Weather and climate predictions are very important topics because of their impact on several activities of society. One essential feature of a prediction system is to compute the best initial condition for starting a forecasting cycle. The procedure to identify the best initial condition is a method by combining observation data from a dynamical system with data from a previous prediction, and this process is called data assimilation (DA). Some non-linear time evolution differential equations present dynamics very sensitive to any tiny changes of initial conditions, exhibiting a chaotic dynamic. Hence, our experiments applying Cell-NN are performed by using the classical Lorenz chaotic model. The methodology is described, where the Cell-NN is presented, then the Lorenz model is shown, and methods for data assimilation (variational and Cell-NN) are explained. The configuration of algorithms is introduced and results for numerical experiments are shown.