Figure 3.

Relationships between phenological indicators derived from remote sensing (here, date at which JRC fAPAR product reached its half-maximum value in spring and autumn) and from eddy covariance CO₂ flux time series (here, date at which the net ecosystem productivity, NEP, source–sink or sink–source transition occurred). In (a,b), relationships are shown across sites; note that ENF are offset from the 1 : 1 line in spring (a), whereas DBF fall on the line. In autumn (b), most deciduous sites fall along the 1 : 1 line, but the evergreen sites are very scattered. In (c,d), relationships are shown in the context of interannual variation in the timing of these events, i.e. anomalies relative to the site mean. In spring (c), source–sink transition date anomalies are well correlated with JRC fAPAR half-maximum date anomalies for both deciduous and evergreen forests, but in autumn (d), there is no clear relationship between the two variables. (a) Open circles: ENF (r = 0.92, p < 0.01); filled circles: DBF (r = 0.81, p < 0.01); (b) open circles: ENF (r = 0.29, p = 0.35); filled circles: DBF (r = 0.68, p = 0.04); (c) open circles: ENF (r = 0.26, p = 0.02); filled circles: DBF (r = 0.55, p < 0.01); (d) open circles: ENF (r = 0.06, p = 0.59); filled circles: DBF (r = −0.12, p = 0.35).