Abstract: Decomposition of plant litter is a fundamental ecological process, integral to soil organic matter chemistry and biogeochemical cycling. However, much of our understanding of decomposition dynamics focuses on rates of litter mass loss and therefore carbon dynamics. Most studies only report initial litter chemistry rather than chemical changes throughout decay, and particularly not chemistry beyond carbon and nitrogen. A better understanding of chemical patterns throughout decay will ultimately inform models of nutrient recycling and carbon storage in organic matter. Existing studies that do investigate chemical changes over time often report idiosyncratic results when compared across litter species and ecosystems. Therefore, the existence of broad-scale patterns of litter chemical changes throughout decomposition are still unclear. We used archived litterbag samples and data from across the U.S. Long-Term Ecological Research Network to investigate the trajectory of a comprehensive array of litter chemistry, including not only C and N but also micronutrient, structural, and metabolic compounds throughout the first 70% of mass loss. Data were analyzed using machine learning techniques to explore (1) the nature of litter chemical changes throughout decomposition across diverse ecosystems and plant functional types; and (2) the predictability of decay rates throughout decomposition based on litter chemistry, functional type, and ecosystem.
Our results do not reveal a universally common pattern of litter chemical trajectories characteristic of functional types or ecosystems. Contrary to existing hypotheses, there is only very limited evidence of convergence or divergence in chemistry over time at this broad scale. Instead there is a general persistence of unique chemical trajectories across ecosystems and functional types, though the relationship among those unique trajectories shifted throughout decay. Given the lack of consistent patterns, the consequences of changes in plant communities driven by global change for nutrient cycling and SOM formation will be difficult to predict. Further, while initial litter chemistry reliably predicted decay rates when calculated through 30% mass remaining, the models could not reliably predict decay rates at early- and mid-stages of decay (75% and 50% mass remaining). Using random forest modeling, we identify which chemical parameters best explain decay rates across these stages, at this broad geographic scale.