This dataset provides a set of spatial distribution products of permafrost thickness based on geothermal gradient model and machine learning inversion, targeting the complexity of the spatial distribution of permafrost thickness in the Greater and Lesser Khingan Mountains in Northeast China. The research is based on limited deep hole (>20 m) ground temperature data, and uses a ground temperature gradient model to invert the deep ground temperature of shallow holes, calculate the depth of permafrost floor, and construct a basic training dataset containing 104 stations based on this. On this basis, precipitation (PRE), surface melting index (TDD), and terrain position index (TPI) are selected as key environmental prediction factors, and the Random Forest (RF) algorithm is applied to simulate and generate them. The results showed that the average thickness of permafrost in the study area was 47.71 ± 10 m, showing a significant spatial distribution pattern of "thick in the north and thin in the south, thick in the west and thin in the east, and thick in mountainous areas and plains". This dataset provides an important reference for estimating the thickness of permafrost in restricted areas of deep hole geothermal gradient measurement.
Observation data: including ground temperature observation data from 104 stations (deep hole>20m and shallow hole data).
Environmental factors: precipitation (PRE), surface melting index (TDD), terrain location index (TPI), etc.
Ground temperature gradient modeling: Establish a ground temperature gradient model using deep hole ground temperature data.
Data inversion: Invert the deep ground temperature of shallow hole stations and calculate the depth of permafrost floor for many years.
Spatial mapping: Using the Random Forest algorithm, combined with environmental factors, to spatially map the thickness of permafrost in the entire region.
Accuracy evaluation: The classification accuracy of the random forest model is 0.74; The simulation results show that the standard deviation of the average frozen soil thickness is about ± 10 m.
This data is modeled using machine learning methods to calculate confusion matrix and overall accuracy. The results show that the model has high consistency.
| # | number | name | type |
| 1 | 2022FY100700 | Survey of Permafrost Conditions and Freeze-Thaw Damage in the High-Latitude Regions of Northeast China | Basic Resource Survey Project |
This work is licensed under
CC BY 4.0 (Creative Commons Attribution 4.0 International License).
CSTR:11738.11.NCDC.NIEER.DB7273.2026
10.12072/ncdc.nieer.db7273.2026
