Studies of land use impacts on stream flow and water quality have generally concentrated on the role of surface water runoff. However, in typical upper Midwestern watersheds, a large majority of annual stream flow is derived from shallow groundwater. Like overland flow, groundwater flow integrates solute inputs from a range of land use types from source areas to streams; groundwater flow however has longer transport times ranging from days to hundreds of years. Therefore, water recharging to streams could be decades old, potentially originating from presettlement landscapes.

We have built models that attempt to characterize the influence that long-term groundwater transport has on current water quality signals. This has been accomplished using a backcast model, which creates historical land use/cover maps using a reverse-time driven land change algorithm, coupled to a MODLFOW-based groundwater travel time model. This spatial-temporal model is used to create “land use legacy" planning maps that link the contribution of historic land uses to the groundwater signal arriving at streams.  

In developing these models, we have recognized that several sources of uncertainty exist. This paper presents the results of an uncertainty analysis of this coupled model system. We conducted a “parameter sweep” for ground water travel time model settings across a likely range of values and coupled this to a “rule-sweep” of various land use transition rules in order to quantify the impact that model uncertainty has on these land use legacy planning maps.  Model uncertainties in several classes – parameter, input, structural and application -- were examined across local units of government responsible for land use planning. We address the research question: how do each of these sources of uncertainty impact the use of land use legacy maps for planning and natural resource management? 

Results/Conclusions Our results show that all four uncertainties classes impact the composition and structure of land use legacy maps.  However, we found that a majority of local planning units are not impacted by parameter, input and structural uncertainties making our model a robust for quantifying the role that historical land use has on current water quality. In some cases, even a moderately poor fit model (using AROC) produces potentially satisfactory results.  Finally, with regard to structural uncertainty, we found that a simple land change model produced the same outcome as a complex model composed of multiple transitions rules.  We discuss how this model can be used for planning and natural resource management.