The question is sometimes asked whether it is necessary to apply any correction for the effect of the atmosphere. The answer depends upon the wavelength region being sensed, the height of the platform and the purpose for which the data will be used. The following discussion relates to data from aircraft and satellite sensors operating in optical wavelengths (0.4 - 2.4µm). In the simplest case, where a single image is being analysed and it is not necessary to generalise the results to other sensors, other times or other places, it is feasible to analyse remotely sensed data as raw digital number (DN) values, with no conversion to any physically meaningful units. Although this approach may be acceptable for a one-off image classification, in most applications there is much to be gained from (a) converting DN values to 'at sensor' radiance (radiometric calibration) and (b) considering the influence of the atmosphere on the measured signal. The aim of atmospheric correction is to convert the 'at sensor' or 'top-of-atmosphere' (TOA) radiance to ground-leaving radiance. The benefits of atmospheric correction include:
- The potential to substitute data from different sensors or platforms, should the primary choice not be available due to cloud cover, orbital limitations or technical problems;
- More reliable results, especially those based on vegetation indices, as the atmosphere introduces different levels of error and uncertainty into the individual bands used to create the index;
- Some methods of atmospheric correction also improve the spatial definition of objects and edges as they include correction for 'adjacency effects'. This sharpening of the image benefits visual image interpretation, so it is not only computer-based interpretation that benefits from atmospheric correction.
Although atmospheric correction is important, it still does not achieve the overall aim of making remotely sensed measurements completly independent of the conditions of measurement. For that goal to be reached, we have to convert the data from ground-leaving radiance to surface reflectance. Some methods of atmospheric correction provide the means to perform this additional step, but it is important to recognise that this often involves major and often untested assumptions which are described in more detail in the section on brdf.
Procedures for atmospheric correction are often included in professional image processing software packages, and there are also a number of stand-alone software products that provide more flexibility and power, but the average user is often uncertain of the merits of different methods, or of the validity of the various models and algorithms. This is especially true when a user wishes to process a data set collected under very different conditions to those which prevailed during the developing and testing of the model. This is often the case for those working in humid temperate latitudes as most of the models were developed and tested under clearer skies. In this section of the NCAVEO website we will describe the application of a number of different atmospheric correction procedures to data acquired by aircraft and satellite sensors from test sites in the UK.
Example 5. Atmospheric correction of an airborne hyperspectral image. (in preparation)
Staff in the Underpinning Technologies Group at CEH Monks Wood have developed a user-friendly front-end for MODTRAN, details of which are available here. For those interested in a more interactive model of the general interactions that take place in the atmosphere, an Excel implementation of the Bird-Riordan Simple Solar Spectral Model can be freely downloaded from http://rredc.nrel.gov/solar/models/spectral.
This work is licenced under a Creative Commons Licence.