ABSTRACT
   Spatial scale effects are an important issue in hydrological modeling, because the need to resolve scale differences between processes, observations and modeling. The main factors driving the scale problem are the existence of particular dominant processes at different scales, the nonlinear behavior of the hydrological systems and a strong spatio-temporal variability at different scales. Knowledge improvement related to scale effects and scaling is valuable for a better understanding and representation of processes.
   The first part of this thesis deals with the spatial aggregation problem, spatial organization, and the relationship of representative elementary area with the existence of an optimum cell size for distributed hydrological modeling. Infiltration conceptualization of TETIS hydrological model has been used to carry out this analysis. A synthetic experiment was conducted with randomized random parameter fields. Results showed that the scaling of effective parameters from microscale to mesoscale derives in non-stationary effective parameters, which depend on input variables and microscale parameters heterogeneity. The stochastic simulations showed that the variance of the estimated effective parameters decreases when the ratio between mesoscale cell size and correlation length at microscale increases. For a ratio greater than 1, we found cell sizes having the characteristics of a representative elementary area (REA); in such case, the microscale variability pattern did not affect the system response at mesoscale
   In the second part, the problem of scaling is driven by using non-stationary effective parameters. To this end, analytical and empirical scaling equations were developed. Scaling equations performance was tested by Monte Carlo simulations and conducting hydrological modeling in an experimental watershed. The incorporation of these scaling equations in hydrological modeling and the application in Goodwin Creek experimental catchment revealed the importance of sub-grid variability representation. Particularly, the use of scaling equations implies a better model performance in validations at internal stream-gaging stations and for the smallest storm events.