Image complexity can provide useful information about the texture or material properties of the acquired scene. In agriculture, the complexity of remotely-sensed images of soil or land cover may reveal underlying properties of soil and/or vegetation. However, the perceived complexity may vary along scales, as a function of the altitude of the sensor in a remote sensing scenario. In this paper, we investigate how the fractal complexity of multi-spectral images acquired using an unmanned aerial vehicle varies with the altitude, in a soil roughness estimation application. We adapt a definition of the multi-spectral fractal dimension to assess the fractal complexity of images with 5 spectral bands and analyze the computed fractal complexity as a function of both altitude and the number of considered spectral bands.