Abstract

OCEANOGRAPHIC FEATURE DETECTION AND LOCATION FROM HIGH-RESOLUTION SATELLITE IMAGES Ramprasad Balasubramanian, Ayan Chaudhuri, Sourish Ray(Computer and Information Science), Avijit Gangopadhyay(Physics and SMAST)

NASA satellites, Terra (EOS AM) and Aqua (EOS PM) view the entire Earth's surface every 1 to 2 days, acquiring data in 36 spectral bands. Moderate Resolution Imaging Spectroradiometer (MODIS), the key instrument aboard these satellites plays a vital role in collecting observations that will help in predicting global changes accurately and concisely. Our research focuses on analysis and processing of high-resolution (250 meters/pixel) MODIS Ocean color (Chlorophyll) and Sea surface temperature (SST) data. Our objective is to detect and locate oceanographic features, utilizing their dominant patterns and variability. Our approach is to identify, spatially segment and temporally track these features. One such feature – an Eddy, in the Gulf of Mexico is studied here. An automatic eddy identification procedure is presented. The eddy structuring element construction method is presented. This approach is demonstrated on a level-2 MODIS ocean color image of Chlorophyll_a.

Abstract

The MODIS Ocean Level 2 files represent 5 minutes of spacecraft viewing at 1 km spatial resolution. There are 144 files per day of the ocean color products (daytime only) and 288 files per day of the sea surface temperature products (day time and night time).

Introduction

---The MODIS files are grouped together into several HDF ( Hierarchical Data Format) files.---The following HDF files have been used for our research: MODOCL2B Ocean Color Group 2 Products. MODO28L2 Sea-surface temperature and related products. MODO3 Corresponding geolocation file.---The Chlorophyll_a product (MOD21, parameter 27) is based on a semi- analytic algorithm, involving the inversion of a spectral reflectance model to solve for chlorophyll in the presence of other substances that affect the water’s optical properties. ---SST(Sea Surface Temperature) is derived from the MODIS IR channels using two channels in either the thermal IR (11-12 um)or channels in the mid-IR region (3.8-4.1 um).

MODIS Level2 Ocean Color Derived Products were ordered from the Goddard Earth Sciences Distributed Active Archive Center ( GEC DAAC), by accessing the following website : http://daac.gsfc.nasa.gov/data/dataset/

--- Focus was on data granules for March 2002. --- Spatial Coordinates were supplied to the system.--- A list of granules were returned via the Web Interface.--- The granules most suited for our research were ordered for and acquired for each day using FTP Pull.--- The data was processed using the ModisGui and relevant MATLAB Functions. --- SST and Chlorophyll_a plots were extracted from the data. --- The images were cropped to our analysis region.--- The best images were selected and converted into MPEG files for tracking the flow of eddies.

Data Acquisition and Processing

Chlorophyll_a (March 17, 2002)

Chlorophyll_a ( March 18, 2002)rim

Structure of the EddyStructure of the Eddy

core filament

Sea Surface Temperature

Can the eddy be resolved using SST?

March 17, 2002 March 18, 2002

T(r) = TT(r) = Tc c - [T- [Tcc - T - Tkk ] {1-exp(-r/R)} ] {1-exp(-r/R)}

‘‘r’ is the radial distance from the center of the eddy.r’ is the radial distance from the center of the eddy.‘‘TTkk’ is the background temperature .’ is the background temperature .‘‘TTcc’ is the core temperature. ’ is the core temperature. R= 5*R0, where R0 is the Rossby radius.R= 5*R0, where R0 is the Rossby radius.

Eddy Pattern EquationEddy Pattern Equation

Image Analysis TechniqueAnalysis began with the extraction of the visible eddy from the March 18 chlorophyll image.This was done by specifying the latitude(19.6 to 21.2 degrees) and longitude(-94.8 to -93.2 degrees) range for the eddy feature. Thereafter, the above equation which is an eddy tracer was used on this sub-image to derive the range of T(r) temperature values. In the equation the value of R, Tc and Tk were empirically taken to be 75 km, 0.6520 (maximum value) and 0.3700 respectively. The radius (r) was calculated by finding the root of the squares of the longitudinal and latitudinal differences of each point with respect to the core (Tc).The range of T(r) values obtained were high around the core, however diminishing while approaching the outer boundary. This conformed to the characteristics of an eddy that has a strong intensity in the center and decays out towards the edges. Subsequently a generic pattern was to be created to encapsulate our analysis. The entire range of T(r) values was divided into a 5x5 grid by taking their average to populate each element of the grid. This 5x5 grid became our structuring element or generic template for eddy detection.

SchemaThe Structuring element was tested onto the entire March 18, 2002 chlorophyll image to see its effectiveness. The testing was done using the following methodology:

• The structuring element was traversed over the entire image. • For each set of 5x5 pixels on the image, corresponding pixels were subtracted from the element.• These differences were then squared and added ; the analogy being if the point is an eddy center, then the sum of squares of the differences between the image and the element would equate to zero. 

The GOMex Eddy Detection

The Structuring Element 0.4652 0.4813 0.4852 0.4728 0.4510 0.4999 0.5397 0.5592 0.5268 0.4824 0.5068 0.5602 0.6036 0.5544 0.4966 0.4771 0.5056 0.5224 0.5090 0.4774 0.4420 0.4552 0.4624 0.4589 0.4457

The figure alongside shows the plots of the four minimum values obtained from applying the structuring element on the March 18, 2002 chlorophyll image. The error percentage was about 2.5%. From observation we could see that the structuring element was quite successful in determining the location of the eddy, although it was not very accurate and also detected a few false negatives.

Future Objectives• One of our primary concerns will be to make the structuring element larger so that it would detect eddy features more accurately.• We will proceed to test the structural element out on other chlorophyll images,thus training our algorithm even further in eliminating false negatives.• Another concern will be to investigate eddy detection hgiven sea surface temperature information,correlating the effects of sea surface temperature on chlorophyll and vice versa.

(5x5)Structuring element

columns

rows

Acknowledgements:UMD Foundation Grant Jasmine Nahorniak, Oregon State University