IKONOS Process Document for the King County Department of Natural Resources

For Impervious Surface for Multispectral Imagery for Eastern King County

 

June 25, 2002
Table of Contents

I.     Introduction.. I-2

II.    Pre-processing.. II-2

ERDAS Imagine Procedures. II-2

Step 1:  Assess Impervious Surface Quality. II-2

Step 2:  Mosaic IKONOS Image Access Satellite Overpass. II-2

Step 3:  Verify Positional Accuracy with Ancillary Datasets. II-2

III.   Impervious Strata.. III-3

Goal. III-3

ERDAS Imagine Procedures. III-3

Step 1:  Threshold NDVI III-3

Step 2:  Potential Thresholds Effectiveness Evaluation. III-3

Step 4:  Create Impervious File. III-3

Step 5:  Isodata Cluster. III-3

IV.   Unsupervised Classification.. IV-3

Goal. IV-3

ERDAS Imagine Procedures. IV-4

Step 1:  Identify Possible Pervious Clusters. IV-4

Step 2:  Assign Category Labels. IV-4

Step 3:  Cluster Bust IV-4

V.    MMU.. V-4

Goal. V-4

ERDAS Imagine Procedures. V-4

Step 1:  Recode. V-4

Step 2:  Clump. V-5

Step 3:  Elimination. V-5

Step 4:  Mask. V-5

VI.   Mosaic Panels Together (1s and 50s) VI-5

Goal. VI-5

ERDAS Imagine Procedures. VI-5

Step 1: Mosaic Products from V.. VI-5

Step 2: Focal Filter. VI-5

Step 3:  Minimum Value Filter. VI-5

Step 4:  Clipping. VI-5


IKONOS Process Document for the King County Department of Natural Resources

For Impervious Surfaces for Multispectral Imagery for Eastern King County

I.                   Introduction

The following instructions are the process steps taken by Marshall and Associates, Inc. in pre-processing and processing the IKONOS product for the King County Department of Natural Resources. 

II.                Pre-processing

Goal 

The goals of the Pre-processing phase are to identify any problems with the data that might lessen the effectiveness of proposed techniques or render certain data unusable, geometrically verify, and, if necessary, correct the data (accurately) and register the different acquisitions.

ERDAS Imagine Procedures

Step 1:  Assess Impervious Surface Quality

1.1   Examine each acquisition for clouds, cloud shadows, haze, fog, smoke, excessive image striping or drop-outs, snow, ice etc.

1.2   If any of the above are found in significant amounts, discard the acquisition and obtain an alternative acquisition.  If necessary, use an acquisition with clouds in it and eliminate the areas covered by clouds and cloud shadows from further consideration.  Eliminate these areas by hand digitizing an AOI around the affected areas.  Label all the pixels inside the AOI as UNKNOWN.

Step 2:  Mosaic IKONOS Image Access Satellite Overpass

2.1   Mosaic images to panels.

2.2   Reproject Imagery into final desired projection

2.3   Check for clouds.

2.4   Create NDVI in float single format from channels 4 and 3 (Channel4-Channel3/Channel4+Channel3).

2.5   Digitize clouds by onscreen interpretation; low NDVI values can be used to help identify potential clouds.

Step 3:  Verify Positional Accuracy with Ancillary Datasets

Potential Ancillary Datasets:

a)      GPS Points

b)      GIS Layers

c)      Other Imagery

III.              Impervious Strata

Goal

The goal of this phase is to spectrally stratify all of the pixels in the study area that are likely to be impervious, thus creating a file that enhances the contrast between impervious and pervious features.

ERDAS Imagine Procedures

Step 1:  Threshold NDVI

Threshold the NDVI to determine a spectral impervious stratum containing all impervious pixels (roads, sidewalks, roofs, etc.) with as few commission errors as possible (e.g., non-constructed bare surfaces).  The NDVI normalizes for the effects of shadows, reducing the differential spectral response of the sunlit and shadowed portions of the imagery.

Step 2:  Potential Thresholds Effectiveness Evaluation

Overlay the DNR training sets on the impervious stratum file.  Assess the proportion of pixels in each training set that are in the stratum.  Most of the pixels in the >80% impervious training sets should be included.

Step 4:  Create Impervious File

For each mosaic, create a file that contains all of the pixels on the impervious side of the threshold (different thresholds for each image). 

Step 5:  Isodata Cluster

Cluster the impervious file with the Isodata algorithm.  Clustering options in ERDAS Imagine include ‘Initialize from the statistics,’ using the principle axis to initialize and scaling the range of signatures with ‘Automatic.’  The other parameters, such as number of clusters, number of processing iterations, and convergence threshold are determined on the basis of the image feature variability (in the case of the IKONOS panel 79, 130 clusters were set at 15 iterations with a convergence threshold of 0.950).  A trade off in image analyst cost exists between creating too many clusters and not creating enough clusters to identify all image features.

IV.            Unsupervised Classification

Goal

The goal of this phase is to refine the spectral impervious stratum created in phase III. Impervious Strata, ultimately removing spectral clusters consisting entirely of non-impervious and vegetative surfaces.

This activity can consume much time and energy if it is not properly limited.  One way to limit it is to assume that the DNR’s training sets are a representative sample to be categorized.  Given this assumption, tentative labels can be assigned to clusters based on the types of training sets with which they are most closely associated.

ERDAS Imagine Procedures

Step 1:  Identify Possible Pervious Clusters

It is possible, though unlikely, that one or more of the impervious stratum clusters could be uniquely associated with a pervious surface.  This proposition can be tested by observing whether any of the pixels associated with a cluster from the impervious stratum are found only in pervious category training sets and not in any of the impervious training sets.  Shadow may cause non-impervious features to look like some types of impervious.  If pixels in the impervious stratum occur in shadow, a representative sample should be analyzed to determine if they are true impervious or rather shadowed versions of pervious categories (e.g., forest and open water).  If they are true impervious, nothing needs to be done.

Step 2:  Assign Category Labels

Assign the value of 1 for High Confidence impervious, 50 for Low Confidence Impervious, and 0 for other.  This also tracks error of commission, allowing for further revision either with ancillary data, heads up analysis, or cluster busting.

Step 3:  Cluster Bust

Clusters remaining in the impervious stratum that contain mixtures of pervious surfaces and impervious surfaces can be further refined with cluster-busting.

3.1   Refine confused clusters through the following methods.

3.1.1          For each confused cluster, mask out the pixels associated with that cluster.

3.1.2          Recluster those pixels with Isodata, creating ten clusters.

3.1.3          Relabel those ten clusters.

Note:  The high spectral sensitivity of IKONOS imagery records many impervious features with high reflectivity not always distinctly captured by the Isodata clustering algorithm.

3.1.4          For each cluster bust, if necessary, create supervised signatures of anomalies, such as very bright impervious surfaces.  Append signatures from step 3.1.3 with supervised signatures when clean cluster busting is not developed.  Run the supervised/cluster bust signatures on the pixels from step 3.1.1 with the maximum likelihood classifier.

3.1.5          Overlay the relabeled/cluster busted pixels on to the original impervious strata labeled clusters, overwriting the confused cluster pixels.

V.               MMU

Goal

The goal of this phase is to produce recoded/refined clumps at the desired MMU, 0.025 acres, regardless of impervious surface classification confidence.

ERDAS Imagine Procedures

Step 1:  Recode

Create two files, recoding the cluster busted, labeled files to one file of 0s (background and pervious surfaces) 1s, and 50s and another file to 0s (background), 1(impervious high and low confidences), and 2 (pervious surfaces, including those excluded from the impervious stratum).

Step 2:  Clump

Clump the 1s and 2s file, using the ERDAS clump function, utilizing eight connected pixels.

Step 3:  Elimination

Run the clump file through ERDAS Elimination command and eliminate clumps seven pixels or less (determined by the desired MMU).

Step 4:  Mask

Take elimination file as a mask to the 0s, 1s, and 50s file, removing from it all cells that are not 1 on the 1 and 2 eliminated file.

VI.            Mosaic Panels Together (1s and 50s)

Goal

The goal of this phase is to produce the final product.  Due to the spatial footprint of IKONOS and the resampling with an averaging filter applied by the data supplier, some impervious features such as arterial roads may not be distinct or spatially contiguous.  The following steps help create impervious features cluster to what is visually distinguishable without increasing commission error.  This step also allows a limited level of panel edge smoothing.

ERDAS Imagine Procedures

 

Step 1: Mosaic Products from V

Utilizing the feathering option for image overlap, mosaic the categorized panels into one file. 

Step 2: Focal Filter

Do a 3x3 focal filter neighborhood analysis, returning majority values to the center pixels ignoring 0s in the focal window.  This first filter strengthens feature identification, essentially filling in “holes.”

Step 3:  Minimum Value Filter

Do a 3x3 minimum value filter on the analysis.  This second filter shrinks features back to original extent.

Step 4:  Clipping

Clip the final product to the tiling scheme of the County.