Scholarlink

<p>Urban land transformation and environmental challenges are emerging urban issues globally and regionally. Accurate simulation of future land use land cover (LULC) change is essential for understanding urbanization-driven environmental transformation in urban landscape processes. Available existing models often failed due to their inability to effectively capture and the lack of spatiotemporal integration. This study investigates spatiotemporal LULC dynamics and estimates future transformations in the Haldia urban-industrial region, India, using an integrated approach of Cellular Automata–Artificial Neural Network (CA-ANN) and AI-based Recurrent Neural Network (RNN) models processed in Geographical Information System (GIS). Multi-spectral and multi-decadal Landsat data (1991–2021) were classified into five major LULC categories—agriculture, vegetation, built-up, waterbody, and fallow land—using the Maximum Likelihood Classification (MLC) algorithm implemented in ArcGIS, QGIS, and Google Earth Engine (GEE). Classification accuracy was validated with overall accuracies ranging from 84.00% to 90.00% and Kappa coefficients from 78.95% to 86.99%. The study highlights rapid urban expansion, with built-up area increasing from 22.87 km<sup>2</sup> (22.03%) in 1991 to 53.37 km<sup>2</sup> (51.40%) in 2021, primarily driven by industrial growth. Future LULC estimates for 2031 and 2041 were simulated using CA-ANN and Long Short-Term Memory (AI-LSTM) models in TensorFlow/Keras. Results indicate continued growth of built-up land (+ 5.09% and + 3.60%), accompanied by significant losses in agriculture (− 11.16% and − 4.07%) and variable changes in vegetation, water bodies, and fallow land (2031–41) for both models, respectively. This dual-simulating approach leverages the spatial suitability of CA-ANN and the temporal sequence learning of LSTM, offering a stout prognostic framework. The findings support the urban planners and concerned authorities in contributing to sustainable urban-environmental management by analyzing and illustrating anthropogenic landscape transformation.</p>

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

Cellular Automata (CA) and AI-based Recurrent Neural Networks (RNNs) Approaches in Land Use Land Cover (LULC) Change Dynamics Using Multi-spectral and Multi-decadal Landsat Data in Haldia, India

  • 期刊论文

  • 发表日期：

    2025年9月3日

查看全文
  • Bikash Das,
  • Janki Prasad

摘要

Urban land transformation and environmental challenges are emerging urban issues globally and regionally. Accurate simulation of future land use land cover (LULC) change is essential for understanding urbanization-driven environmental transformation in urban landscape processes. Available existing models often failed due to their inability to effectively capture and the lack of spatiotemporal integration. This study investigates spatiotemporal LULC dynamics and estimates future transformations in the Haldia urban-industrial region, India, using an integrated approach of Cellular Automata–Artificial Neural Network (CA-ANN) and AI-based Recurrent Neural Network (RNN) models processed in Geographical Information System (GIS). Multi-spectral and multi-decadal Landsat data (1991–2021) were classified into five major LULC categories—agriculture, vegetation, built-up, waterbody, and fallow land—using the Maximum Likelihood Classification (MLC) algorithm implemented in ArcGIS, QGIS, and Google Earth Engine (GEE). Classification accuracy was validated with overall accuracies ranging from 84.00% to 90.00% and Kappa coefficients from 78.95% to 86.99%. The study highlights rapid urban expansion, with built-up area increasing from 22.87 km2 (22.03%) in 1991 to 53.37 km2 (51.40%) in 2021, primarily driven by industrial growth. Future LULC estimates for 2031 and 2041 were simulated using CA-ANN and Long Short-Term Memory (AI-LSTM) models in TensorFlow/Keras. Results indicate continued growth of built-up land (+ 5.09% and + 3.60%), accompanied by significant losses in agriculture (− 11.16% and − 4.07%) and variable changes in vegetation, water bodies, and fallow land (2031–41) for both models, respectively. This dual-simulating approach leverages the spatial suitability of CA-ANN and the temporal sequence learning of LSTM, offering a stout prognostic framework. The findings support the urban planners and concerned authorities in contributing to sustainable urban-environmental management by analyzing and illustrating anthropogenic landscape transformation.