Sun-induced chlorophyll fluorescence is superior to satellite vegetation indices for predicting summer maize yield under drought conditions

Abstract

Maize is one of the most important food crops in the world and its demand has increased dramatically in recent decades. Traditional vegetation indices (e.g., normalized difference vegetation index, NDVI, enhanced vegetation index, EVI) reveal a significant lag in response to drought events, which may affect the accuracy of maize yield estimation, especially under drought conditions. Although solar-induced chlorophyll fluorescence (SIF) is recognized to be more sensitive to water stress, the potential of SIF for maize yield prediction under drought conditions requires further investigation. To this end, this study takes a major corn production region (Hebei Province) in China as an example to predict maize yield under both normal and drought years. We have utilized three satellite data (SIF, NDVI, and EVI), meteorological data, and soil data from 2000 to 2019 to predict the summer maize yield at the county level, using a machine learning (ML) method (random forest, RF) and a deep learning (DL) method (long and short-term memory, LSTM). The results showed that the overall accuracy of the predicted yield reached 90% and that the ML method was slightly better than the DL method. In normal years, the three satellite data (SIF, NDVI, and EVI) can generally predict maize yield with comparable accuracy. Nonetheless, the performance using SIF (743.53 kg/ha) was better than EVI (791.88 kg/ha) and NDVI (814.48 kg/ha) in a recognized severe drought year (2002) over the study area, indicating that SIF has a certain superiority in predicting summer maize yield under drought conditions.