| Population estimation is a critical component in urban planning, resource management, and socio-economic analysis. This paper introduces CNNPL, a convolutional neural network-based model that incorporates optimized Point of Interest weights and various geospatial data. The study utilizes a dataset of spatial patches to evaluate the model’s performance against both a pixel-level counterpart (CNNPxL) and established baseline models, including LandScan and GPWv4. Results indicate that CNNPL significantly improves predictive performance, achieving a coefficient of determination R² of 0.8377 and a lower Mean Squared Error (MSE) of 1.07 x10^{10}, thereby demonstrating superior efficacy in capturing population distribution patterns while maintaining computational efficiency. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.