Comparisons between temporal statistical metrics, time series stacks and phenological features derived from NASA Harmonized Landsat Sentinel-2 data for crop type mapping

Authors: LIU, XIAOMI - Shandong University; XIE, SHUAI - Shandong University; YANG, JIANGNING - Chinese Academy Of Sciences; SUN, LIN - Shandong University; LIU, LIANGYUN - Chinese Academy Of Sciences; ZHANG, QING - Chinese Academy Of Sciences; Yang, Chenghai

Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/21/2023
Publication Date: 6/30/2023
Citation: Liu, X., Xie, S., Yang, J., Sun, L., Liu, L., Zhang, Q., Yang, C. 2023. Comparisons between temporal statistical metrics, time series stacks and phenological features derived from NASA Harmonized Landsat Sentinel-2 data for crop type mapping. Computers and Electronics in Agriculture. https://doi.org/10.1016/j.compag.2023.108015.
DOI: https://doi.org/10.1016/j.compag.2023.108015

Interpretive Summary: Spectrotemporal features derived from satellite imagery are useful for characterizing land cover of highly dynamic crops. This research investigated three spectrotemporal features (temporal statistical metrics, time series image stacks, and phenological features) for mapping crop types using combined Landsat-8 and Sentinel-2 satellite data. The results from the study showed that time series image stacks performed better than the other two features in terms of accuracy and other measures. An analysis of the effects of different temporal factors on time series stacks further indicated that the best classification results were achieved when using Landsat-8 and Sentinel-2 data composited monthly for the growing season. The findings from this study provide valuable insights for spectrotemporal feature selection and optimization for accurate crop type mapping.

Technical Abstract: Spectrotemporal features that capture changes in reflectance over time are useful for characterizing land cover of highly dynamic crops. In crop type mapping, three commonly used spectrotemporal features are temporal statistical metrics, time series stacks, and phenological features, which differ in their calculation methods and physical implications. However, there has been limited investigation on the performance comparisons between them for crop type mapping. The objective of this study was to evaluate and compare the effectiveness of the three features derived from Harmonized Landsat Sentinel-2 (HLS) data for crop type mapping. The HLS data were first pre-processed with cloud masking, temporal compositing and gap filling to create the gap-free time series for extracting the three spectrotemporal features. Crop reference data were obtained through a field survey conducted over a study area of 14.5 km by 8 km near College Station, Texas, USA. For the calibration of the Random Forest (RF) classification model with different sets of spectrotemporal features, 30% of the total reference data were used, and the remaining 70% were used for quantitative accuracy assessment. Results showed that although all three spectrotemporal features could yield accurate crop type maps, time series stacks (with overall accuracy (OA) of 96.62% and Kappa of 0.95) had superior crop classification performance to temporal statistical metrics (with OA of 92.19% and Kappa of 0.88) and phenological features (with OA of 90.87% and Kappa of 0.86). In addition, time series stacks also outperformed temporal statistical metrics and phenological features for any individual crop type mapped in terms of user’s accuracy, producer’s accuracy and F1-score. Moreover, the effects of temporal density, interval and depth on time series stacks were further analyzed. The analysis suggested that the optimal crop mapping results for time series stacks could be achieved by using the time series input data from the combined Landsat-8 and Sentinel-2 that were composited on a monthly basis covering the period from March to October. Supplementary experiments from two additional areas confirmed the consistency of the results from this study, demonstrating the scalability of the methods used. This research provides valuable insights into spectrotemporal feature selection and optimization for accurate crop type mapping.