ANALYSIS OF BOREHOLE JACK ICE STRENGTH DATA USING QUANTILE REGRESSION

摘要

The magnitude of ice strength is a topic that impacts the design of offshore structures and the use of ice as a construction material (ISO, 2010). There is a large body of confined compressive strength values on old ice, first year and fresh water ice obtained using Borehole Jack (BHJ) measurements in Ihe field. In some cases the corresponding values of ice temperature and ice salinity have been obtained. The ice strength measurements have often been analysed by plotting the BHJ ice strength as a function of the brine volume and performing least-squares analysis on the data. This approach has had limited success mainly due to the large scatter in the data. In least-squares analysis, the scatter is treated as an aspect that is to be minimised so that the underlying trend can be quantified. Using quantile regression in contrast, the scatter is treated as an essential part of the data. In this paper we use the quantile regression method to generate values of the BHJ strength at quantiles, from 0.05 to 0.95, using 236 strength measurements from multi-year, first-year and flooded ice. The analysis showed that the ice strength can be modelled using a linear function of root brine volume. The Normal, Log-normal, Gamma and Gumbel probability distributions were tested as suitable models for the generated quantiles. The Normal distribution was found to match the quantiles with the lowest RMS error. From the calibrated probability distribution, BHJ pressure at other quantiles can be obtained. For example at the 0.99 quantile (1% probability of exceedence) and at zero brine volume, the ice strength is 53.5 MPa. The calibrated probability distribution of the BHJ ice strength is compared with the recommendations contained in ISO 19906.