Modeling urban expansion in the Philippines using a cellular automata–integrated spatiotemporal machine learning framework

Abstract. Urbanization in the Philippines rapidly shift towards a certain critical level, yet there are only few established frameworks that provide localized forecasts of urban growth across the Philippines, making it difficult to identify which specific cities are likely to thrive or be left behind, a gap that poses significant challenges for effective urban planning. This study introduces a spatiotemporal modeling framework that combines Long Short-Term Memory (LSTM) networks with Cellular Automata (CA) to simulate urban expansion at a national scale. Using open-source geospatial data on land use, slope, transportation, and urban footprints, the model generates dynamic urban transition probabilities through LSTM, which are then fed into a CA system that simulates urban sprawl. After simulation, the comparison of the generated maps with the actual maps showed that hybrid approach outperformed conventional methods such as standalone LSTM, and traditional Logistic-Regression CA, achieving a Fuzzy Similarity Rating (FSR) of 38.20% and a figure of merit (FOM) of 55.37%, highlighting emerging urban hotspots such as Calamba, Carcar City, and Davao City by 2030. The integrated LSTM-CA model captures spatial interactions and temporal dynamics more effectively than static models, offering improved realism in simulating pixel-level transitions. By offering data-driven forecasts of urban growth, this study supports a more informed spatial planning decisions, including infrastructure development and land conservation. With more sustainable and inclusive urbanization, this aims to ensure that no city is left behind as the country moves toward its urban future.