High spatiotemporal resolution time series of leaf area index (LAI) are essential for monitoring crop dynamics and validating coarse-resolution LAI products. The optical satellite sensors at decametric resolution have historically suffered from a long revisit cycle and cloud contamination issues that hampered the acquisition of frequent and high-quality observations. The 16-m/four-day resolution of the new-generation Gaofen-1 (GF-1) and Gaofen-6 (GF-6) satellites provide an unprecedented opportunity to address these limitations. Here, we developed an effective strategy to generate daily 16-m LAI maps combining GF-1/6 data and ground LAINet measurements. All high-quality GF-1/6 observations were utilized first to derive smoothed time series of vegetation indices (VIs). Second, a random forest regression (RF-r) model was trained to link the VIs with corresponding field LAI measurements. The trained RF-r was finally employed to generate the LAI maps. Results demonstrated the reliability of the reconstructed daily VIs (relative error (RE) < 1%) and the derived LAI time series, which greatly benefited from GF-1/6 high-frequency observations. The direct comparison with field LAI measurements by LAI-2200/LI-3000 showed the good performance of retrieved LAI maps, with bias, root mean square error (RMSE), and R2 of 0.05, 0.59, and 0.75, respectively. The LAI time series well captured the spatiotemporal variation of crop growth. Furthermore, the continuous GF-1/6 LAI maps outperformed Sentinel-2 LAI estimates both in terms of temporal frequency and accuracy. Our study indicates the potential of GF-1/6 to generate continuous decametric-resolution LAI maps for fine-scale agricultural monitoring.

https://doi.org/10.1109/tgrs.2023.3257290