Improved inference of tropical vegetation properties using seasonal Landsat Vegetation Indices

Abstract

Spatio-temporal quantification of vegetation diversity and structure is important for accurate monitoring of terrestrial ecosystems from space. Landsat data have been employed to provide estimates of vegetation biophysical properties due to their medium (30 m) spatial resolution, sufficient (16 days) temporal resolution, and spectral sensitivity to biophysical properties. This study has explored the use of Landsat-derived vegetation indices (VIs) related to greenness, moisture content and fire severity, recorded during the peak and late growing seasons, to estimate in situ observed species number, basal area and aboveground biomass (AGB) in tropical biomes that are affected by fire, and were surveyed at various stages of post-fire recovery. Linear and logarithmic regressions and coefficients of determination (R2) were computed to assess the relationships of species number, basal area and AGB with ten broadband VIs, with goodness of fit measured by root mean squared error (RMSE). Best fits were obtained using peak-season Green Chlorophyll Index (CI Green), Normalized Difference Moisture Index (NDMI) and Normalized Burn Ratio (NBR2) for species number (R2 = 0.50–0.68, RMSE = 3–4), basal area (R2 = 0.23–0.37, RMSE = 1.0–1.1 m2 ha–1) and AGB (R2 = 0.66–0.74, RMSE = 1.1–1.2 Mg ha-1) in open savanna and savanna forest. Late-season Normalized Difference Vegetation Index (NDVI), NDMI and NBR showed stronger relationships for species number (R2 = 0.88, RMSE = 5.72), basal area (R2 = 0.24–0.68, RMSE = 0.03–9.7 m2 ha–1) and AGB (R2 = 0.20–0.73, RMSE = 1.4–19.2 Mg ha–1) in most of the more complex forest biomes. These results are promising for the wider application of Landsat data especially from Landsat-8 Operational Land Imager (OLI) multispectral sensor to infer post-fire vegetation recovery in tropical ecosystems.