Mapping surface cover types using ASTER data


Decorrelation Strech:
Decorrelation stretch is applied to the ASTER data. The decorrelation stretched is to reduce the inter-channel correlation and stretch the dynamic range to the full extent which enhances the color variation and improve the visualisation for interpretation (Gillespie et al, 1986, Gillespie, 1992). Before and after decorrelation stretched VNIR, SWIR and TIR images are shown below.







Before and after decorrelation stretched RGB (NIR, VisRed, VisGreen) images of VNIR suggest that the D-Stretched image has reduced the mist/cloud, which is found above the dam at the bottom-left of the image. The D-Stretched images of SWIR (RGB ASTER 6,7,9) and TIR (RGB ASTER 13,12,10) images have substantially improved and enhanced the color variation of the bands combination for visual interpretation.

Lithologies / Minerals:
To be able to analyze spectral responses of surface cover types using SWIR ASTER data it is necessary to apply log residual algorithm, which reduces noises from topography, instrument and sun illumination (Green and Craig, 1985). The resultant data is assumed to be more representatives of the soils or lithologies of the exposed areas. Spectrum convolved from log residual applied data can be compared to the library spectrum. Two steps are taken in the log residual algorithm. Step one is generating an addition band to the bands of the ASTER data. For instance, to apply log residual to the 6-band SWIR ASTER data, an additional seventh band (AVG) is generated, which is the average of the all the input six bands (SWIR 6 bands). Step number two is applying the log residual formula to the SWIR bands. Instead of geometric means arithmetic means are used in the log residual algorithm. The formula used is as below: (i1 * mean(AVG))/(AVG * mean(i1)) where i1 = SWIR band 1 to 6 It is not recommended to apply log residual on VISNIR bands as there may be problem with additive atmospheric effects. It is also not recommended to apply log residual on TIR, as there may be non-linear emissivity temperature relationship of different lithologies (Dr Robert Hewson; Personal communication).

To reduce any bias from water, water is nullified prior to applying Log Residual on the SWIR 22 March, 2001 ASTER data. The classified image of the Log Residual applied ASTER data using Unsupervised classification is shown below:



There is no priori knowledge of the study area and soil/rocks are simply classified as Class1 to Class4. The spectrum of vegetation and the 4 soil/rock classes are displayed below. For comparison John Hopkin University (JHU) spectrum of grass, dolomite and calcite are also displayed. There are some similarity between vegetation and grass and between the 4 soil/rock classes and carbonates. Ground truth is necessary for verification.



Change detection:
A simple example of vegetation change detection is demonstrated. Decorrelation Stretch algorithm was applied to both dates imagery and resultant stretched images were rescaled back to unsigned 8 bit integer. NDVI formula ((NIR-Gr)/(NIR+Gr)) was applied to both dates imagery and a threshold of 0.3 was used to map vegetation.



Vegetation found only in the 22 March, 2001 imagery, vegetation common in both dates imagery and vegetation found only in 7 April, 2001 imagery are mapped using the following procedure:
  1. Map vegetation of ASTER 22 March, 2001 as veg1 and the remaining as non-veg1
  2. Map vegetation of ASTER 7 April, 2001 as veg2 and the remaining as non-veg2
  3. Combine the above 2 imagery as an integrated Virtual Dataset
  4. Map vegetation found only in ASTER 22 March, 2001 VNIR imagery using the formula (If veg1 and not veg2 then veg-22Mar-only else null)
    (NOTE: veg1, veg2, veg1-22Mar-only are variables)
  5. Map vegetation found only in ASTER 7 April, 2001 VNIR imagery using the formula
    (If veg2 and not veg1 then veg-7Apr-only else null)
  6. Map vegetation in both dates imagery using the formula
    (If veg1 and veg2 then common-veg else null)


Conclusion:
Providing there is no or only minor cloud, surface water and vegetation can be easily mapped using ASTER data. Decorrelation Stretch algorithm can be applied to reduce inter-channel correlation and enhances VNIR, SWIR and TIR ASTER data to gain more spectral variation for visual interpretation. Log Residual algorithm can be applied on SWIR to reduce noises from the sun illumination, topography and instrument, after which 6 ASTER SWIR bands can be convolved for representative spectrum of surface cover types. If multi-temporal data are available of the same area, change detection can also be easily carried out.

References:
  • Gillespie, A.R., Kahle, A.B. and Walker, R.E., (1986); “Color enhancement of highly correlated images. Decorrelation and HIS contrast stretches.” Remote Sens. Environ., v. 20, pp. 209-235.
  • Gillespie, A.R.,(1992); “Enhancement of multispectral thermal infrared images: decorrelation contrast stretching.” Remote Sens. Environ., v. 42, pp. 147-155.
  • Green, A.A and Craig, M.D.,(1985); ”Analysis of aircraft spectrometer data with logarithmic residuals." JPL Publ. 85-41, pp. 111-119
  • Hewson R.D., Cudahy T.J. and Huntington J.F. (2001); “Geologic and alteration mapping at Mt Fitton, South Australia, using ASTER satellite-borne data.”
  • Lillesand and Kieffer, 1994; “Remote Sensing and Image Interpretation”, 3rd Edition, 750 p, John Wiley & Sons, Inc Publisher
  • Yamaguchi, Y.; Fujisada, H.; Kahle, A.B.; Tsu, H.; Kato, M.; Watanabe, H.; Sato, I.; and Kudoh, M.; (2001); “ASTER Instrument Performance, Operational Status, and Application to Earth Sciences”, IEEE Trans. Geosci. Remote Sens., 2001

Page 2 of 2
| Previous |