518 888.7214482 0 Table 7. Flowrates in channels 1 and 2 when θ1 = 75 θ θ q Q 3 Q 3 1 2 r 1 , m /s 2 , m /s 75 0 0.8509646114 850.9646114 149.0353886 75 15 0.8710508212 871.0508212 128.9491788 75 30 0.8830122328 883.0122328 116.9877672 75 45 0.8856383590 885.6383590 114.3616410 75 60 0.8716827337 871.6827337 128.3172663 75 75 0.8063916606 806.3916606 193.6083394 75 90 0.0542840796 54.28407957 945.7159204 0 Table 8. Flowrates in channels 1 and 2 when θ1 = 90 θ θ q Q 3 Q 3 1 2 r 1 , m /s 2 , m /s 90 0 0.6742997042 674.2997042 325.7002958

90 15 0.7031325671 703.1325671 296.8674329 Fi gure 3. Graph of flowrate ratios, q with respect to θ wh en 90 30 0.7164798741 716.4798741 283.5201259 r 2 00 0 0000 90 45 0.7131229768 713.1229768 286.8770232 θ1 = 0 , 15 , 30 , 45 , 60 , 75 , 90 90 60 0.6837385036 683.7385036 316.2614964 Figure 3 shows the graph of q versus θ where the 90 75 0.5844868749 584.4868749 415.5131251 r 2 0 0 90 90 0.8198591028 819.8591028 180.1408972 values of θ1 are fro m 0 to 90 . The horizontal

56 Mathematical Modeling for Flood Mitigation: Effect of Bifurcation Angles in River Flowrates

dashed line at qr = 0.8 represents the critical flowrate b : Bottom width of the channel Width ratio ratio. For any θ1, it can be seen that qr approaches to Br : 0 T : Top width of the channel the critical flowrate ratio except when θ2 = 90 . To U : Momentum transfer avoid over-flow in channel 1, both θ1 and θ2 cannot be V : Flow velocity 900 . y : Flow depth From the results obtained, it can be observed that the yr Flow depth ratio right-angled bifurcation at one of the branches (either : ρ Specific gravity : θ = 900 or θ = 900 ) would be efficient to reduce the 1 2 γ Specific weight amount of flowrates in channel 1 significantly. However, : 0 θ : Bifurcation angles of channels 1 and 2 T-junction (when both θ1 and θ2 are 90 ) is not recommended. Subscripts

5. Conclusions 0 : Main channel (Sungai Nenggiri) 1 Channel 1 (Sungai Nenggiri after bifurcation) This study provides insightful information for : understanding of the open-channel flow and assists 2 : Channel 2 (Sungai Anak Nenggiri) engineering design of river bifurcation. The mathematical r : Ratio model is derived based on continuity equation, momentum principle and some algebraic manipulations to predict the bifurcated river flowrates with different bifurcation angles. Acknowledgements The model equation consists of Froude number and various important parameters such as bifurcation angles, width of We would like to take this opportunity to acknowledge th channels, depth of flows and flowrates in branches of river. and thank the 5 International Conference on the Thus, it can be applied for other rivers with different Application of Science and Mathematics (SCIEMATHIC geometric properties. The analysis of the results reveals 2019) secretariat for their role in the success of the that the river flowrate after the bifurcated junction is below conference. This research was supported by Universiti the critical flowrate if an appropriate bifurcation angles are Tenaga Nasional (UNITEN) and Universiti Teknologi considered. The implementation of right-angled MARA (UiTM) Negeri Sembilan, Seremban Campus. The bifurcation at Sungai Nenggiri can be an alternative action UTM Centre for Industrial and Applied Mathematics to mitigate flood. (UTM -CIAM), Universiti Teknologi Malaysia provided In future study, the mathematical model for river topic problem for Malaysia Industrial Mathematical bifurcation with different bifurcation angles can be Modeling Challenge 2019 (MIMMC2019). investigated when the problem concerning the recirculation region is understood. Other interesting features that can be observed are the hydraulic jumps and the surface discontinuity. Furthermore, other minor factors affect the river bifurcation flowrate such as wall frictions and REFERENCES external forces that should be considered. [1] S. W. Law. Dividing Flow in Open Channel, Master Thesis, McGill University, Montreal, 1965.

[2] A. S. Ramamurthy, D. M. Tran, L. B. Carballada. Dividing Nomenclatures Flow in Open Channels, Journal of Hydraulic Engineering, ASCE, Vol. 116, No. 3, 449–455, 1990. A : Cross-sectional area of the channel [3] C. C. Hsu, C. J. Tang, W. J. Lee, M. Y. Shieh. Subcritical C : Constant 900 Equal-Width Open-Channel Dividing Flow, Journal of F : Froude number Hydraulic Engineering, Vol. 128, No. 7, 716-720, 2002. g : Gravitational acceleration k : Side slope x flow depth to bottom width ratio [4] I. M. H. Rashwan. Dynamic Model for Subcritical Dividing Flows in Open Channel Junction, Eight International P : Pressure force Water Technology Conference, 511-520, 2004. P : Wetted perimeter w [5] N. L. Obasi, J. C. Agunwamba, N. Egbuniwe. Influence of Flowrate Q : Off-Take Angles on Flow Distribution Pattern at Concave Channel Bifurcation, Nigerian Journal of Technology, Vol. qr : Flowrate ratio

Civil Engineering and Architecture 7(6A): 50-57, 2019 57

27, No. 2, 46-57, 2008.

[6] A. Pandey, R. Mishra. Comparison of Flow Characteristics at Rectangular and Trapezoidal Channel Junctions, Journal of Physics Conference Series, Vol. 364, 1-11, 2012.

[7] G. Kesserwani, J. Vazquez, N. Rivière, Q. Liang, G. Travin, R. Mosé. New Approach for Predicting Flow Bifurcation at Right-Angled Open-Channel Junction, Journal of Hydraulic Engineering, ASCE, 662–668, 2010.

[8] D. A. Tholibon, J. Ariffin. Bifurcation Simulation Modeling Review, International Journal of Sciences: Basic and Applied Research, Vol. 8, No. 1, 45-50, 2013.

[9] A. Zahiri, A. A. Dehghani. Flow Discharge Determination in Straight Compound Channels using Anns, World Academy of Science, Engineering and Technology, Vol. 58, 12-15, 2009.

[10] S. A. Mirbagheri, M. Abaspour, K. H. Zamani. Mathematical Modeling of Water Quality in River Systems. Case Study: Jajrood River in Tehran – Iran, European Water, Vol. 27/28, 31-41, 2009.

[11] Y. Yang, T. A. Endreny, D. J. Nowak. Application of Advection-Diffusion Routing Model to Flood Wave Propagation: A Case Study on Big Piney River, Missouri USA, Journal of Earth Science, Vol. 27, No. 1, 9-14, 2016.

Civil Engineering and Architecture 7(6A): 58-70, 2019 http://www.hrpub.org DOI: 10.13189/cea.2019.071407

Predicting the Capability of Carboxylated Cellulose Nanowhiskers for the Remediation of Copper from Wastewater Effluent Using Statistical Approach

Hazren A. Hamid1,*, Hasnida Harun1, Norshuhaila Mohamed Sunar1, Faridahanim Ahmad1, Nuramidah Hamidon1, Mimi Suliza Muhamad1, Latifah Jasmani2, Norhidayah Suleiman3

1Department of Civil Engineering Technology, Universiti Tun Hussein Onn Malaysia, Malaysia 2Forest Products Division, Forest Research Institute Malaysia, Malaysia 3Department of Food Technology, Universiti Putra Malaysia, Malaysia

Received August 2, 2019; Revised October 1, 2019; Accepted December 15, 2019

Abstract This study focused on remediation of spiked 1. Introduction Cu(II) from wastewater effluent obtained from a real wastewater treatment plants (WWTPs) using oxidised Adsorption is a complex process, as it involves the cellulose nanowhisker (CNW) adsorbents. Response interaction of various parameters. Moreover, the surface methodology (RSM) and artificial neural network complexity and variability of the wastewater matrix also (ANN) were used to develop an approach for the have a direct impact on the process performance [1, 2]. In remediation of spiked Cu(II) from wastewater effluent. As addition, studies tend to focus on evaluating one single remediation processes from wastewater are often parameter at a time, which assumes each parameter complicated due to the variation in wastewater operates independently. This is non-practical because compositions, results obtained from the benchmark parameter interactions cannot be elucidated using this experiments are included as one of the independent approach [3]. variables for ANN modelling. This novel approach and the Developing a new mathematical modelling for outcomes are allowed for the first time, since most studies remediation process not only can reduce cost and time in do not consider matrix variability and its impact when wastewater treatment, but also investigate the effects of evaluating the efficiency of an adsorbent. Moreover, to input variables or factors on an output variable or response. confirm the model suitability, additional 10 unseen The capabilities of these models were applied to the case experiments, which were not used in developing both study: remediation of copper from water matrices using models, were chosen to represent the system of conditions modified cellulose nanowhisker (CNW) adsorbents. In this both inside and outside the system. This study found that study, the central composite design (CCD) was selected because as it has been extensively applied in adsorption the ANN model accounting for wastewater variability was studies [4, 5]. The mathematical models, response surface superior to the RSM model and to the ANN model not methodology (RSM), and artificial neural network (ANN) including wastewater variability, in terms of the coefficient are among the most popular models used in research on of determination (R2), the absolute average deviation remediation of heavy metals from the clean water matrix [6, (AAD) and root mean squared error (RMSE) when 7]. predicting the efficiency of Cu(II) removal from the Copper is a naturally occurring element which is found wastewater matrix. in water, air, and soil, and is considered as one of the most dangerous substances found in the environment [8]. The Keywords Artificial Neural Networks, Adsorption, maximum guideline concentration limit for copper Cu(II) ions, Cellulose Nanowhiskers, Response Surface discharge to water has been established by the Water Methodology Framework Directive of water policy discharge to inland surface water directive (2000/60/EC) [9]. With these strict guidelines for the regulated levels of copper for wastewater discharge (1–28 µg/L), proper and suitable treatment is Civil Engineering and Architecture 7(6A): 58-70, 2019 59

required in order to meet these discharge limits. Trent Water Stoke Bardolph WWTP in Nottingham, UK. Cellulose has been identified as a promising adsorbent Wastewater effluent was collected twice a week using a for the remediation of heavy metals from the water matrix grab-sampling approach. Dissolved oxygen (DO) and [10, 11]. Although CNWs have been recognised due to temperature were measured by a DO meter (Jenway 970, their high surface area and high reactive group density on Staffordshire, UK) and thermocouple thermometer the surface, only limited research has been published on (Digi-Sense, Cole-Parmer Instrument Ltd., UK), using CNWs as an adsorbent, as the majority of the respectively at the sampling point to avoid any changes literature has mainly focused on macroscopic during storage. lignocellulosic biomass such as jute, orange peel, and sugarcane bagasse fibres, rather than pure cellulose [12, 2.2.1. Wastewater Characteristics 13]. In recent years, TEMPO In order to perform a physico-chemical characterisation (2,2,6,6-tetramethylpyperidine-1-oxyl)-mediated of the water samples, the American Public Health oxidation has frequently been used to introduce carboxyl Association’s Standard methods for the examination of functional groups on the surface of nanowhiskers or native water and wastewater were applied [18]. In each of the cellulose without affecting the crystallinity or changing the samples, the following water quality parameters: total original fibrous morphology [14]. suspended solids (TSS), pH, conductivity, total dissolved However, most of the previous literature focuses on solids (TDS) and chemical oxygen demand (COD) were ad-sorption studies by using either RSM or ANN, without determined. comparing the performances between these two models. Furthermore, the testing of both RSM and ANN using new 2.2.2. Wastewater Effluent Spiked with Cu(II) sets of experiments not belonging to the training data set The Cu(II) concentration in wastewater effluent was has only been undertaken by a limited number of studies on adjusted to the required concentrations (1–5 mg/L) by biomass adsorption, and without consideration of how the dissolving the appropriate amount of CuSO4.5H2O in a 200 additional experiments represent the system and provide a ml volumetric flask, followed by dilution up to the mark by more accurate indicator of performance [4, 15]. Moreover, the addition of filtered effluent. The effluent was studying the effect of matrix complexity and the variability previously filtered through a standard 1.2 µm glass fibre of the wastewater, along with applying realistic conditions filter. for WWTP, could lead to the establishment of a good knowledge based on adsorption behaviour and provide the 2.3. Determination of the Cu(II) in the Solutions foundation for further studies. The initial and final concentration of Cu(II) in the wastewater effluent was determined by flame atomic 2. Materials and Methods absorption spectrometry (AAS) (Model No: 272, PerkinElmer Inc., USA). The hollow cathode lamp was operated at 10 mA and the analytical wavelength was set at 2.1. Preparation of CNWs 324.8 nm. The standard solutions (2–10 mg/L) that span CNWs were produced from bleached cotton by the working ranges were prepared by using the provided hydrolysis with a mass fraction of 64% sulphuric acid to 1000 mg/L reference standard solution (ROMIL Ltd) for produce a suspension of highly crystalline CNWs Cu(II) with Milli-Q water. A linear regression curve according to standard procedures [16]. The resulting (y=0.03313x+0.00212) was obtained in the Cu(II) CNWs were then reacted with TEMPO, sodium bromide, concentration ranging from 2 to 10 mg/L with a correlation and sodium hypochlorite for 45 min under constant coefficient of 0.999. stirring at room temperature (T=19°C) at pH 10 to introduce the carboxyl groups onto the CNWs’ surface 2.4. Batch Adsorption Studies Using Wastewater followed by freeze-drying [17]. The characteristics and Effluent analyses of CNWs are crucial in understanding the mechanism on the adsorbent surface. The methods that Batch experiments were performed in 100 mL conical have been used in this study are Fourier Transform flasks, in an incubator (Model No: 120, LMS Ltd., Kent, Infra-Red (FTIR) spectroscopy, zeta potential, Brunauer– UK), with temperature control and agitation (150 rpm) Emmett–Teller (BET) method, scanning electron using a mini table shaker (IKA Vibrax VXR, Germany). microscopy (SEM), transmission electron microscopy The contact time (30 min), and the temperature (20ºC) (TEM) and conductometric titration method. were selected on the basis of the results obtained from scoping experiments. The required weight of sorbent (0.5– 2.2. Wastewater Samples 10.0 g/L) was measured separately into the 100 mL conical flask, and then 20 mL of Cu(II) solution with the The sample collection was carried out at the Severn known concentration (1–5 mg/L) was added into the

60 Predicting the Capability of Carboxylated Cellulose Nanowhiskers for the Remediation of Copper from Wastewater Effluent Using Statistical Approach flasks. The initial and final solutions were separated by hidden layer and is commonly applied when predicting the filtration using 0.2 µm surfactant-free cellulose acetate performance of many processes [20, 21]. In order to use membrane syringe filter and Cu(II) concentration the ANN model for predicting Cu(II) removal from the determined using AAS. water matrix, a feed-forward backpropagation was used The percentage of the removal Cu (II) ions by the for modelling the experimental design. In this study, the sorbent and the adsorption capacity (mg Cu(II)/g) was first layers of neurons representing the independent expressed by: variables were identical to the factors considered in the RSM approach. Similar to the RSM modelling, the % removal = x 100 (1) percentage removal of Cu(II) was considered as the output Co−Ce ( ) neurons and was developed in MATLAB (R2009b), = Co (2) Mathwork Inc. Software. Co−Ce V 푒 where Co (mg/L) is푞 the initial� W Cu(II)� concentration and Ce (mg/L) is the equilibrium Cu (II) concentration in solution, 2.7. Data Analysis and Statistical Techniques V is the volume of the solution (L), and W is the mass of adsorbent (g) [19]. 2.7.1. Model Evaluation The regression analysis, graphical analysis, and analysis 2.5. Benchmarks Study of variance (ANOVA) were undertaken using MINITAB 16 Statistical Software. The performance of the ANN and A benchmark study was performed in order to RSM model was statistically evaluated in terms of the understand the complexity and variability of the coefficient of determination (R2), absolute average wastewater matrix on the adsorption performance. For deviation (AAD), and the root mean squared error benchmark studies, 20 ml of 4 mg/L Cu(II) wastewater (RMSE). Both models and the parameters variation were effluent spiked with Cu(II) was agitated with 1 g/L determined based on the minimum value of the RMSE sorbent dosage for 30 minutes at 20 C at pH 6. This and AAD of the training and prediction set (Equation 4 benchmark experiment was performed for each and 5) [22]. wastewater sample (for every sampling trip). The initial and final samples were separated by filtration using a 0.2 RMSE = y y ) (4) 1� µm surfactant-free cellulose acetate membrane syringe 1 n 2 2 p e and the concentration of the samples was determined by = �n ∑i=1� − ×�100� (5) AAS. 1 푛 푦푝−푦푒 푖=1 Where n 퐴퐴퐷is the number�푛 ∑ of� 푦points,푒 �� yp is the predicted 2.6. Prediction Model of Cu(II) Removal from the value, ye is the experimental value. Wastewater Matrix 2.7.2. Test and Validation of the Model 2.6.1. Response Surface Methodology (RSM) For purposes of validation and evaluation of RSM and The RSM is an approach that combines mathematical ANN models, additional unseen experiments were and statistical techniques and can be applied to give a better conducted in addition to those determined by the CCD, overall understanding with a minimal number of consisting of combinations of experimental parameters not experiments. The experimental data were processed using found in the training data set for the models. The MINITAB 16 Statistical Software. The predicted prediction abilities of the newly constructed ANN and percentage of the removal Cu(II) ions is explained by the RSM models were also statistically measured in terms of 2 following quadratic equation: R , AAD and RMSE. Y(%) = + x + x + x x + k k 2 (3) β0 ∑i=1 βi i ∑i=1 βii i ∑i

Civil Engineering and Architecture 7(6A): 58-70, 2019 61

reported range of surface area for CNWs. TEM was used to spiked with Cu(II)). identify individual whiskers, which enabled the determination of their size and shape. The whiskers were Table 1. CCD Experimental ranges and levels of independent variables Independent Factor measured to be 134.4 ± 51.2 nm and 9.0 ± 2.3 nm in length Range and level and width respectively. TEM images confirmed that the variable code oxidised CNW maintain their initial morphological -α -1 0 +1 +α integrity after the oxidation process. pH Y1 5.0 5.6 6.5 7.4 8.0 Sorbent dosage Y 0.5 2.34 5.25 8.16 10.0 (g/L) 2 3.2. Benchmarking the Wastewater Matrix from the Initial Pollutant and Adsorbent Perspective concentration of wastewater Y 1.00 1.78 3.00 4.23 5.00 New wastewater quality parameter (benchmark study) effluent spiked 3 has been developed to quantify the impact of wastewater with Cu(II) composition on the efficiency of Cu(II) removal by (mg/L) oxidised CNW adsorbents. The issue with previous studies 3.3.2. Analysis of Variance (ANOVA) Using RSM Model is the assumption that the actual wastewater composition is the same for each experiment, or has no influence on the Results for the percentage of Cu(II) removal from removal capability of that process. No work to date has wastewater effluent spiked with Cu(II) were obtained by performed benchmark experiments on each fresh performing batch experiments according to the CCD wastewater sample to challenge that assumption. matrix of conditions. The experimental results obtained The results showed that the percentage Cu(II) removal from the various runs, together with the values predicted by by oxidised CNW adsorbents varies for each wastewater the built RSM model, with residual values in the range of sample on different sampling dates. Results over 6 weeks 0.21 to 4.8, which influenced the value of R2. give an average 77.35% ± 4.15. This demonstrates that the In order to test the suitability of the model, the predicted complexity of wastewater, in terms of its composition and and actual experimental values were plotted to provide the variability, affects the capability of the adsorbent to coefficient of determination (R2 = 0.9409). The R2 value in remediate Cu(II) from the wastewater matrix. Moreover, this study was low compared to other studies that used a the complexity and variation of wastewater composition clean water matrix for the adsorption process [23]. For may also affect the accuracy and efficiency of example, the study of a cellulose-based adsorbent for mathematical modelling in predicting the capability of this chromium removal from a clean water matrix, for instance, adsorbent to remediate spiked copper from the wastewater showed a high coefficient of determination (R2 = 0.9959) effluent. [24]. Analysis of variance (ANOVA) for Cu(II) removal from 3.3. Mathematical Modelling of Spiked Cu(II) Removal the effluent was applied to evaluate the quality of fit of the from Wastewater Effluent model. The significance of each term in the equation to the percentage of the adsorbed Cu(II) ions was validated by 3.3.1. Central Composite Design (CCD) this statistical test. The results of the second-order response The pH, sorbent dosage, and initial concentration of surface model fitting in the form of ANOVA are shown in wastewater effluent spiked with Cu(II) were used as Table 2. independent (input) variables and were studied for their Generally, it can be considered that higher Fisher’s impact on the removal of spiked Cu(II) from wastewater F-test values and lower P values indicate the significance effluent. The range of independent variables, with the of the coefficients of the parameters. Values of P that are levels of the experimental factors, is given in Table 1. The greater than 0.10 indicate that the model terms are not final equation in terms of coded factors obtained by the significant [25]. As seen from Table 2, all the first-order application of RSM is given by: main effects in the quadratic model are statistically significant (P<0.05) for their effect on the Cu(II) Z (%) = 81.4753 + 1.5759Y + 4.1651Y + percentage removal from the effluent. However, the 7.1588Y + 0.9451Y 2.3417Y 2.3848Y 1 2 second-order effect of pH ( ) on the Cu(II) percentage 0.3212Y Y 0.30872 Y Y 1.21872 Y Y 2(6) 3 1 2 3 removal is not significant among2 the other second-order − − − 푌1 Where Z is the response1 2 − variable1 2(percentage− removal2 3 of effects. Meanwhile, other variables such as Y1Y2, Y1Y3, copper from wastewater effluent) and Y1–Y3 are the and Y2Y3 also had non-significant effects on the Cu(II) uncoded values of the independent variables (pH, sorbent percentage removal (P>0.10). dosage, and initial concentration of wastewater effluent

62 Predicting the Capability of Carboxylated Cellulose Nanowhiskers for the Remediation of Copper from Wastewater Effluent Using Statistical Approach

Table 2. Analysis of variance (ANOVA) of Cu(II) removal prediction by using RSM model P-value Source Sum of squares DF Mean square F-value Co-efficient Prob > F Model 1120.08 9 124.45 22.72 <0.0001 81.4753

Y1 33.11 1 33.11 6.05 0.036 1.5759

Y2 231.31 1 231.30 42.23 <0.0001 4.1651

Y3 683.31 1 683.31 124.76 <0.0001 7.1588 2 Y1 21.18 1 11.80 2.15 0.176 0.9451 2 Y2 62.59 1 72.42 13.22 0.005 -2.3414 2 Y3 75.11 1 75.11 13.71 0.005 -2.3848

Y1Y2 0.83 1 0.83 0.15 0.707 -0.3212

Y1Y3 0.76 1 0.76 0.14 0.718 -0.3087

Y2Y3 11.88 1 11.88 2.17 0.175 -1.2187 Residual 49.29 9 5.447 Lack of fit 45.74 5 9.148 10.3 0.021 Significant Pure error 3.55 4 0.888 Total 1190.4 19 *DF- degree of freedom In order to improve the accuracy of the model, the experimental design have been increasingly applied in the insignificant terms were removed from the quadratic area of water and wastewater treatment, the operation of a equation. However, there was no improvement to the WWTP is often complicated because of the complexity of accuracy of the model, even after eliminating the the wastewater matrix; this varies both temporally and insignificant terms. The significance of lack of fit indicates spatially. A study by Ebrahimzadeh et al., (2012) showed that the RSM model is invalid for the present work when it good agreement between ANN predictions and has a value of less than 0.05 [26, 27]. Therefore, from the experimental data, with a correlation coefficient of 0.9945, results, the lack of fit obtained is significant due to low whereas this amount decreases to 0.8857 for an RSM probability (P=0.005) and a higher F-test value of 15.31, model. However, there was no obvious reported which is reinforced by the relatively low coefficient of improvement in the determination of metal ions from an determination (R2 = 0.9409) for the overall model. industrial sample, even after applying the optimum Therefore, this result shows that the RSM model is unable conditions suggested by both models [28]. to effectively predict the removal of spiked Cu(II) from Of these two models, ANN is found to be more efficient wastewater effluent. and more suitable for modelling such WWTP processes A possible explanation for the poor fit of the quadratic due to its accuracy and adequacy, and is promising in equation in the RSM model is the variability of wastewater engineering applications [20]. ANN is more appropriate in composition, which is likely to influence the efficiency of the case of complex processes (i.e. WWTP processes) as the adsorption process. Similar observations were found in the model allows for predictions of the output on the basis studies on lead removal from industrial sludge leachate of input data without the need to define the relationship using red mud adsorbent, and in the case of solid-phase between them [20]. extraction of gold from industrial wastewater using A number of high quality reviews have appeared in the modified mesoporous silica. They showed that the literature dealing with the application of ANN-based variability in real wastewater samples cannot be efficiently models in the field of water treatment [29, 30]. However, predicted by the RSM model [22, 28]. no work to date has included the variation in wastewater composition as one of the independent variables (ANN 3.3.3. Artificial Neural Network (ANN) input) for remediation of Cu(II) from wastewater. For An ANN-based model was also built for predicting the example, a study by Aber et al., (2009) observed that the removal of Cu(II) from the effluent by oxidised CNWs. In performance of electrocoagulation processes in removal of a similar way to RSM modelling, the data generated Cr(VI) from synthetic and real wastewater was through CCD were used to determine the optimal successfully predicted by an ANN model [31]. Even architecture of the ANN model. The total of 20 though this study included effective parameters as experiments were divided into three subsets comprising independent variables, the process is often complicated due training (12 data points), validation (4 data points) and to the variety of contamination present in raw wastewater. testing (4 data points). The issue with these studies is the assumption that the Although ANN and RSM models in combination with actual wastewater composition is the same for each

Civil Engineering and Architecture 7(6A): 58-70, 2019 63

experiment, or that its composition has no influence on the copper ions for active sites on the oxidised CNW surface removal capability of that process. Therefore, in order to [32]. However, pH did not significantly affect the study the effect of variation in wastewater composition adsorption removal, contrary to the findings of other through removal of spiked Cu(II) from wastewater effluent, studies, as the pH range studied in this work is narrow (pH benchmark experiments were conducted for each sampling 5.0–8.0). The effect of pH on the adsorption of Cr(VI) was trip. The results obtained from the benchmark experiments investigated by varying pH from 2.0 to 10.0. Due to the showed that the percentage removals (74.41 – 78.76%) wide pH range, pH was found to be one of the main were different from each sampling trip, due to complexity parameters affecting the ad-sorption process [33]. of the wastewater matrix and its variability. Thus, it can be Figure 1(B) shows the interaction effect of sorbent summarised that the variation of actual wastewater dosage and initial concentration of wastewater effluent composition affects adsorption performance. The results spiked with Cu(II) on the removal of Cu(II) from the obtained from benchmark experiments will therefore be effluent, with pH held constant at pH 6.5. The Cu(II) included as the fourth independent variable in ANN removal increases with increasing sorbent dosage, which modelling. The coefficient of determination (R2 =0.9963) may be due to the increase in total surface active sites on for the ANN model with variability indicates good the adsorbent surface. The amount of proton exchange agreement between experimental and predicted results. between the adsorbent and the solutions increases with For a better graphical interpretation of the process of increasing sorbent dosage. Similar observations were Cu(II) adsorption from wastewater, three-dimensional found in studies on Cu(II) removal using alkali-modified response surface plots were generated. The mutual spent tea, and in the case of Cr(VI) removal by modified interactive behaviour between two in-dependent variables, silica [25]. Percentage removal of Cu(II) increases when while the third variable is held constant at its intermediate pH and initial concentration of wastewater effluent spiked value (pH 6.5, 5.25 g/L, 3 mg/L), is shown in Figure 1. with Cu(II) increase, as shown in Figure 1(C). The As shown in Figure 1(A), maximum removal of Cu(II) is increasing initial concentration of wastewater effluent observed at a sorbent dosage of 8 g/L and pH 8. The spiked with Cu(II) provides the driving force to overcome percentage Cu(II) removal increased with the increase of the mass transfer resistance of Cu(II) ions between the pH due to the negative surface charge of oxidised CNW at aqueous and solid phases. Similar observations were alkaline pH values. At pH 8 and above, carboxyl group, – reported in the literature, where the maximum Cu(II) COOH, was changed into –COO-, hence the ion exchange removal by Trametes versicolor fungi was observed when between Cu(II) and potential functional groups increased initial Cu(II) concentration increased from 37 to 60 mg/L at [12]. The low adsorption that takes place in acidic solutions pH of 5.51 [34]. can be due to the competition between hydrogen and

64 Predicting the Capability of Carboxylated Cellulose Nanowhiskers for the Remediation of Copper from Wastewater Effluent Using Statistical Approach

Figure 1. Surface plots (left) and corresponding contour plots (right) showing the effects of adsorption parameters on the spiked Cu (II) removal from wastewater effluent as predicted by the ANN model with initial concentration of wastewater effluent spiked with Cu(II) held constant 3 mg/L (A), pH held constant at 6.5 (B) and sorbent dosage held constant at 5.25 g/L (C)

Civil Engineering and Architecture 7(6A): 58-70, 2019 65

3.4. Multiple Linear Regression (MLR) In addition to the coefficients of determination for ANNs, the AAD and RMSE confirm that the ANN model Multiple linear regression (MLR) is a linear statistical including the variability of the wastewater matrix as the analysis that is applicable for predicting the relationship fourth independent variable is superior in predicting the between a dependent variable and two or more removal of spiked Cu(II) from wastewater effluent. independent variables [35]. In MLR, the dependent A possible explanation for this result is the complexity variable is known as the predictand, while the independent of the wastewater in term of composition and its variability, variables are the predictors [36]. MLR models are used in which can affect the capability of the adsorbent to the prediction of Cu(II) removal from the wastewater remediate Cu(II) from the wastewater matrix. A variety of matrix, being represented by the relationship between the organic and inorganic compounds can be found in the percentage removal and a set of predictor variables. MLR composition of wastewater, and its variability, both is based on least squares fit, where the model is adjusted temporally and spatially (within a WWTP and in different so that the sum of squares of differences of actual and wastewater streams), is likely to influence the efficiency predicted values is minimised. The general MLR equation and capability of oxidised CNW adsorbents. This can be formulated by: explanation is stated that in a real WWTP, the water matrix Ÿ = + X + + X + (7) will be far more complex than clean water. Thus, with the Where Ÿ is the dependent variable, X the independent results obtained from benchmark experiments, it is β0 β1 1 ⋯ βn n n ε variables, n the predicted parameters, and is the error demonstrated that the complexity and variability in term. wastewater composition affect the adsorption performance. Multiple훽 linear regression (MLR) and artificial휀 neural This finding has also been supported by other studies networks (ANNs) were used to predict the removal of that have not included the variations in real wastewater Cu(II) from the effluent by oxidised CNW adsorbents. composition as one of the independent variables (ANN The data used in the MLR and ANN models were input). For instance, although Geyikci and his co-workers obtained from 20 CCD experiments. The MLR model reported that the results of ANN were found to be more (with and without accounting for the variability of the reliable than RSM (R2 = 0.672), a low coefficient of wastewater matrix) gives the mathematical expression of determination (R2 = 0.898) from the ANN model indicated the output of the MLR analysis: that the variation in industrial sludge leachate composition Z(%) = 17.0722 + 1.7352Y + 1.6348Y + 5.7487Y + had an influence on the removal capability of the adsorbent. 0.3218Y (8) The major issue with this study is the assumption that real 1 2 3 wastewater composition is the same for each sample, or has Z(%) = 42.4169 + 1.7352Y + 14.4315Y + 5.8445Y no impact on the removal capability of the adsorbent. (9) 1 2 3 Hence, it can be concluded that including wastewater Where Z is the dependent variable (percentage removal variability as one of the input variables will lead to of copper from wastewater effluent) and Yi the improvements in the predictability of the ANN model. independent variables (pH, sorbent dosage, initial 3.5.1. Model Validation Using Unseen Experiments concentration of wastewater effluent spiked with Cu(II) and benchmark experiment, respectively). For validation, additional 10 unseen experiments were conducted, consisting of combinations of experimental 3.5. Statistical Comparison and Performance of factors that were not considered in the 20 CCD Models for Wastewater Effluent experiments. This was a necessary procedure, since no work has been reported to date in the choice of additional The performance of the built MLR and ANN models experiments to represent the whole system of remediating (with and without accounting for the variability of the Cu(II) from a wastewater matrix. 3D scatter plots for the wastewater matrix) was compared and statistically unseen experiments are displayed in Figure 2, and include a measured by the coefficient of determination (R2), absolute comparison of 3D scatter plots derived from data contained average deviation (AAD), and root mean squared error in two other studies which involved real wastewater (RMSE). samples. The 10 unseen experiments undertaken in this The ANN including variability of wastewater matrix study, illustrated in Figure 2(A), were chosen to represent model fitted the experimental data with excellent accuracy parameter space both inside and outside the system, to and with a better prediction (R2 = 0.9963) than the ANN better understand and test the validity of the models. not including variability (R2 = 0.9945), and than the MLR However, as illustrated in Figure 2(B), Ebrahimzadeh et model including (R2 = 0.7994) and not including al., (2012) designed 10 random experiments, using variability (R2 = 0.7961). The AAD and RMSE for the MATLAB programming, which did not represent the ANN model including variability of wastewater matrix whole system, in order to study the predictability of the were calculated to be 0.30 % and 0.48 respectively, whilst RSM and ANN models for solid-phase extraction of gold those of the ANN model were 0.63 % and 0.69 respectively. ions from industrial wastewater [28]. Furthermore, in order

66 Predicting the Capability of Carboxylated Cellulose Nanowhiskers for the Remediation of Copper from Wastewater Effluent Using Statistical Approach to test the validity of RSM and ANN results, Geyikci et al., concentrated inside the system, as shown in Figure 2(C) (2012) conducted 10 extra experiments that were all [22].

Figure 2. 3D scatter plots showing: (A) comparison of the CCD with unseen experiments within the systems for this work; (B) solid-phase extraction for determination of gold from industrial wastewater [28]; (C) Lead adsorption from industrial sludge leachate [22]

Civil Engineering and Architecture 7(6A): 58-70, 2019 67

Table 3. Validation data for 10 unseen experiments

Inputs Cu(II) removal (%) MLR ANN Data index Run MLR ANN Y1 Y2 Y3 Benchmarks (%) Actual (WW variation) (WW variability) Predicted Residual Predicted Residual Predicted Residual Predicted Residual 1 7 1 3 71.96 74.49 73.53 -0.96 71.26 -3.23 72.98 1.51 77.61 -3.12 2 6 1 2 71.96 65.81 65.95 0.14 63.77 -2.04 60.68 5.13 64.98 0.83 Outside of the system 3 5 8 3 77.39 76.88 80.08 3.20 77.52 0.64 84.14 -7.26 81.92 -5.04 4 7 10 4 77.39 78.72 92.26 13.54 90.01 11.29 87.54 -8.82 86.28 -7.56 5 5 5 2 78.76 68.8 69.94 1.14 67.41 -1.39 71.25 -2.45 72.81 -4.01 6 6 4 4 78.76 77.64 71.67 -4.72 70.04 -6.35 69.93 6.46 74.31 2.08 7 6 5 2 74.41 72.50 83.99 6.09 81.91 4.01 84.92 -7.02 77.07 0.83 Inside of the system 8 5.5 4 4.5 74.41 77.9 75.77 1.38 74.15 -0.24 80.19 -5.80 80.01 -5.62 9 7.5 4 2.5 78.76 74.39 84.98 6.05 82.62 3.69 84.85 -5.92 86.01 -7.08 10 7 9 3 74.15 78.93 79.82 1.36 77.39 -1.07 82.29 -3.83 82.04 -3.58

Y1=pH; Y2=sorbent dosage; Y3=initial concentration of wastewater effluent spiked with Cu(II); WW=wastewater

Table 4. Comparison of the predictive abilities of RSM and ANN model

AAD (%) RMSE MLR ANN MLR ANN Data index MLR ANN MLR ANN (WW variability) (WW variability) (WW variability) (WW variability) 20 CCD 3.23 3.84 0.63 0.30 3.48 3.46 0.69 0.48 10 unseen 4.98 4.38 7.17 5.23 5.43 4.64 5.83 4.57 5 Inside 5.05 3.84 7.53 4.98 4.46 4.46 5.90 4.46 5 Outside 4.90 4.93 6.81 5.49 6.26 5.37 5.75 4.67 AAD= absolute average deviation; RMSE= root mean squared error; WW=wastewater

68 Predicting the Capability of Carboxylated Cellulose Nanowhiskers for the Remediation of Copper from Wastewater Effluent Using Statistical Approach

The actual and predicted values of the responses for the (RMSE) when predicting the efficiency of Cu(II) removal 10 unseen experiments, along with their residual values for from the wastewater matrix. The optimum adsorption the models, are summarised in Table 3. Moreover, the R2 conditions were determined as an initial pH value of 8.0, a for both models (R2 = 0.9644 for ANN including sorbent dosage of 6.45 g/L and initial Cu(II) concentration wastewater variability, R2 = 0.8991 for ANN without), of 4.72 mg/L. At optimum adsorption conditions, the show that the ANN model predicts more accurately when percentage removal of spiked Cu(II) from the wastewater variation in wastewater composition is included as the effluent was found to be 92.11%. Although oxidised CNW fourth independent variable. As shown in Table 4, the adsorbents were able to remove approximately 90% of predictive abilities of the newly constructed MLR and spiked Cu(II) from wastewater effluent, the physical ANN models, with and without wastewater variability, structure of oxidised CNW adsorbents are not suitable for were statistically measured in terms of R2, AAD and use in continuous flow column operations. RMSE. From the results, it is confirmed that the ANN model including wastewater variability predicts more accurately the remediation of spiked Cu(II) from Acknowledgements wastewater effluent, in both the original 20 CCD and the 10 unseen experiments. This is because the ANN model The author (Hazren Hamid) would like to acknowledge allows for predicting the response (percentage of Cu(II) the support from University of Tun Hussein Onn Malaysia removal) without the need to justify the relationship (UTHM) for financial support under Grant Tier 1 (Code between them, which is particularly important in the case Grant: H200) and Government of Malaysia for a of real-world WWTP, where the water matrix will be more scholarship from the Majlis Amanah Rakyat (MARA). complex [20].

This finding has also been supported by others who have used MLR and ANN in prediction studies. For instance, Tiryaki and his co-workers used ANN and MLR for predicting the compression strength of heat-treated woods. REFERENCES The results indicated that an ANN model provided better [1] Hanafiah, M.A.K.M., et al., Enhanced adsorption of Pb(II) prediction results compared to an MLR model. Moreover, on chemically treated neem (Azadirachta indica) leaf ANN models save time and decrease the experimental powder, in Material Science and Engineering Technology costs [35]. Another advantage of the ANN model is the Ii, K.M. Gupta, Editor. 2014. p. 128-133. flexibility to work with more input variables, which is [2] Ileri, O., et al., Removal of common heavy metals from helpful when involving large number of experiments; for aque-ous solutions by waste salvadora persica L. Branches MLR, a large number of input variables lead to a (Mi-swak). International Journal of Environmental polynomial with many coefficients that involve tedious Research, 2014. 8(4): p. 987-996. computation [36]. [3] Turan, N.G., B. Mesci, and O. Ozgonenel, Response surface modeling of Cu(II) removal from electroplating waste by ad-sorption: application of BoxBehnken experimental design. Clean-Soil Air Water, 2013. 41(3): p. 4. Conclusions 304-312. Oxidised CNW adsorbents are capable of removing [4] Bingol, D., et al., Comparison of the results of response spiked Cu(II) ions from wastewater effluent. The RSM and sur-face methodology and artificial neural network for the ANN models were employed to optimise the system and to bio-sorption of lead using black cumin. Bioresource create a good predictive model. No work in the reviewed Technology, 2012. 112: p. 111-115. literature included matrix complexity and the variability of [5] Shanmugaprakash, M. and V. Sivakumar, Development of the wastewater as one of the independent variables in ANN experimental design approach and ANN-based models for modelling. Evidently this novel approach and the outcomes de-termination of Cr(VI) ions uptake rate from aqueous were employed in this study for the first time, as most solution onto the solid biodiesel waste residue. studies do not consider matrix variability and its impact Bioresource Technolo-gy, 2013. 148: p. 550-559. when evaluating the efficiency of an adsorbent. To test the [6] Li, L., et al., Optimization of methyl orange removal from predictive capability of these models, additional 10 unseen aqueous solution by response surface methodology using experiments, not used in developing both models, were spent tea leaves as adsorbent. Frontiers of Environmental Sci-ence & Engineering, 2014. 8(4): p. 496-502. chosen to represent the system of conditions both inside and outside the system. This study (20 CCD and 10 unseen [7] Mandal, S., et al., Artificial neural network modelling of experiments) found that the ANN model accounting for As(III) removal from water by novel hybrid material. wastewater variability was superior to the RSM model and Process Safety and Environmental Protection, 2015. 93: p. 249-264. to the ANN model not including wastewater variability, in 2 terms of the coefficient of determination (R ), the absolute [8] CEC, Council directive on pollution caused by certain average deviation (AAD) and root mean squared error danger-ous substances discharged into the aquatic

Civil Engineering and Architecture 7(6A): 58-70, 2019 69

environment of the Community. 1976, Council of the [22] Geyikci, F., et al., Modelling of lead adsorption from European Communities. industri-al sludge leachate on red mud by using RSM and ANN. Chem-ical Engineering Journal, 2012. 183: p. [9] CEC, Directive 2000/60/EC of the European Parliament 53-59. and of the Council on establishing a framework for Community ac-tion in the field of water policy. 2000, [23] Hamid, H.A., et al., Predicting the capability of Council of the Europe-an Communities. carboxylated cellulose nanowhiskers for the remediation of copper from water using response surface methodology [10] Isobe, N., et al., TEMPO-oxidized cellulose hydrogel as a (RSM) and artifi-cial neural network (ANN) models. high-capacity and reusable heavy metal ion adsorbent. Industrial Crops and Products. Journal of hazardous materials, 2013. 260: p. 195-201. [24] Liu, J., et al., Study of glutamate-modified cellulose beads [11] Alves, G., V. Leandro, and L.F. Gil, Adsorption of Cu(II), for Cr(III) adsorption by response surface methodology. Cd(II), and Pb(II) from aqueous single metal solutions by Indus-trial & Engineering Chemistry Research, 2011. succinylated mercerized cellulose modified with 50(18): p. 10784-10791. triethylenetet-ramine. Carbohydrate Polymers, 2009. 77(1): p. 142-149. [25] Cao, J., et al., Response surface methodology approach for optimization of the removal of chromium(VI) by [12] Reddy, D.H.K., Seshaiah, K.,Reddy, A. V. R.,Lee, S. M., NH2-MCM-41. Journal of the Taiwan Institute of Op-timization of Cd(II), Cu(II) and Ni(II) biosorption by Chemical Engineers, 2014. 45(3): p. 860-868. chemi-cally modified moringa oleifera leaves powder. Carbohydrate Polymers, 2012. 88(3): p. 1077-1086. [26] Zulkali, M.M.D., A.L. Ahmad, and N.H. Norulakmal, Oryza sativa L. husk as heavy metal adsorbent: [13] Eyley, S. and W. Thielemans, Surface modification of optimization with lead as model solution. Bioresource cellu-lose nanocrystals. Nanoscale, 2014. 6(14): p. Technology, 2006. 97(1): p. 21-25. 7764-7779. [27] Hamsaveni, D.R., S.G. Prapulla, and S. Divakar, Response [14] Saito, T., et al., Distribution of carboxylate groups surface methodological approach for the synthesis of introduced into cotton linters by the TEMPO-mediated isobutyl isobutyrate. Process Biochemistry, 2001. 36(11): oxidation. Carbo-hydrate Polymers, 2005. 61(4): p. p. 1103-1109. 414-419. [28] Ebrahimzadeh, H., et al., Optimization of solid-phase [15] Ghosh, A., P. Das, and K. Sinha, Modeling of biosorption extrac-tion using artificial neural networks and response of Cu(II) by alkali-modified spent tea leaves using surface methodology in combination with experimental response sur-face methodology (RSM) and artificial neural design for de-termination of gold by atomic absorption network (ANN). Applied Water Science, 2015. 5(2): p. spectrometry in in-dustrial wastewater samples. Talanta, 191-199. 2012. 97: p. 211-217. [16] Labet, M. and W. Thielemans, Improving the reproducibility of chemical reactions on the surface of [29] Hamed, M.M., M.G. Khalafallah, and E.A. Hassanien, cellulose nanocrystals: ROP of epsilon-caprolactone as a Predic-tion of wastewater treatment plant performance case study. Cellulose, 2011. 18(3): p. 607-617. using artificial neural networks. Environmental Modelling & Software, 2004. 19(10): p. 919-928. [17] Habibi, Y., H. Chanzy, and M.R. Vignon, TEMPO-mediated surface oxidation of cellulose whiskers. [30] Antonopoulou, M., V. Papadopoulos, and I. Konstantinou, Cellulose, 2006. 13(6): p. 679-687. Photocatalytic oxidation of treated municipal wastewaters for the removal of phenolic compounds: optimization and [18] APHA, Standard Methods for the Examination of Water model-ing using response surface methodology (RSM) and and Wastewater. 1998, American Public Health artificial neural networks (ANNs). Journal of Chemical Association: American Public Health Association: Technology & Biotechnology, 2012. 87(10): p. 1385-1395. Washington, D.C. [31] Aber, S., A.R. Amani-Ghadim, and V. Mirzajani, Removal [19] Ghosh, A., K. Sinha, and P. Das Saha, Central composite of Cr(VI) from polluted solutions by electrocoagulation: de-sign optimization and artificial neural network model-ing of experimental results using artificial neural modeling of copper removal by chemically modified network. Journal of Hazardous materials, 2009. 171(1–3): orange peel. Desali-nation and Water Treatment, 2013. p. 484-490. 51(40-42): p. 7791-7799. [32] Rajemahadik, C.F., S.V. Kulkarni, and D.G.S. Kulkarni, [20] Witek-Krowiak, A., et al., Application of response surface Effi-cient removal of heavy metals from electroplating methodology and artificial neural network methods in wastewater using electrocoagulation. International Journal model-ling and optimization of biosorption process. of Scientific and Research Publications, 2013. 3(10). Bioresource Technology, 2014. 160: p. 150-160. [33] Mohan, S., et al., Synthesis of CuO nanoparticles through [21] Pilkington, J.L., C. Preston, and R.L. Gomes, Comparison green route using citrus limon juice and its application as of response surface methodology (RSM) and artificial na-nosorbent for Cr(VI) remediation: process optimization neural networks (ANN) towards efficient extraction of with RSM and ANN-GA based model. Process Safety and artemisinin from Artemisia annua. Industrial Crops and Envi-ronmental Protection, 2015. 96: p. 156-166. Products, 2014. 58: p. 15-24.

70 Predicting the Capability of Carboxylated Cellulose Nanowhiskers for the Remediation of Copper from Wastewater Effluent Using Statistical Approach

[34] Sahan, T., et al., Optimization of removal conditions of cop-per ions from aqueous solutions by Trametes versicolor. Bio-resource Technology, 2010. 101(12): p. 4520-4526. [35] Tiryaki, S. and A. Aydın, An artificial neural network model for predicting compression strength of heat treated woods and comparison with a multiple linear regression model. Construc-tion and Building Materials, 2014. 62: p. 102-108. [36] Arulsudar, N., N. Subramanian, and R.S.R. Muthy, Compari-son of Artificial Neural Network and Multiple Linear Regres-sion in the Optimization of Formulation Parameters of Leuprolide Acetate Loaded Liposomes. Journal of Pharmacy and Pharmaceutical Sciences, 2005. 8(2): p. 243-258.

Civil Engineering and Architecture 7(6A): 71-76, 2019 http://www.hrpub.org DOI: 10.13189/cea.2019.071408

Determining the Chaotic Dynamics of Hydrological Data in Flood-Prone Area

Adib Mashuri, Nur Hamiza Adenan*, Nor Zila Abd Hamid

Department of Mathematics, Universiti Pendidikan Sultan Idris, Malaysia

Received July 30, 2019; Revised October 4, 2019; Accepted December 15, 2019

Abstract Flood-prone areas are associated with However, this study only chooses Cao method [4] to hydrological time series data such as rainfall, water level identify the dynamics of all the hydrological data involved. and river flow. The possibility to predict flood is to relate This is because the Cao method has successfully identified all the three data involved. However, in order to develop a the presence of chaotic dynamics for hydrological data [5], multivariable prediction model based on chaos approach, [6]. In this study, the presence of the most rapidly detected each datum needs to identify chaotic dynamics. As such, dynamics will be identified by comparing the results of the the Sungai Galas, Dabong in Kelantan, Malaysia which is a study on the analysis using the Cao method to the flood disaster area has been selected for the analysis. hydrological data involved. Additionally, the study on this Rainfall, water level and river flow data in this area were chaotic dynamics has recently been implemented in collected to be analysed using the Cao method to identify Malaysia. This research only involves river flow [1], the presence of chaotic dynamics. The hydrological data is ozone[7], suspended particles [8] and rainfall [9] uncertain, which is difficult to predict because the data independently. However, for the next study, the only involved is located in the area of flood disaster. The chaotic dynamically identifiable data will be used in further analysis showed the presence of chaotic dynamics on studies to develop a multivariable prediction model based rainfall, water level and river flow data in the Sungai Galas on a gradual approach. The construction of this which involved uncertain data located in flood affected multivariable prediction model is to predict flood in area of areas by using Cao method. Therefore, a multivariable Sungai Galas. flood prediction model can be implemented using a chaos approach. 2. Data Keywords Cao Method, Chaos Approach, Chaotic Dynamic, Flood Area, Hydrological Time Series Data Rainfall, water level and river flow time series that are hydrological data from Sungai Galas are used in the analysis. The area has been selected for this study because the area was involved during the worst flood disaster in 2014 [10]. Time series data were taken from 2 stations 1. Introduction which were (1) station 5320038 (rainfall observation station) and (2) station 5320443 (water level and river flow This study is to identify the presence of chaotic station). All data were obtained from the Department of dynamics of hydrological data located in flood prone areas Irrigation and Drainage Malaysia (DID). There were 2068 for the purpose of constructing multivariable flood data being used for analysis, taken from February 1, 2005 prediction model based on a chaos approach. The to September 30, 2010. Table 1 sows the percentage of hydrological data involved are rainfall time series data, missing data. The missing data were determined by using water level time series data and river flow time series data the mean. at Sungai Galas, Dabong in Kelantan, Malaysia. This area is a region where flood happens that causes uncertain Table 1. Percentage of Missing Data hydrological data. The uncertain data refer to the data that Time Series Data Percentage of Missing Data are fluctuating as seen in Fig. 1. Rainfall 0.87% There are several methods that can help in identifying the presence of chaotic dynamics i.e.; phase space plot Water level 0.00% method [1], Cao method [2] and exponent Lyapunov [3]. River flow 5.66% 72 Determining the Chaotic Dynamics of Hydrological Data in Flood-Prone Area

Fig. 1 shows the distribution of data involved in this 140 analysis. Referring to Fig. 1 (a), the maximum value of the 120 rain was 140mm and there was also non-rainy day. Meanwhile, Fig. 1 (b) shows that the water level data had a 100 value of over 36m for four times over the period of 2005 to 80 2010. According to DID, the Sungai Galas level is classified as danger level when it reaches 38m. Next, Fig. 1 60

Rainfall (mm) Rainfall (c) shows the flow of Sungai Galas which had the same 40 fluctuation as water level data. If viewed more closely, the river flow reached over 1500 m3s-1. Hence, uncertain data 20 are involved which are referred to the high and low peaks 0 in each hydrological datum involved. 0 500 1000 1500 2000 Days The dynamics of the data were divided into two types; namely, deterministic and random. The choatic dynamics (a) data were located between the deterministic and random [11]. The deterministic data moved from the starting point 38 and were in accordance to the rules. Meanwhile, the choatic dynamics data were seen as being random but 36 actually in accordance to the rules. It can be noted here that 34 the hydrological data involved looked like random but analysis was needed to identify the presence of chaotic 32 dynamics on the hydrological data involved.

30 Water Level (m) Level Water 28 3. Methodology 26 The Cao method [4] was chosen to analyze the presence 24 of chaotic dynamic on the hydrological time series data 0 500 1000 1500 2000 involved. There are two advantage methods compared to Days other methods; (1) this method does not depend on the (b) number of parameters required except the delay time () and (2) this method does not depend on the number of data. There are two parameters involved in this method; Qd1( ) 2500 and Qd2( ). Both parameters need to be plotted together with embedding dimension ()d to identify the presence 2000 of chaotic dynamics. There are two indications of chaotic 1500 dynamics available for the analysis of hydrological data involved using this method. The first indicator is that if is saturated when embedding dimension is River FlowRiver (m3/s) 1000 increased, then the time series are chaotic dynamics. 500 Whereas, chaotic dynamic is present if Qd2( ) 1 is present for any embedding dimension , which is the 0 500 1000 1500 2000 Days second indicator. The basis for the use of this Cao method is the phase space reconstruction. The phase space (c) reconstruction involves each time series datum to be used. Figure 1. Hydrological data at Sungai Galas at Dabong, Kelantan: (a) All hidrological data involved need to be reconstructed into rainfall, (b) water level and (c) river flow d  dimensional. Assuming the data is:

Civil Engineering and Architecture 7(6A): 71-76, 2019 73

. (1) shows the result of AMI calculation for the determination x12, x ,..., xN . AMI's first minimum value is the optimal value for x1 refers to the first data while x2 will be the second Referring to Table 2, the optimum values for rainfall data. Next xN is the data that refer to the last number of data and water levels data are 10 with each having the first data N. Subsequently, the data is reconstructed into d  minimum values of AMI at 3.897 and 3.347. Meanwhile, dimensional as in the following equation: AMI's first minimum value of river flow data is 3.419 with   7 All these values are used for the phase space Y( d ) x , x ,..., x , i  1,2,..., N  d  1  (2) reconstruction as in equation (2). Furthermore, the result of i i i id 1    phase space reconstruction for each hydrological time where d is embedding dimension and  is time delay. series data will be used to calculate the parameters The value of varies while the value is from the and Qd2( ) . calculation of Average Mutual Information (AMI) [12]. Table 2. Average Mutual Information (AMI) Ydi () is the result of the reconstruction of the vector i th for each embedding dimension. After the phase sapce Average Mutual Information (AMI) reconstruction is performed, the value Qd1( ) is calculated Time Time Series Data by using the equation: delay () Rainfall Water level River flow Qd( 1) Qd1( )  (3) 1 4.263 3.811 3.850 Qd() , and 2 4.136 3.586 3.633

Nd  Y d11  Y d  1 i   j i, d    3 4.085 3.488 3.515 Qd()  (4) Nd  n1 Yi  d  Yj i, d   d  4 4.015 3.467 3.500 5 3.988 3.447 3.473 with ||•|| is Eucledian distance between Ydj i, d   1 and 6 3.958 3.420 3.436 the nearest neighbour for Ydi  . Ydi  1 is the result of 7 3.907 3.389 3.419 the phase space reconstruction of vector for d 1 dimension with : 8 3.904 3.350 3.429 9 3.901 3.349 3.384 Y d1 x , x ,..., x i   i i id   . (5) 10 3.897 3.347 3.369 where j(,) i d has value1j ( i , d )  N  d . In identifying 11 3.923 3.354 3.371 the presence of chaotic dynamics, the graph against 12 3.873 3.321 3.337 d needs to be plotted in order to determine chaotic 13 3.880 3.304 3.347 dynamics. Subsequently, computations of Qd2( ) are 14 3.884 3.282 3.300 continued in order to confirm the existence of chaos in the time series data involved. The calculations of are: 15 3.904 3.290 3.318

1 Nd  Fig. 2 shows the results for calculations for each Q  x x  id  j i, d d and (6) hydrological data used. Each calculation is plotted Nd  i1 together with the increasing value of d. Referring to the Qd  1 Qd2.   (7) figure , the value increases as the value d increases Qd   for each hydrological data involved. However, at one stage, the value will reach the saturated value. Fig. 2 (a) In order to identify the presence of chaotic dynamics, graph against needs to be plotted. shows the value of the rainfall data begins to saturate when the value d  9 which is between 0.9 to 1.0. Although the value fluctuates, it will eventually 4. Results and Discussion saturate. Whereas, Fig. 2 (b) and Fig. 2 (c) showed the Qd1( ) value starts to saturate when d  5 when The results of this study are divided into 3 parts; (1) AMI approaching the value of 1 for water level and river flow calculation result for determination of time delay  value, data. Therefore, all the hydrological data analysed are (2) result for indicator, and (3) result for Qd2( ) chaotic dynamics with all values continuing to indicator for all three hydrological data involved. Table 2 saturate with the increase of d.

74 Determining the Chaotic Dynamics of Hydrological Data in Flood-Prone Area

dynamics indication slower than water level data and river

flow by using Qd1( ). 1.6

1.4

1.2 1

Q1(d) 1 0.9

0.8 Q2(d)

0.6

0.4 1 2 3 4 5 6 7 8 9 10 Embedding dimension (d) 1 2 3 4 5 6 7 8 9 10 (a) Embedding dimension (d) (a) 1

0.9

0.8

0.7

0.6 1

Q1(d) 0.5

0.4 Q2(d)

0.3

0.2

1 2 3 4 5 6 7 8 9 10 Embedding dimension (d) 1 2 3 4 5 6 7 8 9 10 (b) Embedding dimension (d) (b) 1

0.9

0.8

0.7

0.6

1 Q1(d) 0.5

0.4 Q2(d) 0.3

0.2

1 2 3 4 5 6 7 8 9 10 Embedding dimension (d)

(c) 1 2 3 4 5 6 7 8 9 10 Embedding dimension (d) Figure 2. Qd1( ) versus embedding dimension d for (a) rainfall, (b) water level and (c) river flow. (c) Next, Fig. 3 shows the findings for Qd2( ) estimation of Figure 3. Qd2( ) versus embedding dimension for (a) rainfall, (b) water level and (c) river flow the analysed hydrological data. All values Qd2( ) 1 for each d. Hence, it is believed that the analysed Application of Cao method for rainfall time series data hydrological data in Sungai Galas are chaotic dynamics. was also conducted in Changchun, China [13]. The results Although data on the same date used the determination of showed that both graphs Qd1( ) against d and chaotic existence, the rainfall data can provide chaotic against gave the same results with the analysis of rainfall

Civil Engineering and Architecture 7(6A): 71-76, 2019 75

data at Sungai Galas, in which Qd1( ) was saturated when system. d increased and Qd2( ) 1 for at least one value of d. Thus, the Cao method has successfully identified the presence of chaotic dynamics in rainfall data. Acknowledgements Meanwhile, the graph Qd1( ) also continued to be The authors thankfully acknowledged the financial saturated with the increased value of for data analysis in support provided by the Ministry of Education Malaysia Sungai Muda [14]. Furthermore, the findings of this study (2019-0009-102-02: FRGS/1/20/2018/STG06/UPSI/02/3) also showed that river flow data was used to provide as well as Department of Irrigation and Drainage Malaysia for at least one value of . Therefore, the for providing the hydrological data. findings were in line with the analysis of chaotic dynamics in Sungai Muda [14] where the river flow was chaotic dynamics by using Qd2( ) . REFERENCES [1] N. H. Adenan, M. S. M. Noorani. Multiple Time-Scales 5. Conclusions Nonlinear Prediction of River Flow using Chaos Approach, Jurnal Teknologi, Vol. 78, No. 7, 1-7, 2016. The purpose of analyzing the presence of chaotic dynamics for hydrological time series data which involve [2] W. N. A. W. M. Zaim, N. Z. A. Hamid. Peramalan Bahan rainfall data, water level data and river flow data is to build Pencemar Ozon (O3 ) di Universiti Pendidikan Sultan Idris, a multivariable flood prediction model by using a chaos Tanjung Malim, Perak, Malaysia mengikut Monsun dengan Menggunakan Pendekatan Kalut, Sains Malaysiana, Vol. approach. Therefore, the first step before the prediction is 46, No. 12, 2523–2528, 2017. to identify the presence of chaotic existence on every hydrological data involved. [3] X. Li, J. Sha, Y. Li , Z. L Wang. Comparison of Hybrid Sungai Galas was chosen as a study area as this area was Models for Daily Streamflow Prediction in a Forested Basin, involved during the worst flood disaster in 2014. Therefore, J. Hydroinformatics, Vol. 20, No. 1, 191–205, 2018. it is seen on the data involved that it is uncertain and always [4] L. Cao. Practical Method for Determining the Minimum fluctuating. The hydrological time series data are located in Embedding Dimension of a Scalar Time Series, Phys. D the same area of Sungai Galas, Kelantan in Malaysia. Nonlinear Phenom., Vol. 110, No. 1–2, 43–50, 1997. Basically, there are two parameters from Cao methods involved which are and For parameter [5] D. X. She, X. Yang. A New Adaptive Local Linear Prediction Method and Its Application in Hydrological water level data and river flow data show quick Time Series, Math. Probl. Eng., Vol. 2010, 1–16, 2010. respond compared to rainfall data in order to determine the [6] N. H Adenan, N. Z. A. Hamid, Z. Mohamed, M. S. M. chaotic dynamics. Meanwhile, all hydrological data shows Noorani. A Pilot Study of River Flow Prediction in Urban for at least one value of d for parameter Area Based on Phase Space Reconstruction, The 24th and this result can be evidence for the presence of National Symposium on Mathematical Sciences (SKSM24), vol. 1870, 040011, 2017. chaotic dynamics for all data involved. The results for both analyses showed the presence of chaotic dynamics for [7] N. Z. A. Hamid. Application of Chaotic Approach In rainfall data, water level data and river flow data in Sungai Forecasting Highland’s Temperature Time Series, IOP Galas which involved uncertain data as the area was Conf. Ser. Earth Environ. Sci., Vol. 169, No. 1, 012107, 2018. located in flood affected areas using the Cao method. Next, with the intention of construction a multivariable flood [8] N. Z. A. Hamid, M. S. M. Noorani. Suatu kajian perintis prediction model by using a chaos approach can be menggunakan pendekatan kalut bagi pengesanan sifat dan implemented. Basically, the multivariable flood prediction peramalan siri masa kepekatan PM10, Sains Malaysiana, model is based on phase space reconstruction using all Vol. 43, No. 3, 475–481, 2014. hydrological data involved in this study. [9] M. L. Sapini, N. S. Adam, N. Ibrahim, N. Rosmen, N. M. In addition, the implication for this study could Yusof. The Presence of Chaos in Rainfall by using 0-1 Test contribute to identifying the chaotic dynamics in order to and Correlation Dimension, AIP Conference Proceedings, , understand the hydrological system in flood area especially Vol. 1905, No. 1, p. 050040, 2017. in Sungai Galas. By comparing all hydrological data [10] Bernama, Terkini: Banjir di Kelantan, Terengganu, Pahang involved, it is proved that Cao method could identify the tambah buruk, Nasional, Berita Harian, 23 Dec 2014. chaotic dynamics for uncertain data, in flood area and use daily data. Besides that, Cao method could be suggested to [11] H. D. I Abarbanel. Analysis of Observed Chaotic Data. New be used in other hydrological data such as sediment York: Springer New York, 1996. transport, evaporations and water quality with the purpose [12] N. Z. A. Hamid, N. H. Adenan, M. S. M. Noorani. of determining the chaotic dynamics in hydrological Forecasting and Analyzing High O3 Time Series in

76 Determining the Chaotic Dynamics of Hydrological Data in Flood-Prone Area

Educational Area through an Improved Chaotic Approach, AIP Conference Proceedings, Vol. 1870, 2017.

[13] W. L. U. Ouyang, X. Xin, Y. Zhang, W. Cheng, T. Yu. Monthly Rainfall Forecasting using EEMD-SVR Based on Phase-Space Reconstruction, Water Resour. Manag., Vol. 30, No. 7, 2311–2325, 2016.

[14] N. H. Adenan, M. S. M. Noorani. Peramalan Data Siri Masa Aliran Sungai di Dataran Banjir dengan Menggunakan Pendekatan Kalut. Sains Malaysiana, Vol. 44, No. 3, 463– 471, 2015.

Civil Engineering and Architecture 7(6A): 77-85, 2019 http://www.hrpub.org DOI: 10.13189/cea.2019.071409

Comparison between Multiple Gradient and Pole Dipole Array Protocols for Groundwater Exploration in Quaternary Formation

A. K. Abd Malik1, A. Madun1,*, M. F. Md Dan1, M. K. Abu Talib1, F. Pakir1, S. A. Ahmad Tajudin1, M. N. H. Zahari2, M. E. Z. Mat Radzi2

1Faculty of Civil and Environment Engineering, University Tun Hussein Onn Malaysia, Malaysia 2Preston GeoCEM (M) Sdn. Bhd., Taman Universiti, Malaysia

Received August 6, 2019; Revised October 6, 2019; Accepted December 15, 2019

Abstract The demand for groundwater supply has contents of the subsurface of the earth. One of the been increasing in recent year. In the past, Electrical non-destructive subsurface exploration methods are by Resistivity Method has been one of the well-known using electrical resistivity tomography. There are multiple non-destructive methods for groundwater surveying using configurations [1] that can be used to obtain the electrical various protocols to obtain the tomography of the soil resistivity tomography and of the typically used electrode subsurface. This study interpreted the difference of configurations that are the Pole-dipole and Multiple imaging results between two different protocols, the Gradient array. In the subsurface field study, Pole-dipole Pole-dipole array and the Multiple Gradient array using array is one of the configurations which have the largest ABEM TERRAMETER LS 2 due to the complexity of depth of detectability value in relation to noise level [2] geological setting. The geophysical survey shows the which is very useful for subsurface explorations especially imaging result of the resistivity and induced polarization in an urbanized area. One of the most recognized for the Pole-dipole array that differs from the imaging characteristics of the Pole-dipole array is the use of the results from Multiple Gradient array because of the remote current electrode which in theory should be placed different path the electric current C2 passing through at a distance of infinity or at least located at a distance of different soil types before reaching to the other side of the five times the maximum spacing from each survey line and current electrode C1. The positioning of the remote cable perpendicular to it [3]. It means that the injected current is was practically perpendicular to the alignment of the outside the electrodes spread line. The placement of the electrode spacing for Pole-dipole and parallel for Multiple remote current electrode at infinity is usually hard to be Gradient array and this causes the data collected at both achieved because of the conditions and the obstacle of the protocols to have different values because of the difference earth surface that the placement of the remote current in composition of soil that the current has to go through. electrode [4]. Meanwhile, for Multiple Gradient array, the This study has verified the differences of resistivity and multichannel data is very well suited and this will enhance induced polarization imaging results by using a drilled tube the speed of data acquisition in the field while at the same well to identify the soil types. time giving a higher data density [5]. Gradient array also does not require remote current cable, thus the injected Keywords Electrical Resistivity Method, Induced current lies within the electrodes spread line. This study Polarization, Multiple Gradient Array, Pole Dipole Array, will evaluate the effect of the position of the injected Resistivity current by comparing the results of the resistivity and induced polarization imaging between the Pole-dipole and Multiple Gradient arrays. In addition, the outcomes from both imaging results are compared with the tube well 1. Introduction drilled on site. The positioning of the tube well is decided based on the potential of the availability of groundwater There are multiple subsurface exploration that is determined through the results of the resistivity and methodswhich can help in interpreting the conditions and chargeability by using both of the information obtained 78 Comparison between Multiple Gradient and Pole Dipole Array Protocols for Groundwater Exploration in Quaternary Formation from Pole-dipole and Multiple Gradient array protocol. Kg. Parit Kaspan Parit Raja Batu Pahat, Johor.as shown in The type of soil obtained from the drilled borehole is Figure 1. The distance of the site to the central of Batu observed and collected to be compared with the 2-D Pahat is about 18 km. The site topography is relatively flat imaging data from the resistivity and induced polarization and surrounded by residents and palm oil plantation. The testing. general geology of peninsular Malaysia has been well documented by Minerals and Geoscience Department of Malaysia and based on the geological map shown in Figure 2. Materials and Methods 1, the study area is located in a quaternary period consisting of unconsolidated deposits from clay and silt (marine). In The study was conducted at Madrasah Darrul Hikmah general, the presence of quaternary aged soil will exhibit located in Batu Pahat to find the availability of soft soil phenomenon due to young and high-water content groundwater. The electrical resistivity method was used to derived from high water tables of lowland areas [6]. Since locate the possibility of groundwater by analyzing the Malaysia is a country with high rainfall intensity it is also tomography of the studied location. The availability of very common for this country to be having a high storage groundwater in the area is proved through drilling and the of groundwater at the study area. Based on the ground location of the proposed drilling position that are chosen surface observation, soft soil conditions related to wet clay based on the analysis of the resistivity and chargeability and silt geomaterials can be easily found in this area. This value can be evident with the compositions of the subsurface profile which consists of thick clay and silt layers, and thus 2.1. Study Area not having potential for usable groundwater. Electrical resistivity surveys are to be performed to further evaluate The study area was located at Madrasah Darrul Hikmah, the subsurface geomaterials.

Figure 1. Location and geology of study area

Civil Engineering and Architecture 7(6A): 77-85, 2019 79

2.2. Electrical Methods study the current electrode is placed at 270 meters away from the geophysical survey line as it is the furthest Electrical method (electrical resistivity and induced distance the current electrode can be placed during the polarization) was performed using ABEM Terrameter LS 2 survey being conducted. After the acquisition of the to obtain the electrical resistivity and induce polarization Pole-dipole array data was completed, the Multiple imaging on site. A maximum of 61 number of electrodes Gradient array then continuously started using the same was peg at the ground surface based on four resistivity land electrode placement used by the Pole-dipole array. After cables and 2 meters of equal electrode spacing. The total obtaining the resistivity and chargeability imaging value of length of 2D resistivity test was 160 meters and the survey the surveyed line, the type of soil within the subsurface can traverse oriented west to east direction as shown in Figure be predicted by referring to the resistivity and chargeability 2. The raw data obtained from the data acquisition was chart [7] which is presented in Table 1. To validate the processed using commercialized RES2DINV software to interpretation of the results from the imaging data, a chosen provide an inverse model that approximates the actual position along the west-east line will be drilled (90 meter subsurface structure. The inversion algorithm of from west to east direction) to obtain the actual type of soil. RES2DINV was used to process the data in order to obtain The drilling data will only represent single point the 2-D electrical results. The two protocols used for this information (1D) at the actual drilling location thus to some study were the Pole-dipole array and Multiple Gradient extent the data obtained can be compared with the results array and both of the tests were conducted continuously from the resistivity and chargeability [8]. The testing one after another. The Pole-dipole array protocol has a results from the Pole-dipole array and Multiple Gradient particular characteristic of having its current electrode (C2) array will be compared with the type of soil obtained from to be placed as far away as possible. In the case of this the drilling process.

Figure 2. Alignment (west-east) of electrical resistivity performed at Madrasah Darul Hikmah, Kg. Parit Kaspan Parit Raja Batu Pahat, Johor

Table 1. Resistivity and chargeability value

Material Resistivity (ohm-m) Chargeability (ms) Groundwater(fresh) 10 to 100 0 Alluvium 10 to 800 1 - 4 Sandstone 8 - 4 × 103 3-12 Sh ale 20 - 2 × 103 50 - 100 Limestone 50 - 4 × 103 10 - 20 Gran it e 5000 to 1,000,000 10 - 50

80 Comparison between Multiple Gradient and Pole Dipole Array Protocols for Groundwater Exploration in Quaternary Formation

2.3. Results and Findings A total of two geophysical surveys were conducted in the study area. The imaging data obtained for both Pole-dipole array and Multiple Gradient array were presented in Figure 3 and Figure 4 respectively. The result indicates that most of the subsurface resistivity values are below 60 Ω-m from both Pole-dipole and Multiple Gradient arrays. The resistivity values for sand and wet silty sand are in the range of 10-800 Ω-m and 100-250 Ω-m respectively [9,10]. Pandey, Shukla and Habibi [11] added that the resistivity of sandy soil decreased rapidly with an increase in water content. The results from the electrical resistivity imaging show, the condition of the subsurface on site has a relatively low resistivity which may indicate that there might be high water presence throughout the study area. This condition corresponds with the lithology of the study area which indicates a high water table presence because the study location is located within a quaternary period consisting of unconsolidated deposits from clay and silt (marine). In term of the induced polarization results by referring to Figure 3 and Figure 4, the chargeability value in both tests showed a high difference in value especially in the lower-mid section of the induced polarization imaging. Starting from the depth of 20 meters and below the Pole-dipole array chargeability value indicates a range of value of 6-13 ms and the Multiple Gradient array having a general chargeability value ranging from1-3 ms. The chargeability value for both induced polarization at this section of the imaging is far–off from each other even though the same survey line was used to conduct both the tests. The two contrasting results indicate a presence of two different soil profiles under the same geophysical survey line. This anomaly can be explained by referring to the geological model of the area as sketched in Figure 5. The positioning of the remote electrode (C2) of the Pole-dipole array was in the area of domination with clay. Meanwhile current electrode (C1) is located on the sandy lenses layer as shown in Figure 5(a). For Multiple Gradient array, all the current electrodes were positioned on top of the sandy lenses layer (Figure 5(b), thus taking a major role in influencing the difference of chargeability value between the two arrays. The difference may also be due to the difference in the geometry factors of the arrays. To further improve the interpretation of the result, a drill tube well was performed on the 90th meter of the survey line and the drilled soil was terminated at 30 meters depth. However, the collected soil samples taken from the water returned only up to 20 meter as shown in Table 2. After 20 meter depth, mixture of silt and clay mixed with water returned was observed.

Civil Engineering and Architecture 7(6A): 77-85, 2019 81

Figure 3. Electrical resistivity and chargeability using the Pole dipole array protocol

82 Comparison between Multiple Gradient and Pole Dipole Array Protocols for Groundwater Exploration in Quaternary Formation

Figure 4. Electrical resistivity and chargeability using the Multiple Gradient array protocol

Civil Engineering and Architecture 7(6A): 77-85, 2019 83

Figure 5. The geological model of the study area illustrate (a) the remote cable C2 positioned at a dominated clay for Pole dipole array and (b) the current electrodes placed on top of a sand lenses layer for Multiple Gradient array

Table 2. Drilled soil on site about 20 meters depth

Collected Soil Depth (Meter) Type of soil

0.0 – 4.5 Clay

4.5 – 9.0 Fine sand

84 Comparison between Multiple Gradient and Pole Dipole Array Protocols for Groundwater Exploration in Quaternary Formation

9.0 – 15.0 Coarse sand

15.0 – 20.0 Fine sand

3. Conclusions Both of the electrical resistivity surveys were successfully performed at Darul Hikmah, Kg. Pt. Kaspan REFERENCES Parit Raja Batu Pahat, Johor. Based on the data obtained from both of the geophysical surveys, the electrical [1] Szalai, S., & Szarka, L. (2008). On the classification of resistivity values show little or no difference in term of surface geoelectric arrays. Geophysical Prospecting, 56(2), 159–175. https://doi.org/10.1111/j.1365-2478.2007.00673. resistivity value which ranging from 0-60 Ω-m. However, x the induced polarization results showed varied chargeability values between the two different arrays for [2] Szalai, S., Novák, A., & Szarka, L. (2011). Which the Pole-dipole array, the dominant chargeability values geoelectric array sees the deepest in a noisy environment? Depth of detectability values of multielectrode systems for are in the around 0.2 and 13 ms and for the Multiple various two-dimensional models. Physics and Chemistry of Gradient array the value ranges from 0.2 – 1.5 ms. The the Earth, 36(16), 1398–1404. https://doi.org/10.1016/j.pce reason for the difference in results is because of the .2011.01.008 positioning of the Remote electrode (C2) which lies far [3] Syukri, M., & Saad, R. (2017). Seulimeum segment away and is almost parallel to the geophysical survey line characteristic indicated by 2-D resistivity imaging method. causing the current path in the clay area for Pole-dipole NRIAG Journal of Astronomy and Geophysics, 6(1), 210– array to be different from the Multiple Gradient array. 217. https://doi.org/10.1016/j.nrjag.2017.04.001 Hence, this study has successfully shown the differences [4] Razafindratsima, S., & Lataste, J. F. (2014). Estimation of geological setting within the electrode currents influence the error made in Pole-Dipole Electrical Resistivity the imaging results between the Pole-dipole array and the Tomography depending on the location of the remote Multiple Gradient array protocol. electrode: Modeling and field study. Journal of Applied Geophysics, 100, 44–57. https://doi.org/10.1016/j.jappgeo. 2013.10.008 Acknowledgements [5] Dahlin, T., Zhou, B. (2006). Multiple-gradient array measurement for multichannel 2D resistivity imaging. Near The authors would like to thank Ministry of Higher surface Geophysics, 4(2), 113-123 Education and Universiti Tun Hussein Onn Malaysia for [6] Hazreek, Z. A. M., Raqib, A. G. A., Aziman, M., Azhar, A. T. their financial support on FRGS vot. K049 and TIER 1 vot S., Khaidir, A. T. M., Fairus, Y. M., … Izzaty, R. A. (2017). H183 respectively. Preliminary Groundwater Assessment using Electrical

Civil Engineering and Architecture 7(6A): 77-85, 2019 85

Method at Quaternary Deposits Area. IOP Conference Series: Materials Science and Engineering, 226(1). https://doi.org/10.1088/1757-899X/226/1/012042 [7] Keller G.V. and Frischknecht F.C., 1996, Electrical methods in geophysical prospecting. Pergamon Press Inc., Oxford. [8] Abidin, M. H. Z., Madun, A., Tajudin, S. A. A., & Ishak, M. F. (2017). Forensic Assessment on Near Surface Landslide Using Electrical Resistivity Imaging (ERI) at Kenyir Lake Area in Terengganu, Malaysia. Procedia Engineering, 171, 434–444. https://doi.org/10.1016/j.proeng.2017.01.35 [9] Telford W M, Geldart L P and Sheriff R E 1976 Applied Geophysics (Cambridge: Cambridge University Press) [10] Lee T S 2002. Slope Stability and Stabilization Methods (New York: John Wiley & Sons, Inc.) [11] Pandey, L. M. S., Shukla, S. K., & Habibi, D. (2015). Electrical resistivity of sandy soil. Géotechnique Letters, 5(3), 178–185. https://doi.org/10.1680/jgele.15.00066

Civil Engineering and Architecture Call for Papers

Civil Engineering and Architecture is an international peer-reviewed journal that publishes original and high-quality research papers in all areas of civil engineering and architecture. As an important academic exchange platform, scientists and researchers can know the most up-to-date academic trends and seek valuable primary sources for reference. Aims & Scope

Architectural Design Competition Computational Science and Engineering Architectural Drawing Construction Engineering Architectural Style Control Engineering Architectural Theory Earthquake Engineering Atmospheric Sciences Environmental Engineering Biomechanics Forensic Engineering Building Materials Geodesy Coastal Engineering Geophysics Editorial Board

Elizabeth Martin-Malikian Southern Polytechnic State University, USA Chia-hui Lin The University of Melbourne, Australia Janendranath Mandal Indian Institute of Technology, India Engin Eroglu Duzce University Faculty of Forestry, Turkey Hasan Tahsin Ozturk Karadeniz Technical University, Turkey Antonio Brencich University of Genoa, Italy Santina Di Salvo University of Palermo, Italy Hongyuan Liu University of Tasmania, Australia Aldo Renato Daniele Accardi Architectural Department, Italy Halil Zafer Alibaba Eastern Mediterranean University, North Cyprus Adel Francis École de technologie supérieure (ÉTS), Canada Contact Us Amira Elnokaly University of Lincoln, UK Damian Beben Opole University of Technology, Poland Horizon Research Publishing Mohamed Mostafa Hassan Mostafa Central University of Technology, South Africa 2880 ZANKER RD STE 203 Stephen Eromobor Central University of Technology, South Africa SAN JOSE, CA 95134 Ying Wang Deakin University, Australia USA H.M.A. Mahzuz Shahjalal University of Science & Technology Sylhet, Bangladesh Email: [email protected] Tomaz Tollazzi University of Maribor, Slovenia Jonas Jakaitis Vilnius Gediminas Technical University, Lithuania Daniela Staffenova University of Žilina, Slovenia Petros Christou Frederick University, Cyprus George Vasdravellis Heriot-Watt University, UK Lee Pugalis Northumbria University, UK Manuscripts Submission

Manuscripts to be considered for publication have to be submitted by Online Manuscript Tracking System(http://www.hrpub.org/submission.php). If you are experiencing difficulties during the submission process, please feel free to contact the editor at [email protected].