The most important requirements for large-area environmental modeling are a tight integration between models and data, and a close match of the spatial scale at which the model is developed with the scale at which the model is to be applied. To better match the scale of data with that of the model, we propose a set of principles for the development of stochastic modeling systems based on linkage of deterministic models with GIS data. For modeling purposes, a region is usually rasterized into cells and the environmental conditions of those cells are specified by ranges or classes using GIS data layers. It is not necessary to simulate each and every GIS cell in the study area because many cells may have similar environmental conditions and can be grouped together to form cohorts. We define a cohort as the assembly of the cells sharing a unique combination of environmental conditions within the study region. Multiple model simulations can be performed for any given cohort. For each simulation, some of the parameter values can be randomly generated within the specified environmental conditions of the cohort according to a certain statistical distribution which, in turn, can be specified by GIS data layers. By this method the variance and covariance of environmental variables in space and time are integrated into the simulation processes with these modeling systems to make full use of the available data and to assess the uncertainties of the simulated results. An integrated simulation system between CENTURY model and GIS was developed to demonstrate the value of the concepts imbedded in stochastic simulation systems for large area studies.