Detection of tillage areas and periods using high spatial resolution optical image time series

In this paper we present an approach to image time series analysis in order to identify tillage areas as a part of early detection of cereal and crop cycles, over the Merguellil plain in central Tunisia. We used Sentine1-2 (S-2) data to calculate eight spectral indices: Normalized Difference Vegetation index (NDVI), Saturation Index (SI), Soil Color Index (CI), Form Index (FI), Redness Index (RI), Brightness Index (BI), Difference Tillage Index (NDTI) and Soil Tillage Index (STI). Principal component analysis (PCA) and indices time series analysis showed that only the difference of BI between the date D+1 and the date D$\langle\langle\Delta$(BI)$\rangle\rangle$ allows tillage detection. NDVI threshold (0.2) allows distinction between vegetation and bare or poorly covered soils.Spectral based tillage indices such as Brightness Index, the normalized difference tillage index and Normalized difference of Bands red and 2 discriminated tillage practices in March and May of 2018 over the Merguellil plain, with an overall accuracy of 92.3%. Sentine12 data appears to be efficient and effective in classifying tillage practices over large areas.