In the permafrost zones along a linear extended object the thawing halos arise leading to a change in its spatial position and deformation. It is advisable to divide the forecast of the ground temperature of the route of the linear extended object into four stages.
At the first stage, a statistical analysis of the ground temperature measurements by spatial and temporal sections is carried out, correlated measurements are revealed layer by layer, time series are checked for stationarity and single roots.
At the second stage, a multidimensional space of features is visualized relative to the identified target features, using, among other things, down sampling and the principal component analysis method.
At the third stage, the most important features that carry the maximum information about the dataset are selected based on the dispersion that uncorrelated features cause in the target variable. Synthetic features are created, for example, by line-by-line construction of a linear regression model.
At the fourth stage, the predictive model of ground temperatures along the route is determined, for which the models are trained considering the time by highlighting the segment from the general data set, for which forecasting is performed. The temperature forecast is made using models considering the trend and (or) seasonal fluctuation.
Based on the experimental data obtained over a long period in the process of field measurements of ground temperatures along the route of a linear extended object, the following was determined: the actual depth of the zone of zero annual temperature fluctuations; monthly average ground temperatures at different depths; a family of models predicting cyclic changes in the ground temperature.
The results of the study allow to increase the accuracy of forecasting changes in the temperature field of ground along the linear extended object by 11 %.