Generation of cloud free image from NOAA AVHRR data

Generation of cloud free image from NOAA AVHRR data

Tsukasa Hosomura*, Haruhisa Shimoda and Toshibumi Sakata
Tokai University Research & Information Center
Tomogaya 2-28-4, Shibuya-ku, Tokyo 151, Japan
*AIT/Division of Computer Science
G.P.O Box 2754, Bankok 10510, Thailand

Introduction
This paper describes about some methods for generation of cloud free image from NOAA AVHRR data. Target area of this study is large paddy fields spreading over Hua Zhong and Hua Nan area, i.e. basin of yang Zi river. This are has a regional scale and it is very difficult to get ground informations in a timely manner. NOAA AVHRR data is optimal for such monitorings. In this study, a system in order to obtain the necessary informations of paddies in this area within certain time interval by using NOAA AVHRR are examined.

This system is composed of the following steps. NOAA AVHRR images are connected in Tokai university Space Information Centre (TSIC) in Kumamoto, Japan. These data re transported to Tokai University Research & Information Center (TRIC) in Tokyo and processed there. Data processing are composed of two steps. First step is preprocessing which include geometric corrections and readiometric correction. Geometric corrections were conducted to precisely co-register each images. Radiometric corrections were performed to eliminate the radiometric distortions mainly caused by incidence angle difference of sub light. In the second step, a cloud free image of the target area was generated by combining images within one or two weeks.

Image data used in this study

Test image data
Test image data used in this study are as follows :

DATA 1	NOAA AVHRR 1988.MAY.15.AM.8:43, NPOAA-10
DATA 2	NOAA AVHRR 1988.MAY.18 PM. 16:52, NOAA-9

A NOAA image of the target area is show in fig. 1 and fig. 2 target area of this study is Hua Zhong and Hua Nan area, i.e basin of Yang Zi river.
Fig. 1 NOAA AVHRR Image 1
(1988.5.15. AM. 8:43)

Fig. 2 NOAA AVHRR Image 2
(1988.5.18. PM. 16:52)

Preprocessing
There are two kinds of preprocessings, geometric correction and radiometric correction. These preprocessings are very important in these processings.

Geometric correction
In this section, geometric correction of AVHRR data are described. In the geometric correction for this study there are following tow problems.
1. Higher Geometric accuracy is required to co-register AVHRR data fro clouds elimination.
2. Faster processing algorithms for geometric corrections are required to process large quantity of data within a limited duration.
Following processings have been adopted to solve these problems. Faster and accurate geometric corrections were performed by three step processings. At the first step, tangential and earth curvature distortions ere corrected using tale look up algorithm.

In the second step, geometric corrections using orbital elements were accelerated with the aid of scan and pixel functions. One dimensional 3trd and 2nd order polynomials for each function were sufficient to maintain within 1 pixel relative accuracy.

Last step is the co-registration process of images. As most of images are largely covered by clouds, cloud free areas of each image were first selected and a correlation technique was used to determine control points. Images were then superimposed with the aid of these control points. With these techniques, co-registration of images were achieved within 1 pixel accuracy.
Radiometric correction
Radiometric distortions mainly caused by incident sun light should be eliminated. In order to eliminate radiometric distortions, sun angle corrections were first applied to geometrically corrected images. However there exists brightness differences between different date images mainly caused by atmospheric conditions. These differences were normalized by histogram normalization process using pixels which can be thought to be in the same conditions.

Clouds elimination
The most primitive idea to generate cloud free image from images is that the channel 1 and 2 values of cloud parts are larger than those of cloud free area. However, the method using this idea also picks up shadows of clouds. I order to avoid this defect, thresholding was introduced to eliminate shadow areas. Thresholding caused another defect that water areas like lakes and rivers were sometimes eliminated as shadows. From these reasons therefore, the method using original data values could not applied directly. In order to eliminate clouds and shadows simultaneously, the method using NVI (Normalised vegetation index) was used. NVI can be calculated by the following equation :

N.V.I. = (Channel 2 - Channel 1) / (channel 2 + Channel 1)

The N.V.I. is then scaled as follows :

Scaled N.V.I. = 240 - (N.V.I. + 0.05) x 350

As shown in the above equation, scaled NVI has larger cloud in cloud or shadow area where the difference of Ch. 1 and Ch 2 is relatively small. On the contrary, scaled NVI in water are has not so large value because there exists some differences between Ch. 1 and Ch. 2 cloud free images could be generated by taking the area which have the smallest scaled NVI value.

Conclusions

The cloud elimination system was established
The accuracy of systematic geometric correction was improved by introducing tangential and earth curvature correction.
Radiometric correction were performed by using sun angle corrections and histogram normalized techniques.
Two kinds of cloud free images were generated fro monitoring the rice growing stages. One is the false color composite image of band 1 and band 2. The other is the use of normalized vegetation index (N.V.I.).

Fig. 3 NOAA AVHRR Cloud Free Image