New Dataset: Continuous Water Quality Data

As part of our research into using satellite imagery to monitor water quality, our team deployed 16 continuous water quality monitors at fish farms across Eluru, Andhra Pradesh, India, over two months. We called this our “Data Campaign”.

This post provides context on why we collected this data, what it includes, and what we’re planning next. Access the full data here:

View Data

Why We Collected This Data

As we’ve shared in previous updates, we have been exploring whether satellite-based remote sensing could allow us to monitor water quality without needing to visit each farm in person. If successful, this approach could significantly increase the scalability and cost-effectiveness of water quality-based fish welfare programs like our farm program.

We have already tested this approach through a proof-of-concept study and two innovation challenges. While we found some initial promise, a core limitation remains the lack of local, high-frequency, ground-truthed water quality data during the exact time of satellite flyovers (~10:30 a.m. IST). Without this, it's difficult to train any remote sensing model.

To make one final, focused attempt to address this gap, we hired a Remote Sensing Lead—Dr. Solomon (Sol) White—and ran a two-month data collection campaign (launch post) between November 2025 and January 2026. The results of this campaign are now shared here.

What This Dataset Contains

To gather sufficient data in a short amount of time, we installed 16 continuous monitoring devices at selected farms. We also used handheld meters (ProDSS) at eight farms. Data collection started with the handheld meters on November 20, 2025. We then installed the monitors in batches through mid December, and continued data collection through January 29, 2026.

We collected the following parameters:

Parameter	Device
Dissolved Oxygen	Continuous monitor & ProDSS handheld meter
pH	Continuous monitor & ProDSS handheld meter
Temperature	Continuous monitor & ProDSS handheld meter
Ammonia	Hanna Spectrophotometer (combined with pH and temperature measurements to get NH3)
Chlorophyll-a	ProDSS handheld meter

Note that:

• The ProDSS data and ammonia values were collected only on satellite flyover days (every 3–5 days).

• The continuous monitors collected values every 15 minutes.

• Four ponds did not have continuous monitors and were monitored only through handheld devices.

• Each flyover day, four ponds with continuous monitors also received a ProDSS measurement to compare the readings. 

• Chlorophyll-a was a secondary focus, as many public datasets already exist. We collected it from 8 ponds each flyover day using the ProDSS.

Left: One of our continuous monitors being installed at a fish farm in Eluru, India. The devices were solar-powered to allow for continuous monitoring in rural areas with frequent power outages. Right: The monitor’s Dissolved Oxygen (left) and pH & temperature (right) sensors.

Limitations Of This Data

While we believe this dataset will be useful to researchers and organizations working on scalable welfare solutions, there are a few important limitations to keep in mind:

• Cleaning frequency: Monitors were cleaned approximately every five days. We prioritized cleaning the day before satellite flyovers, but algae buildup likely reduced sensor accuracy in the days between cleanings. See photo below.

  
  

• Device accuracy: We compared the monitors’ readings to those from our higher-accuracy ProDSS meters. Differences were within our accepted margins for this model, but we encourage others using our data to assess whether these fit their needs. You can also view our initial comparison and analysis results in our data upload.

  
  

• Device reliability: Some devices recorded sudden dips or spikes in values. While these monitors are useful for observing general trends, we advise caution when interpreting individual time points.

Algae residue on a sensor after five days. While we kept the monitors clean for the satellite flyover days, data in between may be affected by this algae residue.

Access the Data

You can access the full dataset by clicking below:

View Data

All pond identifiers have been anonymized, but otherwise, the dataset is complete. Note that the Pond IDs in this dataset are the same as those listed in our public ARA dataset, so that they might be cross-compared.

What’s Next

We have used this data to develop remote sensing models for water quality. The models seem to have middling levels of accuracy, although we are continuing to test and explore them.

We expect to share a full post about our results here in the coming month.