Low-Cloud Fraction and ECVs

Although the above section describes climatic shifts in some ECVs, it is unclear how these variabilities are associated with ΔCF. Our PLSR models evaluated previous results, indicating that ΔCF is strongly related to changes in ECVs (Fig. 4 ). Overall, eight components were required as optimum to explain the ΔCF variability (Fig. 4a ), which was equal to the number of predicting variables. Using this number of components, the PLSR models could predict 55 ± 0.67 % of the ΔCF variance (Fig. 4b ). Such models provided a mean RMSE of 10.16 ± 0.07 ×10-4 CF year-1 (Fig. 4c ) with a relative RMSE close to 8.76 ± 0.06 %. The VIP of these models also revealed that ΔVSWC, all surface temperature trends (i.e., average, min, max), and ΔPressure are highly important for predicting ΔCF among ECVs (Fig. 3d ).