Spatial-aware SAR-optical time-series deep integration for crop phenology tracking

Abstract

Accurate crop phenology information is essential for precision farming and agricultural productivity improvement. In recent years, in-situ equipment on crop phenology observation has been boosted, which generates high-quality real-time pictures on capturing vegetation phenological changes. However, due to the limited number of ground sites, it is impossible to measure large-scale crop phenology with local observations. Complementary, the freely available Sentinel satellites with high revisit frequency provide an opportunity to map accurate crop phenology at an unprecedented fine spatial scale. Because of the differences in viewing angle and range, the consistency of crop phenological stages varies between satellite and ground observations. To fill the gap between satellite and ground observations, we developed a spatial-aware scheme to integrate SAR and optical time-series data for accurate crop phenology tracking. To be specific, we propose a new deep learning model called Deep-CroP framework to improve the alignment between satellite and ground observations on crop phenology. The experiment results on selected ground sites demonstrate that the proposed Deep-CroP is able to accurately identify crops phenology and narrow the discrepancies from 30+ days to as high as several days. In addition, we applied the Deep-CroP to large-scale Sentinel time-series to map spatial patterns of phenology at fine resolution imagery on two study areas (i.e., TA1 and TA2). In general, the potential of satellites time-series for ground-level crop phenology observation is verified. Also, the consistency between satellite and PhenoCam observations is expected to be further improved.