Decreased latency in landsat-derived land surface temperature products: A case for near-real-time evapotranspiration estimation in California

Authors: Knipper, Kyle; YANG, YUN - Mississippi State University; Anderson, Martha; BAMBACH, NICOLAS - University Of California, Davis; Kustas, William - Bill; McElrone, Andrew; Gao, Feng; ALSINA, MARIA MAR - E & J Gallo Winery

Submitted to: Agricultural Water Management
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/11/2023
Publication Date: 4/24/2023
Citation: Knipper, K.R., Yang, Y., Anderson, M.C., Bambach, N., Kustas, W.P., McElrone, A.J., Gao, F.N., Alsina, M. 2023. Decreased latency in landsat-derived land surface temperature products: A case for near-real-time evapotranspiration estimation in California. Agricultural Water Management. 283. Article 108316. https://doi.org/10.1016/j.agwat.2023.108316.
DOI: https://doi.org/10.1016/j.agwat.2023.108316

Interpretive Summary: Acquiring accurate measurements of crop consumptive water use in the form of evapotranspiration (ET) is increasingly important in California cropping systems as demands for water resources shift under a changing climate. Growers use various methods for estimating ET to guide irrigation management, some of which are based on satellite remote sensing. Although this approach has proven reliable, operational applications remain hindered by latency in satellite product delivery, due in part to computationally expensive atmospheric correction steps. This is particularly true in approaches that utilize thermal infrared (TIR) imagery, which is sensitive to atmospheric corrections. The current study evaluates methods based on machine learning to derive corrected Landsat Surface Temperature (LST) in near-real-time (NRT) for ingestion into a satellite-based ET model. The idea is to decrease the latency in TIR acquisition, allowing ET estimation to be made in NRT, aiding in irrigation management decision making. Results suggest that the machine learning approach provides accurate LST estimates with a latency of < 8 hours but more research is required if the approach is to be scaled over larger areas (i.e., the entire Western United States).

Technical Abstract: Acquiring accurate measurements of crop consumptive water use in the form of evapotranspiration (ET) is increasingly important in California cropping systems as demands for water resources shift under a changing climate. Growers use various methods for estimating ET to guide irrigation management, some of which are based on satellite remote sensing. Although this approach has proven reliable, operational applications remain hindered by latency in satellite product delivery, due in part to computationally expensive atmospheric correction steps. This is particularly true in approaches that utilize thermal infrared (TIR) imagery, which is sensitive to atmospheric corrections. The current study evaluates two approaches to derive a pseudo-atmospherically corrected Land Surface Temperature (LST) in near-real-time (NRT) for ingestion into the ALEXI-DisALEXI ET model. Evaluation is done for selected Landsat scenes over California for the year 2022. Both approaches take advantage of the Landsat Collection 2 dataset, including availability of atmospheric correction parameters and atmospherically corrected LST. The first approach is based on the Radiative Transfer Equation (RTE) and atmospheric correction parameters from previous overpass available in the Collection 2 dataset. The second is based on a random forest (RF) regression model, using Landsat Collection 2 atmospheric correction parameters and LST as input for training. Results indicate the RF approach outperforms the RTE approach, with an average mean absolute error of 0.6 K, compared to 2.0 K for the RTE method. The RTE method produces more spatial and temporal variability in LST due to temporal differences in atmospheric transmissivity. When used to estimate ET, we find little difference between NRT LST-based ET estimates and ET derived using the Collection 2 LST product, albeit RF-based ET provides less day-to-day variation. Results suggest promise in using such an approach to derive LST and subsequently ET in NRT, and toward improving daily water management and irrigation efficiency in the vineyard and orchard systems of California.