Ten Years of GLEAM: A Review of Scientific Advances and Applications

During the past decades, consistent efforts have been undertaken to model the Earth's hydrological cycle. Multiple mathematical models have been designed to understand, predict, and manage water resources, particularly under the context of climate change. A variable that has traditionally received limited attention by the hydrological community—but that is crucial to understand the links to climate—is terrestrial evaporation. The Global Land Evaporation Amsterdam Model (GLEAM) was developed ten years ago with the goal to derive terrestrial evaporation from satellite imagery. Since then, GLEAM has been used in a variety of applications, including trend analysis, drought and heatwave studies, hydrological model calibration and validation, water budget assessment, and studies of changes in vegetation. To streamline the development of the model and improve its ability and accuracy in capturing the spatiotemporal patterns of evaporation, while tailoring the development to the needs of stakeholders, it is important to review previous studies and highlight the potential strengths and weaknesses of the model. Therefore, in this study, we provide a literature review of the GLEAM model applications and its accuracy. The results of this metanalysis indicate that GLEAM is preferentially used in climate studies, potentially due to its coarse (25 km) spatial resolution being a limiting factor for its use in water management and, particularly, agricultural applications. Validations to date suggest that, while GLEAM provides a relatively accurate evaporation dataset, its performance over short canopies requires further improvement. Two major sources of uncertainty in the GLEAM algorithm have been identified: (1) the modelling of evaporative stress in response to water limitation, (2) the need to consider below canopy evaporation estimates for a more realistic attribution of evaporation to its different sources. These potential drawbacks of the model could be alleviated by combining the current algorithm with a machine learning-based approach for a next generation of the model. Likewise, ongoing activities of running the model at high (100 m–1 km) resolutions open possibilities to utilise the data for water and agricultural management applications.