Simulating land loss and land gain by integrating neighborhood effect and deep learning with cellular automata

Keywords: Land Loss, Land Gain, Geo-morphology, Neighborhood Effect, Bayesian Network, Coastal Louisiana
Session Type: Virtual Paper Abstract
Day: Saturday
Session Start / End Time: 2/26/2022 02:00 PM (Eastern Time (US & Canada)) - 2/26/2022 03:20 PM (Eastern Time (US & Canada))
Room: Virtual 28

Abstract

Dynamic land cover changes in coastal zones, e.g., land losses and land gains, severely disrupt regional ecosystem balance and affect coastal human communities in profound ways. There is an urgent need to simulate the coastal land change from a spatial perspective and identify areas prone to land loss or land gain in the future. Artificial neural network with cellular automata (ANN-CA) model was generally applied in land-use change simulation in the previous studies. However, ANN-CA is unable to retrieve the spatial features of the driving factors within a neighborhood and to discover their relationships with the land use type/change of the central cell, leading to lower simulation accuracy. Hence, in this study, two methods were designed to extract the neighborhood information of variables. The first one is that we developed the neighborhood effect profile (NEP) algorithm to calculate the optimal linear neighborhood effect of each variable on the center cell which is considered as a new variable and is put into the ANN-CA model. The second one used a convolutional neural network (CNN) to extract the non-linear neighborhood features. Three models which are ANN-CA, NEP-ANN-CA, and CNN-CA were applied to simulate land loss and land gain in the Louisiana Coastal Zone. Land cover images in 1996 and 2006 were used to train the models and the validation of three models was processed based on the land cover image in 2016. This research offers a more applicable method for researchers and managers to investigate land change patterns in coastal regions.