Use-inspired hydrology seeks to address specific water resources management and policy questions by improving fundamental understanding of hydrologic systems. This dissertation consists of two use-inspired hydrologic projects in landscapes experiencing rapid anthropogenic change. The first project examines associations between deforestation and rainfall patterns in the Brazilian Amazon-Cerrado (forest-savanna) transition zone. The second project benchmarks several hydrologic models used for streamflow prediction in a California watershed.

Deforestation in the Brazilian Amazon-Cerrado transition zone can lead to changing rainfall patterns. Simulations suggest that deforestation affects rainfall in a seasonally varying, scale- and direction-dependent manner. Yet, few observational studies have attempted to verify these model findings. Though previous observational studies confirmed scale-dependence, these studies typically focused on several agriculturally relevant rainfall metrics rather than a suite of rainfall metrics that broadly characterize hydroclimatic dynamics. They also did not examine all seasons and relied on the assumption that deforestation’s effects on rainfall were isotropic. In Chapter 1, we related deforestation to seven rainfall metrics characterizing observed rainfall intensity, occurrence, and spatial clustering in a seasonal, multi-scale, and multi-directional statistical analysis. We presented three main findings. First, we found that in wet and transitional seasons, deforestation corresponded to more spatially clustered rainfall, with higher intensity, lower frequency, and less total seasonal rainfall. In contrast, in the dry season, lower intensity rainfall occurred more frequently in association with deforestation. Second, our results indicated little scale-dependence in the strength of the deforestation-rainfall relationship across the examined spatial scales. Third, most deforestation-rainfall associations did not depend on the cardinal direction of forest cover relative to the rainfall sites, though there were exceptions to this during the dry and transition seasons. Our findings from Chapter 1 provide a more holistic description of hydroclimatic dynamics in a region undergoing rapid deforestation, with implications for water resources management.

Then, in Chapter 2, we focus on the association between deforestation and rainfall spatial connectivity in the Brazilian Amazon-Cerrado transition zone. Rainfall spatial connectivity describes the statistical similarity of daily rainfall at different geographic locations. It may be indicative of deforestation-caused changes in the spatial extent of rainfall events and, thus, changes in the spatial scale of dominant rainfall generation processes. However, the relationship between deforestation and rainfall spatial connectivity has not yet been fully examined. Here, we quantified rainfall spatial connectivity as a network. The network was summarized using network statistics, such as the number, length, clustering, and sorting of rainfall connections. We then examined seasonal associations between deforestation and the rainfall network statistics. Overall, we found sparse, weak associations between deforestation and the rainfall network statistics. However, we found some evidence for increased rainfall patchiness with deforestation during the wet-to-dry transition season. We also observed a shift towards longer connection distances during the wet season. Additionally, when rainfall connections were more sorted by forest cover, we observed fewer, less direct connections during the transition seasons. Our findings from Chapter 2 may be the result of changes in moisture recycling, the initiation of mesoscale circulations, or intra-seasonal variability in the large-scale drivers of rainfall.

Finally, in Chapter 3, we shift our attention to streamflow prediction in California, a landscape experiencing increased drought risk and precipitation volatility due to climate change. Under these conditions, accurate streamflow predictions are essential for water resources management, yet challenging to produce. Though recent studies have examined the integration of machine learning tools with process-based hydrologic models to improve streamflow predictions, the possibility of replacing parameter calibration or complex process depictions with machine learning has not yet been investigated. By combining machine learning with a simpler or uncalibrated process-based model, hydrologic model post-processing may improve streamflow predictability and minimize time spent on model construction, without sacrificing process understanding. In this study, we used model benchmarking as a diagnostic tool to examine the added value of hydrologic model calibration, complexity, and post-processing within two watersheds in northern California’s Mediterranean climate. The watersheds were managed for both municipal water supplies and aquatic habitat. Consequently, the models were benchmarked using a suite of error metrics that capture a wide variety of streamflow characteristics at annual, monthly, and daily time scales. Though model performance varied substantially by watershed and error metric, we found that (i) parameter calibration reduced error for 71% of the error metrics; (ii) more complex, calibrated process-based models had lower error than less complex, calibrated process-based models for 48% of the error metrics; and (iii) post-processing could replace parameter calibration and model complexity for 43% and 69% of eligible error metrics, respectively. Furthermore, the calibrated hybrid models (calibrated process-based models that were post-processed using the machine learning algorithm) frequently outperformed both the machine learning algorithm (for 83% of error metrics) and the calibrated process-based models (for 72% of error metrics). Our results demonstrate that post-processing is a promising method for replacing parameter calibration and model complexity. However, additional studies using a larger sample of watersheds are needed to uncover more generalizable guidelines regarding its use.

Throughout this dissertation, we investigate use-inspired hydrologic questions in landscapes undergoing rapid anthropogenic change. In Chapters 1 and 2, we used statistical models to examine the association between observations of deforestation and rainfall patterns in the Amazon-Cerrado transition zone. Then, in Chapter 3, we benchmarked a group of process-based, data-driven, and hybrid streamflow prediction models in a California watershed. In all three chapters, our synthesis of hydrologic models and observations has the potential to improve the hydrologic predictions used for water resources management and policy in landscapes facing increasing risks from deforestation, precipitation volatility, and drought.

Data is one of the essential components in analyzing complex earth system problems. With high-quality data more feasible to the researchers, more details of the system could be revealed by those data-intensive computational and data-driven approaches. The measurement and data collection devices have been developing dramatically, especially those used for earth system science. The high sampling resolution in all spatial, temporal and spectral scale have enabled the analysis of earth system problems into a data-driven era. Meanwhile, the fast development of computational ability and resources allow the emergence of innovative data-driven methods (e.g., information theory, traditional statistical learning models, deep learning models). The data-driven approach is different from the physical-based (or knowledge-based) modeling. It emphasizes learning and generalizing the rules from large amounts of representative data. It tries to fit the probability distribution function, for any questions, with the support of large numbers of observations with little constraining conditions like those from the physical-based model. However, before relying on purely data-driven methods, it is essential to remember that Earth systems are characterized as nonlinear, complex and dynamic systems with couplings and feedback among components and subsystems.

Additionally, these coupled processes change depending on the status of the system and the spatial and temporal scale at which the system is analyzed. To understand the underlying mechanisms that drive complex systems, it is useful to conceptualize the system as a network of variables undergoing interactions and feedback. Traditional statistical analysis methods are ill-suited to capture the key attributes of this type of feedback processes due to the stochasticity of the variables, the nonlinearities of the couplings and the non-stationarity of the system. The limitation of the data (in terms of resolution and length in both spatial and temporal scale) and computational ability further narrow the effectiveness of those methods.

The various science communities are now facing a new challenging problem. On the one hand, you have 1) more and more data being collected, 2) the significantly-improved ability to depict the status of a system and to describe the details of a relationship between the components within the system, and 3) the computational capacity and resources to be able to handle this large number of data, motivating the use of data-driven methods.

In this dissertation, I will examine the potential for integrating data-driven techniques into earth systems science to improve our understanding of earth-surface processes. Specifically, I focus on applying data-driven techniques for resolving causal interactions of the several complex earth systems over multispectral and temporal scales. Four complex earth system problems with different spatial and temporal scales are discussed. First, we implement the data-driven methods in regional and decadal issues, streamflow prediction, as a case study. Our findings suggest that while information-flow identifies dominant streamflow controls, the results should not be limited to only “critical hydrologic timescales;” instead they should guide a range of timescales over which inputs, stores, and losses are filtered into catchment discharge. Second, we analyzed a regional and yearly problem, the feedback process between vegetation and topography in a lake delta ecosystem. The transfer entropy analysis suggests that different vegetation communities play functionally different roles in landscape evolution that should be differentiated in ecogeomorphic models. Within such models, it would be most imperative to resolve detailed flow characteristics at lower to low-middle island elevations.

Furthermore, within elevation zones, it is likely essential to differentiate between the roles of multiple vegetation communities rather than treating the entire elevation zone as a single ecogeomorphic entity. Third, we analyzed global and millennium problems, the interaction among climatically variables over 42,000 years. We show that, during the past 420,000 years, orbital forcings trigger temperature and CO2 responses at short (5 kyr) time lags. Over longer timescales, internal feedback, mediated by interactions with dust, also plays a significant role in governing temperature and CO2 concentrations. The short-term influence of CO2 on temperature was stronger than dust’s long-term impact, consistent with on radiative forcing. However, dust remained an essential driver of temperature over 50-kyr time lags, the amount of time between sequential glacial maxima and minima during the latter portion of the Pleistocene. Last, we analyzed a global and decadal problem, the interaction between ocean and precipitation on land. We quantitatively demonstrate that Sea Surface Temperature (SST) over the Gulf of Guinea controls moisture advection and transport to the West Sahel region; strong bidirectional interaction exists between local vegetation dynamics and rainfall patterns. The spatial distribution map of time lag with most significant transfer entropy also shows the apparent trend of each climate indices tested in this research. The Niño 3+4 and Niño 4 have a relatively short time lag with significant transfer entropy to the west coast and have insignificant information transferred to the middle US. The Niño 1+2 and Niño 3 have a relatively short time lag with significant information transferred to the middle region but insignificant information transferred to the west coast.

By testing the effectiveness and efficiency of the data-driven methods in complex earth system problems over multiple spatial and temporal scales, the results verified the ability of those methods in identifying and quantifying the strength, statistical significance, directionality and critical time lags of feedback (as well as one-way forcing) among variables. With these data-driven methods, we could identify which components comprise the system, and which dominate changes within the system. With the input of that knowledge, we could further predict the behavior of an element of interest or the stationery of the whole system and simulate the future behavior of the system under different scenario after fully understanding the rules and the connections of a system.

Data science is a fundamental tool in hydrology nowadays. The significance of data science lies in its ability to confront the multifaceted challenges posed by global warming, facilitating a deeper comprehension of hydrological processes, and enhancing the accuracy of runoff predictions. This dissertation embarks on a journey aimed at advancing our insights into hydrological processes, refining the physical-consistency of runoff predictions, and addressing the intricate task of forecasting hydrological behaviors in ungauged basins through the application of data science techniques. Comprising three main bodies of work, this dissertation unfolds a comprehensive exploration of these objectives. The first contribution (Chapter 2) centers on the synthesis of extensive hydrological datasets and subsequent analysis of hydrological trends under recent warming. Chapter 3 explores a physics-informed machine learning model designed for predicting streamflow tested across different scenarios. Lastly, the fourth chapter evaluates the potency of various watershed clustering mechanisms for predicting within ungauged basins (PUB).

Chapter 2 addresses a long-standing limitation in comparative hydrology: the scarcity of geographically extensive, inter-compatible monitoring data on comprehensive water balance stores and fluxes. These limitations have, for example, restricted comprehensive assessment of multiple dimensions of wetting and drying related to climate change and hampered understanding of why widespread changes in precipitation extremes are uncorrelated with changes in streamflow extremes. In this chapter, both the requirements of developing a new data synthesis product and using this data product to detect trends in the frequencies and magnitudes of a comprehensive set of hydroclimatic and hydrologic extremes are addressed. The Comprehensive Hydrologic Observatory Sensor Network (CHOSEN), a database encompassing hydroclimatic and hydrologic variables from 30 diverse study areas across the United States is introduced. And a reproducible data pipeline that ensures data quality and accessibility is developed. Analyzing the CHOSEN dataset, the hotspots of hydroclimatic extremes in regions like the Pacific Northwest, New England, Florida, and Alaska are uncovered. The analysis reveals regional coherence in extreme streamflow wetting and drying trends, shedding light on the complex interplay between climate-induced changes and hydrologic processes.

Chapter 3 is built upon the development of the CHOSEN dataset to create subsequent analyses and a new runoff prediction model. The challenge of a lack of interpretability and physical consistency in machine learning models used for streamflow prediction is confronted. To address this issue, a physics-informed long short-term memory (PILSTM) model is proposed, incorporating water balance restrictions for runoff prediction. A physical rainfall-runoff model is combined with the long short-term memory (LSTM) model, and it is applied to eight intensively-monitored watersheds in the United States, selected based on data length and hydroclimatic diversity. LSTM, physical, and PILSTM models are used under non-stationary scenarios and data-scarce situations. Results show that the PILSTM exhibits similar or better performance to the LSTM counterpart in terms of multiple metrics and under various scenarios. Additionally, based on the analysis of feature importance, it is shown that adding physical constraints could potentially guide machine learning models to generate predictions that are more consistent with known physical processes.

Chapter 4 explores the effectiveness of watershed clustering, a conventional practice in watershed regionalization, in combination with neural networks for predicting in ungauged basins. Traditionally, watershed clustering involves grouping basins with similar characteristics to facilitate knowledge transfer from monitored to ungauged basins within the same cluster. Recent advancements in data science, however, suggest that clustering may not be necessary. This study aims to investigate this matter and presents a comparative analysis of various watershed clustering methodologies. The concept is explored by directly integrating static watershed attributes into predictive models for streamflow (entity-aware LSTM). The analysis covers 415 sites from the CAMELS (Catchment Attributes and Meteorology for Large-sample Studies) dataset. Results indicate that pre-clustering generally does not enhance the performance of entity-aware LSTM models for predicting in ungauged basins. Models incorporating clustering results either match or perform worse overall compared to global models that directly integrate clustering features as static inputs. Notably, among the different features used for clustering, hydrological signatures prove most effective in extracting information for use in the LSTM model.

Chapter 2 addresses crucial gaps in data availability, while the subsequent chapters explore novel approaches for forecasting streamflow across diverse scenarios and ungauged basins, leveraging the power of data science. In Chapter 3, the integration of physical and machine learning models is pursued, while Chapter 4 focuses on harnessing data science methodologies for predicting in ungauged basins. Collectively, these chapters offer an exploration of the intersection between data science and hydrology. This dissertation emphasizes the transformative potential of interdisciplinary strategies, which bridge data-driven insights with the dynamics of hydrological systems.

Precipitation and land cover in the West African Sahel have changed dramatically over the past 50 years. Region-wide data on land cover change in the Sahel, however, have been sparse or unreliable. I present a new 30 meter 2000-2016 annual resolution land cover dataset for the West African Sahel. The product is built from hand-classified land cover maps using random forest machine learning methods with Landsat, precipitation, and topography features. The resulting maps confirm the widespread extensification of agriculture in the region over this time period. Contrary to the common narrative of desertification, this increase in agriculture has not been accompanied by an increase in bare soil or sandy area. Land cover change volatility is shown to be spatially heterogeneous, both at local and regional scales. In addition to the new land cover dataset, I present spatial and temporal analyses of precipitation during the recent years of increased variability in the West African Sahel. I examine seasonal trends, interannual variability, and differences among datasets representing precipitation in the Sahel. Region-wide spatial organization of precipitation is identified using the self-organizing mapping pattern recognition technique. The number of days spent in the monsoon transition period is strongly negatively correlated with annual precipitation anomaly` indicating a tradeoff with the peak monsoon period, a result that supports the upped-ante hypothesis of precipitation in the Sahel.