We explore three options to predict subsurface water distribution from trickle emitters using artificial neural networks.  The first step is to prepare input-output databases from a numerical model (HYDRUS-2D).  From the database we evaluate feasibility and advantages of the three alternative options:  water contents at given coordinates, moment analysis describing the shape of the plume, and finally coordinates of individual moisture contours.  The best option is determined somewhat by the application objectives, but preliminary results suggest that predictions based on moment analysis is probably the most versatile and useful for describing water distribution.