⚡ electrical engineering

A Hetero-functional Graph State Estimator for Watershed Systems: Application to the Chesapeake Bay

This paper introduces a novel modeling framework based on Hetero-functional Graph Theory and a Weighted Least Squares Error State Estimator to overcome the structural limitations of existing regional watershed models, enabling the integrated analysis of hydrological, economic, and governance systems for sustainable management of the Chesapeake Bay.

Original authors: Megan S. Harris, John C. Little, Amro M. Farid

Published 2026-03-03

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Megan S. Harris, John C. Little, Amro M. Farid

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the Chesapeake Bay watershed as a massive, complex kitchen where millions of people are cooking, cleaning, and farming all at once. The goal is to keep the "soup" (the Bay) healthy. But right now, the chefs (scientists and policymakers) have a recipe book called CAST that tells them exactly how much salt (nitrogen) and pepper (phosphorus) ends up in the final pot.

The problem? The recipe book is a bit of a "black box." It tells you the final taste, but it doesn't show you how the ingredients traveled from the farms, through the rivers, and into the bay. It's like knowing the cake is sweet, but not knowing which spoonful of sugar went where.

This paper introduces a new way to look at the kitchen using a tool called Hetero-functional Graph Theory (HFGT). Think of HFGT as a super-detailed, 3D map that connects every single ingredient, every river, and every decision made by a farmer.

Here is the breakdown of what the authors did, using simple analogies:

1. The Problem: The "Black Box" Recipe

The current tool (CAST) is great at giving a total number. It says, "Okay, 100 pounds of nitrogen entered the Bay this year." But it treats the journey as a magic trick: Poof! It goes from the farm to the Bay. It hides the middle steps. This makes it hard to fix specific problems because you don't know exactly where the pollution is getting stuck or lost along the way.

2. The Solution: Building a "Digital Twin"

The authors built a digital twin of the watershed. Instead of just looking at the final number, they used a new mathematical framework to reconstruct the entire journey of the nutrients.

The Map (The Graph): They turned the watershed into a giant network of pipes and buckets.
- Buckets (Buffers): These are the land segments (farms, cities) and the river segments. They hold the water and nutrients.
- Pipes (Processes): These are the flows moving nutrients from one bucket to the next.
The Detective Work (State Estimation): Since they didn't have a sensor on every single drop of water, they had to use math to "guess" the missing pieces. They used a method called Weighted Least Squares Error.
- Analogy: Imagine you are trying to figure out how much water is in a series of connected buckets, but you only know how much was poured in at the start and how much came out at the end. Some of the pipes might be leaking, or some buckets might be holding extra water. The math acts like a detective, balancing the equation to find the most likely amount of water in every single bucket, even the ones you can't see.

3. The "Hetero-Functional" Part

This is the fancy name for their map. "Hetero" means different types of things are mixed together.

In a normal map, you might just see rivers.
In their map, they also included decisions. They connected the physical rivers to the economic decisions (like a farmer deciding to use fertilizer) and the governance rules (like county laws).
Analogy: It's like a video game where you can see the terrain (the rivers), but you can also see the player's inventory (the fertilizer) and the rules of the game (the laws) all on the same screen.

4. What They Found

When they ran their new model on the Chesapeake Bay data:

It matched the big picture: The total amount of pollution reaching the Bay was almost identical to the old recipe book (CAST). This proves their new map is accurate.
It revealed the hidden paths: Suddenly, they could see exactly which small rivers were carrying the most pollution and which "buckets" (land areas) were holding onto too much nitrogen.
It showed the gaps: They found that while the total numbers were right, the old model was a bit fuzzy about the middle steps. Their new model highlighted exactly where we need better data.

5. Why This Matters

Think of the old model as a blurry photo of a crime scene. You know the crime happened, but you can't see the details.
This new model is a high-definition, 3D reconstruction of the scene.

For Policymakers: Instead of guessing where to put money to clean the water, they can now see exactly which river segment or county needs help.
For the Future: Because this model is built on a flexible "language" (Systems Modeling Language), it's easy to add new things later. You could plug in a model for how the economy works, or how climate change will affect the rain, and the whole system updates automatically.

The Bottom Line

The authors didn't throw away the old, trusted recipe book (CAST). Instead, they built a transparent, interactive dashboard underneath it. They took the reliable numbers from the old book and used them to fill in the missing details of the journey. This allows us to manage the Chesapeake Bay not just by looking at the final result, but by understanding the entire story of how the water and nutrients travel through our world.

1. Problem Statement

Regional watersheds are complex "systems of systems" involving hydrology, land-use decisions, ecological feedbacks, and overlapping governance jurisdictions. Effective management requires models that capture interdependencies between natural and institutional systems.

The primary challenge identified is the limitation of existing operational models, specifically the Chesapeake Assessment Scenario Tool (CAST), which is the official planning model for the Chesapeake Bay Watershed (CBW). While CAST provides valuable nutrient estimates, it suffers from three major limitations for system-of-systems analysis:

Structural Opacity: Its internal routing structure is not explicitly represented; it relies on aggregated "delivery factors" that map land segments directly to the estuary, obscuring the upstream-downstream network topology.
Static Logic: It is a time-averaged (annual) model not designed for dynamic, time-dependent simulation.
Loose Coupling: It is weakly coupled with estuarine models, limiting the ability to represent cross-domain interdependencies (e.g., between hydrology, economics, and policy).

Consequently, researchers lack an extensible, scalable model of the CBW that can be embedded directly into broader system-of-systems analyses to support sustainable management.

2. Methodology

The authors propose a modeling framework that integrates Model-Based Systems Engineering (MBSE) and Hetero-functional Graph Theory (HFGT) to reconstruct the implicit flow structure of CAST into an explicit, mass-conserving network.

A. Theoretical Foundation: HFGT and MBSE

Ontology: The framework uses a shared ontology based on Resources (infrastructure, land), Processes (transformations, transport), and Operands (nutrients like Nitrogen and Phosphorus).
Reference Architecture (RA): A generic watershed Reference Architecture is defined using SysML (Systems Modeling Language) diagrams:
- Block Definition Diagram (BDD): Defines system form (Land segments as transformation resources; River segments as transport resources; Outlet points/Estuaries as buffers).
- Activity Diagram (ACT): Defines system function (Capabilities such as accepting agricultural nutrients, transporting nutrients from land to outlet, and river-to-river transport).
Mathematical Formulation: The core engine is the Weighted Least Squares Error Hetero-functional Graph State Estimator (WLSEHFGSE).
- This extends the Hetero-functional Network Minimum Cost Flow (HFNMCF) problem.
- It introduces statistical error terms into the state transition constraints, allowing the model to infer unknown or partially observed flows from uncertain, exogenous data while maintaining physical feasibility (mass balance).

B. Implementation Steps

Geospatial Reconstruction: The authors used public geospatial data (DEMs, stream networks, river polygons) to reconstruct the CBW's physical topology. They mapped CAST's "Land-River Segments" (intersections of counties and watersheds) to a graph of connected river segments and outlet points, creating a Hetero-functional Incidence Tensor.
Data Integration: CAST behavioral data (nutrient loads, delivery factors, BMP efficiencies) were used to define the Measurement Function. These data serve as exogenous constraints (boundary conditions) for the state estimator.
Optimization: The WLSEHFGSE solves an optimization problem to minimize the weighted sum of squared errors between the model's estimated flows and the CAST-reported data, subject to the mass balance constraints of the reconstructed network.

3. Key Contributions

Ontology-Driven Framework: Introduces a region-independent reference architecture for watershed systems, formalizing the structure of land, river, and estuarine connectivity using MBSE and HFGT.
Algorithmic Extension: Adapts the WLSEHFGSE to estimate nutrient flows in large-scale watershed systems, enabling the inference of flows from uncertain, aggregated data.
Structural Transparency: Successfully reconstructs the "black box" routing of CAST. The model transforms CAST's statistical delivery factors into an explicit, physically interpretable network of river segments and outlet points.
Integration Capability: Creates a scaffold for future integration with economic and governance models by providing a shared computational ontology that bridges hydrology, policy, and economics.
Case Study Instantiation: Demonstrates the framework on the Chesapeake Bay Watershed, utilizing a reproducible Python-based pipeline to process geospatial data and CAST datasets.

4. Results

The model was executed for the "2024 Progress" scenario (a steady-state annual time step).

Mass Balance Closure: The reconstructed HFGT network achieved nearly exact agreement with CAST's total delivered loads to the estuary (End of Tide - EoT), with errors < 0.01% for both Nitrogen and Phosphorus. This confirms the model preserves system-level mass balance.
Spatial Heterogeneity: The model successfully visualized nutrient accumulation and transport across the dendritic river network, identifying critical intermediate nodes that disproportionately accumulate nutrients—details obscured in the aggregated CAST model.
Sector Tracing: The model linked economic sector inputs (agricultural vs. developed land) to specific transport pathways, enabling sector-specific tracing of nutrient delivery.
Goodness of Fit:
- Applied Loads: Strong correlation with CAST county-level data ( $R^2 \approx 0.91$ for N, $0.84$ for P), though with higher normalized RMSE due to spatial heterogeneity.
- Edge of Stream (EoS): Lower agreement ( $R^2 \approx 0.29$ for N) compared to aggregated totals. This is diagnostic, indicating that CAST's regression-based delivery factors at the sub-watershed level are difficult to uniquely resolve without higher-resolution data.
- Stream-to-Tide: Moderate discrepancies (8.86% for N, 28.3% for P) highlight the difference between CAST's lumped coefficients and the explicit network routing.

5. Significance and Future Directions

From Statistical to Structural: The work shifts watershed modeling from a purely statistical emulator (CAST) to a physically structured network model. This allows for dynamic simulation and the assimilation of real-time monitoring data.
Diagnostic Value: The residuals (errors) between the model and CAST data are not failures but diagnostic tools. They reveal where data is sparse or aggregated, guiding future data collection efforts (e.g., field-scale measurements of attenuation).
Scalability: The framework is extensible. Future work will focus on:
- Coupling with estuarine models (incorporating freshwater delivery and nutrient phase partitioning).
- Integrating dynamic economic and governance models to simulate policy impacts.
- Comparing the HFGT reconstruction against the original dynamic HSPF model to validate spatial routing.

Conclusion: This paper establishes a foundational step toward a "System of Systems" model for the Chesapeake Bay. By rendering the complex socio-environmental dynamics of the watershed explicit and structurally transparent, the HFGT-based approach provides a complementary tool to existing regional models, enabling more robust, interdisciplinary analysis for sustainable management.