Global warming has shown an Arctic amplification effect in recent decades, leading to pronounced changes in pan-Arctic soil surface temperature (SST). SST plays a direct role in energy exchange between soil and atmosphere and serves as an indicator of the land-atmosphere energy balance. Remote sensing land surface temperature (LST) data is able to indicate near-surface temperature, but influences from environment factors, such as vegetation and snow, can introduce biases between LST and SST. In this study, the importances of five environment factors (vegetation, snow, surface soil composition, topography, and solar radiation) to monthly mean SST estimation from MODIS LST in pan-Arctic were analyzed. Then a method for pan-Arctic monthly mean SST estimation from MODIS LST by incorporating these environment factors and monthly-based modeling based on random forest (RF) algorithm was proposed. The results reveal that all the selected environment factors contribute to monthly-based modeling, with vegetation exerting the greatest importance from May to October and snow in March and April. The root mean square error (RMSE) of pan-Arctic monthly SST estimated by the proposed method from 2003 to 2022 ranges from 0.89 to 1.88 degrees C, which is a 42.95---53.35 % reduction compared to the widely used season-based multivariate linear regression (MLR) models based solely on LST (RMSE between 1.56 and 4.03 degrees C). The accuracy is notably improved in areas with lower and no vegetation (grassy woodlands, grasslands, permanent wetlands, and barrens) in the cold season (September to the following April), and in higher vegetation (forests) areas in the warm season (May to August). The proposed method can contribute to producing high-precision monthly mean SST data from LST, estimating permafrost extent and active layer thickness, and understanding the land-atmosphere energy balance in pan-Arctic.

The soil freeze/thaw (FT) state has emerged as a critical role in the ecosystem, hydrological, and biogeochemical processes, but obtaining representative soil FT state datasets with a long time sequence, fine spatial resolution, and high accuracy remains challenging. Therefore, we propose a decision-level spatiotemporal data fusion algorithm based on Convolutional Long Short-Term Memory networks (ConvLSTM) to expand the SMAP-enhanced L3 landscape freeze/thaw product (SMAP_E_FT) temporally. In the algorithm, the Freeze/Thaw Earth System Data Record product (ESDR_FT) is sucked in the ConvLSTM and fused with SMAP_E_FT at the decision level. Eight predictor datasets, i.e., soil temperature, snow depth, soil moisture, precipitation, terrain complexity index, area of open water data, latitude and longitude, are used to train the ConvLSTM. Direct validation using six dense observation networks located in the Genhe, Maqu, Naqu, Pali, Saihanba, and Shandian river shows that the fusion product (ConvLSTM_FT) effectively absorbs the high accuracy characteristics of ESDR_FT and expands SMAP_E_FT with an overall average improvement of 2.44% relative to SMAP_E_FT, especially in frozen seasons (averagely improved by 7.03%). The result from indirect validation based on categorical triple collocation also shows that ConvLSTM_FT performs stable regardless of land cover types, climate types, and terrain complexity. The findings, drawn from preliminary analyses on ConvLSTM_FT from 1980 to 2020 over China, suggest that with global warming, most parts of China suffer from different degrees of shortening of the frozen period. Moreover, in the Qinghai-Tibet region, the higher the permafrost thermal stability, the faster the degradation rate.

Surface temperature is critical for the simulation of climate change impacts on the ecology, environment, and particularly permafrost in the cryosphere. Virtually, surface temperatures are different in the near-surface air temperature (T-a) measured at a screen-height of 1.5-2 m, the land surface temperature (LST) on the top canopy layer, and the ground surface temperature (GST) 0-5 cm beneath the surface cover. However, not enough attention has been concentrated on the difference in these surface temperatures. This study aims at quantifying the distinction of surface temperatures by the comparisons and numerical simulations of observational field data collected in a discontinuous permafrost region on the northeastern Qinghai-Tibet Plateau (QTP). We compared the hourly, seasonal and yearly differences between T omega, IST, GST, and ground temperatures, as well as the freezing and thawing indices, the N-factors, and the surface and thermal offsets derived from these temperatures. The results showed that the peak hourly LST was reached earliest, closely followed by the hourly T-a. Mean annual LST (MALST) was moderately comparable to mean annual T-a (MAAT), and both were lower than mean annual GST (MAGST). Surface offsets (MAGST-MAAT) were all within 3.5 degrees C, which are somewhat consistent with other parts of the QTP but smaller than those in the Arctic and Subarctic regions with dense vegetation and thick, long-duration snow cover. Thermal offsets, the mean annual differences between the ground surface and the permafrost surface, were within -0.3 degrees C, and one site was even reversed, which may be relevant to equally thawed to frozen thermal conductivities of the soils. Even with identical T-a (comparable to MAAT of -3.27 and -3.17 degrees C), the freezing and thawing processes of the active layer were distinctly different, due to the complex influence of surface characteristics and soil, textures. Furthermore, we employed the Geophysical Institute Permafrost Lab (GIPL) model to numerically simulate the dynamics of ground temperature driven by T-a, LST, and GST, respectively. Simulated results demonstrated that GST was a reliable driving indicator for the thermal regime of frozen ground, even if no thermal effects of surface characteristics were taken into account. However, great biases of mean annual ground temperatures, being as large as 3 degrees C, were induced on the basis of simulations with LST and T-a when the thermal effect of surface characteristics was neglected. We conclude that quantitative calculation of the thermal effect of surface characteristics on GST is indispensable for the permafrost simulations based on the T-a datasets and the LST products-that derived from thermal infrared remote sensing.

The accelerated warming of the Arctic climate may alter the local and regional surface energy balances, for which changing land surface temperatures (LSTs) are a key indicator. Modeling current and anticipated changes in the surface energy balance tequires an understanding of the spatio-temporal interactions between LSTs and land cover, both of which can be monitored globally by measurements from space. This paper investigates the accuracy of the MODIS LST/Emissivity Daily L3 Global 1 km V005 product and its spatio-temporal sensitivity to land surface properties in a Canadian High Arctic permafrost landscape. The land cover ranged from fully vegetated wet sedge tundra to barren rock. MODIS LSTs were compared with in situ radiometer measurements from wet tundra areas collected over a 2-year period from July 2008 to July 2010 including both summer and winter conditions. The accuracy of the MODIS LSTs was -1.1 degrees C with a root mean square error of 3.9 degrees C over the entire observation period. Agreement was lowest during the freeze-back periods where MODIS 1ST showed a cold bias likely due to the overrepresentation of clear-sky conditions. A multi-year analysis of LST spatial anomalies, i.e., the difference between MODIS LSTs and the MODIS 1ST regional mean, revealed a robust spatiotemporal pattern. Highest variability in LST anomalies was found during freeze-up and thaw periods as well as for open water surface in early summer due to the presence or absence of snow or ice. The summer anomaly pattern was similar for all three years despite strong differences in precipitation, air temperature and net radiation. Summer periods with regional mean ISTs above 5.0 degrees C showed the greatest spatial diversity with four distinct 2.0 degrees C classes. Summer anomalies ranged from -4.5 degrees C to 2.6 degrees C with an average standard deviation of 1.8 degrees C. Dry ridge areas heated up the most, while wetland areas and dry areas of sparsely vegetated bedrock with a high albedo remained coolest. The observed summer LST anomalies can be used as a baseline against which to evaluate both past and future changes in land surface properties that relate to the surface energy balance. Summer anomaly classes mainly reflected a combination of albedo and surface wetness. The potential to use this tool to monitor surface drying and wetting in the Arctic should therefore be further explored. A multi-sensor approach combining thermal satellite measurements with optical and radar imagery promises to be an effective tool for a dynamic, process-based ecosystem monitoring scheme. (C) 2015 Elsevier Inc. All rights reserved.