Keywords: Cloud-free, Balance, Rank, Mosaic
Abstract
An algorithm is described for automatic generation of "cloud-free" scenes from many SPOT
images over a given region. It is commonly known that remote sensing using optical sensors
often suffers from the presence of cloud covers, especially over the humid tropical regions. This
problem can be partially overcome by acquisition of multiple images within a specified time
interval over a given region. It is assumed that the land covers do not change within this time
interval. By mosaicking the cloud-free areas in the set of images, a reasonably cloud-free
composite image can be made. In this paper, we explicate the approach for generating "cloud-free"
multi-scene mosaics of SPOT images.
1. Introduction
Optical remote sensing always encounters the problem of cloud covers, especially over the
tropical areas. However, if multiple images acquired at different time over a given region are
available, then it is possible to generate a reasonably cloud-free composite scene by mosaicking
the cloud-free areas in the set of images. We have previously reported an operational algorithm
for generating such cloud-free mosaics from multispectral images acquired by the SPOT
satellites (Liew, 1998). In this algorithm, an intensity-thresholding method was used to identify
the best cloud-free and non-shadow pixel among the pixels from the multiple images of a given
region. In this intensity-thresholding method, bright pixels of land surfaces or buildings could be
mistaken as cloud pixels, and they were ranked inferior than the cloud shadow pixels. Hence,
these bright pixels were often replaced by cloud shadows. In areas covered by thin clouds or
haze (especially over low-albedo vegetated areas or water surfaces), the hazy pixels would be
selected instead of the cloud free pixels which were mistaken as cloud shadows.
In this paper, an improved algorithm for generating cloud-free multi-scene mosaics of SPOT
images is reported. This improved algorithm avoids the pitfalls of the previous algorithm by
making use of band ratios to classify the non-cloud and non-shadow pixels into one of three
broad classes: water, vegetation or buildings. The intensity-based ranking rules were then
modified accordingly to favour the selection of these "good pixels" instead of cloud shadows or
thin clouds.
2. Description of the Algorithm
Figure 1 shows a schematic diagram of the operational system for generating cloud-free mosaics
from SPOT images. With minimal modification, this system can also be used to generate cloud-free
mosaics from optical images acquired by other satellites (e.g. Landsat-TM).
Fig. 1: A schematic diagram of the cloud-free mosaic generating system
2.1 Input Images
The inputs to the system are SPOT multispectral images of the same region acquired within a
specified time interval, pre-processed to level 2A or 2B. The images are also co-registered before
being fed into the system. The SPOT HRV sensor, when operating in the multispectral mode,
captures data in three spectral bands, i.e. the green band (Band 1, 0.50 to 0.59 µ m), red band
(Band 2, 0.61 to 0.68 µ m) and near-infrared band (Band 3, 0.79 to 0.89 µ m). In the conventional
false-colour display of a SPOT multispectral image, band 3, band 2, and band 1 are assigned to
the red, green and blue display channels respectively.
2.2 Radiometric Balancing
The radiometric balancing procedure assumes a Lambertian surface. It only makes a correction
for differences in sensor gains, solar incidence angles and solar flux between the acquired scenes
and no attempt has been made to have a correction for atmospheric effects. Suppose that a set of
N images is available for a given scene. For each band-k of the set of images, the pixel values are
radiometrically balanced with respect to each other according to
where the additional subscript n (n = 0, 1, 2, ..., N-1) is an index identifying the individual
image within the set of N images, R
k,n is the pixel digital number of the radiometrically
balanced image, DN
k,n is the pixel digital number of the input image, g
g,n is the sensor
amplifier gain. F
k,n is the extra-terrestrial solar flux,
qn is the solar elevation angle, c
k is a
multiplicative constant (same for all n) which stretches the balanced digital numbers to fill the
range from 0 to 255. The above equation assumes that the ground reflectance did not change
during the time interval within which these N images were acquired.
