Scholarlink

<p>The dynamics between population and vegetation are critical for sustainable development, particularly in fragile mountainous ecosystems threatened by rapid urbanization. Traditional statistical models (e.g., Random Forest, ARIMA) or standalone machine learning approaches (e.g., LSTMs) often fail to capture complex spatiotemporal interdependencies, neglecting the integration of spatial attention mechanisms with recurrent networks. To address this gap, a Spatial-Temporal Graph Attention Network (ST-GAT) is proposed, combining Graph Attention Networks (GATs) and Long Short-Term Memory (LSTM) layers. MODIS-derived NDVI (250&#xa0;m resolution) and WorldPop population datasets (1&#xa0;km resolution) from 2002 to 2020 are processed, with Himachal Pradesh partitioned into <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="41324_2025_622_Article_IEq1.gif" Format="GIF" Height="17" Rendition="HTML" Resolution="72" Type="Linedraw" Width="42" /> </InlineMediaObject> <EquationSource Format="TEX">\(\text {1Km}^2\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mtext>1Km</mtext> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation> grids. GAT layers dynamically assign attention weights to neighboring grids, identifying localized spatial dependencies (e.g., urban sprawl impacting forests), while LSTMs model temporal trends. Compared to Random Forest, ConvLSTM, and FeedForwardNN, ST-GAT reduces prediction errors by 20.25%, 14.13%, and 40%, respectively. The framework provides policymakers with data-driven insights for balancing urbanization with ecological preservation, aligning with UN Sustainable Development Goals. ST-GAT establishes a precedent for scalable, spatially explicit forecasting in mountainous regions globally.</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

Deep spatio-temporal modeling of vegetation index and population density in Himachal Pradesh, India

  • 期刊论文

  • 发表日期：

    2025年6月3日

查看全文
  • Shivesh Singh,
  • Nitu Kumari,
  • Aditya Nigam

摘要

The dynamics between population and vegetation are critical for sustainable development, particularly in fragile mountainous ecosystems threatened by rapid urbanization. Traditional statistical models (e.g., Random Forest, ARIMA) or standalone machine learning approaches (e.g., LSTMs) often fail to capture complex spatiotemporal interdependencies, neglecting the integration of spatial attention mechanisms with recurrent networks. To address this gap, a Spatial-Temporal Graph Attention Network (ST-GAT) is proposed, combining Graph Attention Networks (GATs) and Long Short-Term Memory (LSTM) layers. MODIS-derived NDVI (250 m resolution) and WorldPop population datasets (1 km resolution) from 2002 to 2020 are processed, with Himachal Pradesh partitioned into \(\text {1Km}^2\) 1Km 2 grids. GAT layers dynamically assign attention weights to neighboring grids, identifying localized spatial dependencies (e.g., urban sprawl impacting forests), while LSTMs model temporal trends. Compared to Random Forest, ConvLSTM, and FeedForwardNN, ST-GAT reduces prediction errors by 20.25%, 14.13%, and 40%, respectively. The framework provides policymakers with data-driven insights for balancing urbanization with ecological preservation, aligning with UN Sustainable Development Goals. ST-GAT establishes a precedent for scalable, spatially explicit forecasting in mountainous regions globally.