Evaluation of temporal compositing algorithms for annual land cover classification using Landsat time series data

In this paper, four widely used temporal compositing algorithms, i.e. median, maximum NDVI, medoid, and weighted scoring-based algorithms, were evaluated for annual land cover classification using monthly Landsat time series data. Four study areas located in California, Texas, Kansas, and Minnesota, USA were selected for image compositing and land cover classification. Results indicated that images composited using weighted scoring-based algorithms have the best spatial fidelity compared to other three algorithms. In addition, the weighted scoring-based algorithms have superior classification accuracy, followed by median, maximum NDVI, and medoid in descending order. However, the median algorithm has a significant advantage in computational efficiency which was & SIM;70 times that of weighted scoring-based algorithms, and with overall classification accuracy just slightly lower (& SIM;0.13% on average) than weighted scoring-based algorithms. Therefore, we recommended the weighted scoring-based compositing algorithms for small area land cover mapping, and median compositing algorithm for the land cover mapping of large area considering the balance between computational complexity and classification accuracy. The findings of this study provide insights into the performance difference between various compositing algorithms, and have potential uses for the selection of pixel-based image compositing technique adopted for land cover mapping based on Landsat time series data.