Location: Hydrology and Remote Sensing LaboratoryTitle: Towards scalable within-season crop mapping with phenology normalization and deep learning
|YANG, Z. - University Of Illinois|
|DIAO, C. - University Of Illinois|
Submitted to: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/13/2023
Publication Date: 1/16/2023
Citation: Yang, Z., Diao, C., Gao, F.N. 2023. Towards scalable within-season crop mapping with phenology normalization and deep learning. Geoscience and Remote Sensing Letters. 16:1390-1402. https://doi.org/10.1109/JSTARS.2023.3237500.
Interpretive Summary: Remote sensing data have been routinely used to map crop types after harvest. However, many applications require the crop type map in the early growing season. It’s challenging to map crop types during the growing season due to the lack of training samples from the current year. This paper proposed a new within-season emergence (WISE)-phenology normalized deep learning model using training samples from previous years. The new method was assessed and compared to the two traditional methods in Illinois from 2017-to 2020. Results show that the WISE-phenology normalized deep learning model can be extended to different years. The new model demonstrates superior performances both within the growing season and at the end of the growing season. A crop map from the early growing season provides timely information for crop acreage estimation and production prediction, which is essential for the food market and agricultural statistics.
Technical Abstract: Crop-type mapping using time-series remote sensing data is crucial for a wide range of agricultural applications. Crop mapping during the growing season is particularly critical in the timely monitoring of the agricultural system. Most existing studies focus on within-season crop mapping leverage historical remote sensing and crop type reference data for model building, due to the difficulty in obtaining timely crop type information for the current growing season. However, the diverse patterns of crop phenology across years and locations hamper the scalability and transferability of the model to the current season for timely crop mapping. In this study, we propose an innovative within season emergence (WISE)-phenology normalized deep learning model towards scalable within-season crop mapping. The crop time-series remote sensing data are first normalized by the WISE crop emergence dates before being fed into a one-dimensional convolutional neural network (1D-CNN) classifier. Compared to conventional approaches, the WISE-phenology normalization approach largely helps the deep learning crop mapping model accommodate the spatiotemporal variations in crop phenological dynamics. Results in Illinois from 2017 to 2020 indicate that the proposed model is more scalable and robust in that it requires less training data and can be transferred to different years with better performances both within the growing season and at the end of the season.