DATA DRIVEN ESTIMATES AND ASSOCIATED UNCERTAINTY OF SOIL NITROGEN AND RESPIRATION AT REGIONAL AND GLOBAL SCALES
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Bottom up, data driven estimates of soil characteristics and processes are integral to our understating of regional and global biogeochemical cycles. Using field observations and empirical models we can predict spatial patterns, temporal trends, and use these estimates to quantify ecological relationships. However, bottom- up estimates have been shown to have large uncertainties. How we communicate and report those uncertainties can help constrain our understanding of how soils contribute to these processes. This dissertation develops regional to global scale estimates of soil nitrogen and soil respiration and their sensitivity to changing environmental conditions, while addressing questions of variable selection, importance, model contribution and associated uncertainty.
The first study in this work explores the spatial distribution of soil nitrogen and uncertainty with depth in the conterminous United States using digital soil mapping techniques and Random Forest modelling. An added novelty of this work was the use of uncertainty estimates to determine potential observation sites. The second study of this work builds on the first, continuing exploration within the conterminous United States, but with a focus on the spatial distribution and temporal trends of soil respiration from 2000-2020. This study emphasized the challenges in quantifying both spatial and temporal variability of soil respiration at the regional scale. The final study of this work expands the modeling framework from the second study but shifts the focus from soil respiration to its temperature sensitivity at the global scale. From these estimates, global soil organic carbon decay constant, and soil organic carbon loss were derived. This chapter quantifies changes to soil carbon under a changing climate.
The overarching goal of the research presented here was to develop data informed machine learning models to predict the spatial distribution, temporal trend, and associated uncertainty of soil nitrogen, soil respiration, and its temperature sensitivity. Other avenues explored included quantifying variable importance and contribution, exploring relationships between predicted values with observed values, uncertainty, or ecological factors.
The collective findings of this research have contributed to our overall understanding of bottom-up estimates of soil characteristics and processes at regional and global scales that shape these characteristics. This work emphasizes the growing power of machine learning in predicting soil characteristics with depth, space, and time, highlighting its ability to address knowledge gaps. This dissertation addresses the need for more well distributed field observations of nitrogen and soil respiration especially in under sampled areas.