Evaluation of NLDAS-2 evapotranspiration against tower flux site observations

The North American Land Data Assimilation System project phase 2 (NLDAS-2) has run four land surface models for a 30-year (1979-2008) retrospective period. Land surface evapotranspiration (ET) is one of the most important model outputs from NLDAS-2 for investigating land–atmosphere interaction or to monitor agricultural drought. Here, we evaluate hourly ET using in situ observations over the Southern Great Plains (Atmospheric Radiation Measurement/Cloud and Radiation Testbed network) for 1 January 1997-30 September 1999 and daily ET u-sing in situ observations at the AmeriFlux network over the conterminous USA for an 8-year period (2000-2007). The NLDAS-2 models compare well against observations, with the National Centers for Environmental Prediction's Noah land surface model performing best, followed, in order, by the Variable Infiltration Capacity, Sacramento Soil Moisture Accounting, and Mosaic models. Daily evaluation across the AmeriFlux network shows that for all models, performance depends on season and vegetation type; they do better in spring and fall than in winter or summer and better for deciduous broadleaf forest and grasslands than for croplands or evergreen needleleaf forest.