Scholarlink

<p>Long-term suspended sediment concentration data are essential for understanding the response of rivers in a changing world. Here, we trained and evaluated a model that uses surface reflectance from multiple Landsat satellites (TM, ETM+, and OLI) to estimate suspended sediment concentration (SSC) in rivers over multiple decades on a global scale. First, we built the largest global matchup database recorded, consisting of remote sensing surface reflectance data and in-situ SSC (<i>N</i> = 240,224). Second, we derived empirical models for harmonizing water surface reflectance across Landsat TM, ETM+, and OLI using ~ 88&#xa0;million riverine surface reflectance observations from 1984 to 2021. After applying the harmonization to water surface reflectance in the matchup database, we developed the SSC prediction model using an extreme gradient boosting algorithm (XGBoost). The model was trained and evaluated with in-situ SSC data ranging from low to extremely turbid water (min = 0.1&#xa0;mg/L and max = 5,760&#xa0;mg/L). Our model achieved accuracy in predicting SSC comparable to or better than existing models (RMSE = 5.22&#xa0;mg/L, RMSLE = 0.24, MAPLE = 22.53%, and relative error = 0.75). Our model shows consistent spatial and temporal predictions despite being a single, global SSC algorithm, suggesting successful spatial-temporal cross-validation during model development. The model will support the development of long-term, global riverine SSC records across Landsat missions.</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

Global model for riverine suspended sediment concentration from Landsat

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月16日

查看全文
  • Punwath Prum,
  • Luisa Vieira Lucchese,
  • John Gardner

摘要

Long-term suspended sediment concentration data are essential for understanding the response of rivers in a changing world. Here, we trained and evaluated a model that uses surface reflectance from multiple Landsat satellites (TM, ETM+, and OLI) to estimate suspended sediment concentration (SSC) in rivers over multiple decades on a global scale. First, we built the largest global matchup database recorded, consisting of remote sensing surface reflectance data and in-situ SSC (N = 240,224). Second, we derived empirical models for harmonizing water surface reflectance across Landsat TM, ETM+, and OLI using ~ 88 million riverine surface reflectance observations from 1984 to 2021. After applying the harmonization to water surface reflectance in the matchup database, we developed the SSC prediction model using an extreme gradient boosting algorithm (XGBoost). The model was trained and evaluated with in-situ SSC data ranging from low to extremely turbid water (min = 0.1 mg/L and max = 5,760 mg/L). Our model achieved accuracy in predicting SSC comparable to or better than existing models (RMSE = 5.22 mg/L, RMSLE = 0.24, MAPLE = 22.53%, and relative error = 0.75). Our model shows consistent spatial and temporal predictions despite being a single, global SSC algorithm, suggesting successful spatial-temporal cross-validation during model development. The model will support the development of long-term, global riverine SSC records across Landsat missions.