TS-CRC Student project - Integration of GIS and remote sensing for environmental monitoring

Northern Territory University

Matthew Fegan

Summary | Aims | Key questions | Significance | Progress | Publications | References | Supervisors |

Summary

The potential synergy arising from the integration of remote sensing and GIS is becoming increasingly likely. Given the ongoing trends of increasing spatial and spectral resolution of remotely sensed imagery (with increased complexity of analysis) and of the increased uptake of GIS by a diverse range of users (with increased demand for data), some degree of integration of RS-GIS is desirable because:

  1. It has the potential to facilitate the flow of information from imagery into GIS
  2. Remotely sensed imagery can be analysed more quickly (Ahmad, 1992)

Optical remote sensing has been in use for over 20 years and techniques of image classification, rectification and so on are well established. These are commonly implemented by an experienced remote-sensing analyst, on a "one-off" basis, for each new image, and this potentially limits the flow of information to GIS.

A method of automating the processing of remotely sensed imagery into a suitable form for import into GIS would have clear advantages. Conversely, the information contained within the GIS can possibly be made to inform the image processing with the view to "piggy back" on previous image classification and ground truth data.

In contrast to optical remote sensing, the techniques of processing RADAR remote sensing data are largely undeveloped as yet. Radar images are increasingly attractive for their "year-round monitoring" capacity, but are not particularly amenable to "traditional" optical remote sensing data processing techniques. The complexity of interpreting radar images invites the application of "knowledge based" or "expert system" classification in which the spatial context of a pixel is taken into account in the classification process.

This is a further arena in which the integration of digital image processing with the spatial analysis functionality of GIS may be of use.

Similarly, the image processing methods developed for optical remote sensing data become increasingly difficult to implement when applied to the problem of the ever-increasing number of bands in hyperspectral imagery. The increased demand on the remote sensing analyst may be ameliorated via automation of the RS-GIS process.

This project focuses on the integration of GIS and RS to expedite processing of remote sensing data in the context of environmental inventory and monitoring. The proposed study area in general is the NATT with the object of utilising data already available possibly supplemented (funds permitting) with RadarSat imagery.

Aims

  • To extend RS-GIS integration in the use of GIS to facilitate remote sensing digital image processing.
  • To develop a scheme to enable GIS to be used to determine the degree of spatial correlation between a classification mask and cover type spatial distribution. Using this, determine what cover types correspond to given spectral signatures with the view to automating routine, repetitive classification and image rectification, once a reliable base classification and registration have been established.
  • To investigate using GIS as means of implementing knowledge based classification with the aim of integrating radar with other remote sensing and non-remote sensing data.

Key Questions

  • Is the possible integration of RS and GIS theoretically sound and does it function correctly with simple artificial test data?
  • Does the proposed integration enable classification of SPOT MSS images and LandSat TM images to be informed by a GIS coverage of ground truth data?
  • What is the utility of Fourier analysis when applied to spatial correlation, filtering and pattern recognition, in optical remote sensing imagery and in Radar remote sensing imagery?
  • Does the process of attempting to "piggy back" an image classification by using a previous classification mask obscure or highlight real temporal change?
  • Can GIS be used to implement knowledge based classification of optical remotely sensed imagery?
  • Can a knowledge based classification scheme developed for optical remotely sensed imagery be applied successfully to hyperspectral imagery and to what extent is it extendable to Radar imagery?

Significance

Automating the processing of images for importing to GIS has several benefits:

  • If it were possible to enter remotely sensed data into GIS in a routine economical fashion, then frequent updating of GIS data by remote sensing imagery would improve data currency.
  • Automatic image processing to provide a rapid (if not foolproof) and application specific reconnaissance of an image, enabling automatic identification of areas sufficiently similar to known target areas, would extend the usefulness of remotely sensed images to a wider pool of (possibly non-expert) users.
  • The land managers' demand for increasing number of bands (eg. Hyperspectral imagery) places growing demand on the remote sensing analyst. A degree of automation through RS-GIS integration may facilitate the adoption of hyperspectral data.
  • The development of methods of processing and interpreting of radar images is particularly useful for year round monitoring in the tropics where cloud is endemic for half the year.
  • There is a perception that GIS and RS may offer benefits to indigenous natural resource management (Crerar et al 1998)
  • "The interpretation of a remotely sensed image is facilitated by using a GIS...to determine what if any cover types or physical parameters correspond to given spectral signatures." (Carter et al 1998)

Progress

The automated integration of Remote Sensing and GIS is approached from a DATA FUSION perspective. Data from several sources (image and non-image) is used in combination (fused) with the goal of producing output information of higher quality than obtainable from a single source.

During 2000-1, fusion with GIS data was explored:

  • The spatial correspondence between reference GIS data and spectral classes from an unsupervised image classification were estimated by GIS overlay and cross-tabulation.
  • Cover types were characterised by a texture measure and the similarity between a pixels immediate spatial neighbourhood and cover type texture templates is estimated.
  • Spatial properties of cover type (such as probability of adjacency, average area, perimeter ,width) derived from reference mapping within GIS were exploited in Image Classification.
  • The use of Inductive Learning algorithms was commenced to attempt to select from multiple data set those most discriminating for the cover type mapping required and to infer possible classification rules.
  • The use of SAR imagery to attempt to map inundated areas and thus shore lines on flood plains was commenced with a view to generating effective contours for the generation of a DEM within a GIS.

Publications

Fegan, M. 2000, 'Beginning an investigation into the use of fusion of image and non-image data for the automation of classification of a tropical wetland', Proceedings of the 10th Australasian Remote Sensing and Photogrammetry Conference, Adelaide, 21-25 Aug. Fegan, M., Devonport, C., & Ahmad, W. 2000, 'Beginning an investigation into the use of fusion of image and non-image data for the automation of classification of a tropical wetland', Proceedings of the 10th Australian Remote Sensing & Photogrammetry Conference 10ARSPC, Adelaide SA, 21-25 Aug.

Fegan, M., Devonport, C., & Ahmad, W. 2001, 'Automation of image classification by fusion with GIS data,' Proceedings of the Fifth North Australian remote sensing and GIS conference, NARGIS 01 Darwin NT, 3-5 July.

Fegan, M., Devonport, C. & Ahmad, W. 2001, 'Automating image classification, a preliminary foray using archival GIS data to label pixels', Proceedings of the IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2001, Sydney NSW, 9-13 July.

References

Ahmad, W. 1992, "The use of remotely sensed data in the context of a GIS for monitoring temporal change in a forested region of Australia," Asian-Pacific Remote Sensing Journal, 5(1): 133-143.

Carter, J.L., Devonport, C.C., Hill, G.J.E. (1998) The application of remote sensing and GIS to coastal habitats in northern Australia.

Crerar, J.M., Hill, G.J.E., Devonport, C.C. (1998) The use of remote sensing and GIS by indigenous people for natural resource management.

Supervisors

Dr Dick Williams, CSIRO W&E;

Dr Waqar Ahmad, CDU;

Mr Chris Devonport, CDU

Contacts

Mr Matthew Fegan
PhD Student
Tel: 08 8946 6367

Fax: 08 8946 7088

Faculty of SITE, Bldg 18
DARWIN, NT