Central Amazon Forest Enhanced Vegetation Index Seasonality Driven by Strongly Seasonal Leaf Flush

Abstract:

We used an RGB camera mounted 50m above an upland forest canopy to quantify leaf phenology during 12 months for 267 upper canopy tree crowns at the Amazon Tall Tower site (59.0005ºW, 2.1433ºS). Daily images under overcast sky were selected and radiometrically intercalibrated to remove any seasonal bias from incoming radiant color balance. Seasonality of crown color was then recovered for each individual crown by plotting its greenness timeline (green chromatic coordinate). We detected rapid large-amplitude positive and negative changes in greenness. Rapid increase was attributed to leaf flush and occurred in 85% of all crowns, with 80% showing a single flush per year. The theory of photoperiod control of equatorial tropical forest leaf phenology predicts two annual peaks of leaf flush, so is not supported. Rapid negative change occurred in 42% of individuals and was caused by massive pre-flush leaf abscission (31% of all trees) or other non-green pre-flushing states (11%). Crown flushing was concentrated in the five driest months (55% of trees) compared to the five wettest months (10%). Enhanced Vegetation Index (EVI) for each of three crown phenostages was obtained from a single high spatial resolution QuickBird satellite image.These phenostages were identified using only the visible bands of QuickBird so they could be related to the same crown stages seen in the RGB tower camera images. Relative frequencies of the three crown level phenostages were monitored with the tower camera, allowing a monthly estimate of landscape-scale EVI. Free of the seasonal effects on orbital sensors from clouds, cloud shadows, aerosols or solar illumination angle and corrected for seasonal change in light quality, the camera- and QuickBird derived EVI served as an independent verification of MODIS EVI seasonality. Camera-based EVI was highly consistent with view- and solar-angle corrected MAIAC-EVI of a 3x3 km footprint centered on the tower (R = 0.95 between the two monthly curves). Both reached minimum greenness in May (late wet season) and maximum during October to December (late dry to early wet season). We conclude that seasonal variation in EVI of Central Amazon forest is real and is driven by the seasonally changing mix of leaf age detected in crown-level leaf cohorts.