Uncertainties in physical and tracer methods in actual groundwater recharge estimation in the thick loess deposits of China

•Hydrogeological parameters are main uncertainty source of physical methods•Lag effect of the thick vadose zone limits WTF and dating-based methods•Ignoring the mixing and dispersion may overestimate recharge by 36–48 % for CFC•Aquifer heterogeneity reduces the potential of 14C for recharge estimation•CMB is recommended as the optimal method for GR estimation in the loess area Groundwater recharge (GR) can be estimated with physical, tracer techniques and numerical modelling; however, the uncertainties in these methods have not been fully interpreted. This study selected six methods for actual GR estimation in the saturated zones with different data requirements and method assumptions, including two physical methods (i.e., water table fluctuation WTF and Darcy's law DL) and four tracer techniques (i.e., chloride mass balance CMB, tritium renewal rate TRR, chlorofluorocarbons CFC and 14C dating method CBD). Taking the loess tablelands of China as an example study area, we first analyzed the uncertainties in these methods and then recommended the appropriate methods for GR estimation in regions with thick vadose zones. Overall, the recharge rates from WTF, CMB and CFC (88 ± 24, 98 ± 33 and 53 ± 9 mm/year) were much greater than those from DL, TRR and CBD (10 ± 3, 14 ± 3 and 7 ± 5 mm/year). The differences in estimated GR are closely related to the method parameters, assumptions and temporal scales. Specifically, the model parameters (e.g., specific yield, hydraulic conductivity) exhibit large ranges due to spatiotemporal heterogeneity, and the physical methods are more sensitive to hydrogeological parameters compared with the tracer techniques. The uncertainties related to model assumptions originate from the simplified water/solute movement processes. The lag effect of the thick vadose zone largely limits WTF and TRR, but contributes to CMB. Ignoring the mixing and dispersion effect may overestimate recharge by 36–48 % for CFC, and the heterogeneity of the aquifer reduces the potential of CBD for recharge estimation. Overall, CMB is considered as the most optimal method for estimating GR in the thick loess deposits. This study deepens the understanding of the uncertainties in GR assessment methods in regions with thick vadose zones.