Abstract
Monitoring the snow coverage and understanding the internal snowpack processes is crucial in mountainous areas for the role of snow as water reservoir in hydrological cycle. Traditional methods for snow properties retrieval are based on pointwise observations collected by operators or at nivometeorological automatic stations. However, both acquisition methods are limited in space and cannot provide information in remote areas. To overcome these limitations, a commonly used approach is the use of physically based models. Although extensively tested and validated in gauged areas, these models are still affected by uncertainties in spatially distributed applications, due, among other things, to their connection to meteorological data, which are pointwise observations and need to be spatialized to the regional scale.
Remote sensing (RS) can make a valuable contribution, by providing information at higher spatial and temporal resolution. Nevertheless, remote sensing has some limitations. Snow detection with optical sensors in clouded or shadowed areas and under the canopy is a still open issue for scientific community. Moreover, optical sensors are not suitable to obtain volumetric information about snowpack and for this purpose microwave sensors are needed. In particular, the most commonly used methods for retrieving volumetric snow properties from the space rely on passive microwave sensors whose use is limited by the poor spatial resolution that implies difficulty in considering mixed pixel effects over heterogeneous landscapes, such as in mountainous areas.
Accordingly, the combined use of RS and physical models could be the suitable solution to overcome the cited limitations and exploit the strengths of both methods.
The aim of this work is to develop new methods for improving the estimation of three cryospheric parameters by exploiting the physically based model AMUNDSEN [1], optical RS products and, as reference samples, the collected ground observations. In particular, the three parameters addressed in this study are: the snow cover area (SCA), the snow water equivalent (SWE) and the glaciers mass balance (GMB), over the study area including the Tyrol, the South Tyrol and the Trentino regions. Differently from standards application where RS is used for model calibration of data assimilation, the proposed fusion scheme is based on machine learning techniques, always involving the final products of both RS and model outcomes. Results derived from developed methods are promising for all three parameters, by showing a general performances improvement: for SCA, the average agreement between the fused product and the reference ground data is of 96%, compared to 90% of the RS product and 92% of the AMUNDSEN simulation; regarding the SWE, the proposed method decreases, with respect to the AMUNDSEN simulations, the RMSE and the MAE with ground data, on average, from 154 to 75 mm and from 99 to 45 mm, respectively. Finally, the significant improvement obtained on the annual GMB estimation with respect to the AMUNDSEN model in terms of RMSE (i.e., 195mm vs 475mm), demonstrates the importance of integrating the RS data and the in-situ measurements with the physical model.
Remote sensing (RS) can make a valuable contribution, by providing information at higher spatial and temporal resolution. Nevertheless, remote sensing has some limitations. Snow detection with optical sensors in clouded or shadowed areas and under the canopy is a still open issue for scientific community. Moreover, optical sensors are not suitable to obtain volumetric information about snowpack and for this purpose microwave sensors are needed. In particular, the most commonly used methods for retrieving volumetric snow properties from the space rely on passive microwave sensors whose use is limited by the poor spatial resolution that implies difficulty in considering mixed pixel effects over heterogeneous landscapes, such as in mountainous areas.
Accordingly, the combined use of RS and physical models could be the suitable solution to overcome the cited limitations and exploit the strengths of both methods.
The aim of this work is to develop new methods for improving the estimation of three cryospheric parameters by exploiting the physically based model AMUNDSEN [1], optical RS products and, as reference samples, the collected ground observations. In particular, the three parameters addressed in this study are: the snow cover area (SCA), the snow water equivalent (SWE) and the glaciers mass balance (GMB), over the study area including the Tyrol, the South Tyrol and the Trentino regions. Differently from standards application where RS is used for model calibration of data assimilation, the proposed fusion scheme is based on machine learning techniques, always involving the final products of both RS and model outcomes. Results derived from developed methods are promising for all three parameters, by showing a general performances improvement: for SCA, the average agreement between the fused product and the reference ground data is of 96%, compared to 90% of the RS product and 92% of the AMUNDSEN simulation; regarding the SWE, the proposed method decreases, with respect to the AMUNDSEN simulations, the RMSE and the MAE with ground data, on average, from 154 to 75 mm and from 99 to 45 mm, respectively. Finally, the significant improvement obtained on the annual GMB estimation with respect to the AMUNDSEN model in terms of RMSE (i.e., 195mm vs 475mm), demonstrates the importance of integrating the RS data and the in-situ measurements with the physical model.
Original language | English |
---|---|
Pages | 1 |
Publication status | Published - 31 Jan 2020 |
Externally published | Yes |
Event | International Conference on Snow Hydrology, SnowHydro 2020 - Bolzano, Italy Duration: 28 Jan 2020 → 31 Jan 2020 https://snowhydro.eurac.edu/ |
Conference
Conference | International Conference on Snow Hydrology, SnowHydro 2020 |
---|---|
Abbreviated title | SnowHydro 2020 |
Country/Territory | Italy |
City | Bolzano |
Period | 28/01/20 → 31/01/20 |
Internet address |
Keywords
- ITC-CV