Many studies of the error growth in weather forecasting have focused on the extra-tropical quasi-geostrophic dynamics and often considered the error-free large-scale initial state. In contrast, the operational global numerical weather prediction (NWP) models and ensemble prediction systems  are  characterized by initial uncertainties at all scales, especially in the tropics. I will discuss the evidence of the role of the large-scale error growth in NWP systems early in the forecasts in comparison with the error cascades from the smaller scales. Then I will derive a new parametric model for the representation of the error growth.  In contrast to the existing parametric models, application of the new model does not employ time derivatives of the empirical data. The asymptotic error is not a fitting parameter, but it is computed from the model constants. The model application to the errors simulated by the ECMWF proves the model robust to reliably represent the error growth dynamics across many scales.