Scholarlink

<p>Soil prediction is vital for sustainable agriculture, environmental management, and land-use planning. This systematic review synthesizes research from 2014 to 2024 on advanced soil prediction techniques, including High-Accuracy Surface Modeling (HASM), Machine Learning with Euclidean Distance Fields (ML-EDF), and Bayesian Maximum Entropy (BME). No previous review has compared HASM, ML-EDF, and BME together. Analyzing 190 studies, we identified a predominant focus on machine learning (28% of studies) and significant gaps in data source reporting (34% of studies did not specify sources) and the consideration of soil nutrients (36% of studies omitted them). Addressing a gap in the literature, this systematic review is the first to explicitly compare the three approaches together for soil property prediction. Bibliometric analysis revealed strong collaborative networks, particularly in China, while performance analysis indicated that HASM, ML, and BME are top-performing but context-dependent. Moreover, the plot-level vs. regional performance results BME has strong performance in handling spatial dependency and outliers, but is sensitive to skewness and sample size; HASM is highly dependent on bountiful sample size and less robust to outliers; MLE is robust to skewness and outliers, at large sample size, but its performance is irregular with spatial dependency. This review critically examines the methodologies, applications, and performance of these approaches, highlighting their strengths, limitations, and future directions. Therefore, the review concludes by highlighting the need for improved data transparency and more comprehensive nutrient integration to advance the field. The findings also provide a clear choice of modeling techniques, validation against empirical data, and consideration of spatial relationships are critical for improving soil predictions. Thus, there should be a hybrid approach for digital soil mapping, as it effectively integrates sparse soil samples with remote sensing indices to accurately predict key soil properties across diverse landscapes and continue to explore these avenues, fostering collaboration across disciplines to enhance our understanding and prediction of soil application.</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

A systematic review of high accuracy surface modeling Euclidean enhanced machine learning and Bayesian maximum entropy for soil property prediction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月23日

查看全文
  • Kenenisa Abdisa Kuse,
  • Codjo Emile Agbangba,
  • Romain Glèlè Kakaï

摘要

Soil prediction is vital for sustainable agriculture, environmental management, and land-use planning. This systematic review synthesizes research from 2014 to 2024 on advanced soil prediction techniques, including High-Accuracy Surface Modeling (HASM), Machine Learning with Euclidean Distance Fields (ML-EDF), and Bayesian Maximum Entropy (BME). No previous review has compared HASM, ML-EDF, and BME together. Analyzing 190 studies, we identified a predominant focus on machine learning (28% of studies) and significant gaps in data source reporting (34% of studies did not specify sources) and the consideration of soil nutrients (36% of studies omitted them). Addressing a gap in the literature, this systematic review is the first to explicitly compare the three approaches together for soil property prediction. Bibliometric analysis revealed strong collaborative networks, particularly in China, while performance analysis indicated that HASM, ML, and BME are top-performing but context-dependent. Moreover, the plot-level vs. regional performance results BME has strong performance in handling spatial dependency and outliers, but is sensitive to skewness and sample size; HASM is highly dependent on bountiful sample size and less robust to outliers; MLE is robust to skewness and outliers, at large sample size, but its performance is irregular with spatial dependency. This review critically examines the methodologies, applications, and performance of these approaches, highlighting their strengths, limitations, and future directions. Therefore, the review concludes by highlighting the need for improved data transparency and more comprehensive nutrient integration to advance the field. The findings also provide a clear choice of modeling techniques, validation against empirical data, and consideration of spatial relationships are critical for improving soil predictions. Thus, there should be a hybrid approach for digital soil mapping, as it effectively integrates sparse soil samples with remote sensing indices to accurately predict key soil properties across diverse landscapes and continue to explore these avenues, fostering collaboration across disciplines to enhance our understanding and prediction of soil application.