Using a climate version of a regional atmospheric model, we show that the seasonal evolution of weather is dependent on the initial soil moisture and landscape specification. Coupling this model to a land-surface model, the soil moisture distribution and landscape are shown to cause a significant nonlinear interaction between vegetation growth and precipitation. These results demonstrate that seasonal weather prediction is an initial value problem. Moreover, on seasonal and longer timescales the surface characteristics such as soil moisture, leaf area index, and landcover type must be treated as dynamically evolving dependent variables, instead of prescribed parameters. ¿ 1999 American Geophysical Union