Development of an AI-Enhanced Mobile GIS for Urban Land Use Mapping

Thesis Abstract

Urban land use mapping is a critical component of effective city planning and management, yet existing methods often encounter limitations related to data accuracy, timeliness, and operational efficiency. The rapid expansion of urban areas combined with the increasing availability of geospatial data necessitates the development of innovative, scalable, and intelligent solutions to enhance land use classification and monitoring. In response, this study aims to develop an AI-enhanced Mobile Geographic Information System (GIS) capable of real-time, user-friendly urban land use mapping. The specific objectives include designing an integrated AI-enabled mobile GIS framework, evaluating its classification accuracy against traditional GIS methods, and assessing its usability by urban planning professionals. The research adopts a mixed-methods approach, combining quantitative analysis to evaluate the system’s performance and qualitative assessments to gauge user acceptance and operational practicality. The study population comprises urban planners, GIS professionals, and data analysts within a metropolitan city with a diverse and rapidly changing land use profile. A stratified random sampling technique was employed to select 150 participants across these professional groups. Data collection instruments include a custom-developed mobile GIS application equipped with machine learning algorithms—specifically convolutional neural networks (CNN)—for land use classification, complemented by structured questionnaires and semi-structured interview guides for user feedback. Ground truth data was collected through field surveys involving 300 ground-truth points distributed across different land-use zones, which served as a benchmark for assessing classification accuracy. Data analysis involves multiple statistical techniques, with accuracy metrics such as overall correctness, precision, recall, and F1-score utilized to compare the AI-enhanced system against conventional GIS classifications through confusion matrices. Regression analysis, specifically multiple linear regression, is applied to identify factors influencing the usability and perceived effectiveness of the mobile GIS. Thematic analysis is conducted on qualitative interview data to extract insights into user experience, challenges, and suggestions for system improvement. The AI models were trained with a dataset of over 5,000 georeferenced images representing diverse land use types, with performance evaluated via cross-validation techniques to prevent overfitting. Expected findings indicate that the AI-enhanced mobile GIS will significantly outperform traditional manual and semi-automated land use mapping methods, achieving an accuracy improvement of at least 15%. The system is anticipated to demonstrate high usability scores, with participants citing increased speed, accuracy, and operational convenience. Furthermore, the integration of deep learning models, combined with user-centric mobile interface design, is expected to facilitate near real-time land use classification, thus enabling timely urban planning decisions. This research contributes to the existing body of knowledge by demonstrating the practical application of artificial intelligence within mobile GIS frameworks for urban land use mapping, filling gaps related to real-time data processing and user accessibility in developing urban contexts. The findings underscore the potential for scalable, AI-driven GIS solutions to revolutionize urban planning paradigms, particularly in rapidly expanding cities facing data scarcity and resource constraints. The study concludes that AI-enhanced mobile GIS presents a viable, efficient, and accurate alternative to traditional land use mapping tools, with broad implications for urban management and sustainable development. Recommendations include integrating the system into routine urban planning workflows, emphasizing the need for continuous model training with updated data, and advocating for capacity building among users. Further research is suggested to extend the system’s application to other facets of urban analytics such as environmental monitoring and hazard assessment, as well as exploring the potential integration of satellite imagery and crowdsourced data to enhance the robustness and coverage of land use datasets.