Speaker
Description
Information about the trafficability of gravel roads is of paramount importance for sustainable, cost-effective, and year-round delivery of timber for industrial use. For heavy vehicles like a timber truck, it is essential that the roads are in a trafficable condition when they are being driven on. The trafficability of forest roads varies depending on the season and weather conditions. Many roads become impassable during the spring thaw and following torrential rain. However, there are no models which could predict the bearing capacity of gravel roads in real-time and for the near future, i.e., the next few days.
The aim of this study was to develop and evaluate a simple dynamic bearing capacity model for gravel roads based on weather and soil type data, i.e., inputs which are readily widely available, and thus could be easily upscaled to encompass eventually a whole country. A further aim was to test the sufficiency of weather and soil type data and to evaluate the potential of calibration measurements of roads to improve the predictions.
The road data consisted of ten approximately 100-m long gravel road segments in eastern Finland. The road segments represented typical forest road conditions regarding subsoils and road structures. Data on air temperature and rainfall were recorded by nearby weather stations on hourly basis. Furthermore, bearing capacities of the road segments were sampled using a light-weight deflectometer at 20 m intervals during the 2.5-year data collection period.
The role of temperature, rainfall, soil type, and their interactions were investigated using a linear mixed-effects model. For temperature, we used a 72-hour moving average capped at the freezing point. For rainfall, we used a 48-hour moving average. The optimal number of calibration visits was estimated using a cross-validation approach in which the model was tested using one segment at a time and trained using the other nine segments.
Rainfall, temperature, and soil type alone explained about 19% of the total variance based on the linear mixed effects model. When road segment and specific location are also included, the full model can explain 48% of the variance in bearing capacity. In general, the predicted dynamical behaviour followed the trend of the bearing capacity measurements for all road segments. However, the predictions tend to have less variation than the ground measurements, especially the minima and maxima tend to be over- and underpredicted, respectively. Two calibration measurements improved the prediction accuracy by approximately 15%. After two field visits the benefit of additional visits was marginal.
The relatively simple prediction model can be further improved with additional explanatory variables, such as features extracted from LiDAR data. The calibration is independent from the used prediction model, and the field visit can be done any time. However, to get unbiased estimates when the bearing capacity is at its lowest, more measurements are needed in poor conditions.
| Keywords | “Trafficability; Forest road; Logistics“ |
|---|
