In this work we investigate the mechanism of photodesorption of water from a WO3(001) surface by theoretical calculations, applying an embedded cluster model. Using the CASSCF method, we have calculated both the ground state as well as the energetically preferred charge-transfer state in three degrees of freedom of the water molecule on the surface. The calculated potential energy surfaces were afterwards fitted with a neural network optimized by a genetic algorithm. A final quantum dynamic wave packet study provided insight into the photodesorption mechanism.