Estimation of Forest Canopy Cover by Combining ICESat-2/ATLAS Data and Geostatistical Method/Co-Kriging

Abstract

Accurately estimating forest canopy cover (FCC) is challenging by using traditional remote sensing images at the regional level due to the spectral saturation phenomenon. In this study, to improve the estimation accuracy, a new method of FCC wall-to-wall mapping was suggested based on ice, cloud, and land elevation satellite/advanced topographic laser altimeter system (ATLAS) data. Specifically, one dataset of FCC's observations was combined with preprocessed ATLAS data and topographic factors to build a random forest regression (RFR) model. Moreover, the Co-Kriging method was used to generate spatially explicit values that are required by the RFR from the point data of ATLAS parameters, and then the wall-to-wall mapping of the FCC was conducted. The results showed that the RFR model had an accuracy of relative root-mean-square error (rRMSE) = 0.09 with a coefficient of determination (R2) = 0.91. The best-fit semivariogram models between primary variables and covariates were asr and TR (Model: Gaussian model, R2 = 0.94, the residual sum of squares (RSS) = 1.73 × 10−6), landsat_perc and NDVI (Model: spherical model, R2 = 0.46, RSS = 1.58 × 10−4), and photon_rate_can and slope (Model: exponential model, R2 = 0.77, RSS = 6.45 × 10−4), respectively. FCC validation result showed that the FCC's wall-to-wall mapping was in great agreement with the dataset-2 (R2 = 0.79; rRMSE = 0.11).

Keywords

• Co-Kriging (CK)
• forest canopy cover (FCC)
• Ice, Cloud, and Land Elevation Satellite/Geoscience Laser Altimeter System (ICESat-2 ATLAS)
• random forest regression (RFR)
• semivariogram
• validity validation