Evaporative flux from soils in arid and semi-arid climates can be very high and may substantially reduce soil moisture retained between infrequent rainfall events. Direct measurement of the evaporative losses from soils is technically challenging. However, environmental tracers such as water stable hydrogen and oxygen isotope composition can be used to calculate evaporation rates if the initial signature of the infiltrating rainwater is distinct from the signature of residual soil moisture. Large tropical cyclones typically result in rainfall events of large volume and very negative δ18O signatures that are significantly lower than the signatures of more frequent and smaller rainfall events. These very negative stable isotope signatures are retained in the soil and can be used to understand the depth of water infiltration, retention and subsequent rate of evaporation from the soil. At our study site in dry subtropical northwest Australia, we repeatedly sampled rainwater and soil moisture prior to, during and after tropical Cyclones Heidi and Lua in 2012. Site inundation from Cyclone Heidi (rainfall 213 mm, δ18O −17.6‰) replenished soil moisture in the unsaturated zone for several months, completely replacing soil moisture down to depths of ~3.5 m and contributing to groundwater recharge. The transient momentary evaporative losses from wet soil at the time of sampling varied between 0.21 and 0.60 mm × day−1 (equivalent to 76 to 220 mm × yr−1 recalculated as an annual rate). During the prolonged dry period between cyclones, evaporative losses decreased to between 8 and 30 mm × yr−1. Mean long-term groundwater recharge for the study site was low (<6 mm × yr−1). Recharge is primarily driven by infrequent but high-volume cyclones that are an important source of soil moisture and an essential water source for vegetation in this semi-arid environment. However, variation in lithology, position in the landscape and time since the last inundation contribute to highly heterogeneous patterns of δ18O in the vadose zone, which complicates upscaling observations from a local to a regional scale model of evaporative demand.