Seasonal Changes in the Prediction Accuracy of Hayfield Productivity Using Sentinel-2 Remote-Sensing Data in Hokkaido, Japan

Abstract

In large hayfields belonging to intensive dairy systems, satellite remote-sensing data can be useful to determine the hayfield yield and quality efficiently. In this study, we compared the land survey data of hayfield yield, and its quality parameters such as crude protein and neutral detergent fiber digestibility (NDF), with the Sentinel-2 satellite image data for thirteen hayfield paddocks in Kamishihoro region, Hokkaido, Japan. Commonly used indices derived from the satellite image data, including the Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI), were used to assess the hayfield yield and quality. In this region, hayfields are usually harvested twice yearly, in early summer (first harvest) and late summer (second harvest). As result, the Sentinel-2 data could predict the pasture growth and quality for the first harvest better than those for the second harvest. The EVI and the index based on the bands B8a and B7 were the best predictors for the biomass and NDF for the first harvest, respectively. However, the satellite-image-based predictors were not found for the second harvest. Towards the second harvest season, the color of the hayfield surface became more heterogeneous because of the flowering of weeds and uneven pasture growth, which made it challenging to predict pasture growth based on the remote-sensing data. Our land survey approach (quadrat-based sampling from a small area) should also be improved to compare the remote-sensing data and the pasture with uneven growth.