Scholarlink

Background and aims <p>Soil salinization poses a global environmental challenge, threatening agricultural productivity and ecosystem health. Despite its widespread adoption, the predictive accuracy of remote sensing for soil salinity remains inconsistent. This study employs a global meta-analysis to identify key drivers of remote sensing-based soil salinity prediction accuracy, offering actionable recommendations for optimizing monitoring frameworks and advancing land management strategies.</p> Methods <p>A meta-analysis was conducted on 2,522 peer-reviewed studies retrieved from the Web of Science (2000—2024), with 102 studies meeting stringent inclusion criteria. Key parameters were analyzed using violin plots, boxplots, and statistical tools to investigate their relationships with prediction accuracy (quantified by R<sup>2</sup>).</p> Results <p>The highest prediction accuracy (R<sup>2</sup> = 0.797) was observed in tropical monsoon climates. Optimal performance (R<sup>2</sup> = 0.707) was achieved at shallow sampling depths (0–5&#xa0;cm and 0–10&#xa0;cm). Integrating optical satellite and unmanned aerial vehicle (UAV) data yielded the highest precision (R<sup>2</sup> = 0.840), while spatial resolutions ≤ 1&#xa0;m improved model stability. Among modeling approaches, XGBoost demonstrated superior performance (R<sup>2</sup> = 0.917). Electrical conductivity (ECₑ) measurements exhibited the most stable outcomes. Training sample size showed a non-strict linear correlation with model accuracy, whereas environmental covariates displayed synergistic effects.</p> Conclusion <p>This study recommends integrating UAV remote sensing data, with a preference for shallow soil electrical conductivity data. The modeling process employs ensemble learning models such as XGBoost, combined with key environmental covariates. Particular emphasis is placed on the accuracy of data quality as well as its compatibility with the model.</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

Unveiling critical drivers of soil salinity prediction accuracy in remote sensing: a global meta-analysis

  • 期刊论文

  • 发表日期：

    2025年7月31日

查看全文
  • Zi’ang Cui,
  • Ruiqi Zhang,
  • Wenhui Wang,
  • Zhaoce Peng,
  • Yifan Wu,
  • Ziyue Zhao,
  • Mohan Li,
  • Yutong Cong,
  • Shaoyan Zhang,
  • Zhenhai Li,
  • Lijing Han,
  • Jianli Ding

摘要

Background and aims

Soil salinization poses a global environmental challenge, threatening agricultural productivity and ecosystem health. Despite its widespread adoption, the predictive accuracy of remote sensing for soil salinity remains inconsistent. This study employs a global meta-analysis to identify key drivers of remote sensing-based soil salinity prediction accuracy, offering actionable recommendations for optimizing monitoring frameworks and advancing land management strategies.

Methods

A meta-analysis was conducted on 2,522 peer-reviewed studies retrieved from the Web of Science (2000—2024), with 102 studies meeting stringent inclusion criteria. Key parameters were analyzed using violin plots, boxplots, and statistical tools to investigate their relationships with prediction accuracy (quantified by R2).

Results

The highest prediction accuracy (R2 = 0.797) was observed in tropical monsoon climates. Optimal performance (R2 = 0.707) was achieved at shallow sampling depths (0–5 cm and 0–10 cm). Integrating optical satellite and unmanned aerial vehicle (UAV) data yielded the highest precision (R2 = 0.840), while spatial resolutions ≤ 1 m improved model stability. Among modeling approaches, XGBoost demonstrated superior performance (R2 = 0.917). Electrical conductivity (ECₑ) measurements exhibited the most stable outcomes. Training sample size showed a non-strict linear correlation with model accuracy, whereas environmental covariates displayed synergistic effects.

Conclusion

This study recommends integrating UAV remote sensing data, with a preference for shallow soil electrical conductivity data. The modeling process employs ensemble learning models such as XGBoost, combined with key environmental covariates. Particular emphasis is placed on the accuracy of data quality as well as its compatibility with the model.