This paper addresses the problem of color changes caused by variation in natural illumination, both in intensity and spectral content. Dealing with intensity changes is relatively easy using color ratios, whereas changes in spectral content present more of a problem. Our previous work has shown that natural daylight can be represented in a form whereby it is possible to derive a monochrome image from a 3-band color image that is invariant to illumination spectral changes. Here we illustrate the method with an application from one of our important problem domains, precision treatment of agricultural crops using cybernetic machinery. We collect images of two components, vegetation and soil, and derive statistical models of the pixels that make up the images for three illumination conditions. We show that the models for the same component are quite different when using color ratios but much more similar when using our invariant transformation as the illumination changes. We conclude that cybernetic applications of outdoor machine vision must be able to deal with illumination spectral changes and that our invariant transformation is a suitable method to do this.