Research

Upscaling carbon and water fluxes

This research aims to develop a spatially-continuous assessment of the impacts of the megadrought on dryland ecosystems across the American West. To achieve this, we improve DryFlux, a machine learning model developed to predict the highly dynamic fluxes that characterize dryland ecosystems. We improve the DryFlux model in three ways. First, we expand DryFlux to predict ET in addition to GPP, which will allow us to quantify patterns of water loss. Second, we will run the model at a daily timestep to match the scale of our flux tower analyses, which is a temporal scale better-suited for capturing the temporally heterogeneous nature of dryland fluxes. Lastly, we will improve the original DryFlux model structure by incorporating fundamental ecohydrological principles into the machine learning model.

Evaluation of TRENDY dynamic vegetation model productivity

In this study, we evaluate the ability of 18 dynamic global vegetation models in ‘TRENDY v11’  to capture spatiotemporal patterns of dryland GPP in western North America. To do this, we use the newly developed ‘DryFlux’ GPP product that better captures spatiotemporal GPP patterns compared to existing, non-dryland focused upscaled flux tower products or satellite derived GPP products (Barnes et al., 2021). More specific objectives of this study are to- a) identify certain dryland regions where models have interannual variability in GPP that better matches the observations; b) separate out the models that perform better than others, and c) determine why those models are performing better. To identify which models are performing better or worse in which regions, a diverse set of statistical metrics are used to compare models to DryFlux v1.0 data. These regions include both different dryland types (arid vs semi-arid) and different dryland regions worldwide (by continent). Furthermore, we investigated whether certain processes in the model (e.g., prescribed vegetation fractional cover) can explain the poor model performance. The findings of this study will provide a clear roadmap for dryland process-specific developments that the DGVM community can make to better estimate dryland vulnerability in general and under a changing climate. 

Improved fractional cover detection

 Accurate and reliable detection of shrubs versus grass cover is very important for identifying shrub spatial distribution patterns and for detecting how shrub cover is changing over time. Shrub cover change has an impact on ecosystem biodiversity, soil texture and moisture, ecohydrology, the carbon cycle, and energy budget. The impacts on carbon and energy budget will also alter climate change feedbacks. Whereas global deforestation and their impacts on carbon budget have been quantified with higher accuracy, we have relatively little information about woody plant encroachment, especially its spatial and temporal patterns, drivers, and its impacts on carbon sequestration. Accurate and reliable identification of the magnitude, patterns, and drivers of shrub cover will help us to properly measure its impacts on terrestrial biosphere processes. My research developed an automated classification method that accurately and reliability detected shrubs/woody plants in mixed shrub-grass ecosystems.