Predicting motorized tricycle emissions using spatiotemporal deep learning: evidence from Kano, Nigeria

Abstract Urban air pollution poses a major challenge in African cities, where informal transport modes, such as motorized tricycles, are prevalent but under-regulated. This study proposes a spatiotemporal deep learning framework to predict emissions from tricycles in Sabon Gari, Kano, Nigeria. Integrating geographic information system, traffic flow, and meteorological data, we develop a hybrid spatiotemporal convolutional neural network-long short-term memory (CNN-LSTM) model and benchmark it against five machine learning methods. The proposed model achieves superior accuracy (RMSE = 4.00, MAPE = 7.8%, R2 = 0.92) and reveals emission hotspots aligned with commercial corridors during peak hours. These findings highlight the strong spatiotemporal dependence of emissions on traffic patterns and land use, supporting the development of spatially targeted, data-informed policies for managing emissions in rapidly urbanizing informal mobility sectors. Beyond predictive accuracy, the model maps time-resolved emission surfaces and hotspot patterns, providing the spatiotemporal evidence necessary for targeted intervention design. By isolating and modelling emissions from motorized tricycles, this study contributes to urban air quality management and low-carbon transport planning in data-scarce, rapidly urbanizing regions.