01 Dec 2022
01 Dec 2022
Status: this preprint is open for discussion.

Improving numerical snowpack simulations by assimilating land surface temperature

Esteban Alonso-González1, Simon Gascoin1, Sara Arioli2, and Ghislain Picard2 Esteban Alonso-González et al.
  • 1Centre d’Etudes Spatiales de la Biosphère, Université de Toulouse,CNRS/CNES/IRD/INRA/UPS, Toulouse, France
  • 2Univ. Grenoble Alpes, CNRS, Institut des Géosciences de l’Environnement (IGE), Grenoble, France

Abstract. The assimilation of data from Earth observation satellites into numerical models is considered as the path forward to estimate SWE distribution in mountain catchments. The land surface temperature (LST) can be observed from space, but its potential to improve SWE simulations remains underexplored. This is likely due to the insufficient temporal or spatial resolution offered by the current thermal infrared (TIR) missions. However, three planned missions will provide global-scale TIR data at much higher spatio-temporal resolution in the coming years.

To investigate the value of TIR data to improve SWE estimation, we developed a synthetic data assimilation experiment at five snow-dominated sites covering a latitudinal gradient in the northern hemisphere. We generated synthetic true LST and SWE series by forcing an energy-balance snowpack model with the ERA5-Land reanalysis. We used this synthetic true LST to recover the synthetic true SWE from a degraded version of ERA5-Land. We defined different observation scenarios to emulate the revisiting times of Landsat 8 (16 days) and the Thermal infraRed Imaging Satellite for High-resolution Natural resource Assessment (TRISHNA) (3 days), while accounting for cloud cover. We replicated the experiments 100 times at each experimental site to assess the robustness of the assimilation process. We performed the assimilation using two different approaches: a sequential scheme (particle filter) and a smoother (particle batch smoother).

The results show that LST data assimilation using the smoother reduced the normalized Root Mean Square Error (nRMSE) of the simulations from 57 % (open loop) to 7 % and 3 % for 16 day revisit and 3 day revisit respectively, in the absence of clouds. We found similar but higher nRMSE values by removing observations due to cloud cover but with a substantial increase of the standard deviation of the nRMSE of the replicates, highlighting the importance of revisiting times in the stability of the assimilation output. The smoother largely outperformed the particle filter algorithm, suggesting that the capability of a smoother to propagate the information along the season is key to exploit LST information for snow modeling. These results suggest that the LST data assimilation has an underappreciated potential to improve snowpack simulations and highlight the value of upcoming TIR missions to advance snow hydrology.

Esteban Alonso-González et al.

Status: open (until 25 Feb 2023)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse

Esteban Alonso-González et al.

Esteban Alonso-González et al.


Total article views: 216 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
147 65 4 216 2 1
  • HTML: 147
  • PDF: 65
  • XML: 4
  • Total: 216
  • BibTeX: 2
  • EndNote: 1
Views and downloads (calculated since 01 Dec 2022)
Cumulative views and downloads (calculated since 01 Dec 2022)

Viewed (geographical distribution)

Total article views: 213 (including HTML, PDF, and XML) Thereof 213 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
Latest update: 27 Jan 2023
Short summary
Data assimilation techniques are a promising approach to improve snowpack simulations in remote areas that are difficult to monitor. This paper studies the ability of satellite-observed land surface temperature to improve snowpack simulations through data assimilation. We show that it is possible to improve snowpack simulations, but the temporal resolution of the observations and the algorithm used are critical to obtain satisfactory results.