Web Applications for Interactive Environmental Modeling

PhD Defense

Jeffrey D. Walker

Dept. Civil & Environmental Engineering

Tufts University

March 28, 2014

Presentation Tutorial

This presentation was made with the reveal.js HTML/JavaScript presentation framework.

If you are unfamiliar with this format, click here to walk through the offical tutorial.

In order to navigate, note the following:

Unlike Powerpoint, slides are arranged in two-dimensions
The 'compass' in the lower-right corner indicates which directions you can move (in this case, left/right/down)
Use the arrow keys to change slides
If you get lost, press the Esc key to see all the slides
Press the down arrow key to see the slide below (do this now)

Good!

Each section of the presentation is a column of slides

So always move down before moving right

When you get to the bottom, then go right to go to the next section

Now press the right arrow to begin the presentation

Overview

Problems
Methodology
Project 1: Water Quality Model
Project 2: Watershed Model
Project 3: Living Model
Discussion & Conclusions

The Problems

Model Understanding
Model Accessibility
Model Life Cycle

Model Understanding

Kuljis (1994). User interfaces and discrete event simulation models

The model understanding problem is sometimes called the black box problem
You give the model some input, it performs some computations, and generates some output.
But how do we know how the model converts the input data into outputs?
Most people would say: well, you have to know the equations
And this is often called the model theory
But are the equations and the theory really the same thing?
I would argue that the equations are really a representation of theory.
Equations are based on mathematical notation and abstract symbols to represent quantities, processes and relationships
But those abstract symbols really have an underlying meaning
For example, consider a model that represents the decay of a pollutant as a first order process
Most of us could write down a differential equation representing that process, but what that equaiton means is that the rate of change is linearly proportional to the concentration
But unless you've studied this equation before, it may not be entirely obvious what the behavior is
So it's sometimes hard to deduce based on the equations alone, what the underlying theory is
And so there's another representation of theory called visualizations
And when you look in textbooks or journal articles, you'll find a variety of visualizations designed to help understanding the shape of an equation, or the components of the model
And you can really think of these visualizations as an alternative representation of model theory

Model Accessibility

Chapra (2003). Engineering Water Quality Models and TMDLs

The second problem is model accessibility
There are two aspects to this problem
There is the physical accessibility, that is whether someone can obtain and execute the model software
There is also technical accessibility, that is how easy it is to use the software
When you read in the literature about the use of models for management and decision making, you find a lot of discussion about the need to make models more accessible, we need to put them in the hands of stakeholders and decision makers so that they can use the model to explore the system behavior and decision space.
So this is really where the web can play a major role. By making models accessible over the web, avoiding the need to install or purchase special software, eliminating compatibility problems, and allowing anyone with an Internet connection and standard web browser to access and interact with the model.

Model Accessibility

Chapra (2003). Engineering Water Quality Models and TMDLs

Model Life Cycle

USEPA (2008). Handbook for Developing Watershed Plans to Restore and Protect Our Waters

The model life cycle is often described as a cyclical process.
A model is created to support some management plan or make some decision, a lot of financial resources and human labor goes into configuring the model. It is applied to compare alternative scenarios or decisions, and then the project is complete.
But after that initial project is complete, the model goes into a state of hibernation. It is basically left idly sitting on someone's hard drive, not doing anything.
A few years go by, and another project comes along and the model is resurrected to evaluate the effectiveness of the previous plan or decision or to support some new project. And it goes through a process of model updates. New input and observation data are compiled to extend the simulation period, and the model is run over more recent years.
So this is clearly a cyclical process, but it is also somewhat sporadic or fragmented. We create a model, apply it, let it sit, then update it and apply again, and so on.
And so all this time that the model is left sitting idle in its state of hibernation is a waste of the resources and knowledge that went into creating the model in the first place.
Furthermore, the process of updating a model is fairly time consuming and expensive.
Today, much of this process can be automated now that most datasets are available and frequently updated right on the web.
So all this data is sitting out there, ready to go. It's a matter of retreiving it, processing it, and feeding into the model to update the input datasets and extend the simulation.
If we can do this, then the model life cycle becomes more stream-lined or continuous.
So instead of this traditional cycle of apply the model, let it sit, update the model, apply it, let it sit, the model can be continuously updated as new data become available, and then it can also be continuously applied.
So this would be a more cost-effective use of the model and increase the rate of return on the initial investment in the model's creation.

Methodology

Client-side web applications built using web standards to create interactive and visual model interfaces

Web Communications

Web-based Simulation Architectures

Byrne et al. (2010) A review of Web-based simulation and supporting tools

Project 1

Web-based Interactive River Model (WIRM)

Goals

Demonstrate client-side approach to simulation and visualization using web standards (HTML, JavaScript)
Store model parameters on server to promote model sharing and collaboration

Walker, J. D., and Chapra, S. C. (2014). A client-side web application for interactive environmental simulation modeling. Environmental Modelling & Software, 55, 49-60

DEMO

wirm.walkerjeff.com

Conclusions

Client-side approach not only possible, but is surprisingly fast
Strong potential for interactive interfaces
Raises new questions:
- Are interactive interfaces useful? How?
- What else can we do? Add data? Calibration?

Project 2

Web-based Interactive Watershed Model (WIWM)

Goals

Use interactive visualizations to understand model theory
Incorporate input and observation data for dynamic simulations
Interactive interfaces for model calibration

Cognitive Perspective

Can interactive visualizations help us understand how models work? If so, how?

Mental model formation
Amplify cognition

Mental Models

"Mental models are the mechanisms whereby humans are able to generate descriptions of system purpose and form, explanations of system functioning and observed system states, and predictions of future system states."
- Rouse & Morris (1986)

Mental Models

Liu and Stasko (2010) Mental models, visual reasoning and interaction in information visualization: a top-down perspective

Amplify Cognition

Card et al. (1999). Information Visualization: Using Vision to Think

abcd Water Balance Model

Developed by Thomas (1981)

Mass Balance

$$\Delta Storage = \sum{Inflow_t} - \sum{Outflow_t} $$

$$Storage_t - Storage_{t-1} = \sum{Inflow_t} - \sum{Outflow_t} $$

$$Storage_{t-1} + \sum{Inflow_t} = Storage_t + \sum{Outflow_t} $$

Mass Balance Diagram

$$ Storage_{t-1} + \sum{Inflow_t} = Storage_t + \sum{Outflow_t} $$

Soil Moisture Diagram

$$ Soil_{t-1} + Precip_t = Soil_t + ET_t + Runoff_t + Recharge_t $$

Soil Moisture Diagram

$$ Soil_{t-1} + Precip_t = Soil_t + ET_t + Runoff_t + Recharge_t $$

DEMO

abcd.walkerjeff.com

Conclusions

Interactive visualizations are a powerful way of understanding model theory and behavior
Client-side applications can maintain application state and perform data input/output using text files
Next question:
- Can we couple a model to input data sources?

Project 3

Living Model

Goals

Couple client-side application to server-side data management system
Interface to inspect and manage the data
Interface to explore and refine the model

Architecture

DEMO

living.walkerjeff.com

Conclusions

Living model provides an interactive interface for updating and evaluating model simulations
Important to provide interfaces for supervising the process
Potential to fundamentally change how we use models
Future Research
- Data processing algorithms (cleaning, filling, ...)
- Additional data management interfaces

Discussion

Limitations

Not all models will run in the browser
Performance limitations
Need more research on the "science of interaction"

Future Research

Try different models
Geospatial visualizations
Data exploration and analysis
Decision support systems

Conclusions

Standard web languages provide the performance and features necessary to support client-side simulation modeling
Client-side applications support highly interactive modeling interfaces
Visual interaction improves understanding of model theory and behavior
Use web services to couple models to data sources and improve the model life cycle
Lots more to explore...

Acknowledgements

Advisor
Dr. Steven Chapra (Tufts)
Committee
Dr. Alva Couch (Tufts)
Dr. Richard Vogel (Tufts)
Dr. Peter Shanahan (MIT)
Friends & Family:
Dr. Elizabeth Halliday
Dr. William Walker & Mrs. Martha Walker

Thanks!

phd.walkerjeff.com