Comparison of three algorithms for tree crown area and available pruning biomass monitoring

Fast and rigorous assessment of tree characteristics from earth observation products has many environmental applications, including monitoring of the canopy biomass available for pruning and utilisation as soil amendment or energy source. Here we explore the efficiency of three supervised classification algorithms in assessing canopy area of olive trees, the staple food crop of the Mediterranean that annually produces an estimated 2,82 Μt ha^-1 of residual biomass (Velázquez-Martí et al., 2011) which is currently largely unexploited and often an environmental hazard due to on-site fires. The algorithms include (a) a thresholding algorithm (Daliakopoulos et al., 2009) processing Normalized Difference Vegetation Index values, (b) a supervised machine learning algorithm comprised on an Artificial Neural Network (ANN) with 4 hidden layers, and (c) the AdaBoost supervised deep learning algorithm. Following Yang et al. (2009), the latter two methods use image colour, texture, and entropy as inputs. Ground truth was developed by manually producing a binary mask where pixels depicting tree crown were marked with 1 and otherwise 0, and classification results were evaluated using the Dice similarity coefficient (DSC; Nisio et al., 2020). The three algorithms were tested on assessing olive tree crown projected surface area on a WorldView II image of resolution 0.5 × 0.5 m of a rural area of Heraklion, Crete, Greece, acquired on November 10, 2020. Masking was performed in 42 olive tree plots including a total of 1,080 olive trees, including on-site visual validation of the masking results. Results show that the ANN performed better than AdaBoost and NDVI thresholding, scoring 81.98%, compared to 75.06 and 70.03%, respectively. The trained ANN is currently used to provide olive tree canopy estimates, used as input to assess canopy biomass available for pruning for the CompOlive system, an online platform that facilitates matchmaking of olive tree farms, olive mills, and mobile composting equipment, to optimise on-farm compost production and utilisation.

Acknowledgements

This research is co-financed by the European Union and Greek national funds through the Operational Program CRETE 2014-2020, under Project “CompOlive: Integrated System for the Exploitation of Olive Cultivation Byproducts Soil Amendments” (KPHP3-0028773).

References

Daliakopoulos, I. N., Grillakis, E. G., Koutroulis, A. G., & Tsanis, L. K. (2009). Tree Crown Detection on Multispectral VHR Satellite Imagery. Photogrammetric Engineering and Remote Sensing, 75(10), 1201–1211. https://doi.org/10.14358/PERS.75.10.1201

Velázquez-Martí, B., Fernández-González, E., López-Cortés, I., & Salazar-Hernández, D. M. (2011). Quantification of the residual biomass obtained from pruning of trees in Mediterranean olive groves. Biomass and Bioenergy, 35(7), 3208–3217. https://doi.org/10.1016/J.BIOMBIOE.2011.04.042

Yang, L., Wu, X., Praun, E., & Ma, X. (2009). Tree detection from aerial imagery. GIS: Proceedings of the ACM International Symposium on Advances in Geographic Information Systems, 131–137. https://doi.org/10.1145/1653771.1653792