Reliably Flattened Radar Backscatter for Wet Snow Mapping from Wide-swath Sensors
Small, David1; Miranda, Nuno2; Ewen, Tracy3; Jonas, Tobias4
1University of Zurich, SWITZERLAND; 2European Space Agency, ITALY; 3University of Zurich, Hydrology and Climate, SWITZERLAND; 4WSL Institute for Snow and Avalanche Research SLF, SWITZERLAND

Modern SAR sensors generally include ScanSAR modes with wide swaths (WS). Such imagery typically has reduced spatial resolution in comparison to conventional strip-map (SM) radar modes, but offers significantly better temporal resolution by enabling denser regional revisit rates. It is true that an operator of a radar satellite with only one small target of interest can image his point just as often by mixing an optimised selection of strip-map acquisitions tuned to his target's position. However, coverage of a wider region of interest at similar temporal resolution can only be achieved using WS modes. Managers of finite satellite resources serving monitoring applications that require frequent revisit therefore increasingly use WS imagery.

ENVISAT's Advanced Synthetic Aperture Radar (ASAR) was the first spaceborne sensor to make WS imagery a commonplace, providing imagery to a broad spectrum of users. Together with ASAR's unprecedented tiepoint-free geolocation accuracy [1][5], the ASAR WS and global monitoring modes opened up new types of applications. The highest revisit rate from wider swaths is achieved by combining information from multiple satellite tracks, rather than simply overlaying backscatter values from repeat-pass imagery acquired from a single track. Combining measurements of backscatter from hilly areas acquired on multiple tracks is only possible if the effects of terrain on the radiometry are first modelled and corrected. The terrain-flattened ã0 [4] radiometric terrain correction (RTC) method [3] was applied to a time-series of hundreds of ASAR WS images acquired of the territory of Switzerland from 2002-2012. The robustness of the RTC-process was tested by comparing backscatter estimates derived from ground and slant-range input products. Local resolution weighting (LRW) [2], an added step used to trade off temporal resolution for improved homogeneity in quality, resolution, and noise properties of backscatter estimates using ascending/descending combinations was applied. Composite backscatter maps were generated from moving time windows and used to illustrate seasonal changes. Given the resulting relatively "level playing field", the backscatter estimates became intercomparable, and were applied to wet snow mapping during the springtime melt period. Backscatter is compared with a dry reference to map affected areas [7].

Beginning on 01-MAR-2012, the EOPI project 10331 began an unprecedentedly dense series of ASAR WS acquisitions over Switzerland. A dense time series enables optimal application of a combination of RTC and LRW techniques: typically approximately a 3-day composite revisit rate was achieved. From March to April 2012 ASAR WS products were downloaded from ESA and processed to the RTC level within a day. The orbital state vectors used were those included in the product, typically AUX_FRO (restituted), or occasionally AUX_FPO (predicted). There were exceptional cases where an old predicted quality state vector set was included in the product, and this negatively impacted the quality of the resulting RTC products. However, within 3 days, the DORIS preliminary orbit products were always available: these were used to regenerate RTC images for the complete dataset with a separate 3-day lag. A single RTC image is shown in Fig. 1.

The wet snow maps were used to constrain runoff generation for several unmanaged Alpine catchments using a conceptual rainfall-runoff model. For each ASAR image update, the snowmelt process in the model was adjusted according to whether or not snowmelt was detected in the wet snow map. Model performance was assessed based on observed versus simulated runoff with and without adjustments to snowmelt timing derived from the wet snow maps. Preliminary results from three Alpine test catchments, Wägital, Ova dal Fuorn and Verzasca in the Swiss Alps show improved runoff forecasts using updates from the wet snow maps during spring snowmelt periods.

Additionally, wet snow maps at the national scale were generated and integrated for the first time on an operational basis within the Swiss Institute for Snow and Avalanche Research SLF. Two separate streams (1- & 3-day lags) of RTC products were uploaded continually to SLF for integration in their operational analysis. These products were integrated with a snowmelt model to map areas that release snowmelt runoff. These maps are part of a series of operational products that SLF provides to the national flood forecasting authorities.

Unfortunately, it was not possible to complete the project's planned programme of acquisitions due to the loss of contact with ENVISAT on 08-APR-2012. Although that cut short the planned project, it was nevertheless possible to learn lessons concerning near-real time integration.

  • Poor state vector quality (e.g. through use of predicted state vectors generated 3 days before the acquisition) can delay provision of higher level products that depend on accurate geolocation
  • No other sensor was able to substitute for the loss of the dense ASAR time series after the demise of ENVISAT. Access to Radarsat-2 ScanSAR data is expensive and uncertain due to many conflicts in central Europe.

    The lessons will be considered in future projects. The Sentinel-1 (S-1) satellites will use the TOPSAR approach [4]: Interferometric wide swath mode (IW) with a 250km swath and 20m resolution, and extra wide (EW) mode with over 400km swath and 50m resolution. As S-1 acquisitions will emphasise IW acquisitions over SM, the mission will be able to offer higher spatial and temporal resolution than was possible with ASAR. Implications for future WS-capable missions such as S-1, ALOS2 and the Radarsat Constellation Mission are discussed together with possible experiments using CoReH2O data.

    [1]Miranda N. et al., 2013, The Envisat ASAR Mission: A Look Back at 10 Years of Operation, Proc. ESA Living Planet Symposium, Edinburgh, Scotland.

    [2]Small, D., 2012. SAR backscatter multitemporal compositing via local resolution weighting. IGARSS Munich, pp. 4521-4524.

    [3]Small, D., 2011. Flattening Gamma: Radiometric Terrain Correction for SAR Imagery. IEEE TGRS, 49(8), pp. 3081-3093.

    [4]Small, D., Miranda, N. & Meier, E., 2009. A revised radiometric normalisation standard for SAR. IGARSS Cape Town, South Africa, pp. 566-569.

    [5]Schubert, A. et al., 2008. ASAR product consistency and geolocation accuracy. Proc. CEOS WGCV SAR Cal/Val Workshop. Oberpfaffenhofen, 6p.

    [6]De Zan, F. & Monti Guarnieri, A.M., 2006. TOPSAR: Terrain Observation by Progressive Scans. IEEE TGRS, 44(9), pp. 2352-2360.

    [7]Nagler, T. & Rott, H., 2000. Retrieval of wet snow by means of multitemporal SAR data. IEEE TGRS, 38(2), pp. 754-765.

    Fig. 1 ASAR WS 07-APR-2012 Terrain-flattened γ0 Backscatter of Switzerland from the day before contact was lost with ENVISAT