OpenET is a software system that makes satellite-based multi-model estimates of evapotranspiration (ET) accessible at multiple spatial and temporal scales over the U.S. Large-scale ET estimates fill a critical data-gap for irrigation management, water resources management, and hydrological modeling and research. We present the methods and results of the second phase of an intercomparison and accuracy assessment between OpenET satellite-based models (ALEXI/DisALEXI, eeMETRIC, PT-JPL, geeSEBAL, SIMS and SSEBop) and a benchmark ground-based ET dataset with data from nearly 200 eddy covariance towers across the contiguous U.S. Processing steps for the benchmark dataset included gap-filling, energy balance closure correction, calculation of closed and unclosed daily ET, and multiple levels of data QA/QC. The dataset was split into three groups, phase I and II of the intercomparison and a reserve dataset for future studies. To sample satellite-based ET pixels, static flux footprints were generated at each station based on dominant wind speed and direction. Where data allowed, two dimensional flux footprints that are weighted by hourly ETo were developed and used for ET pixel sampling. A wide range of visual and statistical comparisons between satellite and ground-based ET were conducted at each station and against stations grouped by land cover type. Based on key performance metrics including bias, coefficient of determination, and root mean square error, model results show promising agreement at many flux sites considering the inherent uncertainty in station data. Remote sensing models show the highest agreement with closed station ET in irrigated annual cropland settings whereas locations of native vegetation with high aridity and some forested stations show relatively less agreement. The benchmark ET dataset was used to explore different approaches to computing a single ensemble estimate from the six model ensemble, with the goal of reducing the influence of model outliers and selection of weighting and data sampling schemes to reduce the influence of flux stations with sparse or extensive data records. We present the results from the model intercomparison and accuracy assessment and discuss model performance relative to accuracy requirements from the OpenET user community.

Landsat-based monitoring of seasonal and near real-time evapotranspiration (ET) in California vineyards is currently challenged by its low temporal revisit period and missing data under cloudy conditions. Gap-filling approaches, such as data fusion with high-temporal resolution images (e.g., MODIS) and interpolation of actual to potential ET ratio (ET/ETo) between image acquisition dates are now commonly used to overcome this challenge. However, these methods may not fully capture non-linear changes in crop condition due to scheduled irrigation, and other management decisions affecting ET during days when satellite images are unavailable and can lead to biased ET estimates. In this study, we combined Landsat-8 and Sentinel-2 data to develop a Shuttleworth-Wallace (SW) based near real-time ET modeling framework for mapping daily ET across three California Vineyard sites. In addition, we utilized daily Leaf area index (LAI) products derived from the Harmonized Landsat and Sentinel-2 (HLS) surface reflectance and MODIS LAI data products to constrain key resistance parameters in the SW model and tested the model across nine flux towers covering three vineyard sites in California. Results suggest that compared to the linear interpolation-based ET/ETo approach, this framework can help reduce biases and root mean squared error of estimated daily ET by over 10%. Results point to a potential utility of the combined Landsat-8 and Sentinel-2 based approach to monitor near real-time ET and complement ongoing thermal remote sensing-based ET modeling approaches to better characterize near real-time crop water status in California vineyards.

Efficient use of available water resources is key to sustainable viticulture management in California (CA) and other regions with limited water availability in the western US and abroad. This requires remote and frequent field-scale information on vineyard water status. Though the Sentinel-2 sensors offer good spatial (10-60m) and temporal (~5 days) coverages, their utility in monitoring vineyard evapotranspiration (ET) has not been considered viable primarily due to the lack of a thermal band. However, recently, a new spectral-based Shuttleworth Wallace (SW) ET model, which uses a contextual framework to determine dry and wet extremes from the Sentinel-2 (SW-S2) surface reflectance data, has shown promise when tested over a single GRAPEX (Grape Remote-sensing Atmospheric Profile and Evapotranspiration eXperiment) site in CA. However, current knowledge on its applicability across a climate gradient in CA with different topography, soils, trellis design and vine variety is lacking. Moreover, how the selection of modeling domain and meteorological forcing data influence model output is limited. Consequently, this presentation expands the evaluation of the SW-S2 model across multiple domains and meteorological inputs covering all three GRAPEX vineyard sites spanning a north to south climate gradient over three recent growing seasons (2018-2020). In comparison with flux tower observations, the size of the modeling domain and the source and quality of meteorological forcing data on the performance of the SW-S2 model as well as application to the three different vineyard study sites will be presented. Future research on merging output from more-frequent spectral and less-frequent thermal-based ET models to reduce latency in ET monitoring of California vineyards will also be discussed.

Most remote sensing-based surface energy balance (SEB) models are limited by data availability and physical constraints to fully capture the non-linear and temporally varying nature of atmospheric, biophysical, and environmental controls on evapotranspiration (ET). As such, currently, no single SEB model is considered to work best under all conditions particularly in irrigated croplands where surface moisture conditions could change dramatically in a short amount of time. Hence, irrigation water management based on a single remotely sensed ET model is often required to cope with model limitations and data latency issues, which could lead to unsustainable and unreliable accounting of water use over time. The recent inception of ensemble-based ET modeling takes the advantage of the strengths of the several SEB models under different conditions and is found to perform better as compared to an Individual model. Yet, challenges remain in how high-temporal ET outputs from different models are accurately assembled in a way that yields the most reliable estimates of ET across any environmental and surface conditions. Specifically, existing simple or Bayesian average and machine learning-based ensemble approaches have not been able to optimally utilize the comprehensive suite of existing SEB models and the availability of multiple remotely sensed datasets. Here, we discuss the utility of convolutional neural networks (CNNs) to assemble the outputs from a host of SEB models that can robustly capture the non-linear dynamics of ET under all conditions. We will also discuss the advantage and potential limitations of using the CNN-based ensemble ET modeling framework with respect to the individual, simple or Bayesian average, and other machine learning approaches and their implications for use in allocating water use across critically dry regions. Several ensemble models will be trained using eddy covariance flux data globally and will be evaluated based on their ability to estimate ET from MODIS and Landsat sensors with both individual and fused products and minimal weather inputs. The results can provide useful insights into how multiple datasets and SEB models could be optimally utilized to accurately monitor crop water status and support sustainable water resource management in drylands.