*3.2. Validation Results*

When using linear fitting at the grid level, 1000 cases were randomly selected to analyze the agreement between the Sentinel-2 derived cropping intensity data and the estimated cropping intensity in the two test regions. The coefficient of determination (R2) of the linear fitting reached 0.85 and 0.94 in the two test regions, respectively (Figure 10), suggesting that our cropping intensity result well captures the variations in the reference samples.

The agreement between the three cropping frequency datasets (MCD12Q2, GCI and GCI30) and estimated cropping intensity at the county level is presented in Figure 11. Their R2 all achieved 0.93, demonstrating that the performance of our method is satisfactory (Figure 11).

**Figure 9.** Visual comparison of cropping intensity in 2018; (**a**,**b**) are MODIS false-color composites (DOY 065, 145, 225); (**c**,**d**) are cropping frequency from MCD12Q2; (**e**,**f**) are the estimated cropping intensity for the test regions A and B.

**Figure 10.** Linear fitting of the Sentinel-2 derived cropping intensity and estimated cropping intensity. (**A**,**B**) are two test regions.

**Figure 11.** Linear fitting of the estimated cropping intensity and three cropping frequency datasets at the county level: (**a**) MCD12Q2, (**b**) GCI and (**c**) GCI30.
