Relating NDVI to ecosystem CO2 exchange patterns in response to season length and soil warming manipulations in arctic Alaska

Climate change in the Arctic will differentially affect physiological rates, leaf phenology, and species composition of tundra, resulting in changing patterns and magnitudes of ecosystem CO2 flux. The normalized difference vegetation index (NDVI) provides a potential means to infer changes in CO2 flux, but whether relationships developed between NDVI and flux components can be generalized across the entire growing season and in response to changes induced by climate warming is uncertain. To investigate how well such changes might be assessed using multispectral digital images, ecosystem CO2 fluxes and NDVI were compared throughout the 2002 growing season on experimental plots with increased growing season length and soil temperature at Toolik Lake, Alaska. Season length was increased by snow removal early in the season and soil temperatures were increased using heating cables. Carbon dioxide fluxes were measured using static chamber techniques and corresponding NDVI images were taken with an agricultural digital camera. The seasonal patterns of NDVI in all treatments showed an increase to a peak in early August followed by an abrupt decline, with the snow removal plots phenologically advanced compared to the controls. Net ecosystem production (NEP) showed uptake of CO2 early in the season leveling out to a slight loss of CO2 at peak season for both control and extended season plots. Gross primary productivity (GPP) closely followed the pattern of NDVI and the pattern of ecosystem respiration (Re) mirrored that of GPP. NDVI was significantly correlated to GPP and ecosystem respiration (R2 = 0.50 and 0.36 respectively) across plots, dates, and treatments combined. However, most of the covariation was across dates. After accounting for seasonal variation, NDVI never accounted for more than 25% of the remaining variation in flux measures. Analysis of covariance showed that a given NDVI value corresponded to different flux rates on different dates and to different Re among treatments after correcting for date. The slopes of the NDVI-GPP and NDVI-Re relationships were much steeper across dates than across plots. These plot-scale results suggest that NDVI alone is not sufficient to estimate carbon flux rate responses to climate change across space or years.