Water-balance predictions made using four codes ͑UNSAT-H, VADOSE/W, HYDRUS, and LEACHM͒ are compared with water-balance data from a test section located in a semiarid climate simulating a monolithic water-balance cover. The accuracy of the runoff prediction ͑underprediction or overprediction͒ was found to affect the accuracy of all other water-balance quantities. Runoff was predicted more accurately when precipitation was applied uniformly throughout the day, the surface layer was assigned higher saturated hydraulic conductivity, or when Brooks-Corey functions were used to describe the hydraulic properties of the cover soils. However, no definitive or universal recommendation could be identified that would provide reasonable assurance that runoff mechanisms are properly simulated and runoff predictions are accurate. Evapotranspiration and soil-water storage were predicted reasonably well ͑within Ϸ25 mm/ yr͒ when runoff was predicted accurately, general mean hydraulic properties were used as input, and the vegetation followed a consistent seasonal transpiration cycle. However, percolation was consistently underpredicted ͑Ͼ3 mm total͒ even when evapotranspiration and soil-water storage were predicted reliably. Better agreement between measured and predicted percolation ͑or a more conservative prediction͒ was obtained using mean properties for the soil-water characteristic curve and increasing the saturated hydraulic conductivity of the cover soils by a factor between 5 and 10. Evapotranspiration and soil-water storage were predicted poorly at the end of the monitoring period by all of the codes due to a change in the evapotranspiration pattern that was not captured by the models. The inability to capture such changes is a weakness in current modeling approaches that needs further study.

Predictions of surface runoff ͑R͒, evapotranspiration ͑ET͒, soil-water storage ͑S͒, and percolation obtained using three numerical codes ͑LEACHM, HYDRUS, and UNSAT-H͒ employed to simulate the hydrology of water-balance covers are compared to measured water-balance data from a lysimeter used to monitor a capillary barrier cover profile in a subhumid climate. All of the codes captured the seasonal variations in water-balance quantities observed in the field. LEACHM and HYDRUS predicted total R during the monitoring period with reasonable accuracy ͑within 18 mm using general mean parameters͒, but the timing of predicted and observed R events was different. In contrast, UNSAT-H consistently overpredicted R by at least 239 mm. Evapotranspiration was predicted reliably ͑within 60 mm͒ with all three codes when data from the first year were excluded. However, all three codes overpredicted ET in late winter and early spring, when snowmelt was occurring and S was accumulating in the field. Consequently, S generally was underpredicted by all three codes. Predicted and measured percolation were in good agreement ͑within 1 mm/ year͒, except during the first year. Results of the comparison indicate that cover modelers should scrutinize runoff predictions for reasonableness and carefully account for snow accumulation, snow melt, and ET during snow cover.