Tree cover and carbon mapping of Argentine savannas: Scaling from field to region

Programs which intend to maintain or enhance carbon (C) stocks in natural ecosystems are promising, but require detailed and spatially explicit C distribution models to monitor the effectiveness of management interventions. Savanna ecosystems are significant components of the global C cycle, covering about one fifth of the global land mass, but they have received less attention in C monitoring protocols. Our goal was to estimate C storage across a broad savanna ecosystem using field surveys and freely available satellite images. We first mapped tree canopies at 2.5 m resolution with a spatial subset of high resolution panchromatic images to then predict regional wall-to-wall tree percent cover using 30-m Landsat imagery and the Random Forests algorithms. We found that a model with summer and winter spectral indices from Landsat, climate and topography performed best. Using a linear relationship between C and % tree cover, we then predicted tree C stocks across the gradient of tree cover, explaining 87% of the variability. The spatially explicit validation of the tree C model with field-measured C-stocks revealed an RMSE of 8.2 tC/ha which represented ~ 30% of the mean C stock for areas with tree cover, comparable with studies based on more advanced remote sensing methods, such as LiDAR and RADAR. Sample spatial distribution highly affected the performance of the RF models in predicting tree cover, raising concerns regarding the predictive capabilities of the model in areas for which training data is not present. The 50,000 km2 has ~ 41 Tg C, which could be released to the atmosphere if agricultural pressure intensifies in this semiarid savanna. In this study, we demonstrated the benefit of using high resolution imagery for regional tree cover and C analysis, increasing available training data when there is paucity of field data.