Bridging the Gap: Visual Evaluation of AI-Based Land Use Maps in Support of Local Development Planning
Nurul Hidayah Y., Nurhaziyatul Adawiyah Y. and Norzailawati M.N.

Kulliyyah of Architecture and Environmental Design, International Islamic University of Malaysia (IIUM), Malaysia

Abstract

Existing land use maps are one of the most crucial elements in preparing development plans, such as the Local Plan. These maps are typically used in conjunction with proposed land use zoning maps, which illustrate the intended future land uses based on the area^s planning policy, to assist urban planners in making informed decisions that shape the direction of urban development. Therefore, having an accurate existing land use maps is very important. Currently existing land use maps in local plans are derived from various sources, such as field surveys, satellite imagery, and existing databases from the planning department. However, these methods rely heavily on human intervention, is time-consuming, labour-intensive, and often not updated regularly. With the growing accessibility of readily available AI-based products that require no pre-processing, such as the Dynamic World dataset (developed by Google and the World Resources Institute) and the Global Forest Change dataset, this paper explores the potential of these off-the-shelf AI-based products in supporting local planning practices in Malaysia. This study uses Kuala Lumpur as its case study. These AI-based land cover datasets were accessed and analysed using Google Earth Engine, along with Kuala Lumpur zoning map, which served as the official reference. A visual assessment was conducted to identify the spatial mismatches between the AI-based land cover datasets and the Kuala Lumpur zoning map. Preliminary findings show that while both the Dynamic World dataset and the Global Forest Change dataset are able to perform reasonably well in detecting the zoning categories, their fixed classification categories, limited thematic detail, and the medium-resolution satellite imagery present significant limitations to be integrated into the urban planning practice. This study contributes to a deeper understanding of the potentials and limitations of current AI-based products and the requirements needed for them to support the preparation of development plans.

Keywords: AI-based, Land Cover Classification, Pre-trained Land Cover Maps, Urban Planning, Zoning Map

Topic: Topic D: Geospatial Data Integration