Exploiting the capabilities of the Sentinel-2 multi spectral instrument for predicting growing stock volume in forest ecosystems

• Published in: International journal of applied earth observation and geoinformation 2018-04, Vol.66, p.126-134
• Main Authors: Mura, Matteo, Bottalico, Francesca, Giannetti, Francesca, Bertani, Remo, Giannini, Raffaello, Mancini, Marco, Orlandini, Simone, Travaglini, Davide, Chirici, Gherardo
• Format: Article
• Language: English
• Subjects: Forests; Growing stock volume; Multi spectral instrument; Prediction; Random forest; Sentinel-2
• ISSN: 1569-8432; 1872-826X
• DOI: 10.1016/j.jag.2017.11.013

•Sentinel2-MSI imagery was evaluated for the growing stock volume estimation in two forest areas in Italy.•For comparison Landsat 8 OLI and RapidEye images were used.•k-Nearest Neighbors algorithm based on RandomForest distance matrix was successful for the estimations.•Sentinel2 MSI and Landsat 8 OLI gave similar results, always better than RapidEye. The spatial prediction of growing stock volume is one of the most frequent application of remote sensing for supporting the sustainable management of forest ecosystems. For such a purpose data from active or passive sensors are used as predictor variables in combination with measures taken in the field in sampling plots. The Sentinel-2 (S2) satellites are equipped with a Multi Spectral Instrument (MSI) capable of acquiring 13 bands in the visible and infrared domains with a spatial resolution varying between 10 and 60m. The present study aimed at evaluating the performance of the S2-MSI imagery for estimating the growing stock volume of forest ecosystems. To do so we used 240 plots measured in two study areas in Italy. The imputation was carried out with eight k-Nearest Neighbours (k-NN) methods available in the open source YaImpute R package. In order to evaluate the S2-MSI performance we repeated the experimental protocol also with two other sets of images acquired by two well-known satellites equipped with multi spectral instruments: Landsat 8 OLI and RapidEye scanner. We found that S2 worked better than Landsat in 37.5% of the cases and in 62.5% of the cases better than RapidEye. In one study area the best performance was obtained with Landsat OLI (RMSD=6.84%) and in the other with S2 (RMSD=22.94%), both with the k-NN system based on a distance matrix calculated with the Random Forest algorithm. The results confirmed that S2 images are suitable for predicting growing stock volume obtaining good performances (average RMSD for both the test areas of less than 19%).