MCA: Improving understanding of controls over spatial heterogeneity in dryland soil carbon pools in the age of big data

点击翻译按钮获取中文摘要

Soils may alter the trajectory of climate change because of their potential to store or release large amounts of carbon, thus altering the concentration of atmospheric carbon dioxide. However, understanding and predicting current and future soil carbon dynamics requires the capability to accurately describe spatial patterns of soil carbon and forecast changes via reliable models. At present, patterns and controls over soil carbon cycle processes are poorly resolved in dryland (arid and semi-arid) ecosystems, which cover nearly half of Earth's terrestrial surface and store one third of global soil carbon. The ‘big data’ revolution has dramatically increased data available to address ecological problems such soil carbon dynamics. However, effective use of big data requires sophisticated data handling skills and use of emerging analytical tools such as machine learning and application of these tools to process modeling. This project will advance the investigator’s research skills in big data handling and will enhance their ability to mentor students in modern approaches to data-intensive ecological problems. Machine learning and process modeling will be used to increase understanding of patterns and controls over spatial heterogeneity in dryland soil carbon. This information is critical for scientifically based evaluation of dryland management strategies of soil carbon storage. This project will explore patterns and mechanistic controls over spatial heterogeneity in dryland soil organic carbon pools. Exploring patterns in two contrasting dryland settings, a semi-arid grassland with well-documented long-term management and vegetation change and a poorly characterized hyper-arid system, will provide deeper understanding of the relationships between environmental variables and soil organic carbon across drylands. Coupling this exploration of soil organic carbon spatial patterns with process modeling will enhance understanding of the mechanistic drivers of soil organic carbon heterogeneity. Spatial downscaling of a deep learning enhanced earth system modeling approach will provide insight into the fine scale mechanisms that drive soil organic carbon heterogeneity and how they respond to environmental change. This mid-career advancement grant will enable the primary investigator to: develop skills for handling and analyzing large and complex data sets; use machine learning approaches to describe spatial patterns of heterogeneity in soil organic carbon pools in two contrasting dryland field sites where the primary investigator has extensive prior experience and data, and; apply a deep learning enhanced earth system model to a dryland site and use this model to explore mechanistic drivers of carbon cycling. This project will build mutually beneficial partnerships between the primary investigator and two research partners, and an engineer with expertise in machine learning and remote sensing and an expert in ecological process models and deep learning enhanced earth system modeling.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.