Flux variance similarity-based partitioning of evapotranspiration over a rainfed alfalfa field using high frequency eddy covariance data

Abstract Although the eddy covariance (EC) technique provides direct and continuous measurements of evapotranspiration (ET), separate measurement of evaporation (E) and transpiration (T) at the ecosystem level is not possible. For partitioning ET into E and T, high frequency (10 Hz) time series EC observations collected from Apr 2016 to May 2018 over a rainfed alfalfa (Medicago sativa L.) field in central Oklahoma, USA were analyzed using the open source software Fluxpart. Fluxpart partitions ET by examining the correlation (Rqc) between water vapor (q) and carbon dioxide (c) fluxes as prescribed by the Flux Variance Similarity (FVS) partitioning method. Patterns of Rqc and partitioned E and T were consistent with expected trends associated with vegetation dynamics and short-term transient features (i.e., hay harvesting and rainfall events). The Rqc grew stronger with increasing alfalfa leaf area and exhibited a strong anti-correlation (Rqc close to -1) during peak growth when T and photosynthesis (P) were dominant and co-regulated by the leaf stomata. Consequently, a strong linear relationship (R2 = 0.96) was found between monthly midday average values of Rqc and monthly average Moderate Resolution Imaging Spectroradiometer (MODIS)-derived leaf area index (LAIMOD). Decorrelation of q and c or dominance of non-photosynthetic (e.g., E and respiration, R) fluxes resulted in less negative or positive Rqc values during winter, hay harvest, rainy, and nighttime periods. Growing season (Apr-Oct) average T:ET was approximately 0.82 and 0.77 in 2016 and 2017, respectively. Diurnal cycles and temporal variations of leaf-level water use efficiency (WUE, an input of the FVS method) estimates were consistent with the seasonal dynamics of ecosystem WUE, computed from EC-derived gross primary production (GPP) and EC-measured ET. These results validate the performance of the FVS ET partitioning method using high frequency EC data.