Adaptation of satellite-based vegetation indices for different land use types

Remote sensing is an important data collection method for farmers and researchers to obtain up-to-date information on the state of vegetation. Due to efficiency and low cost of method, the satellite-based remote sensing also allows analysis at the catchment level. The aim of present study was to investigate the usability of satellite-based vegetation indices for different land use types while using ground-truth measurements for comparisons. The study area was a small agricultural catchment in Balaton Upland, Hungary. Four different land use types (forest, grassland, vineyard and cropland) were investigated, located on different angled slopes. In the vineyard there were three different inter-row managements investigated.

In 2023 the field measurements were taken in every two weeks during vegetation period. Hand-held (H) sensor set was used to measure vegetation indices on the slopes of grassland, cropland, and the three vineyard sites. The Normalized Difference Vegetation Index (NDVI) and Photochemical Reflectance Index (PRI) sensors were used to measure leaf reflectance. A hemispherical sensor set was used for each measurement. Additional handheld instruments were used for Leaf Area Index (LAI), and soil water content (SWC) measurements. The source of satellite-based data was Sentinel-2 (S2). At the same time as the field measurements 8 out of 13 available spectral bands were collected from S2 and used to calculate different spectral indices (e.g. NDVI or green chlorophyll index - GCI).

The vegetation of different land use types varies considerably, which also affects the applicability of the vegetation indices. In 2023, the strongest correlation between NDVI of field measurement and satellite NDVI was for grassland (r=0.76). The highest overall NDVI values for both methods were observed in the vineyard with cover crop inter-row (H NDVI: 0.76, S2 NDVI: 0.56). PRI values for all land use types were most strongly correlated with the Red Edge 2 band (e.g. r=0.65 for grassland, r=0.69 for Cropland, r=0.70 for Vineyard C). The highest average leaf area index was measured for the forest (3.36), and the lowest in grassland (0.86). LAI showed good correlation with cropland GCI (0.86), moderate correlation with forest and tilled inter-row vineyard (0.50 and 0.55, respectively). PCA analysis showed that the cover crop and grassed inter-row did not, but most other land use types grouped distinctly.

Acknowledgments: This material is based upon work supported by the Hungarian National Research Fund (OTKA/NKFI) project OTKA FK-131792. The research presented in the article was carried out within the framework of the Széchenyi Plan Plus program with the support of the RRF 2.3.1 21 2022 00008 project. The research was funded by the Sustainable Development and Technologies National Programme of the Hungarian Academy of Sciences (FFT NP FTA).