Assessment of soil organic matter in the post-fire period using VNIR spectroscopy

Abstract

Wildfires profoundly impact ecosystems and soil organic matter (SOM), a critical factor in soil quality and carbon cycling. This research aimed to assess the impact of wildfire severity on SOM and the potential of visible-near infrared spectroscopy (VNIR) spanning the 350 - 1050 nm wavelength range for monitoring SOM in a post-fire landscape using two modelling approaches (i) Partial Least Squares Regression (PLSR) and (ii) Artificial Neural Networks (ANN). Following a comprehensive two-year investigation in Zadar County, Croatia, where a 13.5 ha mixed forest was moderately to severely affected by a wildfire, spectral reflectance analysis revealed that SOM content strongly influenced soil reflectance. High-severity samples exhibited the lowest reflectance compared to those with moderate severity and the control group. The critical region for SOM information in post-wildfire soil estimation models was between 550 and 700 nm. ANN consistently outperformed PLSR, achieving a ratio of performance to deviation (RPD) values from 1.74 to > 2.5. In contrast, PLSR achieved values between 1.62 and 2.29, demonstrating ANN's capability to provide accurate predictions of SOM content in complex post-fire SOM dynamics conditions. This research indicates that VNIR spectroscopy, particularly coupled with ANN-based models, offers a reliable and non-destructive method for assessing SOM content in post-fire environments, facilitating informed land management decisions for ecosystem recovery.

Keywords

• wildfire
• hyperspectral data
• linear modelling
• nonlinear modelling