ABSTRACT One of the steps in the construction of thematic products is the categorization. This process is done by partitioning the range of values of a given feature into several subranges and associating a new value, in this case ordinal, to all instances that are in that subrange. The objective of this work is to propose an innovative pre-categorization process that applies a computational search method to maximize the accuracy of thematic products during the classification stage. The theme feature used is the Above Ground Biomass (AGB), which estimation model is built over synthetic aperture radar features. A system is developed in Java script using the Weka data mining class library. The proposed heuristic is the hill climbing greedy, with the Kappa coefficient as the objective function. The results obtained shows that the proposed Categorization Optimization algorithm demonstrated the ability to obtain new states with subintervals of categories that increased the Kappa agreement index to 1.0 with much lower computational cost than the exhaustive search. The thematic products constructed maintained the representativeness of the study area while increasing in thematic accuracy.