Weathering Intensity Model of Australia

Weathering intensity or the degree of weathering is an important characteristic of the earth’s surface that has a significant influence on the chemical and physical properties of surface materials. Weathering intensity largely controls the degree to which primary minerals are altered to secondary components including clay minerals and oxides. The degree of surface weathering is particularly important in Australia where variations in weathering intensity correspond to the nature and distribution of regolith (weathered bedrock and sediments) which mantles approximately 90% of the Australian continent.

The weathering intensity prediction has been generated using the Random Forest decision tree machine learning algorithm. The algorithm is used to establish predictive relationships between field estimates of the degree of weathering and a comprehensive suite of covariate or predictive datasets. The covariates used to generate the model include satellite imagery, terrain attributes, airborne radiometric imagery and mapped geology. Correlations between the training dataset and the covariates were explored through the generation of 300 random tree models. An r-squared correlation of 0.85 is reported using 5 K-fold cross-validation. The mean of the 300 models is used for predicting the weathering intensity and the uncertainty in the weathering intensity is estimated at each location via the standard deviation in the 300 model values.

The predictive weathering intensity model is an estimate of the degree of surface weathering only. The interpretation of the weathering intensity is different for in-situ or residual landscapes compared with transported materials within depositional landscapes. In residual landscapes, weathering process are operating locally whereas in depositional landscapes the model is reflecting the degree of weathering either prior to erosion and subsequent deposition, or weathering of sediments after being deposited.

The weathering intensity model has broad utility in assisting mineral exploration in variably weathered geochemical landscapes across the Australian continent, mapping chemical and physical attributes of soils in agricultural landscapes and in understanding the nature and distribution of weathering processes occurring within the upper regolith.

风化强度（weathering intensity），或称风化程度，是地球表层的重要特征之一，对地表物质的化学与物理性质具有显著影响。风化强度在很大程度上控制着原生矿物向次生组分（包括黏土矿物与氧化物）转化的程度。地表风化程度在澳大利亚尤为关键：该国约90%的陆地被风化基岩与沉积物构成的风化壳（regolith）所覆盖，而风化强度的变化与该风化壳的性质及分布直接相关。

本风化强度预测结果基于随机森林决策树机器学习算法（Random Forest decision tree machine learning algorithm）生成。该算法用于构建风化程度野外估算值与一系列完整的协变量（covariate，即预测数据集）之间的预测关联关系。用于构建模型的协变量包括卫星影像、地形属性、航空放射性影像与填绘地质数据。研究通过生成300棵随机树模型，探究了训练数据集与协变量之间的相关性。经5折交叉验证（5 K-fold cross-validation），模型决定系数（r-squared）可达0.85。本研究以300个模型的均值作为风化强度的预测结果，并通过300个模型预测值的标准差，估算每个采样点处风化强度的不确定性。

本预测风化强度模型仅针对地表风化程度进行估算。对于原地残留景观与沉积景观中的搬运物质，风化强度的解读方式存在差异。在残留景观中，风化过程仅在局地发生；而在沉积景观中，模型所反映的要么是侵蚀与后续沉积前的风化程度，要么是沉积物沉积后的风化程度。

该风化强度模型具有广泛的应用场景：可辅助澳大利亚大陆不同风化程度的地球化学景观中的矿产勘探，绘制农业景观中土壤的化学与物理属性分布图，以及探究上层风化壳内风化过程的性质与分布。