In this study, a machine learning algorithm was introduced to fuse gridded snow depth datasets. The input variables of the machine learning method included geolocation (latitude and longitude), topographic data (elevation), gridded snow depth datasets and in situ observations. A total of 29,565 in situ observations were used to train and optimize the machine learning algorithm. A total of five gridded snow depth datasets—Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E) snow depth, Global Snow Monitoring for Climate Research (GlobSnow) snow depth, Long time series of daily snow depth over the Northern Hemisphere (NHSD) snow depth, ERA-Interim snow depth and Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2) snow depth—were used as input variables. The first three snow depth datasets are retrieved from passive microwave brightness temperature or assimilation with in situ observations, while the last two are snow depth datasets obtained from meteorological reanalysis data with a land surface model and data assimilation system. Then, three machine learning methods, i.e., Artificial Neural Networks (ANN), Support Vector Regression (SVR), and Random Forest Regression (RFR), were used to produce a fused snow depth dataset from 2002 to 2004. The RFR model performed best and was thus used to produce a new snow depth product from the fusion of the five snow depth datasets and auxiliary data over the Northern Hemisphere from 2002 to 2011. The fused snow-depth product was verified at five well-known snow observation sites. The R² of Sodankylä, Old Aspen, and Reynolds Mountains East were 0.88, 0.69, and 0.63, respectively. At the Swamp Angel Study Plot and Weissfluhjoch observation sites, which have an average snow depth exceeding 200 cm, the fused snow depth did not perform well. The spatial patterns of the average snow depth were analyzed seasonally, and the average snow depths of autumn, winter, and spring were 5.7, 25.8, and 21.5 cm, respectively. In the future, random forest regression will be used to produce a long time series of a fused snow depth dataset over the Northern Hemisphere or other specific regions.

Snow cover is a key parameter of the climate system and its significant seasonal and annual variability have significant impacts on the surface energy balance and global water circulation. However, current snow depth datasets show large inconsistencies and uncertainties, which limit their applications in climate change projections and hydrological processes simulations. In this study, a comprehensive assessment of five hemispheric snow depth datasets was carried out against ground observations from 43,391 stations. The five snow depth datasets included three remote sensing datasets, i.e., Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E), Advanced Microwave Scanning Radiometer-2 (AMSR2), Global Snow Monitoring for Climate Research (GlobSnow), and two reanalysis datasets, i.e., ERA-Interim and the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2). Assessment results imply that the spatial distribution of GlobSnow and ERA-Interim exhibit overall better agreements with ground observations than other datasets. GlobSnow and ERA-Interim exhibit less uncertainty during the snow accumulation and ablation periods, respectively. In plain and forested regions, GlobSnow, ERA-Interim and MERRA-2 show better performances, while in mountain and forested mountain areas, GlobSnow exhibits the best performance. AMSR-E and AMSR2 agree better with ground observations in shallow snow condition (0–10 cm), while MERRA-2 shows more satisfying performance when snow depth exceeds 50 cm. These systematic and integrated understanding of the five representative snow depth datasets provides information on data selection and data refinement, as well as data fusion, which is our next work of interest.

Snow albedo feedback is one of the most crucial feedback processes that control equilibrium climate sensitivity, which is a central parameter for better prediction of future climate change. However, persistent large discrepancies and uncertainties are found in snow albedo feedback estimations. Remotely sensed snow cover products, atmospheric reanalysis data and radiative kernel data are used in this study to quantify snow albedo radiative forcing and its feedback on both hemispheric and global scales during 2003–2016. The strongest snow albedo radiative forcing is located north of 30°N, apart from Antarctica. In general, it has large monthly variation and peaks in spring. Snow albedo feedback is estimated to be 0.18 ± 0.08 W∙m⁻²∙°C⁻¹ and 0.04 ± 0.02 W∙m⁻²∙°C⁻¹ on hemispheric and global scales, respectively. Compared to previous studies, this paper focuses specifically on quantifying snow albedo feedback and demonstrates three improvements: (1) used high spatial and temporal resolution satellite-based snow cover data to determine the areas of snow albedo radiative forcing and feedback; (2) provided detailed information for model parameterization by using the results from (1), together with accurate description of snow cover change and constrained snow albedo and snow-free albedo data; and (3) effectively reduced the uncertainty of snow albedo feedback and increased its confidence level through the block bootstrap test. Our results of snow albedo feedback agreed well with other partially observation-based studies and indicate that the 25 Coupled Model Intercomparison Project Phase 5 (CMIP5) models might have overestimated the snow albedo feedback, largely due to the overestimation of surface albedo change between snow-covered and snow-free surface in these models.

Snow cover is an important part of the Earth’s surface and its changes affect local and even global climates due to the high albedo and heat insulation. However, it is difficult to directly compare the results of previous studies on changes in snow cover in the Northern Hemisphere mainland (NH) due to the use of different datasets, research methods, or study periods, and a lack comparison in terms of the differences and similarities at high latitudes and high altitudes. By using snow depth datasets, we analyzed the spatio-temporal distributions and variations in snow depth (SD) and snow phenology (SP) in the NH and nine typical areas. This study revealed that SD in the NH generally decreased significantly (p < 0.01) from 1988 to 2018, with a rate of −0.55 cm/decade. Changes in SD were insignificant at high altitudes, but significant decreases were found at high latitudes. With regard to SP, the snow cover onset day (SCOD) advanced in 31.57% of the NH and was delayed in 21.10% of the NH. In typical areas such as the Rocky Mountains, the West Siberian Plain, and the Central Siberian Plateau, the SCOD presented significant advancing trends, while a significant delay was the trend observed in the Eastern European Plain. The snow cover end day (SCED) advanced in 37.29% of the NH and was delayed in 14.77% of the NH. Negative SCED trends were found in most typical areas. The snow cover duration (SCD) and snow season length (SSL) showed significant positive trends in the Rocky Mountains, while significant negative trends were found in the Qinghai–Tibet Plateau. The results of this comprehensive comparison showed that most typical areas were characterized by decreased SD, advanced SCOD and SCED, and insignificantly increasing SCD and SSL trends. The SCD and SSL values were similar at high latitudes, while the SSL value was larger than the SCD value at high altitudes. The SD exhibited similar interannual fluctuation characteristics as the SCD and SSL in each typical area. The SCD and SSL increased (decreased) with advanced (delayed) SCODs.

Litter decomposition is a vital link between material circulation and energy flow in forest ecosystems and is intensely affected by global change factors, such as increased nitrogen (N) deposition and altered precipitation regimes. As essential nutrients, calcium (Ca), magnesium (Mg), and manganese (Mn) play crucial roles in plant energy metabolism, photosynthesis, and membrane transport of plants, and the major source of these nutrients is litter decomposition. However, the dynamics of Ca, Mg, and Mn during decomposition have been largely ignored. Thus, to better understand Ca, Mg, and Mn dynamics during leaf litter decomposition in the scenario of increasing N deposition and decreasing precipitation, we carried out a two-year field litterbag experiment in a natural evergreen broad-leaved forest in the central area of the rainy area of Western China. Two levels of N deposition (ambient N deposition and 150 kg·N·ha⁻¹·y⁻¹) and precipitation reduction (no throughfall reduction and 10% throughfall reduction) were set, i.e., control (Ctr: without nitrogen deposition or throughfall reduction), N deposition (N, 150 kg·N·ha⁻¹·y⁻¹), throughfall reduction (T, 10% throughfall reduction), and N deposition and throughfall reduction (NT, 150 kg·N·ha⁻¹·y⁻¹ and 10% throughfall reduction). We found that leaf litter Ca concentration increased in the early decomposition stage and then decreased, while Mg and Mn concentrations generally decreased during the whole period of decomposition. The amount of Ca showed an accumulation pattern, while Mg and Mn generally showed a release pattern. N deposition and throughfall reduction affected the Ca, Mg, and Mn dynamics, varying with different decomposition stages; i.e., N deposition significantly affected the concentration and amount of Ca, regardless of the decomposition stages, while throughfall reduction significantly affected the Ca concentration in the whole and early decomposition stages. N deposition significantly affected the concentration and amount of Mg in the whole and early decomposition stages, while throughfall reduction had no significant effects. Throughfall reduction significantly affected the concentration and amount of Mn in the whole and late decomposition stages, while N deposition had no significant effects. Ca concentration generally showed a significant positive linear relationship with mass loss in the early decomposition stage; Mg concentration showed a significant positive linear relationship with mass loss in the Ctr and N treatments in the early and late decomposition stages; Mn generally showed a significant negative linear relationship with mass loss, regardless of the decomposition stage. Overall, the results suggest that Ca accumulation is more likely affected by N deposition, while Mg and Mn releases are more likely affected by N deposition combined with throughfall reduction, particularly in the early decomposition stage.

Volume scattering (VS) estimation plays a critical role in microwave emission modeling of the snowpack. However, it is challenging to obtain VS accurately for different frequencies by using the microwave emission model of layered snowpacks (MEMLS), which is one of the representative microwave emission models. This article develops a new VS method to consider frequency and exponential correlation length based on a snowfield campaign from November 2015 to April 2016 in Altay, China. Compared with the commonly used empirical and improved Born approximation (IBA) algorithms, the proposed VS algorithm exhibits better performances at both 18 and 36 GHz with a wide range of snow grain sizes. The bias of brightness temperatures at vertical polarization from the proposed algorithm against the observed brightness temperatures are 1.1 K and −0.4 K at 18 and 36 GHz, respectively; the root mean square errors (RMSEs) are 1.8 K and 2.6 K, respectively. The RMSEs decreased by 16.2 K at 18 GHz and 6.5 K at 36 GHz compared with those from the empirical methods and by 2.1 K and 22.2 K compared with those from the IBA. This work demonstrates that the VS difference between 18 and 36 GHz is larger and the dependence of VS on grain size is weaker than those represented by existing methods.