Application of hyperspectral imaging of peat profiles to the case of fen-bog transition in aapa mires

Hyperspectral imaging (HSI) is a promising precision tool for analysing chronological peat strata from vegetation transitions in peatlands. We explored the potential of HSI in identifying transitions in peat-forming vegetation and degree of peat humification. The changes in aapa mire complexes during recent decades have been assessed by various remote sensing methods (aerial image time series, satellite data and high-resolution UAV multispectral imaging) and HSI methods have been developed to support the data from other sources. Rapid growth of Sphagnum mosses over string-patterned aapa mires in the north-boreal zone has immense significance, since it can alter ecosystem structure and functions such as carbon sequestration. HSI is well suited for analysis of recent ecosystem changes, since it can be applied for large sample sets with extremely fine spatial detail. Additionally, peat layers have complex 3D structures that can be overlooked by other sampling methods.

The HSI data was collected in laboratory conditions with two spectral imaging cameras, covering the visible to near-infrared range (VNIR 400-1000 nm), short-wave infrared range (SWIR, 1000-2500 nm). We used various methods such as PCA, k-means clustering and support vector machines for both quantitative and qualitative analysis of peat.  Our analyses revealed detailed spectral changes that matched with transitions in peat quality and composition. Methodological issues unique to peat samples, such as the effect of oxidation and water content, were assessed for method development. We also used HSI to estimate quality changes that would easily be overlooked or only found by most laborious conventional techniques, like high-frequency microscopic counting of plant remains. Here, the spectral results can be used to guide sampling for microscopic routines, for example.

Results with Carex and Sphagnum peat proved that efficient image-based classification methods for identifying peat transitions can be developed. Our SVM models in the VNIR and SWIR regions were able to distinguish Sphagnum and Carex peat with overall accuracy of validation 80 % and 81 %, respectively. We also developed simple NDI indices for the estimation of von Post humification index that worked with accuracy of 86 % and 59 % for VNIR and SWIR, respectively. In combination with data collected from other sources (remote sensing, ground-truthing, conventional laboratory analysis), peat spectral analyses give strong inference of changes. In our study system, results indicate high sensitivity of northern aapa mires to ecosystem-scale changes.