A Time Series Approach for Wheat Crop Harvest Detection using Multispectral Data

Abstract

In this paper, a time series approach for the detection of winter wheat harvest is proposed. The use of satellites to monitor crops and predict crop yield and estimated harvest dates is gathering attention now a days. The present study undertakes the task of finding and predicting the accurate time for the harvesting of wheat crops based on the change in the spectral signature of the field during observation and the growth timeline of the crop. For this purpose, Sentinel-2 multispectral time series data is used for which a timely ground observation is done. High emphasis is given to the selection of the optimal set of bands among 13 available bands, and after a rigorous pattern observation, Red, Blue, and NIR bands are found to be the optimal band set. Although Red and Blue bands together are able to identify various crop growth stages, using the NIR band in the band set is an added advantage because it is used to generate NDVI time series with Red band. The only contrast between Red and Blue bands for this specific study is that Red band is more aggressive towards changes in the state of the crop. The selected band set is used for the observation of the field to detect an accurate harvest period. The proposed approach is validated using Sentinel-1 SAR coherence time series and found to be accurate.