The accuracy of the number (k) of Nearest Neighbor (NN) technique in estimating the aboveground carbon (AGC) pool of a naturally grown pine forest as affected by reflectance normalization and data fusion was evaluated in this study. Accurate AGC map provides detail information for the measurement, reporting, and verification of the greenhouse gases (GHG) that are accumulated in the atmosphere. This study utilized the established 56 plots and Landsat 8 OLI & TIRS. The reflectance normalization was conducted with the use of the FLAASH technique and the data fusion process using the NNDiffuse technique was performed in the Landsat image for enhancement. 

This study indicated that the AGC estimation accuracy of k-NN method using Landsat image with combination of reflectance normalization and data fusion is 81.82% with an average AGC of 71.00 tC/ha as compared with that of reflectance normalization conducted in the original image without data fusion which has an accuracy of 54.54% and with an average AGC of 69.20tC/ha. This study concludes that reflectance normalization and data fusion of Landsat images provide relevant accuracy on AGC estimation using the k-NN method. Besides, the k-NN method could offer reliable AGC estimates addressing the hindrances of conducted field inventory in a tropical forest with irregular topography.

Ground data of the reference plots for the k-NN method

Fifty-six plots with an area of approximately 400 m² (11.3m radius) and with a distance of more or less 2km from one plot to another were determined through a random sampling method and served as reference plots for this study. 100% of the trees larger than 5cm in diameter found inside the plot was measured at 1.3m above the ground for the diameter at breast height (D). The GPS receiver was used to determine the geographical coordinates of each plots. The AGB estimates of the plots were calculated with the use of the allometric equation developed by Brown (1997) while the AGC of the reference plots were computed using the formula recommended by the IPCC (2006). 

Satellite image of the study site

The satellite image that was used in this study is the Landsat-8 Operational Land Imager and Thermal Infrared Sensor (Landsat 8 OLI &TIRS). The prepocessing namely reflectance normalization and data fusion was performed in the satellite image before data analysis. 

AGC estimation of the target plots using the k-NN method

The k-NN method in determining the AGC of the target plots of the natural pine forests was used in this study. It is a nonparametric algorithm that had been extensively used for calculating forest resources with the use of Landsat images (Yim, Kong, Kim, & Shin, 2007; McRoberts, 2012; Lumbres & Lee, 2014; Doyog et al., 2018).

Accuracy assessment using confusion matrices

The 56 plots in this study were randomly split into two: 80% (45 plots) for estimating the AGC and the 20% (11 plots) for confirmation in the confusion matrix for the accuracy assessment. This study used the confusion matrix to evaluate the accurateness of the k-NN method in estimating the AGC of the natural pine forest.