Decomposition of long time-series fraction of absorbed photosynthetically active radiation signal for distinguishing heavy metal stress in rice

Crops are often affected by heavy metal stress and other types of stresses present in farming. Distinguishing heavy metal stress from other types of stresses can be addressed through agricultural remote sensing. Unlike color and morphology, monitoring physiological functions can lead to detect heavy metal contamination even in early stages or at low stress levels by considering characteristics that are not visible to the naked eye. We aimed to distinguish heavy metal stress in rice using physiological function variability that shows spatiotemporal stability based on the multitemporal fraction of absorbed photosynthetically active radiation (FAPAR). Zhuzhou City in the Hunan Province of China was selected as the study area, where Sentinel-2 remote sensing images and FAPAR data were collected. Overall, 121 Sentinel-2 remote sensing images during the April–September season from 2018 to 2020 were collected along with FAPAR data in the canopy from 25 rice paddy fields in the same region. First, the FAPAR was calculated using the biophysical processor of the Sentinel Application Platform software, and the time-series FAPAR data were determined using cubic spline interpolation. Second, different intrinsic mode function components of the FAPAR time series were derived by applying the complete ensemble empirical mode decomposition with adaptive noise at multiple scales. Third, the local spatiotemporal Moran index of the interannual and residual components was calculated. The results show that the FAPAR estimated from Sentinel-2 data suitably correlates with the measured FAPAR and can be used for detecting heavy metal stress in rice. In addition, the adopted decomposition method can successfully extract heavy metal stress signals by eliminating intra-annual and interannual components. Moreover, the local spatiotemporal Moran index of interannual and residual components correctly indicates heavy metal stress, suitably characterizing the temporal stability of spatial aggregation of heavy metal stress in rice. Therefore, combining spatiotemporal features with signal decomposition for describing a physiological function is a promising approach for identifying heavy metal stress from complex stress sources.