Identifying crop phenology using maize height constructed from multi-sources images

In agriculture, crop height is an important indicator that is commonly applied for monitoring physiological-related traits such as above-ground biomass and grain yields. Timely and precisely acquiring information on crop height at a regional scale still remains challenging, and its potential effectiveness for identifying crop phenology is under studied. In this work, unmanned aerial vehicle (UAV)-based RGB and multispectral-based images, and maize height were collected at critical growth stages in 2019, 2020, and 2021. Direct method of extracting maize height using D-value in digital surface models (DSM), and indirect methods using linear re-gressions by RGB-based vegetation indices (VIs), RGB-based texture indices, and multispectral-based VIs were separately applied to extract maize height. The results indicated that the optimal variables for extracting maize height were DSM and RGB-based VIs, and these variables were then used to construct maize height through a multi-linear regression. The multi-indicators, namely, constructed maize height, RGB-based VIs, and multispectral-based VIs were filtered using a single logistic model (SLM) and HANTS, respectively. The heading and tasseling dates of maize were identified using threshold methods and the results were compared with measured ones. The average of RMSE was 6.83 (7.14) days for constructed maize height, 10.19 (12.02) days for RGB-based VIs, and 8.02 (7.92) days for multispectral-based VIs filtered by SLM and HANTS, respectively. In conclusion, the constructed maize height well described maize's growth stages and can serve as an important complement means for extracting maize phenology compared to traditional remote sensing-formed VIs.