Effects of evapotranspiration on baseflow in a tropical headwater catchment

SummaryDiel cycles in stream discharge during baseflow periods in a headwater stream in La Selva Biological Station, Costa Rica, a tropical wet forest site, appear to be associated with groundwater withdrawal by the forest for evapotranspiration (ET). Analysis of the cycles indicates a strong correlation of stage change with ET demand, similar to the variation found in riparian water table elevation by previous researchers. Links between daily forest ET demand cycles and stream discharge cycles have been reported in temperate humid and semi-arid regions, but the frequent flood hydrographs of the wet tropics tend to obscure this daily signal. This study modifies and combines two established empirical methods for analyzing the diel ET signal in streamflow which lead to estimates of riparian ET derived from groundwater (ETG) at hourly time scales and spatial extent of the riparian area. The model has a direct dependence on the estimate of specific yield, a difficult to constrain parameter, which we estimate from previously published soil analyses. For the six baseflow periods analyzed, the model estimates groundwater ET losses ranging from 1.8 to 3.9 mm/day within the riparian area. These estimates are 52–81% of the total ET estimated with the Penman–Monteith equation (ETPM). The signal of ETG in the stream lags ETPM by 1.5–3 h, with apparent peak decay and signal duration lengthening during propagation. Model results indicate that the area of the riparian zone that influences streamflow by means of ET withdrawal increases with stream stage and ranges from 2.5% to 6.6% of the total basin area. Variations in the rate of change of nightly stream stage recovery suggest possible variations in the relative importance of subsurface hydraulic properties. At high stages, the rate of stream stage recovery from ET losses decreases throughout the night, whereas at low stages the rate of stream stage recovery increases throughout the night. Future work with numerical models could explore mechanistic controls on these empirically-derived recovery functions.