Quantifying the effect of changes in the Earth’s surface on regional water cycles is essential for water security. Hydrologists have traditionally used manipulative experiments in individual or paired river basins, or model-based analyses, to identify water cycle responses to land use or land cover change. Limitations in these approaches leave important gaps in understanding: (i) existing studies are not representative of biomes worldwide, and there is a deficit especially for tropical regions; (ii) individual site-specific experiments generally do not provide a representative sample of regional river systems in a way that can inform policy; and (iii) regional-scale analyses are largely based on simulation models, and are therefore limited by parameterizing assumptions and calibration uncertainty.
Larger-sample, empirical analyses are needed for more accurate modeling and policy-relevant understanding – and these analyses must be supported by regional data. The Amazon-Cerrado (tropical savanna) transition region in Brazil is a global agricultural and biodiversity center, where regional climate and hydrology are projected to have strong sensitivities to land cover change. Dramatic land cover change has and continues to occur in this region, and its effects on streamflow (as an overall indicator of water cycle function) are not empirically understood at regional scales – in part due to data uncertainties. Therefore, analyses of hydroclimate in the Amazon-Cerrado region of Brazil exhibit many of the challenges of interest to evaluation of regional hydrological change, as well as a suite of important applications.
This thesis explores data uncertainties implicit in the study of hydrological systems, and the hydrological effects of anthropogenic land use change. Specifically, this thesis: describes a novel data collection effort supporting empirical analysis at multiple-basin scales in Amazon-Cerrado Brazil; evaluates rainfall data uncertainty embedded in the study of the hydroclimate; and measures the effect of agricultural-driven deforestation on regional streamflow. This work required the harmonization of multiple in-situ and remotely-sensed (e.g. satellite-derived land use and climate) data products, and novel application of empirical statistical analysis methods developed in other fields (i.e. public health and economics) to establish causality in complex observational data settings. At the time of writing, this research is the first application of these methods to a geoscience inquiry.
Firstly, this research contributes a novel hydrological dataset, including processing and quality control of more than 1,000 rain gauges, over 300 streamflow gauges, and associated GIS data (rain and streamflow gauge locations, river basin delineations, and large reservoir locations and drainage area delineations) across eight states in Brazil. Secondly, the research demonstrates that the magnitude of uncertainty from rainfall ``data selection uncertainty'', or uncertainty across multiple in-situ and remotely-sensed (satellite) rainfall data products, is comparable to estimated bias in global climate model projections, and provides practical recommendations for addressing this problem. Third, the research provides a series of regression analysis-based quantifications of the causal effects of deforestation on stream flow, which show that (a) streamflow, and especially ecologically-important dry-season low flow, has significantly increased across Amazon-Cerrado Brazil, and (b) that annual average increases in streamflow due to land cover change (agricultural development and corresponding forest loss) accounts for nearly half of total streamflow increases in the region over the past half century.
