Downscaling of the SMOS Freeze / Thaw Product using Land Surface Models
Menard, Cecile B.1; Ikonen, Jaakko2; Rautiainen, Kimmo2; Pulliainen, Jouni2; Drusch, Matthias3
1FMI, FINLAND; 2Finnish Meteorological Institute, FINLAND; 3ESA, NETHERLANDS
Soil temperature and soil moisture content are important characteristics of the energy, water and carbon balance. Warmer air temperature, changes in snow amount and thermal characteristics are changing the soil regime in the Arctic. Increases in soil temperatures and water content are associated with thawing of the permafrost and deepening of the active layer, the layer at the top of the permafrost subject to freezing and thawing on an annual basis. As most exchanges of energy, moisture and gases between the atmosphere and the permafrost occur through this layer, changing its thermal regime could trigger the release of significant amounts of greenhouse gases to the atmosphere. Changes to the frozen soil and permafrost regimes are also likely to affect flow paths and flow rates. In zones of continuous permafrost (defined as 90-100% permafrost) the ice-rich layer acts as an impermeable layer where surface runoff causes large snowmelt-induced peak flows which contribute little to groundwater recharge. Changes in the ratio of surface to sub-surface runoff are likely to exacerbate water stresses in basins whose water supply heavily depends on snow thus forcing water management bodies in all affected regions to adapt in order to secure water availability.
ESA's water mission Soil Moisture and Ocean Salinity (SMOS) provides a unique opportunity to monitor the water balance of the Arctic region. As part of the SMOS + Innovation Permafrost project, an algorithm to detect the timing of the soil freezing and thawing was developed and successfully evaluated against ground-based L-band radiometer over the boreal forest in Finnish Lapland. However, the average spatial resolution of SMOS is 43 km (30 to 50 depending on the viewing angle) which may be appropriate for global scale studies but too low resolution for regional or local investigations. As a consequence there is a need to understand the soil freeze / thaw processes at sub-pixel scale, over different soil and land surface types.
In order to gain a better understanding of the information provided by SMOS at sub-pixel resolution, downscaling of the data is required. This is performed in a two-step method using two land surface models (LSMs), namely the Joint UK Land Environment Simulator (JULES) and the Variable Infiltration Capacity (VIC) model. Firstly, the models are evaluated at a point over single land surface types (forest, open canopy and bog) characterizing the Sodankylä area. This evaluation is performed against data collected in the FMI Pallas-Sodankylä supersite area, Finnish Lapland. The site is one of the 28 Global Station for the World Meteorological Organization Global Atmosphere Watch program, thus guaranteeing intensive and high quality meteorological and manual observations. In addition, since October 2009, the supersite has been hosting a ground-based L-band dual-polarization radiometer, the ESA-owned ELBARA-II instrument, an official reference instrument for the SMOS mission. The ability of the models to reproduce the thermodynamics and hydrological states of each surface type was evaluated by comparing modeled soil temperature, soil moisture, snow depth and water equivalent against automatic measurements for 4 years in the forested and open sites and 2 years at the bog site. Evaluation against the ELBARA-II measurements enabled the direct evaluation of the model performance against L-band radiometer data and the freeze / thaw algorithm developed as part of the SMOS + project. The two models performed well at the forest and the open sites and managed to capture diurnal, seasonal and inter-annual variations in snow water equivalence, depth and soil temperature. On the other, LSMs often fail to appropriately represent storage or high flow attenuation by ponding, wetlands or organic soils and processes are usually over-simplified, reliant on the coupling of the LSM with non-physically based models (e.g. TOPMODEL) or altogether inexistent. Therefore, in the absence of the representation of explicit wetland physical processes in the models, a number of configurations were tested before a compromise between model performance and physical processes could be reached.
The second stage of this study consisted in performing distributed model runs taking the SMOS pixel covering the Sodankylä supersite as the study area. Following the quantification of model errors in the previous stage, the aim of this phase was to gain a better understanding of sub-SMOS pixel processes using the LSMs and reconcile the high resolution model results with the lower resolution SMOS data. The models were run at different resolution (1, 5, 10, 25 and 50 km) to test the sensitivity of the downscaling process to the grid size. This method was used to identify non-linear sub-grid processes and investigate the relationship between the information provided by the SMOS data and the responses of the two models to different spatial configurations.