Optimizing harmonics from Landsat time series data: the case of mapping rainfed and irrigated agriculture in Zimbabwe

Abstract

Explicit and large-scale information on farming systems is important for many applications such as crop production estimates, drought impact assessments or water footprint analysis. This contribution investigated the possibility of optimizing harmonic functions fitted to cloud-corrected Landsat NDVI (Normalized Difference Vegetation Index) time-series imagery to effectively map rainfed and irrigated croplands in Zimbabwe. The following harmonic optimizations were investigated: the effect of the linear trend, length of the time series data and the harmonics degree. The verification accuracy scores for mapping the two farming systems were used to ascertain the most optimal harmonics settings. The most accurate classification results (overall accuracy of 97%) were produced when the linear trend was excluded and the full time period (2013 to 2018) with a 7th degree harmonics fitting function was used. Optimized harmonics provide an effective way to compute vegetation seasonality signals while accounting for data gaps and residual noise inherent in Landsat time-series data.