📄 earth_science

Mapping physicochemical properties of soils and reservoir sediments by multi- and hyperspectral satellite data in the Brazilian semiarid region to support the sediment reuse technique

This study demonstrates that combining VNIR-SWIR spectroscopy with multivariate statistical techniques using satellite data (Sentinel-2, PRISMA, and EnMAP) and laboratory spectra effectively maps key physicochemical properties of soils and reservoir sediments in Brazil's semiarid region, thereby supporting sediment reuse strategies for agricultural management.

Original authors: Brennda Braga, Arlena Brosinsky, Kathrin Ward, Pedro Medeiros, Felipe Nogueira, Adunias Teixeira, Saskia Foerster

Published 2026-06-29

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Brennda Braga, Arlena Brosinsky, Kathrin Ward, Pedro Medeiros, Felipe Nogueira, Adunias Teixeira, Saskia Foerster

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: Turning "Mud" into Gold

Imagine a region in Brazil that is very dry, like a giant sponge that sometimes soaks up rain and sometimes dries out completely. In this area, there are hundreds of small reservoirs (man-made lakes) that fill up during the rainy season. When the dry season hits, these lakes evaporate, leaving behind a thick layer of mud at the bottom.

Farmers in this region have a clever idea: What if we dig up this dried-up mud and use it to fertilize their crops?

This mud (sediment) is actually full of nutrients, organic matter, and clay, which can make soil healthier. However, there's a catch. Not all mud is the same. Some might be too salty (which kills plants), while others might be perfect. To figure out which mud is good and which is bad, scientists usually have to take samples, dry them, and run expensive, time-consuming lab tests. If you have 1,000 reservoirs, that's a lot of work!

This paper is about finding a faster, cheaper way to check the quality of this mud using satellite photos instead of a lab.

The "Super-Eye" Satellites

The researchers used three different types of satellites to take pictures of the dried-up reservoirs:

Sentinel-2: A standard camera that sees the world in a few broad colors (like a basic digital photo).
PRISMA & EnMAP: These are "super-cameras" (hyperspectral) that can see hundreds of tiny, specific shades of color that the human eye can't even detect. Think of them as having a vision so sharp they can see the difference between two shades of blue that look identical to us.

How They "Read" the Mud

Just like a doctor looks at a patient's skin to guess if they are healthy, the scientists looked at the colors reflected off the mud to guess its properties.

The Analogy: Imagine the mud is a musical instrument. When light hits it, it "sings" a specific song (a spectrum).
- If the mud is salty, it sings a certain note.
- If it has organic carbon (fertilizer), it sings a different note.
- If it has clay, it sings another note.

The researchers tried to teach computers to listen to these "songs" and translate them into numbers (like "this mud has 5% salt" or "this mud has 10% clay").

The Two Methods They Tried

To decode these songs, they used two different strategies:

The "Spot Check" (Spectral Indices): This is like using a simple rule of thumb. "If the mud looks this shade of red compared to that shade of blue, it's probably salty." It's quick and easy to apply to any camera.
The "Deep Dive" (PLSR): This is like a complex math model that listens to the entire song, looking for subtle patterns across hundreds of notes to make a prediction. It's powerful but can be tricky to tune.

What They Found (The Results)

1. The "Super-Cameras" vs. The "Basic Camera"

For Salt and Clay: The "Super-Cameras" (PRISMA and EnMAP) were the winners. Because they can see so many tiny colors, they were much better at spotting salt and clay. It's like trying to find a specific needle in a haystack; the super-cameras have a better flashlight.
For Organic Carbon (Fertilizer): Surprisingly, the "Basic Camera" (Sentinel-2) actually did a decent job here, sometimes even better than the complex models when looking at the mud specifically.

2. The "Separate vs. Mixed" Rule
The researchers found that mud from the bottom of a reservoir and regular farm soil are like apples and oranges. They look different and behave differently.

When they mixed the data from both together, the computer got confused.
When they built separate models for "Reservoir Mud" and "Farm Soil," the predictions got much more accurate. It's like having a specific recipe for apple pie and a different one for orange cake, rather than trying to make one "fruit pie" that tastes like neither.

3. The "Pre-Processing" Puzzle
Before feeding the data to the computer, they tried cleaning it up (removing noise, smoothing the lines).

For the simple "Spot Check" method, cleaning the data helped a lot.
For the complex "Deep Dive" math model, cleaning the data sometimes actually made it worse. It's like polishing a diamond too much and accidentally scratching it.

The Final Map

Using the best methods they found, the researchers created maps of a specific reservoir. These maps showed exactly where the mud was too salty to use and where it was rich in nutrients.

Why does this matter?
Instead of sending a team to dig up mud from every single corner of a reservoir and wait weeks for lab results, farmers and officials can now look at a satellite map. They can instantly see: "Hey, the mud in the north corner is perfect for our crops, but the south corner is too salty. Let's only dig up the north!"

The Bottom Line

This study proves that we can use satellite "super-vision" to quickly and cheaply check if reservoir mud is safe and useful for farming. It saves time, saves money, and helps farmers in dry regions get the nutrients they need without the heavy lifting of traditional lab testing. However, to make these maps even better in the future, they will need more samples and clearer photos of the dried-up reservoirs.

Technical Summary: Mapping Physicochemical Properties of Soils and Reservoir Sediments in the Brazilian Semiarid Region

Problem Statement
The reuse of sediments accumulated in surface water reservoirs offers a sustainable alternative to chemical fertilizers in subsistence farming, providing nutrients and organic matter while reducing the agricultural carbon footprint. However, the effective integration of sediment reuse into agricultural management requires comprehensive knowledge of physicochemical properties (specifically electrical conductivity, organic carbon, and clay content) across large spatial scales. Traditional laboratory analysis is time-consuming, costly, and often unfeasible for the hundreds of reservoirs in semiarid regions like Northeast Brazil, where sediment properties exhibit high spatial heterogeneity. While satellite remote sensing offers a scalable solution, there is a gap in applying these techniques specifically to reservoir sediments, which differ from agricultural soils in deposition processes and composition. Furthermore, existing studies often focus solely on agricultural soils, leaving the potential of multi- and hyperspectral data for sediment characterization in semiarid environments underexplored.

Methodology
The study was conducted in the Banabuiú River Basin in Northeast Brazil, an area characterized by intermittent rivers and a dense network of over 1,000 reservoirs. The research utilized a multi-source data approach:

Satellite Data: Imagery was acquired on December 4, 2022, from three sensors: Sentinel-2 (multispectral), PRISMA (hyperspectral), and EnMAP (hyperspectral).
Ground Truth: A field campaign in January 2023 collected 31 topsoil samples (20 agricultural soils, 11 reservoir sediments). These were supplemented by a historical database containing over 168 sediment samples and 200 soil samples collected between 2015 and 2022.
Laboratory Analysis: Samples were analyzed for Electrical Conductivity (EC), Soil Organic Carbon (SOC), and clay content. Laboratory reflectance spectra were also measured using an ASD FieldSpec Pro spectroradiometer (350–2500 nm).
Data Processing: Bare soil pixels were isolated from satellite imagery using spectral indices (e.g., NDVI, NDRBI, NBR) to exclude vegetation and water. Spectral pre-processing techniques included Savitzky–Golay smoothing of first-derivative spectra (FD-SG) and continuum removal (CR).
Modeling: The study employed two main modeling approaches:
1. Spectral Indices: Linear regression models using Difference (DI), Normalized Difference (NDI), and Ratio (RI) indices for EC; specific indices (SOC1-3) for SOC; and SWIRFI/CRAD for clay content.
2. Partial Least Squares Regression (PLSR): Multivariate models developed for soil-only, sediment-only, and combined datasets, utilizing leave-one-out cross-validation (LOOCV).

Key Contributions

Applicability to Sediments: The study evaluates the transferability of spectral indices and PLSR models, traditionally used for soils, to reservoir sediments in a semiarid context.
Sensor Comparison: It provides a direct performance comparison between multispectral (Sentinel-2) and hyperspectral (PRISMA and EnMAP) satellite data for predicting soil and sediment properties.
Methodological Recommendations: The research investigates the impact of dataset composition (separating soil vs. sediment) and spectral pre-processing on model accuracy.
Spatial Mapping: It generates spatial maps of EC, SOC, and clay content for a specific reservoir, demonstrating the feasibility of using remote sensing to guide sediment reuse decisions without exhaustive ground sampling.

Results

Electrical Conductivity (Salinity): Spectral indices generally outperformed PLSR models for satellite data. The Ratio Index (RI) combined with FD-SG pre-processing yielded the highest accuracy, particularly with EnMAP data ( $R^2 = 0.91$ ). PLSR models for satellite data showed poor performance ($RPD < 1.2$), whereas laboratory-scale PLSR models were reliable ($RPD = 1.41$).
Soil Organic Carbon (SOC): Prediction of SOC was challenging across all scales. Spectral indices performed poorly on satellite data ($RPD < 1.4$). Interestingly, a PLSR model using Sentinel-2 data on a sediment-only dataset achieved the best SOC performance ( $R^2 = 0.6$ , $RPD > 1.4$), suggesting that multispectral data may be sufficient for specific subsets.
Clay Content: The CRAD index performed best for clay estimation. Hyperspectral data (EnMAP and PRISMA) outperformed Sentinel-2. The best results were obtained with the CRAD index on raw EnMAP spectra for sediment-only datasets ( $R^2 = 0.54$ , $RPD = 1.47$).
Pre-processing Impact: While FD-SG and CR improved the performance of spectral indices for salinity, they often degraded the performance of PLSR models. Untransformed (raw) spectra frequently yielded better PLSR results for SOC and clay.
Dataset Separation: Models built separately for soil and sediment generally performed better than combined models, reflecting the distinct physicochemical and spectral characteristics of the two materials.

Significance and Claims
The authors claim that the combination of reflectance spectroscopy with spectral indices and multivariate techniques provides an effective, cost-effective approach for mapping soil and sediment properties in semiarid regions. The study demonstrates that while hyperspectral data (EnMAP, PRISMA) offers superior accuracy for salinity and clay mapping due to higher spectral resolution, multispectral data (Sentinel-2) can still be effective for specific properties like SOC when applied to targeted datasets.

The primary significance of this work lies in its support for the "sediment reuse technique." By enabling the identification of areas with suitable sediment properties (low salinity, high organic carbon, and adequate clay) and flagging unsuitable areas (high salinity), remote sensing can guide the excavation and application of sediments as soil conditioners. This reduces the need for costly, large-scale laboratory sampling in regions with dense reservoir networks, thereby facilitating the implementation of sustainable agricultural practices in water-scarce environments. The authors modestly note that model accuracy is currently limited by the scarcity of exposed sediment pixels in satellite imagery and recommend future work to expand spectral libraries and separate soil/sediment mapping as more data becomes available.