Extreme weather events are inherent in climate variability and they can cause ecosystem alterations, infrastructure damages, suspension of food supply chains, and loss of lives. Today's highly populated and urbanized society is more vulnerable than ever to natural hazards and their disruptive consequences. Projected population growth and changes in climate variability are expected to exacerbate the societal and economic impact of climatic extremes. The scientific community, global organizations, and other stakeholders have all recognized the urgency of improving our understanding of both natural and human-induced climate variability.

The overarching goal of this thesis is to advance the current methods for nonstationary analysis of climatic extremes and their attributions. Here we propose a methodological framework for investigating hydroclimatic extremes over time and in response to a physical driver/covariate. The Process-based Nonstationary Extreme Value Analysis (ProNEVA) framework is unique in that it allows for incorporating a physical component into traditional frequency analysis techniques to account for observed or process-based changes in the variable of interest. The model can be used for both stationary and nonstationary analyses of extremes and includes a Graphical User Interface (GUI) for easier implementation.

We then shed light on the uncertainty inherent in the estimation of climatic extremes. We propose a generalized approach for including uncertainty information in the recurrence interval of extremes. The approach offers insights on how information about extremes should be interpreted by planners and decision-makers under conditions of uncertainty.

Using the method developed in this thesis, we show how extreme precipitation is expected to change in the future. We also highlight the importance of merging information from observations and climate model simulations for risk assessment purposes.

Finally, we outline a methodological framework for attribution of changes in multiple extremes or multiple features of an extreme event. We show the potential of copula functions for attribution of changes in both the magnitude and the dependence structure between characteristics of a natural phenomenon or, more generally, between two dependent variables.

ABSTRACT OF THE THESIS:

Information Content of Hydrologic Signatures and their Impacts on Watershed Model Calibration

By

Mohammad Sassani

Master of Science in Environmental Engineering

University of California, Irvine, 2018

Professor Amir AghaKouchak, Chair

Approximate Bayesian computation, also known as ABC, is a simulation-based method with its roots in Bayesian analysis, which is increasingly used in the past couple of years in the field of hydrology for parameter estimation and uncertainty analysis. ABC relaxes the need for an explicit likelihood function and uses multiple hydrologic metrics (signatures) to evaluate model simulations. If the hydrologic signatures are sufficient, and the distance between simulated metrics and their observed counterparts are less than a nominal threshold, model parameters are assumed to be derived from the posterior distribution. Since finding a sufficient set of summary statistics is difficult for complex environmental systems and high-order models, the inferred posterior parameter and predictive distributions are constrained on the set of selected metrics and may not necessarily mirror the "true" posterior parameter distributions.

In this thesis, we have strived to shed some light on how different hydrologic signatures constrain watershed models. More specifically, we are interested to understand which portions of the hydrograph is constrained by each individual hydrologic signature, and ultimately move toward finding a set of sufficient metrics with least overlapping information. Throughout the years significant strides have been made in the field of calibration and uncertainty assessment of hydrologic models, but the constraining power of different hydrologic metrics have not yet been characterized. Using information theory, likelihood, entropy, and conventional metrics such as NSE, RMSE, and percent bias; we try to characterize constraining power and model parameter sensitivity to each metric. Sensitivity analysis of model parameters is performed using Kullback-Leiber divergence metric and Two Sample Kolmogorov Smirnov test. One interesting finding of this study is that hydrologic signatures are model specific and cannot be readily transferred between different models. The result of this study also depict that that each watershed possesses different and unique characteristics, which is also manifested in the hydrologic signatures that are selected for each watershed. In other word, hydrologic signatures are model- and watershed- specific. We also observe that more flexible watershed models, such as GR4J, react differently to hydrologic signatures as compared to less flexible model structures, such as HyMod. In more detail, more flexible model structures are associated with higher information content, which might ultimately help the hydrologic signatures in constraining model behavior.

Extreme climatic events have significant impacts on society and the environment, especially when multiple hazards occur concurrently (e.g., drought and heat waves) or consecutively (wildfires and extreme precipitation). A large number of indicators have been developed to detect and study changes in extreme events across space and time. While the current climate extreme indicators provide useful information, most do not provide any information on compound/concurrent events. A compound event corresponds to a situation in which multiple (often interrelated) hazard drivers lead to an extreme outcome. Therefore, current univariate methods used for frequency analysis and risk assessment may underestimate the risk or occurrence probability of extreme events. After a comprehensive review of the existing methods, this study outlines frameworks for detecting, modeling, and analyzing inter-related events and processes including compound extremes.

Fire occurrence and spread are influenced by various factors including climate, vegetation, land use, and human activities. Understanding the underlying mechanisms of wildfires and quantifying their impacts on the Earth system is crucial for improving our resilience and preparedness for future wildfires. However, the complexity of interactions among different drivers makes it challenging to comprehend the underlying causes, effects, and feedback. Wildfires alter the land cover of hydrologic basins and significantly increase overland flow and debris movement, reducing the time of concentration. Additionally, wildfires can have adverse effects on the soil properties in burned areas, including reduced moisture content and soil stability, increased risk of post-wildfire landslides, mud and debris flows, erosion, and excessive runoff. This is especially concerning when extreme precipitation falls over burned areas, leading to floods and debris flows, such as the 2018 debris flow in Montecito, California. Such compound events, where two consecutive/concurrent events result in severe societal impacts, pose a significant threat. In this study, one objective is to collect post-fire data to investigate changes in soil characteristics in burned areas and their applications to hydrology, debris flow modeling, and land surface evolution. In this dissertation, I employed statistical and machine learning techniques in conjunction with large-scale meteorology, vegetation, topography, and social datasets to investigate fire-climate-land interactions. This dissertation includes original field data collected for analysis of small-scale processes, assessment of large-scale satellite observations, and development of a machine learning model for improving fire prediction.

Droughts are among the most common and devastating natural disasters. Reducing damages associated with droughts relies on monitoring and prediction information as well as plans to cope with droughts. The overarching goal of this dissertation is to improve current capabilities in drought monitoring using space-based observations, with a focus on integrating remotely sensed data products that are not commonly being used for drought monitoring. The first chapter of this dissertation, surveys current and emerging drought monitoring approaches using remotely-sensed observations from climatological and ecosystem perspectives. Current and future satellite missions offer opportunities to develop composite and multi-sensor (or multi-index) drought assessment models. While there are immense opportunities, there are major challenges including data continuity, unquantified uncertainty, sensor changes, and community acceptability. One of the major limitations of many of the currently available satellite observations is their short length of record. A number of relevant satellite missions and sensors (e.g., Atmospheric Infrared Sounder (AIRS), Gravity Recovery and Climate Experiment) provide only slightly over a decade of data, which may not be sufficient to study droughts from a climatological perspective. However, they still provide valuable information about relevant hydrologic and ecological processes linked to this natural hazard. Therefore, there is a need for models and algorithms that combine multiple data sets and/or assimilate satellite observations into model simulations to generate long-term climate data records. To address this gap, Chapter 2 introduces Standardized Drought Analysis Toolbox (SDAT), which includes a generalized framework for deriving nonparametric univariate and multivariate standardized drought indices. Current indicators suffer from deficiencies including some prior distributional assumption, temporal inconsistency, and statistical incomparability. Different indicators have varying scales and ranges and their values cannot be compared with each other directly. Most drought indicators rely on a representative parametric probability distribution function that fits the data. However, a parametric distribution function may not fit the data, especially in continental/global scale studies. Particularly, when the sample size is relatively small as in the case of many satellite precipitation products. SDAT is based on a nonparametric framework that can be applied to different climatic variables including precipitation, soil moisture and relative humidity, without having to assume representative parametric distributions. The most attractive feature of the framework is that it leads to statistically consistent drought indicators based on different variables. We show that using SDAT with satellite observation leads to more reliable drought information, compared to the commonly used parametric methods.

We argue that satellite observations not currently used for operational drought monitoring, such as near-surface air relative humidity data from the Atmospheric Infrared Sounder (AIRS) mission, provide opportunities to improve early drought warning. In the third chapter of this dissertation, we outline a new drought monitoring framework for early drought onset detection using AIRS relative humidity data. The early warning and onset detection of drought is of particular importance for effective agriculture and water resource management. Previous studies show that the Standard Precipitation Index (SPI), a measure of precipitation deficit, detects drought onset earlier than other indicators. Here satellite-based near surface air relative humidity data can further improve drought onset detection and early warning. This chapter introduces the Standardized Relative Humidity Index (SRHI) based on the NASA's AIRS observations. SRHI relies on SDAT's nonparametric framework, introduced in Chapter 2. The results indicate that the SRHI typically detects the drought onset earlier than SPI. While the AIRS mission was not originally designed for drought monitoring, its relative humidity data offers a new and unique avenue for drought monitoring and early warning. Early warning aspects of SRHI may have merit for integration into current drought monitoring systems.

One of the research opportunities identified in Chapter 1 is using current (and future) satellite missions to develop composite and multi-indicator drought models. In Chapter 4, we outline a framework for assessing impacts of droughts on forest health using a multi-sensor approach. This framework relies on the relationship between climate conditions (e.g., temperature, precipitation, relative humidity, Vapor Pressure Deficit) and forest health based on greenness of vegetation. Wildfires, tree mortality and forest productivity increase during drought periods. Using the proposed multi-index approach, Chapter 4 aims to investigate the effects of recent summer, dry-season and winter droughts on the forest health in western United States. We use Vapor Pressure Deficit (VPD) as an indicator that combines temperature and relative humidity for forest stress assessment. Normalized Difference Vegetation Index (NDVI) is commonly used for assessing vegetation health. During summer and growing season, VPD values are generally high. The results show that the VPD and NDVI provide consistent information on forest health. In addition to VPD, we use conditional probability of NDVI in high temperature and low relative humidity percentiles over the summer and the growing season. We show that combining temperature and relative humidity using a conditional probability approach offers multi-sensor information on forest condition. During winter, on the other hand, VPD and temperature is relatively lower. NDVI distributions in winter were found to be more associated with precipitation as opposed to relative humidity and temperature. We believe the a joint indicator based on temperature and relative humidity can be considered as a link between climate condition and actual impact on the ecosystem.