Remote sensing of evapotranspiration over cotton using the TSEB and METRIC energy balance models

Authors: French, Andrew; Hunsaker, Douglas - Doug; Thorp, Kelly

Submitted to: Remote Sensing of Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/11/2014
Publication Date: 3/1/2015
Publication URL: http://handle.nal.usda.gov/10113/61101
Citation: French, A.N., Hunsaker, D.J., Thorp, K.R. 2015. Remote sensing of evapotranspiration over cotton using the TSEB and METRIC energy balance models. Remote Sensing of Environment. 158:281-294.

Interpretive Summary: Improved detection and mapping of water stress and evapotranspiration (ET) is important for managing scarce water supplies while maintaining high yields. Imaging land surface temperatures with remote sensing is emerging as a practical and operational tool to meet these needs. When crops are water stressed, their temperatures are elevated above normal. These temperatures can be detected from aircraft and satellites. Research in recent years has been underway to model the correspondence between crop temperature and crop water use so that remote sensing data can be used to monitor plant conditions at any time. However, the models can be complex to implement and sometimes do not provide the same values for water use when it is known that they should. To help modelers understand which models are not too complex yet accurate enough for farm water decision-making, we tested two prominent ET models using actual crop water use for a large field of cotton in 2009 and 2011. Results showed that the more complex model, known as TSEB, was more accurate than the less complex one, METRIC. But they showed that if little supplemental data are available, that crop water use can still be accurately mapped with METRIC. This information will be useful for farmers, irrigation district personnel and consultants that wish to use remote sensing to help improve crop water management.

Technical Abstract: Remote sensing of evapotranspiration (ET) can help detect, map and provide guidance for crop water needs in irrigated lands that cannot be done in other ways. Remote sensing with thermal infrared (TIR) provides the potential to rapidly detect water-related plant stress that would otherwise be missed when using maps of vegetation indices. Two remote sensing ET models based on TIR, TSEB and METRIC, have been widely tested, reported and are strongly influencing remote sensing satellite development and on ways ET data can be used in climate and drought models. However questions continually arise about their relative accuracies, biases, complexity and ease of implementation. This study investigated these questions using airborne data collected at Maricopa, Arizona in 2009 and 2011. The site was a 4.9 ha irrigated cotton study focused on use of remote sensing for irrigation scheduling. TSEB and METRIC models both estimate ET, but do so in very different ways. TSEB is strongly tied to biophysics on the ground, while METRIC links the entire processing chain from satellite to ground. Here, aspects not directly related to turbulent flux estimation on the ground were set aside. Thus models were compared with respect to core algorithms that convert land surface temperatures (LST) into ET. Using image and ground data from seven airborne surveys, net radiation and soil heat fluxes were computed with the more physically based TSEB approach. These fluxes were then used as input layers for both models. Based on soil moisture profile observations at 112 locations, METRIC was found accurate to 2 mm/day in a majority of cases, while TSEB was similarly accurate at a 1.5 mm/day threshold. These accuracies were representative for emergent, full canopy, and late season cotton growth phases. TSEB and METRIC were similarly biased, ~ -0.7 mm/day, an outcome likely indicative of error in net radiation estimation and not to ET estimation. Extrapolation of instantaneous ET estimates to daily time steps was tested with the constant evaporative fraction (EF) approach and with an approach incorporating ground-based LST data; although the ground-based approach has potential to improve ET estimates on cloudy days, this study found the EF approach more consistent. Based on soil moisture depletion observations, ET results from both TSEB and METRIC were physically reasonable, though variability in ET on a plot-by-plot basis was 1/2 of the observed typical variation of +/-5 C. Considering model complexity, input data requirements, and ease of implementation, the recommended model choice for irrigated ET mapping is contingent upon the quality and extent of ancillary meteorological and agronomic observations. This study had rich data sets and hence TSEB is recommended. However, if such data were lacking, METRIC would also be strongly recommended.