The sea ice information system : integrating multisensor remote sensing data with in situ and historical measurements

Symposium International de Cartographie Thématique Dérivée des Images Satellitaires, Saint-Mandé 2-4 octobre 1990 0 Intégration de la télédétection dans les SIG

Integration of remote sensing into GIS

THE SEA ICE INFORMATION SYSTEM :

INTREGATING MULTI-SENSOR REMOTE SENSING DATA WHITH IN SITU AND HISTORICAL MEASUREMENTS

Abstract

par Joseph M. PIWOWAR David G. BARBER et Ellsworth F. LEDREW Earth Observations Laboratory lnstitute for Space and T errestrial Science

Department of Geography University of Waterloo, Canada

ln this paper, we report on sorne of the issues involved in deve/oping a spatial data exploration and ana/y sis too/ which will en able us to examine the linkages between sea ice parameters - both measured and derived from remotely sensed imagery - with atmospheric variables which are required for the monitoring of climate change and variability. This too/, cal/ed the Sea lee Information System (SI/S), is being designed to integrate remote sensing data with in situ measurements and historical data extracted from scanned maps in a coherent research application which will encourage data exploration.

The design issues discussed are georeferencing, remo te sensing data integration, in situ data integration, and historical ice chart data integration. We a/so describe sorne of the design functiona/ity under the headings of display, temporal modelling, algebraic mode/ling, texture measures, and classification. ft is our intent that this discussion forms a framework for this program's implementation.

+ Introduction

Most existing GIS are designed for managers who are charged with making decisions based on map products which have been produced following sorne prescribed procedure (De Cola, 1989; Edwards et al., 1990). These systems leave little room for experimentation. There is a very different GIS user group, however, who want to model our environ ment by integrating a variety of disparate data sets with an appropriate set of assumptions. These are the research scientists. They prefer to approach problems more informally, exploring different views and combinations of the data in search of spatial relationships (Haslett et al., 1990). Openshaw et al. (1990) predict th at as more digital spatial data become available and the complex procedures for handling them become fairly trivial in conventional GIS, there will be an increase in the desire of researchers to use GIS as a tool to explore geographical data in a search for interesting results. The challenge is to incorporate the ability to explore different linked views of spatial data into a coherent package.

Atmosphere-cryosphere-hydrosphere interactions in the Arctic have been identified as a key research theme by the International Geosphere-Biosphere Programme (IGBP). One part of our involvement in this research agenda involves the development of a spatial data exploration and analysis tool, called the Se a lee Information System (SIIS), to en able us to examine the linkages between sea ice parameters- both measured and derived from remotely sensed imagery- with atmospheric variables which are requiered for the monitoring of climate change and variability.

ln this paper, we report on sorne of the issues involved in developing SIIS with the intent of designing a framework

We th en outline the data sources with which SIIS is expected to ope rate, sorne critical design concerns, and a summary of its anticipated functionality.

+ ^{The Sea} lee Information System

The primary objectives of this research is to gain an understanding of the physical processes active in the floating ice regime and to develop methods by which certain ice related indicator variables can be extracted from remote sensing imagery. The Sea lee Information System will provide a computer-based framework by which the various data sets can be efficiently manipulated and to develop a methodology by which historical data sets can be integrated with current measurements for monitoring climate change. The mandate of SIIS is to:

• provide a flexible research environment whereby in situ observations can be linked with concurrent remote sensing imagery and where synergistic image data sets can be integrated for reduction to sea ice geophysical and climatological variables; and

• provide a research environment whereby sea ice geophysical or climatological variables extracted from rem ote sensing imagery can be incorporated into climate modelling studies of the arctic floating ice regime at both local and regional scales.

Because the ïelationships between se a ica, oceans and the atmosphere are presently only crudely mode lied (Ucar, 1988) and the extent to which remote sensing data can be used as proxies for key ice and atmospheric parameters is poorly understood (Barber, et al., 1990), SIIS wil have to promote data exploration and experimentation. Commercially avai.lable GIS do not meet these requirements. We propose to build SIIS by assembling components from existing ••off-the-shelf, data analysis and management programs with custom modules do be developed as required.

SIIS will have two principal components: a data module and an analysis module (Figure 1 ). The data module will be used to assemble data from the various input file formats into the data structures required by the analysis functions.

ln Piwowar et al (1990) we describe a prototype design which transparently integrates spatial data from a variety of sources without intervention from the user. This design has the capability of making foreign data available to an analysis procedure as if they were already in that procedure's native format.

The key to flexible functionality within SIIS lies in the design of the analysis module. We note th at many of the routine functions of GIS are readily available from a variety of sources. We will incorporate as much existing technology as possible. This encourages a modular design strategy and enable function substitution and/or replacement if better programs are developed. This approach has been successful in the past (Barber, et al., 1989; Haslett, et al., 1990).

Requiered functionality which does not presently exist will be developed in-house.

REMOTE SÊNSING

1

IMAGERY

1

^{IR SitU}

J

MEASUREMENTS

, .. ---0---"'\

1 DATA 1

1 ANAlYSES 1

HlstoAicXL

ICE CHARTS

l._ - - - -

-0---.)

..---.~""L,...,~""'~'""~"=~=:----.1 ,...,--V-A-l-ID_A_TI_ON_--..1

STATISTICAL

1

¹

ANALYSES - . CLASSIFiCATION

1

Fig. 1 : The Sea lee Information System

1

+ Data Sources

SIIS will be expected to integrate spatial data from three principal sources:

in situ

measurements, remote sensing imagery, and historical ice charts:

• ln Situ Measurements : The Sea lee Monitoring Site

The Earth Observations laboratory has developed a sea ice monitoring site (SIMS) in lancaster Sound, Northwest Territories, Canada (Figure 2) to characterize the physical processes of atmosphere-cyrosphere-hydrosphere interactions and to develop the capability to measure the pertinent variables using remote sensing data (Barber,

et

al., 1990). Annual field programmes are planned for the next five years to validate the type and range of sea ice parameters which can be extracted from SAR imagery. The 1990 SIMS experiment lasted from May 15 to June 8.

Eighteen

in situ

measurements were taken at four sites on representative samples of different ice conditions (Table 1 ). Ali of the field data collected has been ente red into computer spreadsheets in preparation for the ir integration with other data sets within SIIS.

Fig. 2: Sea lee Monitoring Site: lancaster Sound, NWT, Canada

• Remote Sensing lmagery

Remotely sensed imagery, particularly in the microwave portions of the spectrum, have been identified ta be of primary importance in this study. Active microwave, Synthetic Aperture Radar (SAR), data are weil suited for the classification of sea ice because of their sensitivity to changes in a target's dielectric properties and the ir ability ta be acquired in varying illumination and weather conditions (Gudmandsen, 1985; Drinkwater and Cracker, 1988). To date, orbital SAR missions have been ad hoc and short lived th us inhibiting the creation of a consistent archive of radar imagery of arctic ice. At least two new initiatives are planned ta fill this void. ln 1991, the European Space Age ney plans ta launch the ir first satellite, ERS-1, which will carry a SAR sensor. Canada plans ta launch RADARSAT in 1994 ta provide continuous radar caver age of the arctic basin for the latter half of this decade. lt is expected th at images from these sensors will play a vital role in SIMS;

Until data from the new orbital platforms can be obtained, SAR data from Environ ment Canada's Challenger Jet will be utilized. lt is expected that these data will provide the testing grounds for much of SIIS development. Special considerations will have ta be made in the system's design ta accommodate the mosaicking of adjacent flight lines, and the data volume.

Passive microwave data have also been shawn ta be useful in identifying sea ice parameters (Carsey, 1985;

Steffen and Schweiger, 1990). lmagery from the SMMR on the Nimbus platform and the SSM/1 sen sor on the DMSP satellites will be incorporated into analysis procedures. Other remote sensing data, acquired at visible and infrared

ô"l-1

< -·

oe>Q

g~ ~~

t:l"l(")~t"l1

(") 3 ::r x

e.. .... e:

0 Vl d

0 ~~'"0

...,::r~- 0 '"0 0

:-:-- s· ::r o

~.,.. t:l"l ~ ...

§~Vl~

~ ~~g

§ ^O::s3 ~>-~'"8

. Vl B.tn

0 r::: ~·

3 ( " ) 0

oS.~

~g(")

::r3::r

3

⁰

^~-

^Vl"

^~

~0

o.

::::1.

0

a.s.

-::s

ê;e

~

ra.

s.~ oo o ...

::t

~

OS· x e_Vl

0

ss

Sorptll" -ro

5,.-d.,.,- -3

Sc. John• -1

8•111• H1rboul'

_,.,

~ ~ - ~ ~­

\b' ~1

7()'

-ro

-+

-3 -9

· »

_65"

~ i

^{19 ~}

D. E.

3X

ICE CONDITIONS FOR ÉTAT DES GLACES POUR

·o - 3 MAR / MAR 1983

"'''""'"10 ,., HM NM .SLT ICI POfllfC.AST1~G CI,..HIAl - OnAWA

wavelengths by the Landsat (MSS, TM), NOAA (AVHRR), and SPOT (HRV) missions will be used in synergy with the microwavelength data. A suite of sensors from NASA's proposed Mission to Planet Earth is also expected to provide data for this project in future years.

• Historical lee Charts

lee conditions for much of Canada's coastal waters have been mapped for over thirty years. A typical product, the composite ice chart (Figure 3), shows total and partial ice concentrations by closed polygon. These maps can have a significant impact on Arctic climatic models but they have not been included in past simulation efforts because of their ana log nature. We are developing a scanning software system for the semi-automated digitization of these data.

Once this information has be en digitally archived in an on-line database, it can become an active part of SIIS modelling.

+ Design Issues

Figure 4 highlights sorne of the problems we expect to encounter wh en we begin to integrate remote sensing data with

in situ

measurements and historical ice charts. These issues are developed below.

1.1 + km NOAA AVHRR pixel

30 m Landsat TM pixel

Fig. 4 : Data Integration Issues for SIIS

• Georeferencing

To facilitate georeferencing between the severa! data sets a corn mon coordinate system will have ta be adopted.

The two common systems fou nd in most existing programs are eastings and northings derived from a UTM (Universal Transverse Mercator) projection, and latitude and longitude. Very often the selection of coordinate system is pre- determined by the characteristics of other data sets or applications with which the data are ta be combined.

UTM georeferencing is most common in raster-based systems since a direct correspondance can be made between pixel locations in the raster array and UTM coordinates on the ground. The UTM system does experience sorne difficulty when large areas are involved since the entire grid is realigned every 15° of longitude. This means that large maps may have a discontinuous gr id system, multiple gr id systems, or erroneous gr id systems over part of the ir coverage a rea. For large a reas, therefore, latitude and longitude referencing would be preferable. Since many of the base maps of Canada's Arctic are drawn on a Lambert conie conformai projection (which already uses latitude and longitude) we decided to record coordinates using this system. Projection transformations can be used to alter the coordinate system, if required.

The grid resolution for each file will be selected such that it is close ta the original resolution of the data and th at it is an integrallinear combination of the other grid resolutions in the data set. For example, a grid resolution of 0.0001 ^o (approximately 11 m latitude) might be appropriate forthe SAR data; a resolution of 0.01 ^o (-1.1 km) could be suitable for NOAA imagery; and a grid spacing of 0.1 o (-11 km) could match the SSM/1 data. We note, however, th at ca re must be taken when manipulating raster data stored in a latitude-longitude grid since the difference between degrees of longitude is a non-linear function of latitude.

• Remote Sensing Data Integration

Combining multiple sets of remotely sensed imagery is the easiest integration task to complete since they are already formatted following a raster organization. Once the files ar georeferenced, the spatial integration of high resolution imagery with lower resolution data involves computing the wighted average of ali pixels in the first image which correspond to each cell in the second. The reverse transformation is accomplished by subdividing each lower resolution pixel through replication untilthe required pixel resolution is reached. This procedure must be exercised with caution, however, since it may be easy to geta false sence of accuracy- although the pixel size for the lower resolution image haschanged, the spatial resolution forthat data has not been altered. The re are nu merous problems associated with developing the appropriate methodology, the most fundamental being a requirement to link the spatial resolution of the synergistic data with the geophysical process under investigation.

ln addition to spatial resolution, the spectral bands recorded by each remote sensing system vary. Wh en combining imagery from separate systems, a significant amount of data duplication results. Standard image precessing techniques, such as principal components analysis, have been applied to reduce the dimensionality of the data sets (Chavez, 1989). Other studies have focused on optimal band selection useful for multispectral classification (Sheffield, 1985; Mausel et al., 1990.); they may have applications in SAR validation. We may also find band substitution techniques, such as those described in Chavez and Bowell (1988), useful in replacing a band from one imaging system for a similar wavelength channel in a second to improve the apparent spatial resolution or the spectral coverage of the data.

• lee Chart Data Integration

Linking the digital ice concentration polygons extracted from the scanned ice chart archive could take place at two levels. At a localized scale, if historical data was required at selected point locations only (such ?S the SIMS sampling sites) then this information can be retrieved in a tabular form. Regionally, however, an overall description of the ice distribution will involve, the creation of new raster files from the vector archive. One file would have to be created for each historical ice attribute required. Sin ce the region of interest can be constrained, this is not expected to be a lengthy process.

• ln situ

Data Integration

Data from the SI MS field experiments form the third major data type to be integrated. To be useful in SIIS modelling, we must assume th at these point measurements are representative of the conditions in the surrounding areas. Th at is, we must be able to extrapolate a neighbourhood of values based on one survey location. The field measurements represent ice, snow, and atmospheric conditions (Table 1) at sampling sites which were specifically situated on a variety of ice types. This configuration permits the ice condition data to be extrapolated to the visual edges (from the SAR overflights) of the ice type involved. By further assuming that the snow conditions are closely al lied with the ice parameters, we can use the same boundaries for these data as weil. lt is one of the goals of this research to verity which snow/ice parameters thus correlated with one sample of SAR imagery can be extrapolated across an entire scene. The atmospheric conditions are not constrained at ice boundaries (although the re may be sorne undetermined influences), so they can be expanded by interpolation methods (e. g. Oliver and Webster, 1990). ·

lee

•depth

•salinity profiles

•temperature profùes

•surface roughness

•structure

Snow -depth

•salinity profiles

•temperature profiles

•free water content

•structure

Atmosphere

•spectral albedo .

•incoming and outgoing radiation (short/longwave)

•wind velocity and direction

•ambient temperature

•cloud cover

•humidity

Table 1 : ln Situ Measurements Made During SIMS'90

Functionality

· Display

The hu man mi nd mi nd has unequalled powers of spatial reasoning: the display functionality of any GIS is its most powerful tool. Commercially available systems provide staticdisplays which are inadequate for our investigations. SIIS will en able the researcher to explore the data sets through multiple actively linked windows. For example, Window 1 may be used to show a SAR image overlaid with historical ice polygons and some point measurements; Window 2 could show a histogram of the same SAR image; Window 3- a scatterplot between different ice textures and the ice polygons; Window 4- the results of a proximity analysis between the in situ data and the SAR texture; Window 5 ...

Ali of these windows cou Id be dynamically linked, so that a change in one view is automatically reflected in ali views of the data set.

tt

is only through this type of data synergism that we will be able to meet our research goals.

· Temporal Modelling

To address the issue of «Change••, we must be able to relate our current findings with past information. The role of temporal modelling is a relatively new tapie in spatial data processing. ln ital work done by Langran and Chrisman (1988), Langran (1989), and Priee (1989) indicate that slicing through temporal layers can yield as much new information as slicing through data layers. A variety of parametricand non-parametric approaches will be built into SIIS.

· Algebraic Modelling

Severa! types of modelling are envisioned within SIIS. Generically, we wish to use remotely sensed imagery in severa! aspects of the modelling problem: lmagery would be used to initial ize climate models where pixel coordinates and the ir theme (classification result) would be integrated directly into the climate madel; imagery would also be used as a means of validating madel predictions and thereby fu net ion as a feedback mechanism; and finally E.-M. physical interaction models would be generated within SIIS in a framework whereby a madel image could be generad and comparisons made with observed E-M interactions. Ulaby et al (1986) derived a set of empirical equations to describe the effect of snow cover on a variety of earth surfaces, based on calibration with scatterometer observations.

o" = 0.63 -(0.36- o"is) • exp ( -0.0344 Psds sec 9') o"

=

0.20- (0.20- o"is) • exp ( -0.0198 psds sec 9') o"

=

^{1.26 -( 1.26 -}o"is) • exp ( -0.0273 psds sec 9') o"

=

0.63 - (0.63 - o"is) • exp ( -0.0373 psds sec 0') Where: Ps

=

snow density (grn!cm3)

ds

=

Snow depth (cm)

a'= Angle of refraction due to snow.

+ Conclusions

8.6 GHz, 9 = 20°

8.6 GHz, 9 = 50°.

17 GHz,

e

⁼ 20°

17 GHz, 9 = 50°

[22]

[23) Within SIIS we envision a method of [24]

[25]

computing an image from these models based on in situ observations of the parameters ps, ds, and 0'. Other, more geophysically meaningful E-M interac- tion models would also be integrated with SIIS.

When complete, the Sea lee Information System will be an flexible research utility for the analysis of sea ice as a climate change indicator. lt will have the capability to integrate data from in situ measurements with those acquired by space-borne sensors into a single operation; it will be able to compare eurre nt observations with trends derived from historical charts and tables. Our goal is to make SIIS a multi-format, multi-temporal geographie information system.

The real task in this project is in the development of a spatial analysis system which can improve our capabilities to conduct research with remote sensing, in situ and historical data. A requirement of <<change» studies is that quantitative information be utilized so that differences between <<Variation .. and «Change .. can be precisely monitored.

Dealing with the spatial variables, inherent in remote sensing imagery of the ice surface, will require extensive spatial statistical analysis capabilities, efficient data integration technology. and the capability to evaluate the interrelation ships between these various datasets statistically. Since <<creativity is the mother of invention», it is important that these new spatial analysis systems be easy to use so that impediments to exploratory data analysis are minimized.

+

Acknowledgments

This research was undertaken with the support of a Centre of Excellence grant from the Province of Ontario to Dr.

E. LeDrew through the lnstitute for Space and Terrestrial Science (ISTS), and an NSERC Operating Grant to Dr. E.

Le Drew.

References

BARBER David G., DOUGLAS DUNLOP J., PIWOWAR Joseph M. and LEDREW Ellsworth F. -Image Analysis on a Macintosh 11 ... Proceedings: IGARSS'89/12th Canadian Symposium on Remote Sensing, Vancouver Canada. Ottawa: Canadian Aeronau- tics and Spaee lnstitute,1989, pp. 150-153.

BARBER David G., PIWOWAR Joseph M .. JOHNSON Douglas D., and LEDREW Ellsworth F., 1990.-The SIMS Project: The Study of Sea lee as a Climate Change lndicator", Proeeedings: ISPRS Commission VIl Mid-Term Symposium: Global and Envi- ronmental Monitoring Techniques and Impacts, Victoria Canada. International Society for Photogrammetry and Remote Sensing, in press.

CARSEY F.O.-Summer Arctic Sea lee Character from Satellite Microwave Data ... Journal of Geophysical Research, Vol. 90, No C3, 1985, pp. 5015-5034.

CHA VEZ Pat S. Jr.-Extracting Spectral Contras! in Landsat Thematic Mapper Image Data Using Selective Principal Components Analysis .. , Photogrammetrie Engineering and Remote Sensing, Vol. 55, No 3, 1989, pp. 339-348.

CHA VEZ Pat S., Jr. and BOWELL Jo Ann- Comparison of the Spectrallnformation Content of Landsat Thematic Mapper and SPOT for Three Different Sites in the Phœnix, Arizona Regionu,1988. Photogrammetrie Engineering and Remote Sensing, Vol. 54, N° 12, pp. 1699-1708.

DE COLA Lee - Multiscale Data Models for Spatial Analysis, With Applications to Multifractal Phenomenau, Proceedings: Ninth International Symposium on Computer-Assisted Cartography (AutoCarto 9), Baltimore Maryland. Falls Church, Va.: American Society for Photogrammetry and Remote Sensing and American Congress on Surveying and Mapping, 1989. pp. 313-323.

DRINKWATER M.R. and G.B. CROC KER- Modelling Changes in the Dielectric and Scattering Properties of Young Snow-Covered Sea lee at GHz Frequencies", Journal of Glaciology, Vol. 34, No 118, 1988, pp. 274-282.

GEOFFREY Edwards, BÉDARD Yvan and EHLERS Manfred-Advanced lntegratio[l of Remote Sensing Image Analysis with Geographie Information Systems, Proeeedings: National Conference: GIS for the 1990s; Ottawa Canada. Ottawa: Canadian lnstitute of Surveying and Mapping, 1990, pp. 1574-1584.

Gudmandsen P. -Application of SAR and Other Remote Sensing Data in Studies of the Atmosphere-Ocean-lee Interaction, Proceedings: Satellite Data in Climate Models; Alpback Austria. ESA SP-244., 1985, pp. 107-111.

HASLETT John, WILLS .Graham and UNWIN Antony -SPIDER- an Interactive Statistical Tool for the Analysis of Spatially Distributed Data, lnt. J. Geographicallnformation Systems, Vol. 4, No 3, 1990, pp. 285-296.

LANGRAN Gail -A Review of Temporal Database Research and its Use in GIS Applications, lnt. J. Geographical Information Systems, Vol. 3, No 3, 1989, pp. 215"232.

LANGRAN Gail and CHRISMAN Nicholas R. -A Framework for Temporal Geographie Information. Cartographica, Vol. 25, N°3,1988, pp. 1-14.

MAUSEL P.W., KRAMBER W.J. and LEE J.K.-Optimum Band Selection For Supervised Classification of Multispectral Data, Photogrammetrie Engineering and Remote Sensing, Vol. 56, N° 1, 1990, pp. 55-60.

Oliver M.A. and WEBSTER R.- Kriging: a Method of Interpolation for Geographicallnformation Systems, lnt. J. Geographical Information Systems, Vol. 4, No 3, 1990, pp. 313-332.

OPENSHAW Stan, CROSS Anna and CHARLTON Martin -Building a Prototype Geographical Correlates Exploration Machine, lnt. J. Geographical Information Systems, Vol. 4, No 3, 1990, pp. 297-311.

PIWOWAR Joseph M., JOYCE Stephen P. and LEDREW Ellsworth F. -Designing Multi-FormatSpatial Databases, Proceedings:

GIS for the 90s: Second National Conference on Geographie Information Systems; Ottawa Canada. Ottawa: Can. lnst. Surveying and Mapping, 1990, pp. 1455-1462.

PRICE Stephen-Modelling the Temporal Element in Land Information Systems, lnt. J. Geographical Information Systems, Vol. 3, N° 3,1989, pp. 233-243.

SHEFFIELD Charles- Selecting Band Combinations from Multispectral Data, Photogrammetrie Engineering and Remo te Sensing, Vol. 51, No 12, AP6, 1985, pp. 681.

STEFFEN K. and SCHWEIGER A.J.- A Multisensor Approach to Sea lee Classification for the Validation of DMSP-SSM/1 Passive Microwave Derived Sea lee Products, Photogrammetrie Engineering and Remote Sensing, Vol. 56, No 1, 1990, pp. 75-82. UCAR -Artic Interactions: Recommandations for an Arctic Component in the International Geosphere-Biosphere Programme.

Boulder Colorado: UCAR Office for lnterdisciplinary Earth Studies, Report OIES-4, 1988, 45 p.

ULABY F.T., MOORE R.K. and FUNG A. K.- Microwave Remote Sensing: Active and Passive. Vol Ill. Addison-Wesley Publishing Company. Massachusetts, 1986.