Variational data assimilation is a method for finding an estimate of the state of a system using both noisy observations and a nonlinear, imperfect model that predicts the state of the system. The particular method used in modern numerical weather prediction is four-dimensional variational data assimilation (otherwise known as 4DVar). The very large, ill-posed 4DVar minimisation problem can be written as a nonlinear Tikhonov regularization. We use this equivalence firstly to apply well-known parameter choice methods for Tikhonov regularisation to variational data assimilation. Secondly we adopt a form of TV-regularization for variational data assimilation, which -in the presence of model error and background - produces a more accurate initial condition for the forecast.