Low Resolution Sea Ice Drift dataset (OSI-405)

In the context of the first Continuous Development and Operation Phase (CDOP-1), a sea ice drift product was introduced in the Ocean and Sea Ice SAF product's portofolio. It is based on low resolution (10-15 km) Passive and Active Microwave instruments (PMW and AMW) such as SSM/I, AMSR-E, ASCAT, etc...

The spatial resolution is 62.5 km on a Polar Stereographic Grid similar to the one used for the other OSISAF sea ice products. It is a 2 days (48 hours) ice drift dataset processed on a daily basis, which means that a product file measuring sea ice motion between 16th and 18th of February is available already on 19th of February. The 48 hours duration is possible thanks to the CMCC (Continuous MCC) method that was developped and implemented at the OSI SAF High Latitude centre. Single-sensor products as well as a multi-sensor dataset are made available.

1. Introduction and FAQ

How should I cite this dataset?

This dataset shall be referred to as the low resolution sea ice drift product of the EUMETSAT Ocean and Sea Ice Satellite Application Facility (OSI SAF, www.osi-saf.org).

Which satellite sensors are processed?

The sensors and channels used are AMSR-E (37 GHz H&V pol.) on board NASA platform Aqua, SSM/I (85 GHz H&V pol.) on board DMSP platform F15 and ASCAT (C-band backscatter) on board EUMETSAT platform Metop-A.

What is the spatial resolution of this product?

The low resolution sea ice drift product is a gridded dataset. The grid has 62.5 km spacing on a Polar Stereographic projection mapping. Definitions for the projection parameters can be found in the NetCDF files as well as in the Product User's Manual.

What is the time-span of this product?

Two days (48 hours). This is the time delay between the start and the stop time of the motion described by one vector. For comparison, the merged products from IFREMER/CERSAT is a 3 days lag dataset while the AMSR-E product by the same data centre is 2 days (using 89 GHz channels).

Several datasets are distributed every day, which one should I use?

The OSI SAF low resolution sea ice drift product is indeed composed of several single-sensor products and one multi-sensor analysis, every day. They are all at the same spatial resolution , on the same grid and with a 48 hours time-span.

The multi-sensor (aka merged, multi-oi) is intended for users requiring a spatial covering dataset. In this product, missing vectors are indeed interpolated from the neighbours and each vector is computed from the individual single-sensor products. In this merging process, however, some level of aliasing and averaging is to be expected that slightly degrade the quality of the dataset.

For this reason, single-sensor products are also distributed. They can be more accurate than the merged product, especially concerning the start and stop time of the motion vectors. However, it is not unlikely that some vectors will be missing or, in case of interrupted data flux, that all data will be missing. Those products are thus intended for more advanced users or assimilation techniques that can cope with missing data. The optimal assimilation setup would be to use all the single-sensor products in building the model analysis.

Today's file is empty, what is happening?

Two possible answers:

• Missing input data: In case of interrupted data link with the satellite operating centres, no input swath data are available for processing of sea ice drift data, resulting in an empty grid.
• Summer period: Due to atmospheric noise and surface melting, it is not possible to track sea ice in the Arctic during summer with the sensors/channels we process. Empty product files are distributed from 1st May to 1st October for not interrupting the data flow to operational centres. Those are the same dates as for the IFREMER/CERSAT product.

Scaling of vectors

Dates associated to the displacements

Colour codes for the vectors

Where can I find the ice motion for Antarctic regions?

The processing chain has currently not been implemented for computing sea ice motion in the Southern Hemisphere. Please let us know (contact) if you need this data.

I do not fancy NetCDF, do you have other formats?

The processing chain is currently only producing NetCDF files. Please let us know (contact) if you need another format.

2. Methodology

2.1 Daily maps

Daily maps of satellite signal are aggregated from swath data at the High Latitude processing centre. The average sensing time is also computed in order to access an estimate of the start and stop time of each drift vector. A Laplacian filter is applied on those daily images for ehancing and stabilizing the features to be tracked (similar to the strategy developped at IFREMER/CERSAT).

Example average sensing time (a), brightness temperature (b) and Laplacian filtered image (c) for the AMSR-E 37 GHz (H pol.) on 21 Dec 2008 over Fram Strait. The circular area with missing data in the upper left corner of each image is the polar observation hole, which results in a lack of drift vectors close to North Pole.

(a)

(b)

(c)

The ice mask used is the OSI-402 ice edge product. Two Laplacian filtered images, with central time two days appart from each others are the input data for the ice motion tracking algortihm (next section).

2.2 Motion tracking algorithm

Based on the well known Maximum Cross Correlation (MCC) method, the Continuous MCC (CMCC) relies on a continuous step for optimizing the components of the motion vector, located at the maximum of the cross-correlation function between a reference and a candidate blocks. In practice, virtual image pixels are interpolated from neighbouring pixels in each candidate block. The maximum point is search for in the 2D plane of valid (dx,dy) components by a "simplex" Nelder-Mead algorithm.

The main effect of using the CMCC is the removal of the quantization noise (aka tracking noise), which has hindered the retrieval of ice motion over short time spans from the same sensors. Quantization noise is responsible for MCC vector fields to look quantized, with a poor angular homogeneity and large areas with exactly the same vector. It is an artifact of the MCC algorithm.

Example 48 hours motion field obtained with the CMCC (left) and MCC (right) methods over the Beaufort Sea and Canadian Basin. Both vector fields were obtained from the same AMSR-E 37 GHz images. The quantization noise apparent in the MCC vector field is not observed in the one from the CMCC, which is much more spatially and angularly smooth. The motion vectors are from 29th to 31st Jan 2008.

(CMCC)

(MCC)

Strategies for optimally merging the information content of two polarization channels into a unique motion vector, as well as for filtering and correcting the few erroneous vectors that arise after the motion tracking steps have also been implemented. The main filtering is based on spatial consistency constraints and is built so that to allow for correcting drift vectors and thus avoiding to empty the product grid.

2.3 Multi-sensor merging strategy

1. Gaining confidence on the ice motion by a synergetic use of several instruments
2. Avoid missing data

To meet those objectives, the merging methodology is based on two steps, applied one after the other. First, all grid locations where at least one single-sensor product has a valid motion vector are assigned a merged motion vector, as a weighted average of the available estimates. Weights are related to the standard deviations of single-sensor datasets, as documented below in the validation section. For the time being, there is no auto-correlation between the grid cells.

The first merging step does not allow for filling in the cell positions where no vectors were processed by any of the single-sensor products. Such areas are present in every product files, for example in the Polar observation hole. The second merging step thus consists in a spatial interpolation of the remaining missing vectors from the motion vectors in the closest vicinity. Interpolated vectors are explicitely marked in the status_flag dataset, should a specific user want to discard them.

Although this strategy fulfills the goals introduced above, a certain level of dynamic is bound to be lost and aliased during the merging process. As a consequence, rapidly changing drift patterns, such as when a low pressure system travels over sea ice, might be better described by single-sensor datasets than in the multi-sensor product.

Validation of OSISAF low resolution sea ice drift dataset was conducted in the Arctic against in situ drifters. Trajectories were usually made available as hourly (and sub-hourly) GPS records for an optimum temporal collocation. The validation exercise extended over 3 Arctic winters: 1st October to 30th April 2006-2007, 2007-2008 and 2008-2009. Statistical results and graphs document an excellent agreement between the various OSISAF low resolution sea ice drift datasets and the reference data.

Trajectories from the Ice Tethered Profilers (ITP), the Tara schooner and the Russian manned stations NP-35 and NP-36 (AARI) were processed during this validation exercise.

Matchup statistics are summarized in the following table and graph:

Validation statistics for the single-sensor and merged datasets in the Arctic. The validation period covers 1st Oct 2006 to 30th April 2007, 1st Oct 2007 to 30th April 2008 and 1st Oct 2008 to 30th April 2009. The graph only pertains to the merged (multi-oi) dataset.

Out of those validation studies, the dataset derived from AMSR-E only compares best to ground truth. The Merged dataset has somewhat degraded quality but also much fewer data gaps. More details on the validation methodology can be found in the Validation Report.

• Product User's Manual
• Validation Report
• Algorithm Theoretical Basis Document (updated soon)

5. References