Measuring success with water quality requires balancing monitoring and modeling

Post by Eric Booth

Understanding how and why water quality changes through time is an important activity of scientists and natural resource managers. As pressures on lakes and streams increase from changes in land use and climate, the public and governing bodies are pushing for management practices that will improve water quality.

To know whether those practices are working, it is necessary to evaluate their outcomes. Gathering such information comes in two flavors: monitoring and modeling.


Monitoring involves tracking changes in a system over time, often in response to management efforts.

Monitoring, or observation, is at the heart of the scientific method and is very often what the public sees as the primary activity of scientists. It involves measuring an aspect of a system, such as phosphorus levels in lakes, to see if and how much it changes in response to external influences, such as management practices.

Modeling is the act of constructing a representation of reality – typically using mathematical formulas and computers – so we can explore how the system works under various conditions or scenarios that we can’t observe in real time. Modeling can help us identify possible future conditions of the system as a result of different management decisions – for example, future lake phosphorus levels under different ways agricultural land use might change.

So which one is best for managing water quality? (Spoiler alert: it’s both!)

“If you don’t like the weather…just wait 5 minutes.” – Mark Twain

Monitoring may seem like the early favorite. This tactic is easy to implement for point sources of water pollution, such as a wastewater treatment plant. Managers can upgrade a treatment process and measure the chemical changes in the water coming out of the plant. No problem!

However, for non-point sources, or pollution that doesn’t come out of a single pipe like agricultural runoff, this tactic is much more challenging, as we explored in a previous post.

First is the dilemma of where to measure water quality. Non-point pollution flows across a watershed’s landscape and through its stream networks. But we can’t monitor everywhere, especially since it can be laborious and expensive.

So typically we monitor at strategic and accessible locations. But this is a problem because water quality conditions can change from site to site, making generalizations difficult.


This edge-of-field monitoring station tracked water clarity changes before and after the installation of a grassed waterway, a water quality improvement practice. Credit: USGS

Second, it can be difficult to attribute water quality improvements to a particular action. Reducing non-point pollution across a landscape requires scattered clean-up actions – and that landscape is often large.

Third, even if you know there have been efforts to improve water quality in your watershed, other processes are occurring simultaneously that can create noise in the system. The largest of these is the weather, specifically rainfall.

For instance, the main mechanism for phosphorus, a major freshwater pollutant, to enter water bodies is through runoff driven by rainfall. We all know how variable rainfall can be, and runoff can be even more variable.

This is a major problem for assessing water quality management efforts because a dry year and wet year will give very different pictures of the state of water quality, even if the watershed landscape stays the same. Less phosphorus will enter water bodies in dry years because of less runoff.

Fourth, monitoring fails at predicting the future. Water quality professionals are interested in how changes in the watershed will impact future water quality. Short of a time machine, monitoring won’t be able to help us out with this task.

“All models are wrong…some are useful.” – George Box, statistician

This brings us to modeling. Modeling can help us make generalizations for large areas or predictions for the future. Scientific hypotheses can be tested and challenged, and new theories can emerge.

Watershed pollution models are simplified representations of the things that happen on a landscape that determine water quality – such as rainfall, erosion, and stream transport – to illustrate and, sometimes, predict trends or outcomes. The possibilities for what aspects of a watershed a modeler can manipulate are endless.

For example, models can simulate current conditions to identify areas that are likely contributing most of the pollutants, so that interventions can be better targeted. They can also simulate various scenarios – such as changes in climate, land use, and management actions – to see what impact they will have on downstream water quality in the future.


Model simulation of the amount of phosphorus moving through the Yahara watershed day-by-day from May to August 2008.

But no model can be 100 percent correct. There are always uncertainties, gaps in our understanding and approximations of reality.

So if models are wrong, how can they be useful?

First, we can confirm the accuracy of models based on how well they reproduce historical observations. This is what modelers call “validation.” The match will not be perfect, but it should be reasonably close.

Second, we can develop mathematical formulas to represent biophysical principals – such as the conservation of mass and energy – that are universal even under conditions that a watershed hasn’t seen in the past. These “process-based” models are typically more accurate at predicting future changes than simpler models that are based on statistical relationships between historical observations.

In other words, we will always know that the laws of physics will be obeyed in the future, but history may not always be a good indicator of the past.

The computer models that perform water quality simulations can vary dramatically in complexity. At one end of the spectrum, a simple empirical model based on historical observations could simulate water quality based solely on how much rainfall could occur.

A more complex, process-based model, like those our project has created, could simulate different land uses and ecosystems as they interact with climate, water, energy, carbon, and nutrients on a daily time scale. Using phosphorus again as an example, a process-based water quality model could incorporate all of the factors that determine how much of the nutrient moves downstream in runoff: the amount of rainfall, the type and moisture level of the soil, how much the land is sloped, how much phosphorus is in the soil and applied as fertilizer and manure, and what vegetation grows there.

While the advantage of such a model is higher accuracy, it requires a lot more time, resources and expertise to execute.

The role of models in water quality management

Despite its imperfections, water quality modeling has a long history in water resources management and policy. Today, the Soil and Water Assessment Tool (SWAT) model is often used to help watershed managers comply with the Clean Water Act, which sets limits on the amount of pollution that is allowed in our waters.


This illustrates the data inputs and outputs for the SWAT model. Credit: USGS

Researchers and managers across the world have used SWAT to reproduce historical observations and explore the relative impact of land and climate changes. Most commonly, SWAT helps us determine 1) the starting conditions for a watershed-wide effort to reduce pollutants such as phosphorus and 2) how much sources should reduce the amount of pollutants they contribute to the system in order to meet certain water quality criteria.

Once the model makes these determinations, managers use other, often simpler models to predict the impacts of implementing pollutant reduction practices. These simulations help them track progress as the practices unfold on the landscape.

However, caution needs to be taken in how models are used to predict future water quality outcomes.

A model is only as good as the data it’s fed. While models like SWAT and the one our project developed account for the important biophysical processes that affect water quality, their accuracy will suffer if model users feed climate and land-use data inputs that are unlikely to occur in the future.

For example, current efforts to improve water quality in Wisconsin have been focused on a compliance option called the Wisconsin Adaptive Management Option. This technique involves point and non-point sources – such as municipal sewerage facilities and farmers, respectively – working together to reduce pollutants over several decades.

But a lot can happen over that time period. A landscape could become more urbanized, agriculture could intensify with more dairy and manure production, and rainfall could increase – as has happened in the Yahara watershed and many other parts of Wisconsin.

The problem is often the only change from the starting conditions that the model assumes is the implementation of conservation practices, such as cover crops, in several locations around a watershed. These other external drivers of water quality change are not taken into account, which can lead to inaccurate estimates and, specifically, underestimates of water quality improvement.

Such inaccuracies have been the case in the Yahara watershed, for example. Models have implied water quality should improve due to extensive management efforts over the last several decades. However, there has been no such improvement, because of these unaccounted for drivers of water quality.

But how do we account for these drivers that fall outside current watershed management practices? Incorporating them into models would be especially challenging given the sheer uncertainty of how exactly climate, agriculture and land-use could change into the future.

While no one has a clear road map, our research team continues to be interested in working with managers to utilize techniques such as scenario analyses, which help address the uncertainty by investigating a range of possible outcomes.

Ultimately, the success of implementing watershed improvement programs will come down to using both modeling and monitoring. Collecting long-term monitoring data and, relatedly, assessing the public’s perception of water quality improvement put us in a good position that ensures accountability for the water quality management process.

Modeling helps increase the chance of success. Models that are regularly updated with land-use and climate information complement long-term monitoring and public engagement, bettering our ability to anticipate the cumulative impacts of the many drivers of water quality change.

Booth is an assistant research scientist for the Water Sustainability and Climate project at UW-Madison.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s