Kithsiri Perera

Dr. Kithsiri Perera

BA (Colombo), MSc, and Dr Eng 1995 (Japan), Honorary Senior ...

Research Focus & Metrics

Academic Career Length

15Years (2012-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
Verified
Journal Article
Open access
PDF file
OAOgbaje Andrew
Ogbaje Andrew
Armando ApanArmando Apan
Armando Apan
Dev Raj PaudyalDev Raj Paudyal
Dev Raj Paudyal
Kithsiri PereraKithsiri Perera
Kithsiri Perera
The accuracy of most SAR-based flood classification and segmentation derived from semi-automated algorithms is often limited due to complicated radar backscatter. However, deep learning techniques, now widely applied in image classifications, have demonstrated excellent potential for mapping complex scenes and improving flood mapping accuracy. Therefore, this study aims to compare the image classification accuracy of three convolutional neural network (CNN)-based encoder–decoders (i.e., U-Net, PSPNet and DeepLapV3) by leveraging the end-to-end ArcGIS Pro workflow. A specific objective of this method consists of labelling and training each CNN model separately on publicly available dual-polarised pre-flood data (i.e., Sentinel-1 and NovaSAR-1) based on the ResNet convolutional backbone via a transfer learning approach. The neural network results were evaluated using multiple model training trials, validation loss, training loss and confusion matrix from test datasets. During testing on the post-flood data, the results revealed that U-Net marginally outperformed the other models. In this study, the overall accuracy and F1-score reached 99% and 98% on the test data, respectively. Interestingly, the segmentation results showed less use of manual cleaning, thus encouraging the use of open-source image data for the rapid, accurate and continuous monitoring of floods using the CNN-based approach.

17 June 2023 in Biodiversity and Conservation
Verified
Journal Article
Open access
PDF file
TGThakshila D. Gunawansa
Thakshila D. Gunawansa
Kithsiri PereraKithsiri Perera
Kithsiri Perera
Armando ApanArmando Apan
Armando Apan
NHNandita K. Hettiarachchi
Nandita K. Hettiarachchi
Human-elephant conflict (HEC) is a severe conservation, socio-economic and environmental issue of forests and ecosystems in elephant inhabiting countries, including Sri Lanka. Due to the rapid growth of human and elephant populations, both struggles to share limited land resources. The major causes and contexts of HEC in Sri Lanka include land use change, habitat loss due to human population growth, crop raiding behavior, problem elephants, and changes in agriculture practices. Since 2019, 125 people and 370 elephants have killed annually on average due to the conflict. Also, Sri Lanka has recorded the highest annual elephant deaths and second-highest human deaths due to HEC. The human death rate has increased by approximately 42% over previous three decades. The Sri Lankan government provides compensation for death and disability of the human caused by elephants and for elephant-damaged houses or properties. The Sri Lankan elephant (Elephas maximus maximus) is an endangered subspecies. It’s home range is restricted to 50–150 km2and depends on the availability of food, water, and shelter of the habitat in which they live. Various management strategies have been developed by the government and villagers to prevent and mitigate HEC. Today, Sri Lankan elephants are protected under Sri Lankan law, with punishment by fines and jail terms. This article reviews the history, present status, and traditional conflict management of HEC in Sri Lanka. We suggest a satellite data fusion approach with GIS modeling to identify risk zones of HEC to develop further protective measures for humans and elephants.

18 August 2023 in International Journal of Remote Sensing
Verified
Journal Article
Open access
TGThakshila D. Gunawansa
Thakshila D. Gunawansa
Kithsiri PereraKithsiri Perera
Kithsiri Perera
Armando ApanArmando Apan
Armando Apan
NHNandita K. Hettiarachchi
Nandita K. Hettiarachchi
DBDananjana Y. Bandara
Dananjana Y. Bandara

6 October 2021 in Earth
Verified
Journal Article
Open access
PDF file
Kithsiri PereraKithsiri Perera
Kithsiri Perera
RTRyutaro Tateishi
Ryutaro Tateishi
KAKondho Akihiko
Kondho Akihiko
SHSrikantha Herath
Srikantha Herath
Bushfires are an integral part of the forest regeneration cycle in Australia. However, from the perspective of a natural disaster, the impact of bushfires on human settlements and the environment is massive. In Australia, bushfires are the most disastrous natural hazards. According to the records of the Parliament of Australia, the recent catastrophic bushfires in NSW and Victoria burnt out over 10 million hectares of land, a figure more significant than any previous bushfire damage on record. After the deadly 2009 Black Saturday bushfires, which killed 173 people in Victoria, public attention to bushfires reached a new peak. Due to the disastrous consequences of bushfires, scientists have explored various methods to mitigate or even avoid bushfire damage, including the use of bushfire alerts. The present study adds satellite imagery and GIS-based semi-real-time bushfire contents to various bushfire warnings issued by government authorities. The new product will disseminate graphical bushfire contents to rural Australians through social media, using Google Maps. This low-cost Media GIS content can be delivered through highly popular smartphone networks in Australia through social media (Facebook and Twitter). We expect its success to encourage people to participate in disaster mitigation efforts as contributors in a participatory GIS network. This paper presents a case study to demonstrate the production process and the quality of media GIS content and further discusses the potential of using social media through the mobile network of Australia while paying attention to mobile blackspots. Media GIS content has the potential to link with the public information systems of local fire management services, disseminate contents through a mobile app, and develop into a fully automated media GIS content system to expand the service beyond bushfires.

17 July 2012 in ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Verified
Journal Article
Open access
PDF file
Kithsiri PereraKithsiri Perera
Kithsiri Perera
SHS. Herath
S. Herath
AAA. Apan
A. Apan
RTR. Tateishi
R. Tateishi
Assessing forest cover of Sri Lanka is becoming important to lower the pressure on forest lands as well as man-elephant conflicts. Furthermore, the land access to north-east Sri Lanka after the end of 30 years long civil war has increased the need of regularly updated land cover information for proper planning. This study produced an assessment of the forest cover of Sri Lanka using two satellite data based maps within 23 years of time span. For the old forest cover map, the study used one of the first island-wide digital land cover classification produced by the main author in 1988. The old land cover classification was produced at 80 m spatial resolution, using Landsat MSS data. A previously published another study by the author has investigated the application feasibility of MODIS and Landsat MSS imagery for a selected sub-section of Sri Lanka to identify the forest cover changes. Through the light of these two studies, the assessment was conducted to investigate the application possibility of MODIS 250 m over a small island like Sri Lanka. The relation between the definition of forest in the study and spatial resolution of the used satellite data sets were considered since the 2012 map was based on MODIS data. The forest cover map of 1988 was interpolated into 250 m spatial resolution to integrate with the GIS data base. The results demonstrated the advantages as well as disadvantages of MODIS data in a study at this scale. The successful monitoring of forest is largely depending on the possibility to update the field conditions at regular basis. Freely available MODIS data provides a very valuable set of information of relatively large green patches on the ground at relatively real-time basis. Based on the changes of forest cover from 1988 to 2012, the study recommends the use of MODIS data as a resalable method to forest assessment and to identify hotspots to be re-investigated. It's noteworthy to mention the possibility of uncounted small isolated pockets of forest, or sub-pixel size forest patches when MODIS 250 m x 250 m data used in small regions.

30 September 2022 in Journal of Advances in Engineering and Technology
Verified
Journal Article
Closed access
UUUva Wellassa University
Uva Wellassa University
TGThakshila Gunawansa
Thakshila Gunawansa
Kithsiri PereraKithsiri Perera
Kithsiri Perera
UQUniversity of Southern Queensland
University of Southern Queensland
Armando ApanArmando Apan
Armando Apan
URUniversity of Ruhuna
University of Ruhuna
Sentinel-2 satellite data has been used for forest cover monitoring for almost five years. Mapping with Sentinel data will be a cost-effective solution for Sri Lanka, where the lack of updated land cover maps with high spatial resolution is a significant challenge in the land resource management of the country. A study area of about 5,000 km2 located in southeast Sri Lanka was selected for this study. Agricultural lands, forests including Yala national park, and villages with perennial crops make up the region. A Level-2A Sentinel-2 image with less than 10 percent cloud cover was used in the European Space Agency's (ESA) SNAP software version 8.0.0 for image processing and the forest cover of the study area was mapped through the Random Forest classifier (RFC). Normalized Difference Vegetation Index (NDVI) is also calculated as a Sentinel product to support RFC output. For RFC, ground truth data were collected through the reference of Google Earth high-resolution data. The classification accuracy was assessed using the Google Earth image as the reference dataset. Furthermore, RFC results were compared with NVDI greenness values. The classification accuracy was calculated using a confusion matrix (error matrix) through randomly selected 100 sample points. The overall accuracy of the land cover map was 85 percent, with a 96 percent accuracy for forest cover identification. The study found RFC as an effective method to isolate forest cover in Sri Lanka.KEYWORDS: Sentinel-2, Random Forest Classifier, Land cover classification, Land cover mapping, Normalized Difference Vegetation Index.