Scholarlink

Rapid urbanization in Bangalore demands timely and interpretable forecasting models to guide sustainable land use planning. This paper presents a two-stage hybrid framework combining Random Forest (RF) classification with Cellular Automata (CA) simulation to classify land use/land cover (LULC) and forecast near-future urban growth. We utilize high-resolution, near real-time Sentinel-2–based Dynamic World data, which provides per-pixel probability distributions across nine land cover classes—enabling more realistic training. The RF model, trained on these probabilities, produces a 2025 LULC map with 98.21% accuracy, a Kappa of 0.9752, and a mean Intersection over Union (IoU) of 0.9004. A CA-based spatial model then simulates 2026 built-up expansion using 3 × 3 neighborhood logic, predicting a 6.21% increase in urban area and a tree cover loss of 41.8 sq.km. Backcasting validation for earlier years yielded IoU scores above 0.86. Implemented in Colab using Earth Engine datasets, the pipeline offers a lightweight, interpretable, and scalable tool to support environmental monitoring and urban policy planning.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Urban Growth Forecasting and LULC Dynamics in Bangalore Using Random Forest Classification and Cellular Automata on Dynamic World Satellite Data

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Shruti Nigam,
  • T. N. Bhuvana Mohini,
  • B. S. Rangaraj,
  • Rishitha Kattipallem

摘要

Rapid urbanization in Bangalore demands timely and interpretable forecasting models to guide sustainable land use planning. This paper presents a two-stage hybrid framework combining Random Forest (RF) classification with Cellular Automata (CA) simulation to classify land use/land cover (LULC) and forecast near-future urban growth. We utilize high-resolution, near real-time Sentinel-2–based Dynamic World data, which provides per-pixel probability distributions across nine land cover classes—enabling more realistic training. The RF model, trained on these probabilities, produces a 2025 LULC map with 98.21% accuracy, a Kappa of 0.9752, and a mean Intersection over Union (IoU) of 0.9004. A CA-based spatial model then simulates 2026 built-up expansion using 3 × 3 neighborhood logic, predicting a 6.21% increase in urban area and a tree cover loss of 41.8 sq.km. Backcasting validation for earlier years yielded IoU scores above 0.86. Implemented in Colab using Earth Engine datasets, the pipeline offers a lightweight, interpretable, and scalable tool to support environmental monitoring and urban policy planning.