Soil aggregate size distribution (ASD) is a key soil property for quantitative characterization of soil structure. Typically, dry and wet sieving methods are followed to measure mass fractions of soil aggregates under each aggregate size class and their cumulative fractions are used to estimate ASD parameters. In general, available soil ASD databases are small and a comprehensive assessment of existing models to describe ASD data is not available for soil aggregates. With objectives to comprehensively assess available ASD models, parameterize each model through pedotransfer functions (PTF), and generate ASD parameter maps for large region, we evaluated 26 models using 1373 soil ASD datasets generated from four large Indian states. The lognormal distribution-based Buchan model was observed to best describe most of our datasets. In fact, 1305 ASD datasets met the threshold criteria consisting of ‘coefficient of determination (R²) > 0.95′ and ‘corrected Akaike information criterion (AIC_c) < − 45′ for the Buchan model. Once ASD models were chosen, we used both soil properties and terrain attributes as predictors in the random forest (RF) and extreme gradient boosting models to develop PTFs for their parameters. To check whether soil properties estimated through digital soil mapping (DSM) can replace measured soil properties in the PTFs, we extracted soil properties from available DSM products. With measured soil properties and terrain attributes as predictors, RF-based PTFs yielded R² values in the range of 0.55–0.76 for ASD parameters. Comparable performance was observed when measured soil properties were replaced by the DSM-estimated soil properties. Such a result enabled us to create national digital soil maps for ASD parameters based on DSM-derived soil properties for the first time.

土壤团聚体粒度分布 (ASD) 是定量表征土壤结构的关键土壤特性。通常，按照干筛法和湿筛法来测量每个骨料尺寸等级下土壤团聚体的质量分数，并使用它们的累积分数来估计 ASD 参数。一般来说，可用的土壤 ASD 数据库很小，对现有模型的综合评估来描述 ASD 数据不适用于土壤聚集体。为了全面评估可用的 ASD 模型，通过 pedotransfer 函数 (PTF) 参数化每个模型，并为大区域生成 ASD 参数图，我们使用从四个印度大州生成的 1373 个土壤 ASD 数据集评估了 26 个模型。观察到基于对数正态分布的 Buchan 模型最能描述我们的大多数数据集。实际上，² ) > 0.95' 和 Buchan 模型的“校正 Akaike 信息标准 (AIC _c ) < - 45'。一旦选择了 ASD 模型，我们就使用土壤特性和地形属性作为随机森林 (RF) 中的预测因子，并使用极端梯度提升模型为其参数开发 PTF。为了检查通过数字土壤测绘 (DSM) 估计的土壤特性是否可以替代 PTF 中测量的土壤特性，我们从可用的 DSM 产品中提取了土壤特性。以测量的土壤特性和地形属性作为预测因子，基于 RF 的 PTF 产生 R ²ASD 参数的值在 0.55–0.76 范围内。当测量的土壤特性被 DSM 估计的土壤特性取代时，观察到了可比的性能。这样的结果使我们首次根据 DSM 衍生的土壤特性为 ASD 参数创建国家数字土壤图。