Scholarlink

Background <p>The widely distributed grasslands of the Qinghai-Tibet Plateau play a vital role in the global carbon cycle and climate regulation. Gross primary productivity (GPP), an important indicator of ecosystem carbon sequestration capacity, remains highly uncertain, partly because the memory effects of environmental conditions (i.e., the influence of past states on the current GPP) are neglected. Moreover, existing models have difficulty handling multidimensional spatiotemporal data and dynamic climate responses simultaneously, leading to simulation deviations and exacerbating uncertainties. Here, we developed a deep learning model called CNN–LSTM to simulate the GPP of alpine grasslands on the plateau. By combining convolutional neural networks (CNNs) with long short-term memory (LSTM) networks, the model integrates climate and vegetation data to capture temporal and spatial characteristics.</p> Results <p>The CNN–LSTM model effectively captured spatial patterns using CNNs and temporal dependencies using LSTM, incorporating memory effects of past environmental conditions. This integration increased the GPP simulation accuracy and improved the model's ability to capture interannual variability. The training and optimization of the CNN–LSTM models demonstrated that the comprehensive memory effect length of GPP on historical climate and vegetation dynamics operated on a 4-month timescale, with the memory effects of GPP varying across environmental variables in duration and intensity. From 2001 to 2021, the annual GPP of the alpine grasslands on the plateau had a mean of 332.29&#xa0;g&#xa0;C&#xa0;m<sup>−2</sup>&#xa0;a<sup>−1</sup> and increased at 1.84&#xa0;g&#xa0;C&#xa0;m<sup>−2</sup>&#xa0;a<sup>−1</sup>. Precipitation had longer durations and higher intensities than the other factors did, and the interannual variability in GPP was influenced mainly by water conditions.</p> Conclusions <p>The CNN–LSTM model proposed in this study can effectively simulate the dynamic characteristics of GPP in alpine grasslands on the plateau. The results reflected a 4-month memory effect of the environmental variables&#xa0;on GPP. The spatiotemporal analysis revealed an increasing trend in GPP. This study emphasizes the necessity of incorporating environmental memory and spatial neighborhood features into GPP modeling, which would improve our understanding of the mechanisms driving GPP and the effects of climate change on carbon cycling in terrestrial ecosystems.</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 learning-based identification of environmental memory effects on the gross primary productivity of alpine grasslands on the Qinghai-Tibet Plateau

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月7日

查看全文
  • Qi Zhang,
  • Tao Zhou,
  • Jingyu Zeng,
  • Yajie Zhang,
  • Jingzhou Zhang,
  • Xuemei Wu,
  • E. Tan,
  • Ying Yu,
  • Yancheng Qu

摘要

Background

The widely distributed grasslands of the Qinghai-Tibet Plateau play a vital role in the global carbon cycle and climate regulation. Gross primary productivity (GPP), an important indicator of ecosystem carbon sequestration capacity, remains highly uncertain, partly because the memory effects of environmental conditions (i.e., the influence of past states on the current GPP) are neglected. Moreover, existing models have difficulty handling multidimensional spatiotemporal data and dynamic climate responses simultaneously, leading to simulation deviations and exacerbating uncertainties. Here, we developed a deep learning model called CNN–LSTM to simulate the GPP of alpine grasslands on the plateau. By combining convolutional neural networks (CNNs) with long short-term memory (LSTM) networks, the model integrates climate and vegetation data to capture temporal and spatial characteristics.

Results

The CNN–LSTM model effectively captured spatial patterns using CNNs and temporal dependencies using LSTM, incorporating memory effects of past environmental conditions. This integration increased the GPP simulation accuracy and improved the model's ability to capture interannual variability. The training and optimization of the CNN–LSTM models demonstrated that the comprehensive memory effect length of GPP on historical climate and vegetation dynamics operated on a 4-month timescale, with the memory effects of GPP varying across environmental variables in duration and intensity. From 2001 to 2021, the annual GPP of the alpine grasslands on the plateau had a mean of 332.29 g C m−2 a−1 and increased at 1.84 g C m−2 a−1. Precipitation had longer durations and higher intensities than the other factors did, and the interannual variability in GPP was influenced mainly by water conditions.

Conclusions

The CNN–LSTM model proposed in this study can effectively simulate the dynamic characteristics of GPP in alpine grasslands on the plateau. The results reflected a 4-month memory effect of the environmental variables on GPP. The spatiotemporal analysis revealed an increasing trend in GPP. This study emphasizes the necessity of incorporating environmental memory and spatial neighborhood features into GPP modeling, which would improve our understanding of the mechanisms driving GPP and the effects of climate change on carbon cycling in terrestrial ecosystems.