Mapping fine-scale population distributions at building level using improved geographical random forests with multi-source open data
Topics: Remote Sensing
, Population Geography
, Geographic Information Science and Systems
Keywords: Remote Sensing, Population estimation, Random Forest
Session Type: Virtual Guided Poster Abstract
Day: Sunday
Session Start / End Time: 2/27/2022 09:40 AM (Eastern Time (US & Canada)) - 2/27/2022 11:00 AM (Eastern Time (US & Canada))
Room: Virtual 1
Authors:
Suiyuan Wang, Group Member
,
,
,
,
,
,
,
,
,
Abstract
Knowledge of the fine-scale population distribution is vital for measuring impacts of socio-economic justice, optimizing resource allocation, assessing the risks of environmental exposures. The existing Geographical Random Forest that predicts population in the smallest administrative unit fails to give us a more accurate understanding of population distribution than census data. This research provides an improved Geographical Random Forest to predict population distribution at the building level using LiDAR, census, parcel, and Microsoft footprint data, then generates a population distribution map of Buffalo, NY in 2020. For data preparation, parcel and census data are analyzed by using the Cadastral-based Expert Dasymetric System to down-scale the block-level population distribution to the tax lot level, get theoretical population distribution. Meanwhile, LiDAR data are generated geometric features (height, area, roof curvature) for buildings. Then, we integrate the geometric features and theoretical population data in Microsoft building footprints. The improved Geographical Random Forests takes the socio-economic characteristics as the selection conditions of neighbors to establish a local sub-model for every building to estimate the number of residents in it. Comparing with census data to assess the accuracy, the result shows improved Geographical Random Forests performance well to predict population in building level and address the spatial heterogeneity of population. Moreover, improved Geographical Random Forest can be used in other areas to get fine-scale population distribution maps and provide a greater reference significance for resource allocation, environment protection, and decision-making.
Mapping fine-scale population distributions at building level using improved geographical random forests with multi-source open data
Category
Virtual Guided Poster Abstract
Description
This abstract is part of a session. Click here to view the session.
| Slides