How should rural residential land in metropolitan suburbs be governed in the future? Evidence from Beijing using RGB-encoded multidimensional drivers and CNN prediction

Rural residential land in metropolitan suburbs often expands, stabilizes and contracts within the same region, making static zoning inadequate for anticipatory governance. Existing land-use prediction approaches commonly use attribute vectors or remotely sensed image features, limiting their ability to reduce input dimensionality while preserving the spatial structure and combined effects of locational, accessibility and planning-related drivers. This study converts abstract geospatial drivers into red–green–blue (RGB) images and integrates them with convolutional neural network (CNN) models and ensemble prediction. Using suburban Beijing as a case, we evaluated 78 three-driver configurations across five deep-learning architectures, yielding 390 models. The modified VGG16 model achieved a weighted F1 score of 0.749. The optimal configuration combined village size, distance to the city center and expressway accessibility, indicating that rural residential land change is mainly associated with local development inertia, metropolitan gradients and external connectivity. A model using all nine drivers as nine-band input achieved a weighted F1 score of 0.758, compared with 0.749 for the three-driver RGB model; the latter used substantially fewer drivers while retaining higher precision and clearer variable semantics. Under the 2035 planning scenario, 36.57% of valid grids were predicted to increase, 33.90% to remain unchanged, 0.26% to decrease, and 29.27% were identified as uncertainty areas. These areas require flexible planning, dynamic monitoring and the avoidance of premature rigid decisions. By linking Earth Observation-derived land-change data with spatially structured driver representations, the workflow supports the explanation and prediction of land-use transitions and more adaptive governance of metropolitan suburbs.

Authors

Institutions

Publication Details

Journal
International Journal of Applied Earth Observation and Geoinformation
Published
2026-11-17
DOI
https://doi.org/10.1016/j.jag.2026.105680
Primary Topic
Land Use and Ecosystem Services
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

How should rural residential land in metropolitan suburbs be governed in the future? Evidence from Beijing using RGB-encoded multidimensional drivers and CNN prediction

Zehao Qiao, Maojun Wang, Xuexia Zhang, Weizhong Liu et al.
International Journal of Applied Earth Observation and Geoinformation
Land Use and Ecosystem Services
article

How should rural residential land in metropolitan suburbs be governed in the future? Evidence from Beijing using RGB-encoded multidimensional drivers and CNN prediction

Zehao Qiao, Maojun Wang, Xuexia Zhang, Weizhong Liu, Wenyu Xu
article en

Abstract

Rural residential land in metropolitan suburbs often expands, stabilizes and contracts within the same region, making static zoning inadequate for anticipatory governance. Existing land-use prediction approaches commonly use attribute vectors or remotely sensed image features, limiting their ability to reduce input dimensionality while preserving the spatial structure and combined effects of locational, accessibility and planning-related drivers. This study converts abstract geospatial drivers into red–green–blue (RGB) images and integrates them with convolutional neural network (CNN) models and ensemble prediction. Using suburban Beijing as a case, we evaluated 78 three-driver configurations across five deep-learning architectures, yielding 390 models. The modified VGG16 model achieved a weighted F1 score of 0.749. The optimal configuration combined village size, distance to the city center and expressway accessibility, indicating that rural residential land change is mainly associated with local development inertia, metropolitan gradients and external connectivity. A model using all nine drivers as nine-band input achieved a weighted F1 score of 0.758, compared with 0.749 for the three-driver RGB model; the latter used substantially fewer drivers while retaining higher precision and clearer variable semantics. Under the 2035 planning scenario, 36.57% of valid grids were predicted to increase, 33.90% to remain unchanged, 0.26% to decrease, and 29.27% were identified as uncertainty areas. These areas require flexible planning, dynamic monitoring and the avoidance of premature rigid decisions. By linking Earth Observation-derived land-change data with spatially structured driver representations, the workflow supports the explanation and prediction of land-use transitions and more adaptive governance of metropolitan suburbs.

International Journal of Applied Earth Observation and GeoinformationVol. 155
Peking University (CN), Beijing Forestry University (CN), Capital Normal University (CN)
National Natural Science Foundation of China
Sustainable cities and communities
Land Use and Ecosystem Services
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon