1. What values did you use for bands 2, 3 & 4? Populate the table below. Compile the relevant data for the additional bands in the data below:

Transcription

1 Graham Emde GEOG3200: Remote Sensing Lab # 3: Atmospheric Correction Introduction: This lab teachs how to use INDRISI to correct for atmospheric haze in remotely sensed imagery. There are three models for atmopheric correction in the ATMOSC module in IDRISI, and this lab shows how to compare the results of each model with each other and with pure spectral signatures from the USGS. Results: 1. What values did you use for bands 2, 3 & 4? Populate the table below. Compile the relevant data for the additional bands in the data below: Wavelength DN haze Offset Gain Sun elevation at band center Band Band Band Describe the differences, if any, among the three color composites. Include a copy of each color composite in your figures section. Based on your knowledge of the model, why are the outputs of the models similar/different? The differences between the three color composites are slight. The Cos(t) composite is slightly darker than both the Apparent composite and the Dark Object Subtraction composite. The reds that represent vegetation in the Cos(t) composite are darker than the reds in either of the other two composites, and the water is darker in the Cos(t) composite than in the other two composites. Overall, it appears that the colors in the Cos(t) composite are more saturated than in colors in the other two composites. Compared to the composite of the original images, the atmospherically corrected images are not much different. It is difficult to see the exact difference, but it looks like the Cos(t) composite is slightly darker than the original composite and the other two atmospheric correction composites are slightly lighter than the original. 3. What are the reflectance values extracted for each band in each model? Populate the table below:

2 Cos(t) Apparent Dark Object Subtraction Band Band Band Create a spectral plot of the results of the three atmospheric correction procedures. Plot wavelength in units of micrometers on the x-axis. Use the wavelength values you tabulated for Question 1. Label the y-axis as. Connect the data points for each method using straight lines. On the same chart, plot the values of the pure spectral signature from the table on the previous page. Include this figure in your lab report. Values of Lawn Grass 8.00E E E E E E E E E+00 (Band 1) (Band 2) 0.56 (Band 3) 0.66 (Band 4) 0.83 (Band 5) (Band 7) Pure Cos(t) Apparent Dark Object Subraction Micrometers (µ) 5. Are the spectra from the three different corrected images similar to the pure spectral signature? If not, describe any differences and possible sources of differences associated with the type of correction applied. Cos(t) comes closest to the pure reflectance values for lawn grass. It is exact for Band 2, slightly high for Band 3, and slightly low for Band 4. The other correction models are not as accurate. Apparent is slightly high for Bands 2 and 3 and very low for Band 4. Dark Object Subtraction is low for Band 2, slightly high for Band 3, and very low for Band 4. The differences between the correction models produce the difference between the reflectance values recorded. The Apparent Model is one of the least accurate models in this study, and it also the simplest atmospheric correction model included in the ATMOSC module in IDRISI. This model simply compensates for the elevation of the sun. The Dark Object Subtraction Model is more complicated, considering the Dn of haze in the image, the date and time of the image, the central wavelength of the image band, the sun elevation, and the radiance conversion parameters. Even so, the Dark Object Subtraction model did not produce very accurate results. The Cos(t) model is based on the Dark Object Subtraction model, but it adds

3 estimates the optical thickness of the atmosphere as well as scattering in the atmosphere, which seem to improve the accuracy of the model. 6. What are the DN values extracted for each band in the uncorrected image? Populate the table below. In the Scaled DN column, divide the DN by 255. This will give us numerical values on a similar scale ( ) to reflectance. Add the spectrum of the scaled DNs to the chart you created in Question 4. DN Scaled DN (DN/255) Band Band Band E E E-01 Values of Lawn Grass 5.00E E E E E E+00 (Band 1) (Band 2) 0.56 (Band 3) 0.66 (Band 4) 0.83 (Band 5) (Band 7) Pure Cos(t) Apparent Dark Object Subraction Uncorrected Micrometers (µ) 7. Is the shape of the spectrum (not the numerical values) of the uncorrected data similar to the spectra of the corrected data? If not, describe any differences and possible sources of differences. The uncorrected data closely follows the shape of the corrected data. In particular, the uncorrected data exactly match the shape of the Apparent data, but the uncorrected data are slightly higher. This indicates that the Apparent model simply lowers the reflectance across Bands 2 to 4 (at least for lawn grass). The uncorrected data are higher than the corrected data in Bands 2 and 3 and lower than the pure values and Cos(t) values in Band 4. Apparently atmospheric haze causes higher reflectance in Bands 2 and 3 and lower reflectance in Band 4 in an uncorrected image.

4 Figures: Figure 1: False Color Composite (Bands 2, 3, and 4) of Landsat 5 TM image of Southeastern New England September 16, 1987 Figure 2: Atmospheric Correction of False Color Composite (Bands 2, 3, and 4) of Landsat 5 TM image of Southeastern New England September 16, 1987, using Cos(t)

5 Figure 3: Atmospheric Correction of False Color Composite (Bands 2, 3, and 4) of Landsat 5 TM image of Southeastern New England September 16, 1987, using the Apparent Model Figure 4: Atmospheric Correction of False Color Composite (Bands 2, 3, and 4) of Landsat 5 TM image of Southeastern New England September 16, 1987, using Dark Object Subtraction