The hydrology of the Peruvian Amazon river and its sensitivity to climate change

Abstract

This PhD thesis explores the utility of a land surface model (Joint UK Land-Environment Simulator, JULES) for large-scale hydrological modelling of the Peruvian Amazon - a humid tropical mountain basin where process understanding is poor and data are scarce. A sparse rain gauge network necessitates the use of large-scale data from satellite and global climate model reanalysis to complement ground observations, commanding a closer look at (1) the uncertainties (2) merging techniques to utilise multiple observations in the model forcing. A main outcome of the research is establishing the model’s sensitivity to precipitation error, and at the same time, demonstrating an increasing reliability of global remote sensing products as model forcing, specifically, with data from the Tropical Rainfall Measuring Mission Multisatellite Precipitation Analysis version 7 algorithm. Furthermore, satellite-rain gauge data assimilation techniques such as mean-bias correction, double smoothing residual blending, and Bayesian combination, are shown to reduce the mean errors in the satellite-based product. Secondly, with regional calibration and an offline runoff routing scheme, JULES is shown to be reasonably skillful at reproducing the observed streamflow dynamic and extremes. Representing the subgrid heterogeneity of soil moisture using the probability distributed model (PDM) was key to improving surface runoff generation. However, evapotranspirative fluxes in the lower basin remain poorly reproduced without an adequate floodplain system representation. Finally, under the Intergovernmental Panel for Climate Change’s RCP4.5 future climate scenario, which projects a warming and wetting up to the year 2035, the Peruvian Amazon basin is shown to respond nonlinearly to the increase in wet season precipitation with more than 40% increase in the peak flows compared to the baseline scenario. There is limited confidence in the projections due to climate projections uncertainty and the assumptions of model stationarity.