AbstractEvapotranspiration (ET) is an important component of micro- and macro-scale climatic processes. In agriculture, estimates of ET are frequently used to monitor droughts, schedule irrigation, and assess crop water productivity over large areas. Currently, in situ measurements of ET are difficult to scale up for regional applications, so remote sensing technology has been increasingly used to estimate crop ET. Ratio-based vegetation indices retrieved from optical remote sensing, like the Normalized Difference Vegetation Index (NDVI), Soil Adjusted Vegetation Index, and Enhanced Vegetation Index are critical components of these models, particularly for the partitioning of ET into transpiration and soil evaporation. These indices have their limitations, however, and can induce large model bias and error. In this study, micrometeorological and spectroradiometric data collected over two growing seasons in cotton, maize, and rice fields in the Central Valley of California were used to identify spectral wavelengths from 428 to 2295nm that produced the highest correlation to and lowest error with ET, transpiration, and soil evaporation. The analysis was performed with hyperspectral narrowbands (HNBs) at 10nm intervals and multispectral broadbands (MSBBs) commonly retrieved by Earth observation platforms. The study revealed that (1) HNB indices consistently explained more variability in ET (ΔR2=0.12), transpiration (ΔR2=0.17), and soil evaporation (ΔR2=0.14) than MSBB indices; (2) the relationship between transpiration using the ratio-based index most commonly used for ET modeling, NDVI, was strong (R2=0.51), but the hyperspectral equivalent was superior (R2=0.68); and (3) soil evaporation was not estimated well using ratio-based indices from the literature (highest R2=0.37), but could be after further evaluation, using ratio-based indices centered on 743 and 953nm (R2=0.72) or 428 and 1518nm (R2=0.69).