The prediction and control of river sediment yield (SY) are critical but challenging tasks. Erosion and sediment transfer in river catchments are controlled by different processes, whose relative importance varies in space and time. We thus put forward that SY can be estimated more efficiently by using explicitly the information contained in the similarity within groups. To test this hypothesis, we developed a novel Bayesian hierarchical model, applied it to a sample of heterogeneous river catchments and compared its fixed-effects and mixed-effects performance incorporating different group levels, namely gauges, rivers, basins and clusters of catchments. With a parsimonious linear model consisting of four variables (specific and extreme discharge, elevation and retention coefficient), we achieved good performance criteria in the calibration (NSE: 0.79–0.85) and in the cross-validation for temporal and spatial prediction (NSE: 0.71 and 0.72, respectively). These results support the promising potential of this technique.

预测和控制河流沉积物产量（SY）是至关重要但具有挑战性的任务。流域的侵蚀和泥沙输送受不同过程控制，其相对重要性随时间和空间而变化。因此，我们提出可以通过显式使用组内相似度中包含的信息来更有效地估计SY。为了检验该假设，我们开发了一个新颖的贝叶斯层次模型，将其应用于非均质河流域的样本，并比较了其在不同组级别（即表，河流，流域和流域群）的固定效应和混合效应表现。使用包含四个变量（比容和极限流量，高程和保留系数）的简约线性模型，我们在校准中达到了良好的性能标准（NSE：0.79–0.85）和时间和空间预测的交叉验证（NSE：分别为0.71和0.72）。这些结果证明了该技术的潜力。