Extraction of Blue Ice Area in Antarctica

Chapter 3

METHODOLOGY

High resolution satellite data has made it possible to obtain optimistic results in feature extraction processes. High resolution World-View-2 data is used for mapping blue ice areas (BIAs) in Antarctic regions. World-View-2 provides extensively high accuracy, agility, capacity and spectral diversity. First high-resolution 8-band multispectral commercial/business satellite is World-View-2 launched October 2009. Working at an elevation of 770 kilometres, World-View-2 gives 46cm panchromatic resolution and 1.85m multispectral resolution. World-View-2 has a normal revisit time of 1.1 days and it is able of catching up to 1 million square kilometres of 8-band imagery per day. Satellite pictures generally track seasonal annual variations in BIAs coverage over the past 30 year on the East Antarctic plateau region. In late studies, the distribution of BIAs can likewise mulled over from the SAR (synthetic aperture radar) images. In SAR satellite image, blue ice can likewise be outwardly perceived. The amplitude of blue ice is less than that of snow (white), because the ice surface is smoother than the latter. Yet, distinction is not at all that conspicuous when applying Semi-automatic extraction approach. Blue ice can be distinguished effortlessly in the coherence map got from two SAR pictures in a view of higher coherence of blue ice. It is additionally found that the picture texture data is useful for distinguishing various types of blue ice like rough, smooth and level blue ice. In this study, Atmospheric corrected (QUAC) sharpen calibrated image (World-View-2 data) is used for extracting blue ice areas in Schirmacher Oasis in Antarctic region. Extraction of blue ice area in Antarctica deal with the total area of blue ice areas excluding the other feature (non-target) appearing on or near it. Blue ice areas have some specific qualities that make them of special interest for extraction as they are just 1% of Antarctic region. Many remote sensing approaches have been implemented to monitor and map Antarctic BIAs.

3.1 Methodology Protocol

The extraction of blue ice areas is simplified by the Methodology protocol. As the whole image takes time for processing, as Schirmacher Oasis is with an area of 34km², ranks among the smallestAntarctic oasis and is a typicalpolar desert, so the image is divided in 12 test tiles of different parts of entire World-View-2 image to achieve prior results. Atmospheric correction is done with QUAC (quick atmospheric correction) method to obtain better results. Atmospheric correction to each tile added suitable outputs results to workflow. Calibrated data is also used without applying atmospheric correction to it. Multiband image combination was made from atmospheric corrected data and calibrated data of the study area.

Alternating snow and blue ice bands surface patterns are generally found in East Antarctica due to which it is hard task to clearly extract BIAs. For feature extraction processes region of interest (ROI) is considered in which blue ice is target and white ice appearing on or near the blue ice is considered as non-target. Methodology workflow is prepared in order to achieve good and prior results comparing with the previous studies. Extraction of blue ice is not that easy task as dust and white snow appears on it as non-target. Various Semi-automatic extraction methods like TERCAT, Target Detection Wizard, Mapping Methods, Spectral Matching and Object Base Image Analysis (OBIA) are used for extracting blue ice areas in Antarctica. The initial results obtained were good but not better enough to keep them prior. Many trials were carried out for extracting blue ice in Antarctica. Prior results were kept in workflow of methodology to compare them with every trial results.

Object based and Pixel based both the classification are used in workflow to get good results. From the High resolution World-View-2 data reference data (digitized data) was prepared for blue ice area and extracted blue ice area was obtained from Semi-automatic extraction methods and OBIA. From the extracted blue ice, blue ice is considered as target and white snow appearing on it as non-target. Comparing reference data and extracted data Bias, % Bias and RMSE values were calculated. After that Average for Bias, % Bias and RMSE values is estimated.

BIAS=

% BIAS=

RMSE=

Where,

Ref A is Reference area and Ext A is Extracted area

n= no. of tiles processed.

3.2 Semi-automatic extraction methods

The semi-automatic feature extraction approach intuitively makes endeavours to commonly empowering the insight or data of human perception framework to robustly detect the targeted feature and the computer-aided system to bring fast extraction of targeted feature and exact shape representation. In semiautomatic feature extraction strategy, first target feature is detected by human vision and a couple of estimates in terms of seed points or coaching samples concerning the targeted feature on highlight are typically given. The targeted feature is then portrayed automatically by the PC helped calculations.

3.2.1 TERCAT approach (ENVI 5.1 Exellis Help) [33]

The Terrain Categorization (TERCAT) tool creates an output product in which pixels with similar spectral properties are clumped into categories. These categories may be either user-defined, or automatically generated by the classification algorithm. The TERCAT tool provides all of the standard ENVI classification algorithms, plus an additional algorithm called Winner Takes All. This is a voting method that classifies pixels based on the majority compiled from all of the other classification methods that were conducted. In this research, the sub approaches for TERCAT are Maximum Likelihood, Spectral Angle Mapper, Parallelepiped and Winner Takes All.

3.2.2 Target Detection approach (ENVI 5.1 Exellis Help) [33]

Target detection algorithms work on the principle of extracting target features based on spectral characteristic of initial coaching spectral signatures of target features, and performing end to the background noise using spectral signatures of non-target features. If the users knows that the image contains at least one target of interest, the wizard can be used to find other targets like it in the same image. The workflow can also be accessed programmatically, so the user can customize options if needed. Target detection tools (ENVI 5.1) were executed to perform supervised image processing tasks into workflows (CEM, ACE, OSP, TCIMF, and MT-TCIMF) to extract blue ice areas (BIAs) as target and white ice as non-target.

3.2.3 Spectral Matching approach (ENVI 5.1 Exellis Help) [33]

Spectral matching approaches extract the target features that are described in multispectral imagery based on the target feature’s spectral characteristics. Spectral matching algorithms confirm the spectral similarity or matching between input satellite imagery and reference key points to form an output product within which pixels with similar spectral properties are clumped into target and non-target categories. Spectral Matching (ENVI 5.1) were executed to perform supervised image processing tasks into workflows (MF, SAM, MTMF and SAMBM) to extract blue ice areas (BIAs) as target and white ice as non-target.

3.2.4 Mapping Methods approach (ENVI 5.1 Exellis Help) [33]

Selected hyperspectral Mapping Methods describes advanced concepts and procedures for analyzing imaging spectrometer data or hyperspectral images. Spectral Information Divergence (SID) is a spectral classification method that uses a divergence measure to match pixels to reference spectra. The smaller the divergence, the more likely the pixels are similar. Pixels with a measurement greater than the specified maximum divergence threshold are not classified. End member spectra used by SID can come from ASCII files or spectral libraries, or you can extract them directly from an image (as ROI average spectra). Mapping Methods (ENVI 5.1) were executed to perform supervised image processing tasks into workflows [SID SV (0.05), SID SV (0.07), SID SV (0.1), SID MV (0.05) and SID MV (0.09)] to extract blue ice areas (BIAs) as target and white ice as non-target.

3.2.5 Object Based Image Analysis (OBIA) approach (Ecognition Developer Help) [34]

Object Based Image Analysis (OBIA), is an advanced method used to segment a pixel based image into map objects that can then be classified as a whole. This kind of analysis is ideal for mapping with high-resolution imagery, where a single feature (such as a tree) might have several different shades of pixels. The example of rule-set for Trial 1, 2, 3 and 4 for extracting blue ice areas in this research is stated below;

For Trial 1:

02.063 50 [shape.: 0.8 compact.:0.6] Creating ‘level 1’

Export view to segmentation (no geo)

Unclassified with mean nir-1>=50 and mean nir-1<=2900 at level 1: blue ice

Export view to assign class (no geo)

Blue ice with mean nir-1>=50 and mean nir-1<=2900 at level 1: merge region

Export view to merging (non geo)

For Trial 2:

02.063 60 [shape.: 0.8 compact.:0.6] Creating ‘level 1’

Export view to segmentation (no geo)

Unclassified with mean nir-1>=100 and mean nir-1<=3000 at level 1: blue ice

Export view to assign class (no geo)

Blue ice with mean nir-1>=100 and mean nir-1<=3000 at level 1: merge region

Export view to merging (non geo)

For Trial 3:

02.063 70 [shape.: 0.8 compact.:0.6] Creating ‘level 1’

Export view to segmentation (no geo)

Unclassified with mean nir-1>=150 and mean nir-1<=3100 at level 1: blue ice

Export view to assign class (no geo)

Blue ice with mean nir-1>=150 and mean nir-1<=3100 at level 1: merge region

Export view to merging (non geo)

For Trial 4:

02.063 80 [shape.: 0.8 compact.:0.6] Creating ‘level 1’

Export view to segmentation (no geo)

Unclassified with mean nir-1>=200 and mean nir-1<=3200 at level 1: blue ice

Export view to assign class (no geo)

Blue ice with mean nir-1>=200 and mean nir-1<=3200 at level 1: merge region

Export view to merging (non geo)

The on-top rule-set is employed to extract blue ice areas as well as non-target depending on their mean band values. OBIA is making considerable progress towards spatially explicit information extraction advancement, such as is required for spatial planning as well as for many monitoring programmes.

The Semi-automatic extraction strategies and OBIA utilized in this study to extract blue ice areas (BIAs) are supported differently on different underlying principles. To compare these strategies objectively, we kept the input ROIs (regions of interest or coaching samples) constant for all methods for each tile. ROIs are different for different tiles as the area differs. After classifying the image into target spectra, i.e., blue ice areas, using the Semi-automatic extraction methods and OBIA approaches, the 12 semi-automatically extracted tiles (for BIAs) were vectorized to calculate the area of individual tile.