Integrating SIF and Clearness Index to Improve Maize GPP Estimation Using Continuous Tower-Based Observations

Solar-induced chlorophyll fluorescence (SIF) has been proven to be well correlated with vegetation photosynthesis. Although multiple studies have found that SIF demonstrates a strong correlation with gross primary production (GPP), SIF-based GPP estimation at different temporal scales has not been well explored. In this study, we aimed to investigate the quality of GPP estimates produced using the far-red SIF retrieved at 760 nm (SIF(760)) based on continuous tower-based observations of a maize field made during 2017 and 2018, and to explore the responses of GPP and SIF to different meteorological conditions, such as the amount of photosynthetically active radiation (PAR), the clearness index (CI, representing the weather condition), the air temperature (AT), and the vapor pressure deficit (VPD). Firstly, our results showed that the SIF(760) tracked GPP well at both diurnal and seasonal scales, and that SIF(760) was more linearly correlated to PAR than GPP was. Therefore, the SIF(760)–GPP relationship was clearly a hyperbolic relationship. For instantaneous observations made within a period of half an hour, the R(2) value was 0.66 in 2017 and 2018. Based on daily mean observations, the R(2) value was 0.82 and 0.76 in 2017 and 2018, respectively. Secondly, it was found that the SIF(760)–GPP relationship varied with the environmental conditions, with the CI being the dominant factor. At both diurnal and seasonal scales, the ratio of GPP to SIF(760) decreased noticeably as the CI increased. Finally, the SIF(760)-based GPP models with and without the inclusion of CI were trained using 70% of daily observations from 2017 and 2018 and the models were validated using the remaining 30% of the dataset. For both linear and non-linear models, the inclusion of the CI greatly improved the SIF(760)-based GPP estimates based on daily mean observations: the value of R(2) increased from 0.71 to 0.82 for the linear model and from 0.82 to 0.87 for the non-linear model. The validation results confirmed that the SIF(760)-based GPP estimation was improved greatly by including the CI, giving a higher R(2) and a lower RMSE. These values improved from R(2) = 0.66 and RMSE = 7.02 mw/m(2)/nm/sr to R(2) = 0.76 and RMSE = 6.36 mw/m(2)/nm/sr for the linear model, and from R(2) = 0.71 and RMSE = 4.76 mw/m(2)/nm/sr to R(2) = 0.78 and RMSE = 3.50 mw/m(2)/nm/sr for the non-linear model. Therefore, our results demonstrated that SIF(760) is a reliable proxy for GPP and that SIF(760)-based GPP estimation can be greatly improved by integrating the CI with SIF(760). These findings will be useful in the remote sensing of vegetation GPP using satellite, airborne, and tower-based SIF data because the CI is usually an easily accessible meteorological variable.