| The cryosphere is an essential part of the global terrestrial ecosystem and has an important impact on surface energy,the hydrological cycle,vegetation growth and human social development.Permafrost is an important component of the cryosphere,and in the background of global warming,permafrost has undergone more serious degradation,which has seriously affected the stability of terrestrial ecosystems and carbon cycling processes.Active layer thickness(ALT)is an important indicator of permafrost degradation and plays a particularly important role in global climate change,the water cycle and the carbon cycle.Therefore,an accurate estimation of the ALT of permafrost is extremely important.There are still large uncertainties in the estimation of ALT,and the relationship between ALT variability and individual climate factors on a hemispheric scale remains unclear.In order to investigate the spatial distribution and variability of ALT in the Northern Hemisphere,and to analyse the main influencing factors on ALT variability.In the study,ALT observations,permafrost distribution data and different remote sensing data were used as input data to simulate ALT in the northern hemisphere permafrost region from 1998 to 2015 using three machine learning algorithms.Firstly,the spatial distribution characteristics,temporal and spatial variations of ALT were analysed,secondly,the SHAP method was used to calculate the relationship between ALT in the northern hemisphere permafrost region and the importance and contribution of freezing and thawing index of air temperature,precipitation,snow depth and solar radiation.The results of the study show that the machine learning model is a more accurate method for estimating the permafrost ALT;the ALT in the Northern Hemisphere shows a fluctuating increasing trend from 1998 to 2015.The main conclusions drawn from this paper are as follows:(1)This study obtained the ALT of the Northern Hemisphere permafrost zone from 1998 to 2015 based on simulations by random forest,extreme random tree and categorical boosting algorithm machine learning models.After comparison,we found that the random forest model simulation results had the highest accuracy,with an average RMSE of 87.81 cm;an average MAE of 50.83 cm;and an average MAPE of31.15%.The spatial distribution of ALT simulated in this paper is more consistent with the simulation results of other studies,and the precision of the simulation results is also comparatively high.(2)The distribution of ALT in the Northern Hemisphere was strongly spatially heterogeneous during the study period,with changed differences with latitude.The ALT first decreases with increasing latitude,starts to rise sharply near 40°N,reaches a maximum at 50°N,and then decreases sharply,reaching a minimum at high latitudes The ALT is roughly symmetrical along the 0° longitude axis,and basically shows a symmetrical distribution.The ALT is relatively high in the northeastern hemisphere and varies considerably;it tends to rise sharply with altitude and then to fall slowly,with a marked change at altitudes of 2500-3000 m.(3)The annual ALT of the Northern Hemisphere permafrost zone from 1998 to2015 ranges from 122.28 to 130.43 cm,with a multi–year average of 123.99 cm.The active layer thickness shows a fluctuating trend of growth,with a linear growth rate of0.22 cm/year.The overall spatial variation of ALT accounted for 62.12 % of the total area of permafrost in the Northern Hemisphere,7.4 % of the area with no significant change in ALT,and 30.48 % of the area with a decrease in ALT.(4)The order of importance of climatic factors was: solar radiation > freezing index > precipitation > snow depth > thawing index.Overall,the freezing index,precipitation,snow thickness and solar radiation mainly contributed negatively to ALT during the study period,while the melting index mainly contributed positively to ALT.Solar radiation had the largest percentage of high negative and high positive contribution area. |
PDF Full Text Request