UC Berkeley

The study of watershed processes between the hydrosphere, biosphere and atmosphere in response to climate change is critical for modeling dynamics in the hydrologic and biogeochemical cycles. Quantifying the occurrences of environmental hot spots and hot moments (HSHMs) defined as the rare locations or events that exert disproportionate influence over the environment is essential to improve our understanding of watershed dynamics under gradual climate change. This dissertation consists of three different approaches that develop theories and models to statistically characterize HSHMs, improve flux estimation and better identify linkages among aspects of land surface, atmosphere and groundwater interactions. First, we propose a statistical framework to characterize the spatiotemporal distribution of HSHMs. The statistical framework utilizes indicator random variables to construct a statistical model for HSHMs, which relate the characteristics of HSHMs to the relevant spatial and temporal components. Three categories of HSHMS are identified, including (1) HSHMs defined by only spatial (static) components; (2) HSHMs defined by both spatial and temporal (dynamic) components and (3) HSHMs defined by multiple dynamic components. In order to demonstrate the suitability of the statistical framework, we demonstrate the procedure for constructing the models for each of the category. We further develop a groundwater hydrology example to illustrate the importance of incorporating subsurface heterogeneity in modeling HSHMs occurrences and the corresponding uncertainties. The representation of an HSHM through its spatial and temporal components allows us to relate the HSHM’s uncertainty to the uncertainty of its components. Second, we develop the hybrid-predictive-modeling (HPM) approach to improve estimation of evapotranspiration (ET) and ecosystem respiration (R_ECO), especially at mountainous watersheds. The proposed HPM approach integrates meteorological forcing data and remote sensing data to estimate ET and R_ECO with the deep learning recurrent neural network long short-term memory as the main driver. HPM can easily incorporate measured data from eddy covariance tower and simulated data from physically-based-models (e.g., Community Land Model). In order to test the performance of HPM, 4 different use cases are developed and tested. Furthermore, we focus on estimating ET and R_ECO at the East River Watershed in Colorado and distinguish the role of small-scale meteorological forcing heterogeneity and vegetation heterogeneity in regulating ET and R_ECO dynamics. Estimation results from HPM can then be used as inputs for assessing the occurrences of ecological HSHMs, especially at mountainous watersheds, to improve our understanding of mountainous watershed dynamics. Third, we recognize the necessity to better understand the intra-annual variability of mountainous watersheds dynamics to better improve our water and resources management. We develop the concept of temporal regimes to identify the sub-annual variability in hydroclimate processes and assess its effects over ET dynamics. We select six mountainous watersheds along the central Rocky Mountain ranges to demonstrate the applicability of temporal regimes. Through the employment of temporal regimes, we identify the temporal boundaries and durations of snow regimes, snowmelt regimes, growing season regimes, monsoon regimes and defoliation regimes from 2005 to 2016. We define within-regime ET as the sub-annual ET contributed from each of the regimes, which enables us to distinguish how the timing and duration of watershed processes regulate ET. High correlation between within-regime ET and regime duration is observed, which suggests intra-annual variability is a major control that regulates the temporal variability of ET at mountainous watersheds. The proposed concept of temporal regimes can further advance of our understanding of how mountainous watersheds evolve under gradual climate change and improve water and energy resources management in the future. The proposed approaches in this dissertation provide us theories and models to further advance our understanding of watershed dynamics under a rapidly changing environment. Even if we present limited number of examples (e.g., subsurface HSHMs and intra-annual variability of ET), we expect these approaches can be applied towards other ecosystem dynamics.