Transformation of Atmospheric and Solar Illumination Conditions on the CCRS Image Analysis System

A software package for the transformation of atmospheric and illumination conditions (T ASIC) has been implemented on the Canada Centre for Remote Sensing's Image Analysis System. This package offers three different transformations: (1) transformation to reflectance units; (2) transformation of illumination conditions; and (3) transformation to radiance units under standard atmospheric and illumination conditions. Atmospheric parameters can be derived from four different sources, allowing users considerable flexibility. In particular, the method of obtaining atmospheriC information from the LANDSAT signals from clear water bodies (Ahern et al. 1977 a, b) has been incorporated as one of the options. It offers high accuracy without the requirement of additional information derived from ground observations. The T ASIC procedure, including the acquisition of two-dimensional atmospheric information, is described in detail. Examples are given demonstrating the removal of atmospheric and solar illumination variations in a sequence of six LANDSAT images of the same area obtained between May and September 1976. Systematic errors introduced by the uncertainty in the absolute calibration of the LANDSAT multispectral scanner are the most significant errors. The radiative transfer model used in the T ASIC software also introduces significant systematic errors. Random errors are less important in the present procedures. It is estimated that 60 to 90 percent of the variations due to atmospheric and illumination changes can be removed using the TASIC algorithm. INTRODUCTION In order for remote sensing measurements to have the greatest possible utility, the influence of factors unrelated to the targets of interest must be removed. This greatly facilitates multi-date and multi-sensor inter-comparisons and the interpretation of the remote sensing data in terms of intrinsic properties of the targets of interest. In the reflective part of the electromagnetic spectrum, the intrinsic property of interest is reflectance, while the most important complicating effects are those due to variations in solar illumination and to atmospheric transmission and patti radiance. Transformations taking these solar and atmospheric effects into account can be applied to measured radiances in order to obtain reflectance values or radiances under specified atmospheric and illumination conditons. Three such transformations have been implemented in a software package now available on the Canada Centre for Remote Sensing's Image Analysis System, which is built around a modified General Electric IMAGE 100 (Goodenough 1977, 1978). II. APPROACHES TO ATMOSPHERIC AND ILLUMINATION CORRECTIONS A. TYPES OF OUTPUT IMAGE DATA 1. Transformation to Reflectance Units. In the case of a multispectral imaging sensor such as the LANDSAT multi-spectral scanner, the measured quantity is the radiance L in each spectral band for each of the many pixels in the scene. The desired intrinsic property is usually the hemispherical reflectance p.For a perfectly diffusing reflector viewed through the atmosphere, the reflectanc~ is p = 1T(L L ) P HT (1) where H is the total downwelling irradiance, T is the atmospheric transmission, and L is the atmospheric path radiance. p The most desirable output from a transformation of atmospheric and solar illumination conditions (T ASIC) would be an image calibrated in reflectance units. This would relieve the user of any further worry about atmospheric and illumination effects in the scene of interest. The T ASIC procedure at CCRS can provide this form of output when the user is confident that the necessary accuracy is achievable with the data available. This usually implies using a relatively clear scene with several clear water bodies within a few kilometres of the study area, or independently acquired atmospheric transmission data for the time of satellite overpass. 34 1977 Machine Processing of Remotely Sensed Data Symposium CHI430-8/79/0000-0034$OO.75 © 1979 IEEE 2. Transformation of Illumination Conditions. If detailed atmospheric information is not available, the radiances of a scene can still be transformed to those which would be observed under different illumination conditions. This transformation is of the form Hn (L-L ) + L P pn (2) H where the subscript n refers to the new illumination conditions. 3. Transformation of Scene Radiances Under Standard Conditions. For some applications, the researcher may want a scene corrected to radiances which would be measured under conditions of standard illumination and atmospheric conditions rather than correcting to reflectances. The equation for such a correction is L s H T (L L ) f L ssp ps HT (3) where the subscript s refers to standard conditions. This option is also available at CCRS through the T ASIC software package. For good results, this approach requires the same high quality atmospheric information needed to correct to a reflectance image. In fact, the transformation given by equations (1) and (3) are linearly related through the constant parameters H , T d L • s s an ps B. IDENTIFICATION AND DETERMINATION OF REQUIRED INPUT PARAMETERS ~. Bright/Dark Reflectors. The most straightforward method of converting radiance measurements to reflectance values results from the fact that equation (1) is linear. If a scene contains one or more relatively dark objects of known reflectance and one or more relatively bright objects of known reflectance, these can be used to establish the linear relationship between radiances and reflectances. The rest of the scene can then be converted using the resulting linear equation. This method, fully demonstrated by Stllnz (1978), has not yet been widely applied because the reflectance properties of natural targets are not sufficiently well understood to allow specification of known bright and dark calibration objects in the majority of LANDSAT (or airborne MSS) scenes. Since atmospheric and illumination problems cannot generally be circumvented with calibration reflectors, they must be addressed at a more fundamental level. Therefore, each of the factors in equation (1) must be considered and evaluated by some means. 2. Total Downwelling Irradiance. The total downwelling irradiance H (also known as the scene illumination) consists of a direct component contributed by light coming directly from the sun to the target, and a diffuse component contributed by light from the sky. The direct component is a function of the earth-sun distance, the solar zenith angle, and the atmospheric transmission. The earth-sun distance and solar zenith angle can be determined from the time of data acquisition. Since attenuation of the direct component is primarily due to scattering by molecules and aerosols with single-particle albedos near unity, most of the radiation removed from the direct component of downwelling irradiance becomes part of the diffuse component. To a very good approximation, the total downwelling irradiance is independent of atmospheric transmission even though the direct and diffuse components are sensitive to changes in atmospheric transmission. Band 7 of the LANDSAT MSS is an exception because water vapour absorption is an additional source of direct beam attenuation in that bandpass. This effect has not been taken into account in the present atmospheric correction procedures, but there is no evidence that its ommission gives rise to adverse results. In short, the total downwelling irradiance can be computed to high accuracy from a knowledge of the time of data acquisition and the resulting illumination and viewing geometry. This allows illumination corrections to be made independently of detailed knowledge of the clarity of the atmosphere through which the observations were made. For a complete conversion to reflectance, another multiplicative factor correcting for atmospheric transmission and an additive term correcting for path radiance must also be known. 3. Atmospheric Transmission from Ground Measurements. The atmospheric transmission can be measured from the ground by observing the sun as a standard source of irradiance in a technique known as the Langley method (well explained by Rogers and Peacock 1973 ). Deepak and Box (1978 a, b) have recently shown how to correct these observations for the significant diffuse component included in the field of view of the measuring instrument if there is some knowledge of the scattering particle albedo and size distribution. 4. Path Radiance. (a) Ground Measurement: Rogers and Peacock (1973) have demonstrated that a technique originally suggest by Gordon(1973) can be used to infer path radiance from ground-based observations of sky radiance. This method gives results which agree well with those of accurate model atmosphere calculations (Miller and O'Neill 1977). How1977 Machine Processing of Remotely Sensed Data Symposium 35 ever, the method requires the sun to be less than 45° above the horizon, with the consequence that observations to determine path radiance must often be separated in time from the actual satellite observations. (b) Atmospheric Modelling: Alternatively, it is possible to use a variety of atmospheric models to calculate the path radiance once the atmospheric transmission, the illumination and viewing geometry, and the mean background albedo are known. Unfortunately, in choosing an atmospheric model, a compromise must be made between speed and accuracy. This has partially been overcome at CCRS by the creation of a hybrid between the popular Turner model (Turner and Spencer, 1972) and a discrete ordinate calculation (O'Neill and Miller 1977, O'Neill et al 1978). In the latter work, the discrete ordinate method was found to give good absolute agreement between calculated and observed values of zenith sky radiance and nadir path radiance. The hybrid model computes the ratio of path radiance interpolated from the tables of O'Neill et al (1978) to Turner path radiance calculated for the same conditions. This ratio is then used to correct Turner model calculations for other viewing angles. For input to the model atmosphere calculation, the illumination and viewing geometry can be computed from the time and location of data acquisition, and the atmospheric transmission is provided by ground observations through the Langley method. The average background albedo has second-order influence on the path radiance. Hence, typical or measured values for a given area and date can be used, or satellite data converted to reflectances with a standard atmosphere can be used. All three approaches are available with the T ASIC algorithms. 5. Clear Water Bodies. Ground-based measurements of specific parameters such as atmospheric transmisison are acceptable for research projects where the necessary manpower and instrumentation can be made available. For operational applications of remotely sensed data, it is preferable to extract the necessary information from the satellite data themselves whenever possible. Previous work has shown that it is possible to extract both path radiance and transmisison information from LANDSAT data if clear water bodies are present in the scene (Ahern et al 1977 a, b). The technique has been implemented in the T ASIC software package, enabling users to correct all or part of a LANDSAT scene containing clear water pixels. More specifically, measurements of path radiance over clear water pixels in the scene are fitted by a twodimensional polynomial of order < 2. Then, inversion of the hybrid atmospheric model discussed in the previous section yields atmospheric transmittance. III. OPERATION OF THE T ASIC SOFTWARE The most involved procedure in the software package is the use of clear water bodies for the determination of atmospheric parameters. The more straightforward procedures involving the input of independent atmospheric data or the transformation to other solar illumination conditions are described in section E. The conversion to reflectance units by the use of atmospheric information derived from clear water bodies can be divided into five logical stages, each of which requires a few minutes on the CCRS Image Analysis System (CIAS). Most stages require some judgement by the user or operator, so the procedure has been designed for easy interaction. The five stages are: (a) loading the appropriate portion of the LANDSA T scene of interest from disk or tape into the CIAS memory; (b) setting thresholds to isolate clear water bodies; (c) determination of the average path radiance over water bodies; (d) fitting of a two-dimensional function to path radiance over the scene; (e) creation of a two-dimensional radiometric correction function from path radiance and solar illumination information, followed by transformation of the scene (digital) values with that function. Each of these steps will be discussed in more detail.