Objectives The accurate estimation of leaf chlorophyll content of jujube can not only reflect its growth and nutritional status, but also provide scientific basis for field management. We aim to evaluate the potential of Sentinel-2A data for estimating soil plant analysis development (SPAD) values of jujube leaves during leaf spreading, fruit setting stage and fruit ripening stage.

Methods Five traditional vegetation indices related to chlorophyll content were selected, and based on the framework of five traditional vegetation indices, ten bands of Sentinel-2A data were used to improve the traditional vegetation indices, and five spectral indices were constructed for three growth periods respectively. The correlation between the constructed spectral index and the measured SPAD value was analyzed, and the optimal spectral index was screened by the correlation coefficient. Based on the optimal spectral index, multiple linear stepped-regression model (MLSR), support vector machine regression model (SVR) and random forest regression model (RFR) were used to establish SPAD estimation models. The coefficient of determination R² and root mean square error (RMSE) were used as model evaluation indexes, and the optimal model for estimating the SPAD value of jujube leaves was screened out through model evaluation.

Results The results show that the five optimal spectral indices optimized at three growth stages are mainly composed of red band, red-edge band and near-infrared band, and the five optimal spectral indices at fruit ripening stage have the highest correlation with SPAD value, all of which pass the significance level test of 0.01, and the absolute values of correlation coefficients are all greater than 0.37. The accuracy of estimation models established at different growth stages is different, and the accuracy of estimation at fruit setting stage is the worst. The estimation accuracy of leaf spreading stage and fruit ripening stage varies according to the models. The accuracy of fruit ripening stage is the highest based on MLSR and SVR, while the accuracy of leaf spreading stage is the highest based on RFR. The RFR model of leaf spreading stage is the best model among all the estimation models, R² and RMSE are 0.90 and 1.04, respectively. Among the three regression models, MLSR, SVR and RFR, the estimation results of MLSR and SVR are similar, and RFR is the best estimation model, and the best estimation model has strong universality under different vegetation cover scenarios.

Conclusions The above results show that Sentinel-2A data is suitable for SPAD estimation of jujube leaves, and the RFR model at leaf spreading stage can be used as the optimal model for SPAD estimation of jujube leaves. The results can provide an important reference for the study of estimating SPAD value of jujube based on Sentinel-2A data.