Assessment of Pine Forest Condition Towards Early Detection and Monitoring of Stress Through a Synergistic Use of Sentinel-1 and Sentinel-2 Imagery

Abstract

An integrated use of Sentinel-1 Synthetic Aperture Radar (SAR) and Sentinel-2 multispectral imagery is implemented in this study for the assessment of pine forest condition in Camp John Hay, Baguio City, Philippines. These assessments include: (1) the inspection of distribution of dead trees, (2) trends analysis of Sentinel-derived products, and (3) the generation of anomaly maps. Dead trees were identified using Google Earth Pro imagery – the trees were classified as such if they manifest brown or grey foliage with evident foliage loss. The products used for succeeding analyses are sigma naught VH (σ0VH) and sigma naught VV (σ0VV) backscatter derived from Sentinel-1, Sentinel-2 bands 2, 3, 4, 5, 6, 7, 8, 8a, 11, and 12, as well as Sentinel-2 derived vegetation indices S2REP (Sentinel-2 Red-Edge Position Index), NDWI (Normalized Difference Water Index), NDVI (Normalized Difference Water Index), NDII (Normalized Difference Infrared Index), MSI (Moisture Stress Index), IRECI (Inverted Red-Edge Chlorophyll Index), GNDVI (Green Normalized Difference Water Index), and EVI (Enhanced Vegetation Index). Trends in these products were backtracked to identify patterns they exhibit as tree health deteriorates, and to determine which are most robust for early detection of stress in pine trees. From the analysis of slope decline against time, and their corresponding R2 statistic, Sentinel-2 products found to be robust for early stress detection are band 5, followed by equally important GNDVI, NDII, MSI, and bands 2, 3, and 4, then NDVI. A synergistic analysis of Sentinel-1 and Sentinel-2 products showed that NDVI is well-correlated with σ0VV backscatter; hence, they exhibit concurrent decline in trends due to stress. Among Sentinel-1 products, a delayed decline in σ0VH backscatter occurred compared with σ0VV backscatter. These patterns manifested in three-month intervals. A map of high risk and low risk forest areas was generated from the overall average of negative and positive anomalies of vegetation indices. The relative reliability of the map of at-risk vegetation areas was computed based on the count of layer values with valid data per cell, resulting in a maximum relative reliability of 65.3%. This study showed that the integration of Sentinel-1 SAR and Sentinel-2 multispectral imagery is a promising approach in the comprehensive assessment of pine forest condition towards the early detection and monitoring of stress.