18.06.2015 Views

Flood Risk and Vulnerability Analysis Project - Atlantic Climate ...

Flood Risk and Vulnerability Analysis Project - Atlantic Climate ...

Flood Risk and Vulnerability Analysis Project - Atlantic Climate ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

coordinates of the SPOT imagery using control points. A reasonable root mean squared error<br />

(RMSE) was achieved during the registration process, but it is important to note that the<br />

classified EOSD dataset was processed using coarser resolution than the reference SPOT data;<br />

therefore, spatial shifts may still exist but are minor. After georeferencing was complete, a final<br />

EOSD mosaic dataset was created.<br />

3.3.3 L<strong>and</strong> Cover Classification<br />

L<strong>and</strong> cover classification was performed through a multi-step process detailed below. General<br />

steps include; (1) image preprocessing <strong>and</strong> mosaicing, (2) ISODATA unsupervised classification<br />

<strong>and</strong> aggregation, (3) change detection processes, (4) classification corrections <strong>and</strong> manual<br />

edits, (5) l<strong>and</strong> cover tabulations by watershed, <strong>and</strong> (6) accuracy assessment. Accuracy<br />

assessment results are presented in the results section.<br />

3.3.3.1 Image Preprocessing<br />

Due to the large geographic area <strong>and</strong> the high frequency of clouds <strong>and</strong> ice in the study area,<br />

more than 100 SPOT images <strong>and</strong> image fragments were used in this study. For each<br />

watershed, mosaics of images with similar reflectance characteristics were combined to create<br />

larger images where possible. Histogram matching was used to reduce pixel variation within<br />

aggregated images. For the remaining, adjacent datasets with high levels of variation in pixel<br />

values at overlapping areas (likely due to temporal resolution), performing an atmospheric<br />

correction process to apparent reflectance values would not have provided any benefit (see<br />

additional information in Recommendations). Clouds <strong>and</strong> ice were then masked from areas<br />

where they represented a large proportion of existing imagery in order to remove these skewed<br />

pixel values during the classification.<br />

TA1112733<br />

38

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!