Abstract: Remote sensing has been expected to provide appropriate information for land cover classification and multi-temporal monitoring of natural environment ( agriculture, forest, geology and etc.) as it can cover wide area periodically. In order to utilize satellite images, however, we have to consider many kinds of measurement uncertainty, which are categorized into systematic error and random error. The systematic error should be corrected based on physical models and appropriate data if they are available, and remained uncertainty should be treated statistically.
In the case of optical sensor, we need geometric correction and radiometric correction. The light traveling through the atmosphere is scattered and absorbed not only by molecules but also aerosols and clouds. Slopes facing toward the sun receive more light and appear brighter than slopes facing away from the sun. In the case of rugged terrain like most part of Japan, some topographic effects are entangled with atmospheric effects. Therefore, we first have to precisely lay the images over the map by the geometric correction including ortho-rectification as the satellite images are center-projected. Then atmospheric and topographic effects on radiance detected on board are corrected to obtain ground surface reflectance based on the physical principles using the digital elevation model and meteorological data.
Ten years ago, there were many obstacles to process and analyze the satellite images properly for users in application fields. The images, both geometrically and radiometrically, were not so accurate as they are now provided. The users were not given the ortho-rectified images, nor the information and the digital elevation model to perform the ortho-rectification. There were some practical topographic correction methods, which were only applicable to images over moderately rugged terrain and high sun elevation season. The parameters of the correction were usually estimated statistically. The physical model of atmospheric effects were known as the radiative transfer model, and the computer simulation code such as 6S were available, but they could not estimate the spatial variation of sky light and reflected irradiance from the adjacent slope over rugged terrain. Although these effects were known theoretically, their calculation was thought impossible to perform for the hardware and software at that time.
Since then, there have been much progress in the satellite imagery itself as well as methods and data for image processing. Satellite position and attitude are well monitored and controlled by using GPS and star trackers. In order to ortho-rectify satellite images easily, coefficients of rational polynomials are attached to some high resolution satellite images. Computer improved in quality and efficient algorithms for calculating horizon and view-shed are developed. In order to mitigate the random errors, spatial averaging over segmented regions seems useful. By exploiting these progress, the ground reflectance and land cover class can be estimated much more accurately, but the satellite image with low sun elevation over rugged terrain are still difficult to analyze.

Brief Biography of the Speaker:
Yoshikazu Iikura is a full professor with the Faculty of Science and Technology at Hirosaki University, Japan. His current research interests are basic processing of satellite images : geometric correction, illumination correction, and land cover classification. He received the BS, Ms, and Dr. Eng. degrees in instrumentation physics and mathematical engineering from the University of Tokyo, Japan. As a student, he studied statistics (discriminant analysis) and information theory (rate distortion theory).
He worked for National Institute for Environmental Studies (Japan) and was engaged in a project for lidar monitoring of stratospheric aerosol. In order to retrieve aerosol concentration, he developed some signal processing method: removing systematic noise caused by photomultiplier, calibration of the signal using Rayleigh scattering profile of upper stratosphere, etc.
In late 1980s, he moved to Iwate University and began the study of satellite image processing. At first, he tried to apply the best linear discriminant function to the image but found that atmospheric and topographic effects should be corrected beforehand. He also noticed the need of digital elevation model and radiative transfer code. He invented the automatic geometric correction method using simulated illumination image as a reference and improved the practical topographic correction method based on the physical models. For these achievements (including lider signal processing), he won the best paper prize of the Remote Sensing Society of Japan three times.
He has been an IEEE GRSS member since 1990, and served as a chair of the IEEE GRSS Japan Chapter (2008-2009). He is currently a director of the Remote Sensing Society of Japan.