Monitoring the early growth of forest plantations with Sentinel-2 satellite time-series

Abstract

Monitoring initial growth phases is essential for the success of forest plantations. This study introduces a methodology aimed at characterizing the growth of Eucalyptus short rotation plantations in Brazil during their first 2 years, based on Sentinel-2 satellite imagery. The primary goal is to detect potential anomalies at the pixel level, covering an area of 400 m2, and to feed operational decision-making strategies aiming at characterizing, correcting or mitigating the problem. The approach relies on predictive machine learning models that estimate an integrated growth index, the volume that the trees will reach at 2 years of age (V2Y). The model uses various plantation characteristics such as planting density, genotypic characteristics and environmental factors and incorporates vegetation indices derived from Sentinel-2 data acquired during the first 2 years of the plantation. These anticipation models were calibrated on an extensive dataset comprising more than 9000 inventory plots spread over more than ninety thousand hectares. The Green Normalized Difference Vegetation index (GNDVI) was shown to give the best results among several vegetation indices tested. The accuracy of V2Y prediction improved significantly when longer periods of vegetation indices were included. Our results demonstrate that using the GNDVI data from the first year or from the initial 18 months of plantation growth yields accurate predictions of V2Y, with R2 values of 0.71 and 0.74 and RMSE values of 7.86 and 7.46 m3 ha−1, respectively. The anticipation model with GNDVI outperformed simpler models that solely rely on stand characteristics. The novel approach developed in this study offers an operational means to reliably estimate an early-stage growth indicator for Eucalyptus plantations in Brazil.