Remote sensing and verification of forest growth

• Main Author: Leong, Jing Wen
• Other Authors: Lee Bu Sung, Francis
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/156643
• Institution: Nanyang Technological University

With rapid urbanization, Singapore’s land use has changed greatly over the past years. In this paper, we investigate land use changes, in particular, vegetation changes from the years 2016 to 2021 using remote sensing, i.e., satellite data. The purpose of this study is to come up with an easy pipeline to map the land use of Singapore accurately through publicly available datasets. The study of the land use changes then aims to provide valuable inputs to policy makers and hopefully aid them in making better policies regarding green spaces in Singapore. Technological advances in remote sensing and increasing availability of public datasets such as the Sentinel-2 satellite images provides a great opportunity for land use mapping. There are 3 main components in this paper: 1. Data preprocessing (cloud removal), 2. Visualising and analysing vegetation changes in Singapore, and 3. Investigating how vegetation changes affect Singapore. In the first component, 3 different algorithms for cloud removal were investigated, namely the QA60 band, s2cloudless, and Fmask. The evaluation found that the s2cloudless method was the most suitable application for Sentinel-2 datasets adopted in this project. In the second component of the project, vegetation covers were visualised through the Normalized Difference Vegetation Index (NDVI). Subsequently, vegetation changes were detected and analysed through change detection methods. Three change detection methods: image differencing, post-classification comparison, and a combination method of both image differencing and post-classification comparison were adopted and evaluated. In the post-classification comparison method, land use classifications maps must be first obtained before change detection analysis can be performed. A traditional pixel-based classification was compared to an object-based classification approach whereby the Simple Non-Iterative Clustering (SNIC) algorithm to identify spatial clusters and the Gray-Level Co-occurrence Matrix (GLCM) to calculate cluster textural indices were combined. Both methods used machine learning algorithms (Random Forest and Support Vector Machine) to perform the final classification. Results found that the SNIC and GLCM algorithms helped to increase the overall accuracy of the classifier to 85.8% as compared to the 83.1% using the traditional pixel-based approach. The three change detection methods had different advantages and limitations and it was found that the combination method was the most appropriate in this context given its ability to focus only on certain areas with significant change, while providing detailed information about the nature of change. The third component of the project then uses the change results found in the second component to study the effects of vegetation change on the temperatures in Singapore. Due to the limited information at hand, the analysis was not able to establish a relationship between greenery and temperature, but it found global warming as a significant factor affecting the study of the relationship. To accurately determine the effects of vegetation change on temperatures, methods which account for global warming and other factors can be explored in future research. To achieve more accurate land use maps of Singapore, other approaches such as the use of deep learning can also be explored. This is important since the accuracy of change analysis is highly dependent on the accuracy of the land use classification maps.