Identifying potential rural residential areas for land consolidation using a data driven agent-based model

Abstract

The identification of potential Rural Residential Areas for Land Consolidation (RRALC) is crucial for effective rural planning and land use management. The decision-making processes of key stakeholders, such as local governments and farmers, significantly impact the determination of RRALC. However, an effective method to simulate these behaviours of these stakeholders is still lacking. This study proposed a data driven agent-based model to identify potential RRALC more accurately. Using multi-source spatiotemporal data, gradient boosted regression trees and long short-term memory algorithms were utilized to construct the data driven agent models for governments and farmers, respectively. The model, applied in Hunan Province, China, demonstrated satisfactory performance. The government agent model achieved a mean absolute percentage error of 11.64 % and an R² of 0.9765 in RRALC area prediction. Meanwhile, the farmer agent model achieved an area under the curve of 0.968, an accuracy rate of 90.67 %, and a recall rate of 91.78 % in potential RRALC identification. Simulations suggest that by 2035, the total area of potential RRALC in Hunan Province could reach 360.50 km², accounting for 4.58 % of the total rural residential land of 2020. The potential RRALC identified are primarily located in underdeveloped regions lacking sufficient infrastructure and public services, which is consistent with the actual consolidated rural residential land in Hunan between 2009 and 2020. These findings contribute to our understanding of stakeholder relationships in land consolidation, and provide decision- making support for land consolidation and rural land use planning.