Constraining above- and belowground uncertainties in tropical montane biogeochemistry

Abstract

Sustaining water resources and soil organic carbon (SOC) storage in the face of global change requires understanding how vegetation and soils function across landscapes. Field-based characterization of vegetation and soils is increasingly complemented or substituted by the use of satellite imagery or geospatial products derived from statistical models. This dissertation comprises three studies presenting strategies for drawing inferences on vegetation and soils from field-, satellite-, and model-based sources of information while quantifying associated uncertainties and biases. All studies focused on a mountainous region in central Veracruz, Mexico. The first study evaluated parameter uncertainty in satellite-based analysis of the seasonality, or phenology, of tropical montane vegetation. Phenological parameters and uncertainties were estimated using imagery with high spatial resolution (5 m) but low temporal resolution. The double-logistic phenology model performed well for cloud forest vegetation but poorly characterized the dynamics of other land-cover types, as reflected in large parameter uncertainties. Significant trends were detected in cloud forest phenology across gradients of topoclimate and forest composition. Accounting for parameter uncertainty was critical to the unbiased quantification of these trends. The second study assessed potential improvements in landscape-specific SOC predictions through the integration of regional-to-global statistical models and local soil data. Off-the-shelf models underestimated SOC stocks by a factor of three, on average. Calibration using local soil data included within global databases corrected this linear bias, while calibration using a more representative dataset corrected disproportionate underestimation in SOC storage hotspots. The calibration approach permitted joint prediction of top- and subsoil SOC storage and can accommodate auxiliary field data to reduce prediction uncertainties. The third study quantified bias in SOC stocks and radiocarbon activity due to soil volume change across land-use gradients, using novel and existing approaches to estimate volume change. Ignoring volume change associated with deforestation and grazing inflated SOC stocks and introduced a previously unrecognized negative bias in radiocarbon activity, causing SOC appear to older. Post hoc adjustments for volume change, using the same data required to calculate SOC stocks, may improve confidence in estimates of land-use impacts on SOC dynamics. Collectively, these results underscore the importance of accounting for uncertainty when integrating multiple information sources to characterize the spatial and temporal heterogeneity of vegetation and soils in complex landscapes.