A deep learning method for creating globally applicable population estimates from sentinel data

Abstract

This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.

Modeling population distribution: A visual and quantitative analysis of gradient boosting and deep learning models for multi‐output spatial disaggregation

“…Vargas-Munõz et al (2022) applied a Markov random fields (MRF) method that iteratively enhances the initial estimations of a dasymetric disaggregation method. Fibaek et al (2022) followed a deep learning approach to generate daytime and nighttime population estimates at 10 m resolution using satellite imagery, structural area estimates, and structure classifications. The spatial disaggregation method introduced by Monteiro et al (2018Monteiro et al ( , 2019Monteiro et al ( , 2021 iteratively refines estimates generated by pycnophylactic interpolation or dasymetric mapping, by applying different types of regression models.…”

“…Fibæk et al. (2022) followed a deep learning approach to generate daytime and nighttime population estimates at 10 m resolution using satellite imagery, structural area estimates, and structure classifications. The spatial disaggregation method introduced by Monteiro et al.…”

Symbolic and subsymbolicGeoAI: Geospatial knowledge graphs and spatially explicit machine learning

“…We call this kind of practices spatially explicit machine learning (Janowicz et al, 2020;Li et al, 2021;Mai, Jiang, et al, 2022;Yan et al, 2017Yan et al, , 2019. Some of the unique challenges include how to represent different types of spatial data into the embedding/subsymbolic space , how to achieve geographic generalizability for a given machine learning/deep learning model (Goodchild & Li, 2021;Li et al, 2022), how to perform transfer learning across space and tasks (Fibaek et al, 2022), how to avoid geographic biases in GeoAI models , and so forth. This special collection also includes two innovative articles that tackle two questions we mentioned above.…”

“…The fourth article, from Fibaek et al (2022), presents a deep learning model for population estimation in areas geographically distinct from Northern Europe. The two major contributions of this article are as follows: (1) the authors demonstrate how to use the same deep learning architecture and transfer learning to solve three tasksstructure area prediction, structure type classification, and population prediction for Ghana and Egypt based on Sentinel data; (2) the authors show how to use multi-sensor data to produce high-resolution population estimation for both daytime and nighttime, using deep learning.…”

Disaggregation of population estimates at high spatial resolution with machine learning

“…Examples of direct future projections for gridded population distributions might not be available yet in the literature, but ML has been applied for projecting built-up land based on time series of remote sensing observations [50] and for estimating existing urban structures [7] and population density [44,122,115,132]. These approaches have demonstrated that ML can assist the extraction of information from unstructured, remote sensing data and provide effective solutions for socio-demographic issues.…”