Validation with upward-looking sonar data

The sparsity o f observations o f ice thickness in the Arctic makes validation o f the satellite measurements a difficult task. This problem is com pounded by the fact that the inter-annual variability of thickness in the high Arctic is large [McLaren et a i ,

1994], necessitating the need for time-coincident m easurem ents from independent sources for validation puiposes.

In this section, we compare the altimeter-derived thicknesses with those deduced from moored ULS stations in the Fram Strait region, which were deployed as part o f the Arctic Ice Thickness Project (AIT?) under the Arctic Climate System Study (ACSYS) [75C Study Group on A C SYS, 1992]. Figure 7.8 shows the locations o f the stations w hich are used in this com parison, and their respective periods o f operation. Observations o f ice draft are made by these instruments at intervals o f 4 minutes. However, due to the data distribution policy, we mostly Just had access to monthly averages of ice thickness during the period of interest (M arch 1995 to present), although some o f the high frequency data was available between M arch and July 1995. These data were provided courtesy of Terje Loyning of the Norwegian Polar Institute (NPI). 20 w 1 6 'W 12'W 8 'W 4 " W 4 ’ E 3 'E 1 2 'E 8 1 ‘ N 8 1 ‘N 8 0 ‘ N 8 0 'N (08/94-07/95) ^ Stn. 1; 7 9 “N 7 9 " N Stn. 17(09/95-08/96)^ 78'N 7 8 'N 7 7 ‘ N 7 7 ” N 1 2 " E 2 0 ' W 1 6 * W 1 2 ”W 8'W 4 'W O' 4 'E 8 'E

Figure 7.8 Locations and periods of operation o f the AITP ULS stations used to validate the ice thicknesses obtained from altimetry. The dashed box indicates the region from which altim eter thickness estim ates were obtained to generate the monthly averages.

Vinje et al. [1996] describe the results of an analysis o f the ice draft measurements from these moored ULS stations between August 1990 and August 1994, and observe a minimum of ice thickness during the autumn. Figure 7.9 shows the time series of monthly ice thickness estimates from May 1995 to May 1996 from both the ULS and ER S-1/2 measurements. A draft to elevation ratio, R, of 7.35 was used in both cases, and altimeter observations from the region 78° < 0< 80° and -12° < A < -2° (indicated by the dashed box in figure 7.8) were included in the averages. Two estim ates of draft from the ULS stations are shown, one of which excludes observations made in areas of open water (for which the draft is set to zero). Regions o f open water in the draft time series are distinguished using an interactive method which scans through the observation series [Vinje et a i, 1998].

--- ■ — _ERS-1 --- ■ — _ERS-2 --- ■ — _{ULS (inc ow)} --- □ — _{ULS (ex ow)}

u o u î u î m m m m L o m m c D 0 ) G ) 0 ) 0 ^ 0 ) 0 ) 0 ) G ) 0 ) 0 ) G )

I l i I ^ 1 1 § i I I

Figure 7.9 Time series o f monthly ice thickness estimates from M ay 1995 to May 1996 from the ULS draft measurements and the ERS-1 and 2 elevation measurements. ULS estimates are shown for both the inclusion and exclusion o f regions of open water.

In figure 7.9, we see generally good agreem ent between the altim eter and ULS estimates, with the maximum ice thickness in May 1995 and the minimum in October 1995 being clearly represented. Com parisons betw een the altim eter and ULS estimates of thickness will be performed using the ULS estimates which exclude open w ater regions. W e know that the diffuse w aveform s from w hich the altim eter estimates are derived originate from the ice surface itself and not from open water, and therefore the altim eter measurements will exclude open water. (Note that this may not necessarily be the case at the ice edge where diffuse waveforms from open ocean may be treated as waveforms from ice floes, due to the use o f low resolution SSM/I-derived ice concentration fields to delineate the ice edge).

It w ould ap pear from figure 7.9 that the altim eter gen erally u n d erp red icts ice thickness by about 50 cm w hen com pared w ith the ULS m easurem ents. In order to investigate this further, we generated plots of the ice thickness frequency distributions

for both datasets. The ice thickness distribution, g{h), is defined by:

J

^1 A

w here h is the ice thickness, A is the total area o f a fixed region A about the point o f in terest, and a{hj, A2) is the area in A covered by ice w ith th ick n ess b etw een

} \ < h < ] \ [Vinje et a l , 1998]. For this exercise, we require the high frequency ULS dataset (w ith observations every 4 minutes). U nfortunately due to the sm all am ount o f overlapping altim eter and high frequency ULS data, we are restricted to m aking these com parisons using non-coincident data. (The sm all am ount o f altim eter data available during this overlap period resulted in a very noisy distribution plot). Figure 7.10 show s the d istrib ution s from the ULS m easurem ents (from N o v em b er and D ecem ber o f 1992-94) and the ERS-2 estim ates (from N ovem ber and D ecem ber o f

1995-97).

The ULS data shows a peak in ice thickness at about 2.7 m, w hich is attributed to a m ultiyear (M Y) ice draft m ode [Vinje et a l , 1996]. D ata from ER S-2 reveals a less well defined peak at around the same thickness, indicating that the altim eter is capable o f d etecting this m ode. A nother peak is evident in the ULS d ata at about 0.8 m, w hich corresponds to a first-year (FY) ice draft mode. This is not seen in the ERS-2 distribution, w hich is probably a result o f the spatial resolution o f the altim eter. For a rough surface, the pulse-lim ited footprint (PLF) o f the altim eter has a diam eter o f the order o f 10 km, w hich is effectively its lim it o f spatial resolution. O ur m ethod w ill therefore preferentially sam ple the thickness o f ice floes w hich are larger than this in order to com pletely fill the PLF and produce a diffuse echo. Sm aller, and hence probably thinner floes are therefore less likely to be detected.

(a) 5.00 n 'r > 4.00 - >' 3.00 - 1.00 - 0.00 o d o o cvi oco o OLO oCD O 00 O G) (b) c (U₃ CT 0) 10.00 1.00 0.10 0.01 ERS-2 Thickness (m) ERS-2 4.0 6.0 Thickness (m)

Figure 7.10 Frequency distribution (in percent) of ice thickness derived from both ERS-2 and ULS observations using (a) a linear scale, and (b) a semilogarithmic scale, with a bin size of 0.1 m. 56,638 observations were included in the ULS distribution calculation, and 4,611 in the ERS-2 calculation.

The resolution of the narrow -beam ULS instrum ents used here is three orders of magnitude greater than the altimeter, therefore allowing draft measurem ents from virtually every passing ice floe. For an instrument with an aperture angle of 2° and an operating depth o f 50 m, the corresponding sonar beam footprint diam eter on the underside of the ice would be 1.75 m. However, the ULS preferentially measures the drafts o f the deepest part o f the ice canopy within the footprint, resulting in an overestimate o f the mean draft [Vinje et a l, 1998]. For the system described above, this overestim ate o f draft (or footprint error) would be betw een 10 and 20 cm, depending on ice type. A correction for this was not applied to the ULS data provided by NPI, and is a probable contributor to the discrepancy seen between the altimeter and ULS estimates of ice thickness.

A notable feature in figure 7.10 is the abrupt cut-off in the altim eter thickness distribution at around 5 m, above which the ice is likely to be ridged due to various

deform ation processes [Bourke and Paquette, 1989]. It w ould appear therefore that the altim eter is unable to m ake thickness m easurem ents o f ridged ice. This is not the

case for the ULS observations, w hich can detect drafts extending up to 35 m [Vinje et

a l , 1996]. Figure 7.10(b) shows that above about 5 m, the ULS distribution becom es a negative exponential, w hich is in accordance w ith the observations o f W a d h a m s

[1992] and the m odel results of Plata and H ibler [1995].

It is thought that the absence o f ridged ice in the altim eter distribution is attributable to the com bination o f a geom etric effect, in w hich the radar signal is being scattered obliquely by the ridges and is never returned to the altim eter, and the fact that the altim eter m easures the m edian ice elevation w ithin the PLF to w hich the ridges contribute a relatively sm all surface area. From the ERS-2 thickness distribution in figure 7.10, we calculated a mean thickness o f 2.51 m, com pared w ith a ULS m ean o f 3.31 m, including ridged ice. If we exclude ridged ice by neglecting thicknesses greater than 5 m in the ULS data, we obtain a m ean o f 2.24 m, over 1 m less than w hen ridged ice is included.

In section 4.1, we stated that the altim eter tracks the m edian height o f the scatterers w ithin the PLF, and that for a G aussian height pdf, this is coincident w ith the m ean height. H ow ever, w hen the distribution becom es skew ed, for exam ple w hen ridges are present in the footprint, the m ean and m edian are no longer the sam e, and in this case the altim eter w ill underestim ate the m ean elevation. F or the ULS distribution, th e m edian ice thickness is around 2.6 m, w hich is about 70 cm low er than the corresponding m ean calculated above.

A study of the width and skewness o f the leading edge o f the diffuse w aveform s may le a d to a fu rth er u n d erstan d in g o f the effects o f rid g in g on the ra d a r echo. C om parisons w ith m odel ridging predictions, such as those o f S teiner et al. [1998], m ight also yield further clues as to w hether the ridging is even detectable in the return echoes. D econvolution o f the return echo to yield the surface height pdf, such as was

perform ed by Lipa and B arrick [1981] over the open ocean, could also be considered.

A s stated previously, the convolutional m odel assum es that a hom ogeneous scattering surface is present within the footprint w hich may not hold over deform ed ice.

In sum m ary, we have seen that the estim ates o f m ean ice thickness from ER S-2 are arou nd 50 cm low er than those from ULS observations. W e have suggested four m ain reasons for this:

(i) The altim eter PLF is too large to resolve the sm allest, and hence probably thinnest ice floes, resulting in a preferential sam pling o f thicker M Y ice.

(ii) The detection o f ridges by the altim eter m ay not be possible due th eir oblique scattering of the incident radar signal.

(iii) The altim eter actually m easures the m edian elevation o f surface scatterers w ithin the PLF, and not the mean. W hen ridging is present, the resulting median will be considerably low er than the mean.

(iv) The footprint error present in the ULS m easurem ents, w hich is dependent on ice type and season, will result in an overestim ation o f ice thickness.

Points (i) to (iii) will be the subject o f further investigations. The ULS footprint error (point (iv)) will be taken account of in future validation exercises.