Improved estimation of forest carbon (biomass) using bi-temporal RapidEye data in a low-altitude tropical landscape

Abstract

In the recent past, optical satellite data have been widely used in estimating forest parameters, particularly above-ground biomass (AGB) and carbon (C) stocks, but were not used much in Papua New Guinea (PNG) forest studies. In this study, forest inventories conducted in 2009 and 2014 for ground estimation of AGB and C were linked with bi-temporal high resolution (5m) optical RapidEye satellite data for 2010 and 2014respectively for estimation at spatial levels using an improved strategy in a low-altitude tropical landscape of PNG. In order to improve the overall estimation process, specific spectral indices were derived from the Red band of the RapidEye data along with explorative derivation of such using the Red Edge narrow-bandto act as added variables for correlation with AGB and C. Variable appropriation for the modeling found significance in the Red Edge derived spectral indices over those derived normally from the Red band. Using these spectral parameters, single preeminent variables were identified and utilized to model AGB and C in each forest stratum via a spatial linear regression analysis. This study presents the idea of generating stratum-specific models using RapidEye imageries and merging these models through the notion of model-fitting for effective cross-landscape estimation of C stocks. The two broad forest stratums analysed were undisturbed primary forest (PF) and disturbed secondary forest (SF). Stratum-specific models developed for PF and SF using spectral indices had high confidence levels of p< 0.01 for both PF and SF in 2010 and also sound confidence levels of p< 0.05 for both PF and SF in 2014. The overall root mean square errors (RMSEs) for both temporal models were reasonably low with values <9MgC ha-1 and <29MgC ha-1 across the study area. RMSEs for the model-fitswere attuned and more promising with values < 7MgC ha-1 and < 18MgC ha-1respectively.These results show that the strategy of stratum-specific modeling used here is an effective approach that can be well applied in other low-altitude tropical forest landscapes in PNG with high resolution optical satellite data for efficient C stock estimations for REDD+ implementations.