Arid and semi-arid grasslands are undergoing prolific changes in vegetation due to woody shrub encroachment that causes large-scale shifts in biogeochemistry. Nitrogen-fixing mesquites, a common shrub to invade these areas, fundamentally alters ecosystem processes by changing plant community diversity, net primary productivity, and soil functioning. Although the spread and proliferation of these shrubs is well documented, uncertainties related to landscape-scale biogeochemical consequences and spatial patterns associated with shrub invasion remain. This information is needed to quantify the effects of vegetation changes on nutrient cycling, soil functioning, and other ecosystem services. Furthermore, although the roles soil microbial communities have in nutrient storage and transformations are well recognized, many of the publically available soil maps classify areas according to physical and chemical traits and fail to address biotic functionality. This project offers a novel approach to answering these questions by combining remotely sensed data from an open-access data platform (NEON) to infer foliar chemistry with in situ measures of plant and soil biogeochemistry to quantify the impacts of shrub encroachment on above- and belowground nutrient dynamics at The Santa Rita Experimental Range in southeastern Arizona. This study explores how hyperspectral data can be used to quantify changes in ecosystem services due to differences in vegetation chemistry and associated belowground processes across a landscape. Data are analyzed with random forest machine learning algorithms to investigate the link between soil nutrients, topography and the abundance of shrubs and grasses; the potential effects of shrub encroachment on landscape-scale nutrient dynamics; and the utility of imaging spectroscopy to inform management practices through remote monitoring.
Results/Conclusions
Predictions of landscape scale belowground dynamics were achieved through machine learning analysis of multiple spatial data products. In many cases, modeling of soil nutrients achieved higher accuracy values than models of soil functionality. Due to the impact that mesquites have on belowground nitrogen and microbial biomass concentrations, these variables were predicted most accurately in the random forest analysis. The most important spatial data predictors varied depending on soil response variable analyzed. Across the board, the foliar nitrogen spatial data product emerged as an important predictor of belowground dynamics. Remotely sensed hyperspectral data offers unprecedented promises in the rapid assessment of ecosystems on both spatial and temporal scales never before possible, and can greatly improve our current understanding of spatial distribution of soil processes. The ability to link spectral signatures to belowground processes from remote sensing platforms could revolutionize this field of research.