8/10/10 1 NDVI Calculation of RS Data

13
22/06/26 1 NDVI Calculation of RS Data November Seminar- 28 /11/06 Shamim Akhter PhD Student AIDA Lab Department of Information Processing Tokyo Institute of Technology, Japan

Transcript of 8/10/10 1 NDVI Calculation of RS Data

Page 1: 8/10/10 1 NDVI Calculation of RS Data

23/04/12 1

NDVI Calculation of RS Data

November Seminar- 28 /11/06

Shamim Akhter PhD Student

AIDA Lab Department of Information Processing

Tokyo Institute of Technology, Japan

Page 2: 8/10/10 1 NDVI Calculation of RS Data

23/04/12 2

Aster reflectance image ASTER stands for Advanced Space borne

Thermal Emission and Reflection Radiometer ASTER started monitoring global environment

changes in March 2000 and it provides image. Sensor on-board Terra is having 14 spectral

bands ranging from 15 to 90 m spatial resolution.

A fifteenth band is a backward view of the NIR (15 x 15 m in the 0.76 to 0.86 μm) for stereoscopic imaging.

L1B image can be acquired by ftppull from the EROS Data Center Internet website and costs $50 per image.

Page 3: 8/10/10 1 NDVI Calculation of RS Data

23/04/12 3

Table 1: Aster Sensor Characteristics

SensorProduct

Sub-System Number of bands Spectral range (μm)

Spatial resolution (m)

Aster L1B VNIR 3 (+ 1 backward) 0.52 up to 0.86 15x15

SWIR 6 1.60 up to 2.43 30x30

TIR 5 8.125 up to 11.65 90x90

Page 4: 8/10/10 1 NDVI Calculation of RS Data

23/04/12 4

Modis 8 days reflectance MODIS (Moderate Resolution Imaging

Spectroradiometer) is the key instrument aboard the Terra (EOS AM-1) satellite.

MODIS started functioning in March 2000. Terra MODIS is viewing the entire Earth's surface every

to 2 days, acquiring data in 36 spectral bands, or groups of wavelengths.

Modis sensor on TERRA platform has a spectral resolution featuring 36 bands ranging from 250 to 1000 meters spatial resolution.

Acquisition of surface reflectance image products (MOD09A1) can be done freely through the Eros Data Center Internet web site using ftppull protocol. Extraction of the binary file can be performed for the necessary bands. The date of the satellite overpass for this image of central Thailand was 13th of March 2000.

Page 5: 8/10/10 1 NDVI Calculation of RS Data

23/04/12 5

Table 2: MOD09_L3_8-Day Characteristic

Product Sub-Systems Number of bands Spectral range (μm) Spatial resolution (m)

ModisMOD09A1

VNIR 2 0.62 up to 0.876 500x500

SWIR 5 0.459 up to 2.155 500x500

Page 6: 8/10/10 1 NDVI Calculation of RS Data

23/04/12 6

RS Image Handling Libraries The remote sensing imagery is generally a common type of

binary image format (BIL, BSQ, TIFF, etc) that is not compressed since the digital numbers are of physical meanings to remote sensing scientists.

A particularity of the remote sensing data is that it is georeferenced. Meaning that each pixel of the images are having a unique location on the surface of the Earth.

Honda (1995) developed a C language based image handling library available on the internet (http://www.star.ait.ac.th/~honda/textbooks/advdip/imagelib.zip) and released under the GNU Public License (www.fsf.org).

It has been used for these experiments. It can handle jpeg, tiff, ENVI and ER Mapper image formats. Currently having georeferencing support, the ENVI format is the one used in this processing.

Page 7: 8/10/10 1 NDVI Calculation of RS Data

23/04/12 7

NDVI Calculationvoid NDVI_CALC(int h, int w, u_char *nir, u_char *red, u_char

*ndvi){

int i,j;double wk;for (i=0; i<h; i++ ){

for(j=0; j<w; j++ ){ wk = (double) ( nir[i*h+j] - red[i*h+j] ) / (double)

( nir[i*h+j] + red[i*h+j] ); ndvi[i*h+j] = wk *100.0 + 100.0; } }}

Page 8: 8/10/10 1 NDVI Calculation of RS Data

23/04/12 8

MPI Versions

The NDVI program is distributed with two methodologies and successfully implemented in Cluster computers by using MPI Libraries. Distributed each pixel. Distributed each row.

Page 9: 8/10/10 1 NDVI Calculation of RS Data

23/04/12 9

Modis processing time (by rowmethod) for different sets of

operations

0

120

240

360

480

600

720

840

0 1 2 3 4 5 6 7 8Number of nodes used

Tim

e (s

)

5

500

1000

1500

2500

Aster processing time (by rowmethod) for different sets of

operations

0

60

120

180

240

300

360

420

0 1 2 3 4 5 6 7 8Number of nodes

Tim

e (s

)

5

5000

25000

It is observed that there is a point of convergence of all curves for a given image with an increasing number of operations applied. Aster image experiment at 25000 operations shows a minimum time processing at about 40 seconds with 7 nodes. While the number of operations sets tested in the case of Aster image is not enough to conclude more with certainty. Modis image provides some insights of additional operational limits. The Modis convergence time of processing is found to be about 150 seconds with the maximum node number. From the same graph, it seems that above 1500 operations is the minimum complexity of pixel processing that this Modis image would benefit from being processed in a distributed manner.

Page 10: 8/10/10 1 NDVI Calculation of RS Data

23/04/12 10

Parallel Methodology

MSVI

………

NDVI

RVI

Server 1

MSVI

………

NDVI

RVI

MSVI

………

NDVI

RVI

Server 2

Server n

Client Side

Server Side

Row wise distribution is preferred because of single pixel executions timing is only less

amount of time

Page 11: 8/10/10 1 NDVI Calculation of RS Data

23/04/12 11

Difference with r.vi

r.vi is a full package. r.vi is run able in GRASS environment. Where as this NDVI calculation program is

simple C code.

r.vi extracts a pixel by pixel information and there after distributed whole row at a time.

Where this C program extracts all pixels information (Dr.Honda`s C library) and puts them in 2D arrays. Then the array is going to process.

Page 12: 8/10/10 1 NDVI Calculation of RS Data

23/04/12 12

Grid Version

NDVI_ALL folder…..

Page 13: 8/10/10 1 NDVI Calculation of RS Data

23/04/12 13

December Task

Experiment on the running time for all three versions.

Compare the timing.. Sensei`S requirements…… I will present Introduction about

Remote Sensing …