Introduction: A Year of Transformation

{#cr-intro} When I began the Master of Environmental Data Science (MEDS) program at UC Santa Barbara’s Bren School, I knew I was stepping into a world where environmental science meets cutting-edge computational methods. What I didn’t fully anticipate was how profoundly this one-year intensive program would reshape my understanding of data, science, and the complex environmental challenges facing our planet.

This journey—from foundational mathematics to advanced machine learning, from geospatial analysis to policy evaluation—culminated in a capstone project that seeks to identify sustainable pathways for hydropower development in a net-zero future. Through this closeread document, I invite you to explore the knowledge, skills, and insights I’ve gained across thirteen core courses and a transformative capstone experience.

The Program Structure

The MEDS curriculum is organized into four quarters, each building upon the previous:

Summer Session B: Foundational skills in mathematics, workflows, Python, and scientific programming
Fall Quarter: Working with environmental datasets, statistics, geospatial analysis, and ethics
Winter Quarter: Modeling environmental systems, data visualization, policy evaluation, and capstone initiation
Spring Quarter: Databases, machine learning, and capstone completion

Each course was designed not just to teach technical skills, but to foster a deep understanding of how data science can be responsibly applied to environmental challenges.

Summer Session B: Building Foundations

{#cr-summer} The summer session was an intensive introduction to the tools and thinking that would underpin everything to come. Four courses in rapid succession: Essential Math, Analytical Workflows, Python Programming, and Scientific Programming Essentials.

EDS 212: Essential Math for Environmental Data Science

Instructor: Carmen Galaz García

Mathematics is the language of data science. This course refreshed fundamental skills in algebra, calculus, differential equations, linear algebra, and logical operations—all essential for understanding the models and methods we would use throughout the program.

Key Concepts: - Multivariate functions and their applications - Derivative and integral calculus for environmental modeling - Differential equations for dynamic systems - Linear algebra for data transformations - Logical operations for data filtering and conditional analysis

The course emphasized that mathematical understanding isn’t just about computation—it’s about developing intuition for how environmental systems behave and how we can represent them mathematically.

EDS 214: Analytical Workflows and Scientific Reproducibility

Instructor: Max Czapanskiy

Reproducibility isn’t optional in science—it’s fundamental. This course taught me how to create workflows that are transparent, shareable, and reproducible.

Core Skills: 1. Automation: Scripting every step of an analytical workflow 2. Modularity: Organizing workflow components for reusability 3. Documentation: Clearly documenting components and their relationships 4. Scaling: Adapting workflows for computational performance and large datasets 5. Collaboration: Working in teams to develop shared workflows

The course instilled in me the importance of thinking about analysis as a process, not just a result. Every script, every data transformation, every visualization should be traceable and reproducible.

EDS 217: Python for Environmental Data Science

Instructor: Kelly Caylor

Python became my primary tool for data manipulation and analysis. This course covered foundational programming concepts through the lens of environmental data.

What I Learned: - Data structures (lists, dictionaries, DataFrames) - Programming basics (loops, conditionals, functions) - Data cleaning, subsetting, aggregation, and transformation - Data visualization with matplotlib and seaborn - Application to environmental problem-solving

The course emphasized that Python isn’t just a programming language—it’s a tool for thinking about data. Learning to manipulate data programmatically opened up possibilities for analysis that would have been impossible with spreadsheet software.

EDS 221: Scientific Programming Essentials

Instructor: Max Czapanskiy

Building on Python foundations, this course delved deeper into scientific programming practices using both R and Python.

Advanced Topics: - Structured programming and algorithm development - Flow control and program logic - Advanced data input-output and representation - Functions and object-oriented programming - Documentation, testing, and debugging

This course taught me that good scientific code is readable, testable, and maintainable. It’s not enough for code to work—it needs to be understandable to others (and to your future self).

Fall Quarter: Deepening Analysis

{#cr-fall} Fall quarter brought four courses that expanded my analytical toolkit: Working with Environmental Datasets, Statistics, Geospatial Analysis, and Data Ethics. These courses showed me how to apply foundational skills to real environmental data and questions.

EDS 220: Working with Environmental Datasets

Instructors: Carmen Galaz García, Annie Adams

Environmental data comes in many forms: field measurements, station data, remote sensing products, climate model projections. This course taught me how to work with this diversity.

Key Skills: - Evaluating data collection and quality control methods - Working with time-series and spatial information - Accessing cloud computing databases and environmental data repositories - Basic workflows for selecting, obtaining, and visualizing datasets - Best practices for reliable data intercomparisons

I developed a tutorial Jupyter notebook for a use case of my choice, learning to navigate the complexities of real environmental datasets. The course emphasized that understanding data provenance and quality is as important as the analysis itself.

EDS 222: Statistics for Environmental Data Science

Instructor: Max Czapanskiy

Statistics provides the framework for making inferences from data. This course covered fundamental statistical concepts and their application to environmental questions.

Core Topics:

Random Variables and Distributions

Binomial distributions for discrete data (presence/absence)
Normal distributions for continuous data
Gamma distributions for positive continuous variables (durations, concentrations)
Understanding parameters and how they control distribution shape

Linear Regression

Fitting lines to points and understanding residuals
Quality of fit: R² and variance explained
Population vs. sample regression models
Assumptions about data generating processes (DGP)

Categorical coefficients as intercepts
Additive vs. interactive models
Confounding variables and backdoor paths
DAGs (Directed Acyclic Graphs) for causal relationships

Inference

Permutation and bootstrapping
Hypothesis testing and confidence intervals
Normal approximations of sampling distributions
Linear regression inference

The course emphasized that statistics isn’t just about running tests—it’s about understanding the assumptions behind methods and choosing appropriate approaches for your data and question.

Categorical and Continuous Predictors

EDS 223: Geospatial Analysis and Remote Sensing

Instructor: Annie Adams

Spatial data is fundamental to environmental science. This course introduced geographic information science and remote sensing through deep understanding of spatial data models.

Spatial Data Models:

Vectors (Discrete Data)

Points, lines, and polygons
Spatial subsetting and topological relationships
Filtering, clipping, and spatial operations

Rasters (Continuous Data)

Grid-based data representation
Map algebra: local, focal, zonal, and global operations
Aggregating and resampling raster data

Understanding CRS (Coordinate Reference Systems)
Geographic vs. projected coordinate systems
Datums and map projections
Transforming between coordinate systems

Remote Sensing

Passive vs. active remote sensing
LiDAR and RADAR fundamentals
Energy-matter interactions (absorption, reflectance, scattering)
Image resolutions: spatial, temporal, spectral, radiometric
Land cover classification (supervised and unsupervised)

The course taught me that spatial data requires special consideration—spatial autocorrelation, coordinate systems, and scale all matter profoundly for analysis and interpretation.

Coordinate Systems

EDS 242: Ethics and Bias in Environmental Data Science

Instructor: Jayajit Chakraborty

Data science isn’t neutral. This course examined ethical considerations in collecting, using, and reporting environmental data, and how to recognize and account for biases.

Types of Data Bias: 1. Confirmation Bias: Seeing what you want to see 2. Historical Bias: Deceived by context of data creators 3. Selection Bias: Non-representative samples (coverage, non-response, sampling, recall) 4. Survivorship Bias: Including only surviving data points 5. Availability Bias: Recording data by ease of access 6. Reporting Bias: Dataset frequency doesn’t reflect real-world frequency 7. Observer Bias: Subjectivity of the observer 8. Geographic Data Bias: Where data comes from matters 9. Urban vs. Rural Bias: Spatial representation disparities

The course emphasized that recognizing bias isn’t enough—we must actively work to mitigate it and be transparent about limitations in our data and methods. Ethical data science requires constant vigilance and critical reflection.

Winter Quarter: Advanced Applications

{#cr-winter} Winter quarter brought advanced courses in modeling, visualization, and policy evaluation, while also launching the capstone project. These courses showed me how to apply data science to complex environmental systems and policy questions.

EDS 230: Modeling Environmental Systems

Instructor: Christina Tague

Models are simplifications of reality designed to answer questions. This course covered both selecting and applying existing models and designing new ones.

Model Types: - Stochastic vs. Deterministic: Probability-based vs. fixed outcomes - Lumped vs. Spatially Distributed: Single point vs. spatial variation - Static vs. Dynamic: Single time point vs. evolving through time - Abstract vs. Physically Based: Data-driven (ML) vs. mechanism-based

Key Concepts: - Conceptual models: diagrams of system relationships - Sensitivity analysis: understanding parameter uncertainty - Dynamic models: differential equations and state space trajectories - Feedback loops: positive (runaway growth) and negative (dampening) - Stability: stable, unstable, and chaotic systems

The course emphasized that “all models are wrong, but some are useful.” The goal isn’t perfect representation—it’s useful simplification that helps answer specific questions.

EDS 240: Data Visualization and Communication

Instructors: Samantha Shanny-Csik, Annie Adams

Effective communication is central to data science. This course focused on creating responsible, accessible, and visually compelling visualizations.

The Grammar of Graphics (ggplot2): 1. Data: Tidy format + aesthetic mappings 2. Geometric objects: Type of plot 3. Statistical transformations: Algorithms for new values 4. Position adjustments: Handling overlaps 5. Coordinate systems: Cartesian, polar, flipped 6. Facets: Subplots for data subsets

Additional Layers: - Labels: titles, axis labels, legends - Annotations: textual labels and highlights - Scales: color scales, axis ticks - Themes: Customizing non-data elements - Layout: Combining multiple plots

The course also introduced closeread-html for creating interactive, scrollytelling visualizations—the format you’re reading now! This tool allows for sticky elements, triggers, and dynamic highlighting that guide readers through complex narratives.

EDS 241: Environmental Policy Evaluation

Instructors: Adam Garber, Annie Adams

Causal inference is crucial for evaluating environmental policies. This course presented state-of-the-art program evaluation techniques to identify and measure causal effects.

Research Designs:

Experimental Methods

Randomized Controlled Trials: Random assignment to treatment/control
Natural Experiments: Real-world situations mimicking random assignment

Quasi-Experimental Methods

Difference-in-Differences (DiD): Leveraging time trends with parallel trends assumption
Regression Discontinuity Design (RDD): Utilizing arbitrary thresholds/cutoffs
Instrumental Variables (IV): Exploiting indirect influences

Other Methods

Fixed Effects: The within estimator for panel data
Matching Methods: Creating apples-to-apples comparisons

Key Concepts: - Potential outcomes framework: Yi(1) vs. Yi(0) - Average Treatment Effect (ATE) vs. Average Treatment for the Treated (ATT) - Selection bias and omitted variable bias - Panel data: repeated measures over time - Robust standard errors for heteroscedasticity

The course emphasized that correlation isn’t causation. Design-based research methods, combined with careful statistical analysis, allow us to make stronger causal claims about policy impacts.

EDS 411A: Capstone Project (Part 1)

Instructor: Carmen Galaz García

The capstone project is the culmination of the MEDS program—a real-world application of data science to an environmental problem, conducted in collaboration with external clients.

Project Structure: - 10-12 hours per week dedicated to capstone work - Design and Implementation Plan (DIP): First draft due week 5, final due week 10 - Faculty Review: 12-minute presentation of DIP - Collaboration: Working with clients, advisors, and team members

Ten Simple Rules for Collaborative Data Science Teams: 1. Lead from the front 2. Seeing is believing 3. Skin in the game 4. Comfort through clarity 5. All codes lead to Rome 6. People first 7. Empowerment through ownership 8. Open science 9. Safe learning spaces 10. Have fun

The capstone isn’t just about technical skills—it’s about professional development, collaboration, and applying everything learned to a meaningful environmental challenge.

Spring Quarter: Integration and Completion

{#cr-spring} Spring quarter brought the final core courses and capstone completion. These courses integrated everything learned and prepared us for professional practice.

EDS 213: Databases and Data Management

Instructors: Julien Brun, Greg Janée, Annie Adams, Renata Curty

Data management is foundational to reproducible science. This course taught relational database structure, SQL, and data archiving.

Key Topics: - Relational database structure and schemas - Data relationships and normalization - SQL for creating and querying databases - Metadata concepts and standards - Archiving data products on repositories - Making data available to broader community

The course emphasized that good data management isn’t just about storage—it’s about making data findable, accessible, interoperable, and reusable (FAIR principles).

EDS 232: Machine Learning in Environmental Science

Instructors: Carmen Galaz García, Annie Adams

Machine learning helps process big, complex data and extract knowledge. This course provided a broad introduction to ML and statistical pattern recognition.

Topics Covered: - Supervised Learning: Decision trees, random forests, support vector machines, neural networks - Unsupervised Learning: Clustering, dimensionality reduction, deep learning - Applications framed within environmental science - Programming in R and Python for advanced scientific programming

The course emphasized that machine learning is a tool, not a solution. Understanding when and how to apply ML methods, and interpreting results responsibly, is crucial for environmental applications.

EDS 411B: Capstone Project (Part 2)

Instructor: Carmen Galaz García

The second quarter of capstone focused on completing all project plans and deliverables, developing the project repository and technical documentation, and presenting research to a general audience.

Deliverables: - Completed project implementation - Public GitHub repository with all code and workflows - Technical documentation - Interactive web application (Shiny app) - Final presentation to general audience - Project repository adhering to open science principles

The capstone experience integrated everything learned throughout the program: technical skills, collaboration, communication, and ethical practice. It was both a culmination and a beginning—preparing us for professional careers in environmental data science.

The Capstone: Hydropower’s Low-Hanging Fruits

{#cr-capstone} My capstone project, conducted with teammates Leela Dixit, Sofia Sarak, and Vedika Shirtekar, represents the integration of everything I learned in MEDS. In collaboration with Dr. Jeff Opperman (WWF Global Science) and Dr. Rafael Schmitt (UCSB), we developed a framework to identify sustainable pathways for hydropower development.

The Environmental Challenge

Hydropower presents a fundamental dilemma: it’s a low-carbon, flexible energy source that can support climate action, but dams fragment rivers, disrupt ecosystems, and impact communities. Current projections suggest we may need to double global hydropower capacity by 2050 to meet climate objectives, but 63% of the world’s longest rivers are already fragmented.

The Core Question: Can we identify “low-hanging fruits”—future dam sites that combine lower environmental impact with energy system benefits—to reconcile climate goals with river conservation?

This question isn’t just academic. It’s urgent. As Dr. Opperman wrote in his letter of support: “Current high-level policy debates about hydropower’s role in addressing future climate and energy challenges are not adequately considering the potential tradeoffs with hydropower expansion.”

Project Objectives

Our project aimed to analyze sustainability constraints for expanding future hydropower by:

Identifying topological relationships: Determining where future dams sit relative to existing dams (upstream, downstream, within cascades, or blocking free-flowing rivers)
Screening for conflicts: Identifying “fatal flaws” such as conflicts with protected areas or population centers
Categorizing sustainability: Classifying at least 3,700 global dam sites into sustainability categories
Quantifying capacity: Calculating generation capacity (in gigawatts) within each category

The goal was to enable investors and policymakers to judge the environmental impact associated with different levels of hydropower expansion, closing a critical knowledge gap in energy systems planning.

The Data Science Approach

Key Datasets: 1. HydroRivers: Global vectorized river network with metadata (discharge, etc.) 2. GRAND: Global Reservoir and Dam dataset (7,300 existing large dams) 3. FHReD: Future Hydropower and Reservoir Data (~3,700 future dam sites) 4. Protected Planet: Protected area boundaries for conflict screening 5. WorldPop: Population density data for conflict screening

Analytical Tools: - sf package: Vector data and geospatial operations - tidygraph package: Network topology and connectivity analysis - riverconn package: River fragmentation indices - Shiny package: Interactive web applications

All datasets are vectorized and publicly available, enabling reproducible analysis on personal computers by processing one river basin at a time.

Deliverables

Deliverable 1: Scripted Workflow

A reproducible workflow to: - Identify topological location of future dams relative to existing dams - Screen for conflicts (“fatal flaws”) with protected areas and population centers - Enable future expansion with additional data and screening criteria

Deliverable 2: Updated Dataset

An enhanced dataset of future dams with: - Sustainability category classification (within cascade, upstream/downstream, undammed river, or fatally flawed) - Increase in river fragmentation index for each future dam - Generation capacity by category

Deliverable 3: Interactive Web Viewer

A Shiny application enabling users to: - Visualize results spatially - Filter and analyze by country, capacity range, or sustainability category - Explore dam locations and their impacts

The Research Context

Our work builds on important prior research:

Schmitt & Rosa (2024) found that: - Global demand for hydropower (+38%) and irrigation storage (+73%) by 2050 - Europe and South Asia face “double deficits” where neither energy nor water needs can be met by dams alone - 43% of current hydropower relies on multipurpose dams, creating conflicts - 60-64% of remaining technical potential needed just to meet basic projections

Grill et al. (2019) mapped the world’s free-flowing rivers, showing that: - Only 1/3 of major rivers remain free-flowing - Current dam building would fragment many remaining free-flowing rivers - Large tropical rivers supporting productive fisheries and deltas are particularly at risk

Schmitt et al. (2019) on the Mekong: - Strategic dam placement can achieve same energy with 4% sediment loss vs. 79% under business-as-usual - The 3S Basin (10% of land area) provides 25% of sediment—losing it is catastrophic - Vietnam gets 0% of new hydropower but bears 100% of geomorphic costs

Key Insights and Broader Impact

What Our Analysis Reveals:

Spatial Patterns: Some regions (Latin America, South Asia) have hydropower surplus relative to decarbonization needs, while others face deficits, especially where sites are in ecologically sensitive or populated areas.
Cascade Benefits: Dams built within existing cascades have much lower impacts than new dams on free-flowing rivers—this is the “low-hanging fruit.”
Conflict Identification: Many proposed sites conflict with protected areas or population centers—these “fatal flaws” should be avoided.
Capacity by Category: Understanding how much capacity exists in each sustainability category enables energy system planners to make informed decisions.

Broader Impact:

Our methods contribute to: - Policy Dialogues: Enabling global and national policy discussions based on quantitative data - Energy Systems Modeling: Straightforward integration of sustainability criteria into energy planning - Accessibility: Allowing local NGOs and citizens without resources to engage in science-informed debates - Avoiding Detrimental Effects: Helping minimize negative impacts on rivers and communities

The workflow is designed to be adaptable, allowing for public expansion with additional indicators and downscaling to inform local decisions.

Technical Challenges and Solutions

Challenge 1: Network Topology Understanding spatial relationships between dams in river networks required network analysis. We used tidygraph and sfnetworks to model river connectivity and identify cascade relationships.

Challenge 2: Scale Processing global datasets with millions of river reaches required careful spatial subsetting and processing by basin. We developed workflows that could run on personal computers by processing one basin at a time.

Challenge 3: Reproducibility Creating a workflow that others could use and modify required extensive documentation, modular code design, and clear data dependencies. We followed best practices from EDS 214 (workflows) throughout.

Challenge 4: Visualization Communicating complex spatial and categorical information required thoughtful design. We used principles from EDS 240 (visualization) to create clear, accessible visualizations in our Shiny app.

Reflection: What I Learned

The capstone project was the most challenging and rewarding experience of the MEDS program. It required:

Integration: Bringing together skills from every course—statistics, geospatial analysis, modeling, visualization, ethics, and more
Collaboration: Working effectively with teammates, clients, and advisors, following the “Ten Simple Rules” for collaborative data science
Problem-Solving: Tackling real-world challenges with no clear answers, requiring creativity and persistence
Communication: Translating technical work for diverse audiences—scientists, policymakers, and the public
Responsibility: Working on a project with real-world implications, requiring careful consideration of ethics and impacts

The capstone wasn’t just about applying what I learned—it was about learning how to learn, how to adapt, and how to contribute meaningfully to environmental problem-solving.

Additional Courses: Specialized Skills

{#cr-additional} Beyond the core curriculum, I took two additional courses that expanded my toolkit:

EDS 231: Text and Sentiment Analysis for Environmental Problems

Instructor: Mateo Robbins

This course covered foundations and applications of natural language processing for environmental data sources.

Applications: - Social media feeds (e.g., Twitter) for environmental sentiment - Agency reports and policy documents - Text processing and classification - Semantics and natural language parsing

The course showed me how text data can provide insights into public perception, policy discourse, and environmental communication.

EDS 296-1S: Intro to Shiny - Building Reactive Apps and Dashboards

Instructor: Samantha Shanny-Csik

This short course taught the fundamentals of building interactive web applications with R Shiny.

Key Skills: - Understanding reactivity - Customizing user interfaces - Best practices and workflows - Deploying apps via shinyapps.io

This course directly supported our capstone project, enabling us to create the interactive web viewer for our dam sustainability analysis.

Conclusion: A Year of Growth

{#cr-conclusion} Looking back on this year, I’m struck by how much I’ve learned—not just technical skills, but ways of thinking about data, science, and environmental challenges.

Technical Skills: - Programming in R and Python - Statistical analysis and inference - Geospatial analysis and remote sensing - Machine learning and modeling - Data visualization and communication - Database management and workflows

Professional Skills: - Collaborative data science - Reproducible workflows - Ethical data practice - Policy evaluation methods - Scientific communication

Personal Growth: - Confidence in tackling complex problems - Appreciation for interdisciplinary collaboration - Commitment to open science and reproducibility - Understanding of the responsibility that comes with data science

The MEDS program has prepared me not just to analyze environmental data, but to do so responsibly, collaboratively, and effectively. The capstone project, in particular, showed me how all these pieces fit together to address real-world environmental challenges.

As I move forward in my career, I carry with me not just technical skills, but a framework for thinking about how data science can contribute to environmental solutions. The challenges we face—climate change, biodiversity loss, resource management, environmental justice—require the kind of interdisciplinary, data-driven approach that MEDS embodies.

Thank you for joining me on this journey through the Master of Environmental Data Science program. The work continues, but the foundation is strong.

Acknowledgments

This journey was made possible by:

The Bren School faculty who designed and taught this exceptional program
NCEAS for providing collaborative research support
My capstone team: Leela Dixit, Sofia Sarak, and Vedika Shirtekar
Our capstone clients: Dr. Jeff Opperman (WWF) and Dr. Rafael Schmitt (UCSB)
My MEDS cohort for collaboration and community
The broader environmental data science community for open science and knowledge sharing

This document was created using Quarto’s closeread-html format, which enables interactive, scrollytelling narratives. The styling matches my personal website’s SCSS theme, creating a cohesive visual experience.

All course materials, code, and project deliverables are available in public GitHub repositories, following open science principles.