Volume XLII-4/W18
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-4/W18, 1059–1063, 2019
https://doi.org/10.5194/isprs-archives-XLII-4-W18-1059-2019
© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-4/W18, 1059–1063, 2019
https://doi.org/10.5194/isprs-archives-XLII-4-W18-1059-2019
© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.

  19 Oct 2019

19 Oct 2019

ESTIMATING ABOVEGROUND BIOMASS IN ZAGROS FOREST, IRAN, USING SENTINEL-2 DATA

H. Torabzadeh1, M. Moradi2, and P. Fatehi3 H. Torabzadeh et al.
  • 1Engineering Faculty, Bu-Ali Sina University, Hamedan, Iran
  • 2Omran Tossee University, Hamedan, Iran
  • 3Department of Forestry and Forest Economics, University of Tehran, Tehran, Iran

Keywords: aboveground biomass (AGB), random forest regression (RFR), Sentinel-2, Zagros forest

Abstract. Accurate and reliable assessment of above-ground biomass (AGB) is important for the sustainable forest management, especially in Zagros forests, in which a frangible forest ecosystem is being threatened by anthropogenic factors as well as climate change effects. This study presents a new method for AGB estimation and demonstrates the potential of Sentinel-2 Multi-Spectral Instrument (MSI) data as an alternative to other costly remotely sensed data, such as hyperspectral and LiDAR data in unapproachable regions. Sentinel-2 performance was evaluated for a forest in Kurdistan province, west of Iran, using in-situ measured AGB as a dependent variable and spectral band values and spectral-derived vegetation indices as independent variables in the Random Forest Regression (RFR) algorithm. The influence of the input variables number on AGB prediction was also investigated. The model using all spectral bands plus all derived spectral vegetation indices provided better AGB estimates (R2 = 0.87 and RMSE = 10.75 t ha−1). Including the optimal subset of key variables did not improve model variance but slightly reduced the error. This result is explained by the technically-advanced nature of Sentinel-2, which includes fine spatial resolution (10, 20 m) and strategically-positioned bands (red-edge), conducted in different topographical conditions with an advanced machine learning algorithm. However, assessing its transferability to other forest types with varying conditions would enable future performance and interpretability assessments of Sentinel-2.