Connecting Field Measurements with Satellite Water Monitoring
Suspended sediment presents both an environmental and water-resource management challenge. Sediment accumulation can reduce reservoir capacity and interfere with infrastructure, while elevated SSC increases turbidity, limits light penetration, and can affect aquatic habitats. Monitoring its distribution therefore requires information about both concentration and spatial variability.
Traditional water sampling provides quantitative measurements, but only at discrete locations and times. Satellite sensors address the spatial limitation by observing entire water bodies repeatedly. The challenge is establishing a reliable relationship between satellite-measured reflectance and the physical concentration of sediment in the water.
The Mula Dam study approached that problem through three interconnected datasets: water samples provided gravimetrically measured SSC; field spectroscopy characterized the detailed spectral response associated with those concentrations; and Sentinel-2 imagery extended those spectral relationships across the reservoir and through historical observations. This created a measurement chain from physical samples to hyperspectral reflectance and ultimately to satellite-based SSC mapping. [1]
At a Glance: From Field to Satellite
Water sampling establishes actual SSC → field spectroscopy characterizes the spectral response → relevant wavelengths and spectral combinations are identified → field spectra are related to Sentinel-2 bands → calibrated satellite imagery is used to estimate and map SSC.
Measuring the Spectral Response of Suspended Sediment
Researchers conducted five sampling campaigns between October 2021 and February 2022, collecting 121 surface-water samples from the Mula Dam Reservoir. Following the exclusion of 16 samples because of manual errors, 105 observations were used for model development and validation. Measured SSC ranged from 15.62 to 137.65 mg/L, with an average concentration of 61.43 mg/L. Higher concentrations generally occurred near river inflow areas, where incoming water and turbulence contribute additional sediment. [1]
Key Takeaways
Key Takeaways
- Ground truth spectroscopy connects physical measurements with remotely sensed observations, helping researchers establish what satellite spectral signals represent under field conditions.
- In the Mula Dam study, researchers used an SVC HR-1024i spectroradiometer to characterize suspended sediment reflectance before relating those measurements to Sentinel-2 imagery.
- Green, Red, and Red Edge 1 showed the strongest individual relationships with suspended sediment concentration, while a (Green × Red Edge 1) / Red combination produced the strongest field relationship among those tested.
- The resulting spectral integration model enabled researchers to move from discrete field measurements to reservoir-scale SSC mapping using Sentinel-2 imagery, including analysis of archived satellite observations.
The research demonstrates a broader remote sensing principle: field spectroscopy and satellite observations are complementary measurement tools, not competing alternatives.
At each sampling site, surface reflectance was measured using an SVC HR-1024i spectroradiometer covering 350–2500 nm. The instrument was positioned approximately 0.5 m above the water surface using a 25° field of view. A Spectralon reference panel was used to characterize incoming solar radiation and convert the measurements to reflectance.
The researchers also took several steps to reduce unwanted spectral effects. Measurements were collected approximately 20 meters from shore to minimize shoreline vegetation, bottom effects, and mixed pixels. Observations were made between 11 a.m. and 3 p.m. to reduce variation associated with solar angle. Three spectral measurements were collected at each location and processed before analysis. [1]
These procedures allowed the researchers to investigate how changes in SSC affected the water’s spectral response before attempting to reproduce those relationships using satellite data.
Green, Red, and Red Edge Bands Reveal Strong SSC Relationships
The field spectra showed a clear relationship between increasing suspended sediment and increasing surface reflectance. The response was particularly apparent across portions of the visible and red-edge spectrum.
Among spectral regions corresponding to Sentinel-2 bands, the strongest relationship occurred in the green region from 543–578 nm, with an R² of 0.76. Red Edge 1 from 698–713 nm followed with an R² of 0.71, while the red region from 650–680 nm produced an R² of 0.68. Blue showed a weaker relationship, and NIR and the higher red-edge bands were less strongly correlated with SSC. [1]
The value of the hyperspectral measurements becomes particularly apparent at this stage. Instead of beginning with the relatively broad, predefined bands available from the satellite sensor, researchers could first examine the spectral behavior measured at the water surface and determine which regions were most responsive to changing sediment concentrations.
They then evaluated several spectral indices, ratios, and multiband combinations. One of the strongest was the average of the Red, Green, and Red Edge 1 bands, which produced an R² of 0.82 against observed SSC. However, the highest-performing field relationship was (Green × Red Edge 1) / Red. Using a quadratic model, this combination achieved an R² of 0.86 with measured SSC. [1]
That relationship became the basis for connecting the detailed field observations with Sentinel-2.
From Hyperspectral Measurements to Sentinel-2 Bands
Sentinel-2 offers an important advantage for environmental monitoring: repeatable observations over large geographic areas. Its Multispectral Instrument includes visible, NIR, SWIR, and several red-edge bands, providing a useful spectral basis for monitoring inland waters.
However, satellite observations and ground measurements do not inherently produce interchangeable reflectance values. Atmospheric effects, spatial resolution, observation geometry, mixed pixels, and differences between sensor characteristics all have to be considered.
The researchers therefore used Sentinel-2 Level-2A bottom-of-atmosphere reflectance imagery corresponding with field sampling dates. Cloud and shadow masking, water masking, shoreline exclusion, removal of noisy observations, and band resampling were used to isolate homogeneous water pixels for comparison with the field measurements. [1]
Rather than directly correlating Sentinel-2 reflectance with SSC alone, the researchers established an intermediate relationship between Sentinel-2-derived spectral combinations and those measured by the field spectroradiometer.
This distinction is important. Ground truth spectroscopy is not functioning simply as a final check on satellite estimates. In this methodology, it provides the intermediate spectral information used to construct the satellite model.
Building a Spectral Integration Model
The researchers describe their methodology as a spectral integration framework based on a transitive relationship. In practical terms, it involved two regression stages: observed SSC → spectroradiometer-derived spectral index, followed by Sentinel-2 spectral index → spectroradiometer-equivalent spectral index.
Combining those relationships enabled SSC to be estimated from Sentinel-2 imagery while retaining the empirical relationship established from ground-based spectral measurements.
Three promising spectral functions were carried through the final model evaluation: Revised NDSSI, the averaged Red + Green + Red Edge 1 combination, and the Green × Red Edge 1 / Red ratio.
The final ratio again performed best overall. The integrated (Green × Red Edge 1) / Red model achieved an R² of 0.809, RMSE of 8.58 mg/L, MAPE of 19.41%, mean residual of 2.53 mg/L, and residual standard deviation of 13.78 mg/L. [1]
The authors did more than evaluate model fit against the original calibration data. They reserved 20 percent of the usable observations for validation and assessed performance using regression analysis, Student’s t-tests, RMSE, MAPE, Nash-Sutcliffe Efficiency, AIC/BIC model-selection criteria, residual analysis, and five-fold cross-validation. Five-fold cross-validation again favored the Green × Red Edge 1 / Red model, which achieved the highest average R² and lowest average RMSE and MAPE among the three tested integration functions. [1]
Extending Field Measurements Across Space and Time
Once the field-to-satellite relationship had been established, the researchers could move beyond the individual sampling points.
Using the selected spectral integration model, they generated spatial SSC maps of the Mula Dam Reservoir from Sentinel-2 imagery. These maps showed higher sediment concentrations around river inflows and dendritic areas of the reservoir, consistent with greater sediment inputs and turbulence.
More significantly, the researchers applied the model to archived Sentinel-2 observations predating the field campaign, including imagery from 2019, 2020, and 2021.
The resulting maps revealed temporal patterns in reservoir sediment conditions. SSC was generally higher during October following the monsoon, when runoff increased sediment delivery, and lower during February as sediment transport declined. [1]
This illustrates one of the larger benefits of connecting ground spectroscopy with remote sensing. A field campaign necessarily captures a limited number of locations over a limited period. Once a defensible relationship with a satellite sensor has been developed, remotely sensed observations can potentially extend that information across a much larger area and provide access to an existing archive of observations.
Why Ground Truth Still Matters in Satellite Remote Sensing
Satellite instruments measure electromagnetic radiation. They do not directly measure suspended sediment concentration.
Transforming remotely sensed reflectance into an environmental parameter requires an empirical or analytical relationship between the spectral signal and conditions in the target environment. Ground measurements provide the evidence required to establish and test that relationship.
Hyperspectral field spectroscopy adds another layer of information by allowing researchers to examine spectral behavior at much finer wavelength intervals than a multispectral satellite sensor provides. Researchers can investigate where meaningful spectral responses occur, evaluate band combinations and indices, and then determine how those relationships correspond with the bands available from a particular satellite platform.
The Mula Dam study illustrates that complementary relationship particularly well. The HR-1024i provided detailed ground-level spectral information, while Sentinel-2 supplied the spatial and temporal scale required for reservoir-wide monitoring. The value came from integrating the two measurement approaches rather than treating either as a replacement for the other. [1]
Measurement Framework at a Glance
Physical Measurement
Water samples establish measured suspended sediment concentration.
Hyperspectral Ground Truth
Field spectroscopy characterizes how changing SSC affects spectral reflectance and helps identify responsive wavelength regions.
Satellite Calibration
Field-derived spectral relationships are connected with corresponding Sentinel-2 bands and spectral combinations.
Spatial Monitoring
Once calibrated and validated, satellite imagery can extend the field observations across a much larger geographic area.
Temporal Analysis
Archived satellite imagery can potentially extend a field-calibrated model backward in time, providing additional observations for studying environmental change.
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.
Understanding the Limits of Empirical Spectral Models
The results are promising, but the authors appropriately identify limitations.
The model was developed for a single reservoir during a relatively short sampling period, and most observations represented low-to-moderate sediment conditions. At SSC above approximately 100 mg/L, reflectance in the green and red bands began to saturate, reducing model accuracy.
Other optically active water constituents, including chlorophyll-a and colored dissolved organic matter, were also not independently separated from the SSC signal. The authors therefore recommend multi-site and multi-season validation before assuming that the relationships developed at Mula Dam can be transferred directly to other water bodies. [1]
That limitation reinforces an important principle of ground truth spectroscopy: spectral relationships need to be characterized within the conditions in which they will be applied.
The specific equation developed at Mula Dam is therefore less broadly significant than the measurement framework used to create it.
From Point Measurements to Scalable Environmental Monitoring
The Mula Dam research demonstrates a practical progression from physical sampling to spectral characterization and ultimately remote environmental monitoring.
Ground-based hyperspectral measurements identified the wavelengths and spectral combinations most responsive to suspended sediment. Those measurements were then related to Sentinel-2 observations through a calibrated integration model. Once validated, the satellite data could be used to examine sediment distribution across the reservoir and investigate conditions captured in historical imagery.
The result is a useful example of how ground truth spectroscopy can connect precise field measurements with the geographic and temporal reach of satellite remote sensing.
For researchers working in water quality, hydrology, environmental monitoring, and remote sensing, that connection can turn isolated field measurements into something much larger: a calibrated foundation for observing environmental change across space and time. [1]
Source and Citation Notes
[1] Joshi, J. K., Atre, A. A., Nandgude, S. B., Shinde, M. G., Durgude, A. G., Gorantiwar, S. D., & Patil, M. R. (2025). Empirical modelling of suspended sediments using spectral data from spectroradiometer and Sentinel-2 in Mula Dam Reservoir, Maharashtra, India. Scientific Reports, 15, 34205. DOI: 10.1038/s41598-025-15719-w.
Connect Field Measurements to the Bigger PictureThis is the heading
Frequently Asked Questions
What is ground truth spectroscopy?
Ground truth spectroscopy is the collection of field-based spectral measurements that can be compared with observations from airborne or satellite remote sensing systems. These measurements help researchers characterize spectral signatures under known field conditions and develop or validate relationships between remotely sensed reflectance and physical or environmental parameters.
Why is ground truth important for satellite remote sensing?
Satellite sensors measure reflected or emitted electromagnetic radiation rather than environmental parameters such as suspended sediment concentration directly. Ground measurements provide the empirical data needed to determine what those spectral observations represent and to evaluate the accuracy of remote sensing models.
How can field spectroscopy support water quality monitoring?
Field spectroscopy can characterize changes in water reflectance associated with suspended sediments and other optically active constituents. Researchers can use these measurements to identify responsive wavelength regions, develop spectral indices, calibrate remote sensing models, and validate satellite or airborne observations.
How was the SVC HR-1024i used in the Mula Dam study?
Researchers used the SVC HR-1024i to collect surface-reflectance measurements from sampling locations across the reservoir. The 350–2500 nm measurements were compared with gravimetrically determined suspended sediment concentrations and subsequently related to corresponding Sentinel-2 observations.
Which spectral regions were most responsive to suspended sediment concentration?
The strongest individual relationship was observed in the Green region at 543–578 nm, followed by Red Edge 1 at 698–713 nm and Red at 650–680 nm. The study’s best-performing spectral combination was (Green × Red Edge 1) / Red.
Can field spectroscopy be used to calibrate satellite data?
Yes. The Mula Dam research provides an example of this approach. The researchers developed relationships between measured SSC, field spectroradiometer observations, and corresponding Sentinel-2 spectral data. The resulting integration model was then used to estimate SSC from satellite imagery.
Can a ground-truth model be applied to historical satellite imagery?
Potentially, when suitable archived imagery exists and the model is valid for the conditions being analyzed. In this study, researchers applied their calibrated spectral integration model to earlier Sentinel-2 imagery to investigate spatial and seasonal SSC patterns in the reservoir.
Can the Mula Dam SSC model be used for other reservoirs?
Not without additional validation and appropriate local calibration. The authors specifically identify the single study location, limited sampling period, spectral saturation at higher SSC levels, and possible influence of other optically active water constituents as limitations. The broader value of the research is therefore the field-to-satellite methodology, rather than a universal SSC equation.