Assessment and correction of Sentinel-2 and Landsat-8/9 NDVI using in-situ measurements across rice growth stages in southern South Korea

Abstract

This study aimed to compare in-situ normalized difference vegetation index (NDVI) measurements with satellite-derived NDVI data for rice paddies field in southern region of South Korea and to develop calibration equations for different growth stages using both linear and non-linear regression models. The in-situ NDVI was measured with a portable leaf index meter (Crop Circle ACS-435), and satellite-derived NDVI was obtained from Landsat-8/9 and Sentinel-2 images. All values represented daily average NDVI across a five-year period (2020–2024). Analysis showed that satellite-derived NDVI values were generally lower than in-situ values, primarily because of atmospheric and spatial resolution differences. Both satellite platforms exhibited a strong positive correlation with ground-based NDVI, although stage-specific differences were observed. Landsat-8/9 outperformed in the pre-heading stage, whereas Sentinel-2 performed better in the post-heading stage. For Landsat-8/9, the mean absolute percentage error (MAPE) decreased substantially from 38.6 % before correction to 16.7 % after applying the calibration equations, whereas for Sentinel-2 it decreased from 22.1 % to 15.3 %. This study establishes a foundation for improving the accuracy and reliability of satellite-based NDVI through in-situ calibration, with potential applications in agricultural productivity, environmental monitoring, and climate change adaptation.