Cloud-Native Deployment of a Sinkhole Susceptibility Mapping Platform Using Google Earth Engine and Colab in Urbanised Karst Terrain
Tan Yan Eng and Siti Nur Aliaa Roslan

Faculty of Engineering, Universiti Putra Malaysia, Malaysia

Abstract

Sinkholes are a critical geohazard in rapidly urbanising karst landscapes, where natural subsurface dissolution is intensified by human activities such as construction and groundwater extraction. Despite their increasing frequency, public access to sinkhole risk information remains limited, particularly in developing urban settings like Kuala Lumpur, Malaysia. This study presents a cloud-native platform for sinkhole susceptibility mapping, integrating multi-source geospatial data with machine learning to achieve both technical scalability and public usability. The proposed workflow leverages Google Earth Engine (GEE) for geospatial data processing and Google Colab for model training, enabling a seamless end-to-end pipeline without reliance on desktop GIS tools. A lightweight one-dimensional Convolutional Neural Network (1D CNN) is implemented to classify sinkhole susceptibility based on 14 spatial control factors representing geological, topographic, hydrological, and anthropogenic influences. The model produces a continuous susceptibility surface, allowing for nuanced risk interpretation. Performance evaluation achieved a high AUC-ROC of 0.97, demonstrating strong discriminatory power despite a limited training dataset. The final susceptibility map is deployed via a public-accessible GEE web application that includes toggleable data layers and interactive query functions designed for planners, engineers, and the general public. The platform supports scalable updates and future integration with real-time data sources. Overall, this study demonstrates the feasibility of cloud-based geospatial modelling for hazard communication in data-scarce urban environments and provides a reproducible, user-friendly template for karst risk assessment in other rapidly developing cities.

Keywords: Google Earth Engine, karst, sinkhole susceptibility, Convolutional Neural Network

Topic: Topic B: Applications of Remote Sensing