Solar-induced fluorescence (SIF) provides key information for inferring terrestrial gross primary productivity (GPP). However, there is little research on analyzing the performance of nadir-adjusted SIF (SIF_nadir) on GPP, compared to total SIF emitted by all leaves (SIF_total) for reducing the viewing angle effects and estimating GPP. Besides, there have been controversial opinions on whether SIF_total is better correlated to GPP than raw SIF observations (SIF_obs). Here we systematically analyzed the relationship between raw/nadir/total SIF (i.e., SIF_obs, SIF_nadir and SIF_total) and GPP and investigated the underlying mechanism, using multi-angular field measurements and eddy covariance data in wheat and corn crops at sub-daily scale. We further compared the performance of SIF_obs, SIF_nadir and SIF_total in estimating GPP using the TROPOspheric Monitoring Instrument (TROPOMI) SIF and concurrent AmeriFlux measurements at daily scale. The results indicate that diurnal SIF_nadir has stronger correlations to GPP than SIF_obs for field measurements, with the increase of mean coefficient of determination (R²) by 0.05-0.07 for far-red band and 0.11-0.20 for red band. SIF_nadir shows comparable performance with SIF_total for both far-red and red bands. Although the viewing angle effects dominate the difference in estimating GPP between SIF_nadir and SIF_obs, the correlation between light use efficiency (LUE) and f_esc further determines the different performance of SIF_nadir and SIF_total in estimating GPP. TROPOMI-based analysis further confirms that SIF_nadir overall has higher correlations to AmeriFlux GPP than SIF_obs for different plant functional types and shows similar performance with SIF_total. Compared to SIF_total, the estimation of SIF_nadir independent of viewing angle effects does not require any canopy structure parameters, and thus offers promising potential for reliably estimating regional and global GPP.