Temporal Synchronization Determines the Accuracy of Sentinel-2 Turbidity Retrieval: Evidence from Community-Based Monitoring of the Buriganga River, Dhaka
Abstract
Community-based environmental organizations require continuous water quality data but are constrained by the cost and labor of physical sampling. Satellite remote sensing offers a free alternative, yet validation studies of inland waters frequently report weak agreement between satellite indices and ground measurements, and the cause of that disagreement is rarely isolated. This study analyzed 28 Sentinel-2 observations across 12 dates at three sites on the Buriganga River in Dhaka, Bangladesh, between January and April 2026, alongside eight calibrated field measurements of turbidity, pH, and dissolved oxygen. Three findings emerged. First, the reach adjacent to the Hazaribagh tannery district was more turbid than the Katashur Outfall on every one of 11 valid observation dates (Wilcoxon p = .001), establishing a persistent spatial gradient that discrete sampling campaigns had not resolved. Second, turbidity rose by 88% river-wide between January and February (p = .006), and the cleanest site deteriorated fastest, compressing the gradient between the tannery reach and the outfall from 2.0 to 1.5. Third, repeat same-day observations established a measurement noise floor 2.6 times smaller than between-date variation, indicating that 86% of observed variance reflects genuine day-to-day change. That last result predicts, and the data confirm, that pairing protocol governs retrieval accuracy: comparing generalized field averages against satellite observations from the same approximate period yielded no significant relationship (R² = .04, p = .46), whereas matching each field measurement to a cloud-free overpass on the same calendar date yielded strong agreement (R² = .94, 95% CI [.71, .99], n = 8, p < .001), with a cross-validated retrieval error of 5.4 NTU. Sensor capability was identical under both protocols. Weak validation outcomes in dynamic inland waters may therefore reflect protocol failure rather than sensor limitation.
Related
- Toward fine-scale microwave monitoring of dryland vegetation: an alternative high spatial resolution vegetation optical depth retrieval algorithm from Sentinel-1 over the Sahel region
- Training-Free Monitoring of Kenya with Pre-computed Satellite Embeddings: Lake Expansion, Construction Dating and Crop Retrieval with TESSERA
- Stratospheric turbidity in the period 1600–1800 CE derived from lunar eclipse brightness and colour, and inferences for global volcanism
- Acid-induced demulsification versus alum and polyaluminium chloride coagulation for oil and turbidity removal from concentrated soybean oil-in-water emulsions.
- Intense rainstorms as drivers of lacustrine turbidity currents: implications for paleoflood records
BibTeX
@misc{khan2026temporal,
title = {{Temporal Synchronization Determines the Accuracy of Sentinel-2 Turbidity Retrieval: Evidence from Community-Based Monitoring of the Buriganga River, Dhaka}},
author = {Adneen Khan},
howpublished = {EarthArXiv},
year = {2026},
month = oct,
doi = {10.31223/x5pv4r},
url = {https://eartharxiv.org/repository/view/15393/},
note = {CC BY-SA 4.0},
}
RIS
TY - UNPB
TI - Temporal Synchronization Determines the Accuracy of Sentinel-2 Turbidity Retrieval: Evidence from Community-Based Monitoring of the Buriganga River, Dhaka
AU - Adneen Khan
PY - 2026
DA - 2026/10/07
PB - EarthArXiv
DO - 10.31223/x5pv4r
UR - https://eartharxiv.org/repository/view/15393/
AB - Community-based environmental organizations require continuous water quality data but are constrained by the cost and labor of physical sampling. Satellite remote sensing offers a free alternative, yet validation studies of inland waters frequently report weak agreement between satellite indices and ground measurements, and the cause of that disagreement is rarely isolated. This study analyzed 28 Sentinel-2 observations across 12 dates at three sites on the Buriganga River in Dhaka, Bangladesh, between January and April 2026, alongside eight calibrated field measurements of turbidity, pH, and dissolved oxygen. Three findings emerged. First, the reach adjacent to the Hazaribagh tannery district was more turbid than the Katashur Outfall on every one of 11 valid observation dates (Wilcoxon p = .001), establishing a persistent spatial gradient that discrete sampling campaigns had not resolved. Second, turbidity rose by 88% river-wide between January and February (p = .006), and the cleanest site deteriorated fastest, compressing the gradient between the tannery reach and the outfall from 2.0 to 1.5. Third, repeat same-day observations established a measurement noise floor 2.6 times smaller than between-date variation, indicating that 86% of observed variance reflects genuine day-to-day change. That last result predicts, and the data confirm, that pairing protocol governs retrieval accuracy: comparing generalized field averages against satellite observations from the same approximate period yielded no significant relationship (R² = .04, p = .46), whereas matching each field measurement to a cloud-free overpass on the same calendar date yielded strong agreement (R² = .94, 95% CI [.71, .99], n = 8, p < .001), with a cross-validated retrieval error of 5.4 NTU. Sensor capability was identical under both protocols. Weak validation outcomes in dynamic inland waters may therefore reflect protocol failure rather than sensor limitation.
ER -