A production system performs automatic transformation of remote sensing imagery into useful information (such as biophysical parameters, categorical maps etc). An example can be a preliminary pixel-based classifier that works top-down (deductive, physical model-driven, prior knowledge-based) and arrives at preliminary classes for each pixel of an image. Such a production system does not require interaction of an operator. The process makes use of a decision tree that encodes the prior knowledge for assigning pixels to a class.