Comparison of vineyard evapotranspiration estimates from surface renewal using measured and modelled energy balance components in the GRAPEX project

Abstract

Surface renewal (SR) is a biometeorological technique that uses high frequency air temperature measurements above a crop surface to estimate sensible heat flux (H). The H derived from SR is then combined with net radiation (R_n) and ground heat flux (G) measurements to estimate latent heat flux (LE) as the residual of an energy balance equation. Recent advances in SR theory enabled its use beyond research settings, and led to the development of an inexpensive, stand-alone SR system for use in commercial agricultural settings. However, these commercial applications require replacing expensive net radiometers with clear sky models designed to estimate R_n for the energy balance approach, while also assuming G is zero on a daily basis. The accuracy of substituting R_n measurements with modelled values is unknown, and the assumption of an inconsequential G requires additional testing. Here, we compare the accuracy of the SR derived estimates of H and LE when R_n is either measured directly or modelled, and we compare results to two eddy covariance (EC) LE observations, namely LE measured via EC with an infrared gas analyzer (EC_IRGA) and LE solved as a residual in the surface energy balance (EC_resid). These measurements were collected at the Grape Remote sensing Atmospheric Profile & Evapotranspiration eXperiment (GRAPEX) conducted over a vineyard within the Lodi, CA wine growing region. LE from SR using tower R_n data measured directly onsite was significantly correlated with LE from EC_resid and from EC_IRGA with a least squares regression slope ~ 1. LE derived with the modelled incoming solar radiation (SWi) and DisALEXI R_n approaches were also significantly correlated with LE from EC_resid, but both modelling approaches overestimated LE at higher fluxes. Patterns were similar, but with more scatter for correlations between LE from EC_IRGA and LE from SR using either modelled or remotely sensed R_n. Incorporating direct measurements of G had minimal impact on the agreement of several SR approaches and LE from both EC approaches, however, when differences did occur direct measures of G reduced scatter and bias especially for the empirical SR approach. Our results suggest that LE derived from the new SR method requires fairly accurate R_n modelling approaches to obtain reliable and unbiased estimates of daily LE in comparison to measured LE using EC techniques.