From Binary Transitions to Continuous Intensification: Deep Change modelling of Accra's Urban Growth

Urban growth is rarely a clean “changed / not changed” process: in fast-transforming peri-urban landscapes it unfolds through incremental intensification, patchy infill, and noisy boundaries that binary change labels compress into a single class. This raises a practical question for urban remote sensing and spatial modelling: do we gain accuracy and interpretability by predicting continuous built-up change rather than binary transitions? In Accra, Ghana, the answer is strongly affirmative: a U-Net trained to predict fractional built-up change substantially outperforms both a random-forest increment baseline and conventional benchmarks, achieving a higher Figure-of-Merit with fewer missed urban changes. Overall, these results suggest that learning continuous change provides a more nuanced and planning-relevant signal of urban growth than thresholded transition mapping.