Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Landcover classification and change detection using remote sensing and machine learning: a case study of Western Fiji

Yadvendra Gurjar, Ruoni Wan, Ehsan Farahbakhsh, Rohitash Chandra

TL;DR

This study tackles land-use/land-cover change (LULCC) in Nadi, Fiji, under rapid urbanization by integrating Landsat-8 remote sensing with a hybrid ML framework. It compares unsupervised (K-means) and supervised (CNN, Random Forest, MLP) approaches using 9×9 image chips and a seven-class scheme derived from LUCAS, aided by spectral indices such as NDVI, NDWI, MNDWI, and NDBI. CNN consistently achieves the highest accuracy (~99.4%), while K-means tends to misclassify complex areas and overestimate urban zones; urban expansion is clearly detected from 2013 to 2023, including mangrove-to-urban conversion near coasts. The framework demonstrates a practical, high-accuracy workflow for LULCC modelling and change detection in data-limited, coastal-Pacific contexts, with public code and data to foster reproducibility and application to planning decisions.

Abstract

As a developing country, Fiji is facing rapid urbanisation, which is visible in the massive development projects that include housing, roads, and civil works. In this study, we present machine learning and remote sensing frameworks to compare land use and land cover change from 2013 to 2024 in Nadi, Fiji. The ultimate goal of this study is to provide technical support in land cover/land use modelling and change detection. We used Landsat-8 satellite image for the study region and created our training dataset with labels for supervised machine learning. We used Google Earth Engine and unsupervised machine learning via k-means clustering to generate the land cover map. We used convolutional neural networks to classify the selected regions' land cover types. We present a visualisation of change detection, highlighting urban area changes over time to monitor changes in the map.

Landcover classification and change detection using remote sensing and machine learning: a case study of Western Fiji

TL;DR

This study tackles land-use/land-cover change (LULCC) in Nadi, Fiji, under rapid urbanization by integrating Landsat-8 remote sensing with a hybrid ML framework. It compares unsupervised (K-means) and supervised (CNN, Random Forest, MLP) approaches using 9×9 image chips and a seven-class scheme derived from LUCAS, aided by spectral indices such as NDVI, NDWI, MNDWI, and NDBI. CNN consistently achieves the highest accuracy (~99.4%), while K-means tends to misclassify complex areas and overestimate urban zones; urban expansion is clearly detected from 2013 to 2023, including mangrove-to-urban conversion near coasts. The framework demonstrates a practical, high-accuracy workflow for LULCC modelling and change detection in data-limited, coastal-Pacific contexts, with public code and data to foster reproducibility and application to planning decisions.

Abstract

As a developing country, Fiji is facing rapid urbanisation, which is visible in the massive development projects that include housing, roads, and civil works. In this study, we present machine learning and remote sensing frameworks to compare land use and land cover change from 2013 to 2024 in Nadi, Fiji. The ultimate goal of this study is to provide technical support in land cover/land use modelling and change detection. We used Landsat-8 satellite image for the study region and created our training dataset with labels for supervised machine learning. We used Google Earth Engine and unsupervised machine learning via k-means clustering to generate the land cover map. We used convolutional neural networks to classify the selected regions' land cover types. We present a visualisation of change detection, highlighting urban area changes over time to monitor changes in the map.

Paper Structure

This paper contains 28 sections, 12 equations, 9 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Study area
  4. Data Preparation
  5. Cloud Removal
  6. Index Calculations
  7. Classification scheme
  8. Data collection and labelling
  9. Chip Generation
  10. Machine learning methods
  11. K-means
  12. Random Forest
  13. Multilayer Perceptron
  14. Convolutional Neural Network
  15. Framework
  16. ...and 13 more sections

Figures (9)

  • Figure 1: Location of the study area in Viti Levu, Fijiwiki:xxx. (a) The map of Viti Levu with the study area highlighted in black. (b) A closer view of the selected study area showing its land cover characteristics
  • Figure 2: Cloud and shadow removal process using Landsat 8 imagery. The sequence shows (a) the initial image with cloud cover and (b–d) the effect of applying the cloud and shadow mask on median composites over one month, three months, and one year, respectively. Longer temporal windows result in reduced cloud contamination and enhanced surface feature visibility.
  • Figure 3: Framework for land use and land cover classification. It includes data normalisation, manual labelling with coastal upsampling, supervised (CNN, RF, ANN) and unsupervised (K-Means clustering) training, model evaluation, and selection of the best model for future classification.
  • Figure 4: a) The land cover map plot using different sample sizes
  • Figure 5: The confusion matrices of supervised learning models
  • ...and 4 more figures