Field verification of SAR wet snow mapping in a non-Alpine environment

The James Hutton Institute

This page is no longer updated. The Macaulay Land Use Research Institute joined forces with SCRI joined forces on 1 April 2011 to create The James Hutton Institute.

Please visit the James Hutton Institute website.

This page has been mothballed.

It is no longer being updated but we've left it here for reference.

Further information

Field verification of SAR wet snow mapping in a non-Alpine environment - Introduction
Ron Caves, Andrew Hodson, Owen Turpin, Chris Clark, Rob Ferguson, Shaun Quegan

INTRODUCTION

The HYDALP project aims to use Earth Observation (EO) data to improve the monitoring and forecasting of snowmelt runoff from alpine and higher latitude basins, and to prepare a basis for the operational use of this information. A crucial variable in snowmelt runoff mod-elling is snow covered area (SCA). Optical EO methods have previously been developed for mapping SCA but their application to basins of interest is often limited by cloud cover. Cloud cover does not affect synthetic ap-erture radar (SAR) which has previously been used for mapping wet snow cover in alpine basins [1]. Total SCA must then be inferred from wet SCA. An objective of HYDALP is to determine the applicability of these meth-ods to higher latitude basins. This paper assesses the generality of the SAR wet snow mapping method by applying it to such a basin, specifically the Spey basin in the Scottish Highlands (57° N 4° W, area 1271 km2, and elevation range 198 to 1288 m).

The nature of snowpacks in the Scottish Highlands The nature of snow accumulation and ablation in the Scottish Highlands is quite different from that in the Alps. Methods developed for the latter may not work in the former where snow tends to be transitory, is often thin and unevenly distributed and can change greatly in depth and extent. In addition, the snowpack can experi-ence many melt-thaw cycles during a winter. These cre-ate buried crusts and ice lenses within the snowpack which may affect the backscatter response of dry snow. The backscatter change due to wet snow may be reduced by boulder and vegetation protrusions and snow patchi-ness even at the start of the melt season.

The main melt period in the Scottish Highlands occurs during March and April when low sun angles result in low melt rates (2 to 3 cm/day) from short wave radiative forcing, although energy may be advected into the basin by low pressure systems at any time of the year. This energy advection allows frequent melting to occur dur-ing the winter, such that it is highly unlikely that a ho-mogeneous dry snowpack will develop at any stage during snow cover accumulation and ablation. A further characteristic of snowpacks in the Scottish Highlands is a significant likelihood that the snow cover rests on a wet or even saturated substrate. This is due to the wet climate, low slope angles and poorly drained soils found in the area.

Outline We first briefly describe the SAR wet snow mapping method. The problems of missing coverage caused by layover and other geometric effects due to relief are highlighted. The results from fieldwork conducted dur-ing an ERS pass in March 1998 are then described. These are used to identify a suitable threshold for detec-tion of wet snow. The results of wet snow detection over the whole basin are then presented. Detections are lim-ited by areas of missing coverage. The area affected, its elevation dependence, and consequences are quantified.

THE SAR WET SNOW MAPPING METHOD

At C-band (5.6 cm) the radar backscatter from a surface is usually reduced when it is covered by wet snow. By comparison, dry snow is transparent at C-band and has little effect on the backscatter from the underlying surface. Thus wet snow can be detected in a C-band SAR image by comparing calibrated backscatter values, pixel by pixel, with those in a reference image from a period of no or dry snow cover. This is done by ratioing the two images and then thresholding. For wet snow detection in alpine areas a threshold of less than or equal to -3 dB is applied to intensity ratios derived from ERS C-band images [1,2]. A major aim of this paper is to determine whether this threshold can be used to detect wet snow in a non-alpine basin and if not what threshold should be used instead. As backscatter is dependent on the local incidence angle the imaging geometry must be exactly the same in the wet snow and reference images. Such repeat pass SAR image pairs are provided by the ERS series of satellites and by Radarsat. Registration of repeat pass images re-quires only translation. Prior to ratioing, images are fil-tered to reduce the effects of speckle.

The Effect of Relief In areas of high relief, such as those studied in HYDALP, areas of missing coverage arise in SAR images due to layover and radar shadow. As the intensity ratio method for wet snow detection performs poorly at local inci-dence angles less than 17° and greater than 78° we also class such areas as missing coverage [2]. The lower bound is due to foreshortening and specular effects at low incidence angles and the upper bound is due to poor signal to noise ratio at grazing angles.

It is possible to appreciably reduce missing coverage by combining images of a scene taken from different viewing directions, e.g. images from the ascending and descending passes of a spaceborne SAR. Image combi-nation is based on the optimal resolution approach (ORA) and is applied between the ratioing and thresh-olding operations [3]. It requires prior geocoding of both ratio images to an accuracy of one pixel and cal-culation of the local incidence angle at each pixel in the two geocoded ratio images. The combined ratio image is formed by selecting each pixel value from the ratio im-age with the greater incidence angle (i.e. finer ground range resolution) at that point, except in areas of shadow/grazing where the non-shadow/grazing value is always chosen, if it exist. Areas with missing coverage in both passes are masked out and are excluded from further analysis.

Temporal Constraints For the ORA combination to be used for wet snow cover detection both images must be taken close together in time during a period of little change in snow conditions, e.g. 2 to 3 days in the Alps. For ERS the minimum lag between ascending and descending passes is half a day. This occurs over the Alps, where the wet snow detection method has previously been applied. For the higher latitude basins in Scotland and Sweden studied within HYDALP, the time lags are respectively 1.5 and 6.5 days, during which snow conditions can change substantially in the respective basins. Hence, inferences on wet snow covered area based on ascending/descending pass com-binations can be less valid than in the Alps.

APPLICATION TO NON-ALPINE BASINS

ERS images have been regularly acquired to monitor wet snow cover in the Spey basin during the 1998 melt season (January to May). Fieldwork was conducted to coincide with an ERS descending pass at 11:16 on 11 March 1998 (frame/track 2457/223). There was also an ascending pass over the basin one and a half days later at 22:10 on 12 March (1143/244). No coincident field-work was conducted for this night-time pass.

Last updated: 13th October, 1998 by Roger Dunham