We may not be able to command the water to stop but we can take steps to predict when and
where it will invade and attack our lives, and provide solutions to deal with the problem. The
research project reported in this paper is concerned with a study of unsteady free surface
water flow, a hydrograph, resulting from a watershed just after the outlet station. The quality
of flood predictions by numerical models depends on the accuracy of the inflow hydrograph,
other control variables such as bed roughness, infiltration rate, and channel topography.
However, none of these are well known, the values of each are uncertain. This research
examines what effect these uncertainties have on the flood prediction. That is we
find out how the uncertainties in control values propagate through the model. This is achieved
by calculating the sensitivities of the flood predictions to changes (uncertainties) in control
variables. The adjoint method is used to study the sensitivity of the flow to changes in the
boundary and initial conditions. To achieve this aim we constructed a numerical hydraulic
model to simulate the flow of water in the main stream based on the shallow water equation
(SWE). The sensitivities are determined using the adjoint method which uses a variational
technique to find the relationships between changes in channel flow conditions and changes in
control variables such as the inflow hydrograph. This could be done at significant
computational expense using multiple runs and ensemble techniques however the adjoint
method presented here determines these sensitivities analytically in one run of the model.