Assessing the Impact of Neighborhood Size on Temporal Convolutional Networks for Modeling Land Cover Change

Land cover change (LCC) studies are increasingly using deep learning (DL) modeling techniques. Past have leveraged temporal or spatiotemporal sequences of historical LC data to forecast changes with DL models. However, these do not adequately assess the association between neighborhood size and model capability LCCs, where refers spatial extent captured by each sample. The objectives this research study were to: (1) evaluate effect on capacity models specifically Temporal Convolutional Networks (TCN) Neural (CNN-TCN), (2) auxiliary variables LCCs. First, type setting configuration was assessed derived from multitemporal MODIS for Regional District Bulkley-Nechako, Canada, comparing subareas exhibiting different amounts LCCs trends obtained full region. Next, outcomes compared three other regions. results evaluated three-map comparison measures, real-world next timestep, previous forecasted year used calculate correctly transitioned areas. Across all regions explored, it observed that increasing sizes improved model’s capabilities short-term CNN–TCN most correct several while reducing error due quantity when provided additional variables. This contributes systematic exploration selected techniques geographic applications.