A Spatial-statistical model to analyse historical rutting data

The rutting dataset comprises of the annual rutting for years 2010-2020 (millimetre, calculated as the difference between current and previous year's data), with rut depth measurement from the previous year (millimetre), annual average daily traffic (AADT), lane width (metre), bearing capacity for year 2021 (tonnes), surface curvature index for year 2021, and base curvature index data (2021). The rutting data was collected for 20-metre road segments at specific latitude and longitude locations. The rutting is assumed to be linearly related to known explanatory variables (e.g., lane width) and random and spatial components. Rutting measurements were used to fit spatial-statistical models with random and spatial components in a Bayesian Hierarchical framework. Non spatial-statistical models with random yearly effects were also fitted. We compared these models to determine the importance of accounting for spatial information and to properly account for the rutting variability.

本车辙数据集涵盖2010-2020年的年度车辙数据（单位：毫米，计算方式为当年与上一年数据的差值），同时包含上一年度车辙深度实测值（单位：毫米）、年平均日交通量（Annual Average Daily Traffic, AADT）、车道宽度（单位：米）、2021年道路承载能力（单位：吨）、2021年路面曲率指数以及基层曲率指数数据（2021年）。该数据集的车辙数据采集自特定经纬度坐标处的20米道路路段。研究假设车辙深度与已知解释变量（如车道宽度）、随机分量及空间分量呈线性相关关系。采用车辙实测数据，在贝叶斯分层框架（Bayesian Hierarchical Framework）下拟合包含随机分量与空间分量的空间统计模型，同时构建带有年度随机效应的非空间统计模型。通过对上述两类模型进行对比，明确考虑空间信息的必要性，并合理表征车辙的变异特征。