Preliminary Assessment of Peat Deposits Imaging Using VNIR (400–1000 nm) Hyperspectral Scanning: Organic Matter and Humification Case Study

Wetlands are one of the most valuable yet most threatened terrestrial ecosystems. These ecosystems are essential in regulating the long-term carbon cycle, maintaining hydrological balance, and serving as local biodiversity hotspots. However, human activities have heavily impacted these ecosystems for centuries, changing them from carbon sinks into carbon sources. Current climate change and land use changes speed up this process globally. Conservation and restoration efforts require a better understanding of these negative phenomena. Studying how peat deposits change over time and across different areas is essential to make informed decisions. This research helps link environmental impacts to how wetlands respond, enabling more effective management strategies.

However, typical high-resolution spatiotemporal studies require numerous drill sites, abundant samples, and analytical techniques, substantially limiting the investigation scope. Non-destructive imaging techniques such as hyperspectral scanning imaging (HSI) might help overcome these limitations by allowing rapid analyses of lengthy peat profiles. Therefore, HSI is a promising tool for paleoenvironmental investigations and can potentially be used in conservation efforts of degraded peatlands.

We selected a peatland in northern Poland to test the HSI's ability to accurately trace organic matter accumulation and peat humification in several microhabitats. Peat excavation sites indicate that human activities have impacted this location in the past, and different processes (like drainage) might have been recorded affecting the peat, leading to the lowering of the water table. We cored in 5 sites and obtained 5.2 m of material, capturing undisturbed and heavily humified deposits. Cores were subsampled every 5 cm. We used estimated organic matter content (OM) by loss on ignition and von Post humification degree. We confronted these results with high-resolution hyperspectral scanning imaging (HSI) of peat cores in the VNIR (400–1000 nm) range. Finally, we compared both approaches using linear regression and machine learning approaches, i.e., random forest and gradient boosting, to find associations between the datasets. Overall, machine learning models generalize OM and humification tendencies in the deposits to a satisfactory level. Further investigation with numerous drills per site, more diverse material, and increased training set size might provide an invaluable opportunity to identify the current condition of the peatlands. In the future, they can provide a rapid and independent tool for checking restoration efforts.

The work was supported by the National Science Centre, Poland, under the research project „Exploring methods of hyperspectral imaging of lake sediments: proxy development and calibration,” no UMO-2023/51/D/ST10/00801