Land Use Monitoring Using Remote Sensing and Deep Learning Neural Networks: A Case Study in Brangay Galalan, Pangil, Laguna

Abstract

Timely and reliable information on land use is crucial for monitoring and achieving national sustainability targets. The advances in technology have its benefits in terms of data quality, timeliness, and cost-savings that could positively influence evidence-based policy making on the environment. This study evaluated the use of deep learning neural network, a subset of the broader machine learning techniques, in analyzing satellite imagery in the context of mapping land use change. The study utilized open-source geographic information system (GIS) and programming software, freely available satellite imageries such as Landsat 8 and Sentinel-2 datasets, and existing land use and land cover maps for training the convolutional neural network (CNN) and identifying major land uses in the upland study area through image recognition. The results suggested that medium spatial resolution satellite data used as input in the CNN model supported mapping and monitoring of land use dynamics over time. Adapting a deep pre-trained network such as Residual Network (ResNet) and up-sampling the satellite imageries delivered promising results for both single-label and multi-label land use and land cover classification. Augmenting the information from existing land use and land cover maps with ancillary information derived from ground- truthing survey and remote validation using open-source Google Earth imageries could improve the object recognition rate of the CNN model. The use of vegetation indices in the CNN training was also recommended to increase the classification accuracy of the model in land cover classes that are visually less distinct. Time series datasets that capture different vegetation signatures throughout the plant cycle could also strengthen land cover mapping certainty over time. The use of synthetic aperture radar (SAR) images to address the issue of cloud cover was also viewed as a promising research area in the future. Integrating the use of these tools and data sources into the current work program of the government would require strategic investments on physical infrastructure, strengthened capacity of the human resource, and established institutional support to fully capitalize on the benefits of these innovative technologies for environmental monitoring.