Towards large-scale daily snow density mapping with spatiotemporally aware model and multi-source data

Abstract

Abstract. Snow density plays a critical role in estimating water resources and predicting natural disasters such as floods, avalanches, and snowstorms. However, gridded products for snow density are lacking for understanding its spatiotemporal patterns. In this study, considering the strong spatiotemporal heterogeneity of snow density, as well as the weak and nonlinear relationship between snow density and the meteorological, topographic, vegetation, and snow variables, the geographically and temporally weighted neural network (GTWNN) model is constructed for estimating daily snow density in China from 2013 to 2020, with the support of satellite, ground, and reanalysis data. The leaf area index of high vegetation, total precipitation, snow depth, and topographic variables are found to be closely related to snow density among the 20 potentially influencing variables. The 10-fold cross-validation results show that the GTWNN model achieves an R2 of 0.531 and RMSE of 0.043 g cm−3, outperforming the geographically and temporally weighted regression model (R2=0.271), geographically weighted neural network model (R2=0.124), and reanalysis snow density product (R2=0.095), which demonstrates the superiority of the GTWNN model in capturing the spatiotemporal heterogeneity of snow density and the nonlinear relationship to the influencing variables. The performance of the GTWNN model is closely related to the state and amount of snow, in which more stable and plentiful snow would result in higher snow density estimation accuracy. With the benefit of the daily snow density map, we are able to obtain knowledge of the spatiotemporal pattern and heterogeneity of snow density in China. The proposed GTWNN model holds the potential for large-scale daily snow density mapping, which will be beneficial for snow parameter estimation and water resource management.

Estimating Daily Snow Density Through a Spatiotemporal Random Forest Model

“…The proposed STRF model not only absorbs spatiotemporal covariates capable of depicting spatiotemporal dependent structure inherent in snow density, but also has the ability to handle nonlinear relations and extrapolate spatiotemporally, conducive to constructing accurate snow density, as proven by validations (see Figures 4-7). It is noted that the GTWNN model depending on parametric model structure could also depict the spatiotemporal dependent structure inherent in snow density and fit in-situ snow density, which shows great interpolation ability because of the inputting of snow density observation data (Wang et al, 2023).…”

“…It is noted that the GTWNN model depending on parametric model structure could also depict the spatiotemporal dependent structure inherent in snow density and fit in‐situ snow density, which shows great interpolation ability because of the inputting of snow density observation data (Wang et al., 2023). Considering the limited observation conditions, the GTWNN model are not able to obtain snow density maps outside these years.…”

“…The typical statistical solution to address this problem involves applying suitable parametric models that explain the respective dependent structure, such as spatial or temporal autoregressive models and phylogenetic least squared regression. For example, Wang et al (2023) made progress in mapping snow density by considering the spatiotemporal dependent structure inherent in snow density based on the geographically and temporally weighted neural network (GTWNN). However, the GTWNN model is an interpolation model, which achieves inferencing unobserved snow density by exploring the spatiotemporal similarity between the unobserved site and its nearby observed samples.…”

“…Snow density serves as a pivotal parameter in the snowpack, impacting its thermal, mechanical, and optical characteristics. It plays a prominent role in substantial snow‐related studies and applications, such as water resource management, snow avalanche warning, and weather and climate forecasting (Wang et al., 2023). Snow density exhibits morbid spatiotemporal heterogeneity, which manifests across various stages of snow evolution (McCreight & Small, 2014; Yang et al., 2019).…”

Remote sensing of mountain snow from space: status and recommendations

“…Indeed, the snow density could be estimated from Landsat with an RMSE of 82 kg m −3 (Colombo et al, 2023). A machine-learning approach enabled to estimate snow density from multiple MODIS and reanalyses datasets with an RMSE of 43 kg m −3 (H. Wang et al, 2023). However, as we argue below, a more optimal method to convert snow depth to SWE is through the assimilation in a snowpack model.…”