Constraining dynamic TOPMODEL responses for imprecise water table information using fuzzy rule based performance measures

Abstract

Dynamic TOPMODEL is applied to the Maimai M8 catchment (3.8 ha), New Zealand using rainfall–runoff and water table information in model calibration. Different parametric representations of hillslope and valley bottom landscape units (LU's) were used to improve the spatial representation of the model structure. The continuous time series water table information is obtained from tensiometric observations from both near stream (NS) and hillslope ( P5) locations having different responses to rainfall events. For each location, and within an area equivalent to an effective model gridscale (25 m 2), a number of tensiometer readings at different depths were available (11 for the NS site and nine for the P5 site). Using this information a distribution of water table elevations for each time step at each location was calculated. The distribution of water table elevations was used to derive fuzzy estimates of the water table depth for the whole time series that includes the temporal variability of the uncertainty in the observations. These data were used to constrain the spatial representation of the model having previously conditioned the model using the rainfall–runoff data. Model conditioning was assessed using the Generalised Likelihood Uncertainty Estimation procedure. Results show that many combinations of parameter values for the two LU's were able to simulate the rainfall–runoff data. Further constraining of the model responses using the fuzzy water table elevations at both locations considerably reduced the number of behavioural parameter sets. An evaluation of the distributions of behavioural parameter sets showed that improvements to the model structure for the two LU's were required, especially for simulations of the response at the hillslope location.