|Time, resource, and replication constraints limit the practicality of conducting agricultural experimental studies
at scales larger than plot‐level. Thus, watershed‐level models such as the Soil and Water Assessment Tool (SWAT) are
increasingly used to forecast effects of land management changes on downstream water quality. With the generalization in scale, the question of effect of generalization in input data arises. That is, to what extent does having field‐level, daily input data for a watershed model aid in the ability to predict watershed‐scale, water quality impacts. The study site, FD‐36, is a 39.5 ha agricultural subwatershed of a long‐term USDA‐ARS study watershed in south central Pennsylvania. FD‐36 is characterized by loamy soils with a substantial near‐stream fragipan. Fifty percent of FD‐36 includes 24 row‐cropped fields from three independently managed farms. Two SWAT scenarios were simulated on FD‐36 and compared with each other as well as with measured data over two 4‐year periods (1997‐2000 and 2001‐2004). The high‐resolution scenario modeled seasonal crop, fertilizer, and tillage events of each row‐cropped field continuously over the 8‐year period. The low‐resolution scenario treated all row‐cropped fields as the same generic crop (AGRR in SWAT). Flow depth predictions at the outlet were similar for the two SWAT scenarios. While both scenarios showed higher levels of soil water in the fragipan soils than the surrounding soils, the high‐resolution scenario was able to identify field‐to‐field distinctions due to the increased detail in input data. In general, model results were more defined at the field‐level under the high‐resolution scenario and followed patterns expected from knowledge about soil science, hydrology, P transport, and the characteristics of the study watershed. However, the time spent collecting, understanding, entering, and error‐checking input data required for the high‐resolution scenario was on the order of months, while full data collection for the low‐resolution scenario took several days. Results
suggest that while detailed input data can enable the model to provide valuable water quality information, research efficiency during exploratory and initial problem‐solving efforts might be maximized by using more easily obtained, although more general, data.