Regression Tree Models to Predict Winter Storm Costs

Abstract

Winter maintenance can consume one-third or more of highway maintenance budgets. Tools for estimating winter maintenance costs can enhance allocation, accountability, and management of expenditures. Historical weather forecasts and associated maintenance resources are used to create statistical models to estimate county-level resources to fight a forecast snow or freezing rain event. County-level analysis allows for model refinement for slightly different business practices and areas small enough to assume uniform weather effects. The statistical models are organized as regression trees that accommodate variables for operation of winter maintenance, such as service level expectations, range of county size, and weekend and overtime events. The regression trees fit subsets of data to form families of multiple linear models. In this way, models can be refined for important categorical variables such as service level and county size. The models presented here estimate labor, equipment, and material resources required to fight a storm in counties having 87 to 1,460 lane mi to maintain. The models estimate resources, not cost. Accordingly, the models are independent of unit costs of labor, equipment, and material that change over time and vary from county to county. Unit costs for labor, material, and equipment at each county are needed to convert resource estimates to resource costs. Statewide or regional storm costs can be computed by summing county-level costs.