Study Results: Comparing Pond Edge vs. Interior Water Quality Sampling
- Haven King-Nobles
- Jul 31
- 5 min read
Updated: Aug 1
Summary and Data
Accurate water quality measurements are critical to our organization’s success, both for our farm program—where water quality sampling directly informs our understanding of fish welfare—and for our ongoing development of remote sensing technology.
Historically, our practice has been to take water quality readings from the pond dyke (pond edge), due to logistical simplicity. However, given concerns that water quality may vary throughout the pond, we recently conducted a short study to assess how big an issue this is. This study involved measuring water quality parameters—dissolved oxygen (DO), chlorophyll-a (Chl-a), ammonia, and turbidity—at six unique ponds (some of which were visited twice, some visited three times). This provided 45 “time-matched” samples, allowing us to compare dyke measurements with measurements from within the ponds collected at the same time.
The study revealed notable differences in water quality readings between dyke and interior sampling points. Nevertheless, these variations would not have significantly altered our decisions around issuing corrective actions to farmers. Therefore, we currently do not anticipate changing the sampling approach in our farm program. However, these insights will likely influence the methods we use when developing our satellite-based water quality remote sensing models.
For those interested in exploring further, you can access:
The full study data (with GPS coordinates removed to preserve farm anonymity).
Our internal slideshow created displaying the results.

The Study
We conducted our investigation across 6 different fish farm ponds, with a total of 15 sampling visits. Each visit involved taking simultaneous water quality samples at several key locations within the ponds:
Sample Point 1 (SP1): Located at the pond dyke (edge), this point served as our baseline sampling location.
Sample Point 2 (SP2): An interior location within the pond, sampled simultaneously with SP1.
Sample Point 3 (SP3): Roughly at the pond’s center, again sampled concurrently with a separate SP1 measurement.
Sample Point 4 (SP4): Another interior location, sampled at the same time as yet another SP1 measurement.
Sampling visits were intentionally split between morning and evening times, as certain water quality parameters—particularly dissolved oxygen—are known to vary significantly with time of day. We measured four water quality parameters: dissolved oxygen (DO), chlorophyll-a (Chl-a), ammonia, and turbidity. We had initially planned to measure pH as well, but unfortunately, technical issues with our measurement equipment (YSI ProDSS meter) prevented this from being included in the study.
Findings
Our study revealed notable differences between measurements taken at the pond edge (dyke) and those from interior sampling points:
Dissolved Oxygen (DO): In 58% of comparisons, dyke measurements differed from interior points by more than 20%.
Ammonia: 38% of comparisons showed differences exceeding 20%.
Chlorophyll-a (Chl-a): The greatest variation was observed here, with 71% of comparisons differing by more than 20%. Note, however, that Chl-a is considered a less direct measure of fish welfare, and is thus generally not measured by our farm program.
The significance of these differences highlights the complexity involved in accurately assessing fish welfare based solely on water quality data from a single location within a pond. We believe this study provides important insights for two key areas of our work: our ongoing farm program and our efforts to develop accurate remote sensing models for water quality monitoring.
You can view the complete dataset and our detailed analysis here.

Implications – Farm Program
Our farm program, the Alliance for Responsible Aquaculture (ARA), relies heavily on water quality sampling. When we detect out-of-range (OOR) values, we recommend specific corrective actions to the farmers. Historically, due to logistical simplicity, we’ve primarily taken samples from the pond dyke. Taking multiple interior samples would significantly increase the time spent per farm, especially when a boat is necessary.
For our program, the key question isn’t whether additional sampling points provide greater accuracy—they almost certainly do. Instead, the question is whether this additional accuracy meaningfully changes our programmatic decisions. For example, if our morning dissolved oxygen (DO) threshold for issuing corrective actions is 3 mg/L, readings of both 1 mg/L and 2 mg/L would trigger the same programmatic response, even though one value is double the other.
Our analysis suggests that taking additional interior measurements would only slightly impact the corrective actions we issue:
If we had relied exclusively on dyke measurements for DO, we would have identified 69%* of samples as OOR, compared to 56% identified as OOR from interior points. This suggests dyke measurements slightly underestimate DO levels, representing a conservative approach to DO management.
For ammonia, relying only on dyke measurements would have identified 74% of samples as OOR, compared to 79% identified by interior points.
In our view, the differences observed do not justify the additional time and resources required to take additional samples per farm. Therefore, we will continue our current approach of taking single samples from the pond dyke for our farm program.
*Note that both DO and ammonia levels at the study farms seemed unusually bad. We did not give corrective actions during this study, but have since decided that this was the wrong approach, and plan to give corrective actions during such studies in the future (unless this would adversely affect the study in question).
Implications – Water Quality Remote Sensing
Given the considerable time required even for single-point sampling, we’re actively working to reduce our dependence on manual on-the-ground sampling. To achieve this, we’re developing satellite-based remote sensing models to estimate water quality remotely.
These models will require significantly more—and potentially more precise—data than we’ve previously gathered. Consequently, any future data collection specifically aimed at training these remote sensing models will likely involve multiple interior sampling points to improve the accuracy of the training dataset.
We have already implemented interior sampling at multiple points for collecting “ground-truthed” data used to validate models as part of our ongoing Innovation Challenges (initially launched in December 2024, and relaunched in June 2025). However, our latest findings here indicate that we may want to adopt interior sampling more broadly for any training data as well.
Acknowledgements
We would like to extend our gratitude to our intern Annika for supporting this study, as she spent many hours taking dyke measurements and perhaps many more measuring ammonia back at our lab.
The girls are well groomed and resemble models, so every second of the date was stimulating. The Hari Nagar Escorts Service provide incall and outcall services, obviously the choice is up to you. I feel like spending time with them is like being in a movie. For me it is the best option for passion in Delhi.