Carbon dioxide (CO2) levels have been shown to be rising dramatically as a result of increased anthropogenic activity. One way of countering excessive CO2 emissions is by restoring natural ecosystems that have historically been found to be efficient carbon sinks. In order to be economically viable, these efforts must consider biomes with long-term sustained carbon sequestration capacities. Low interannual variation in this sink capacity minimizes risk of sequestration reversal. The goal of this study was to compare the interannual variability of carbon at four proximate Ameriflux eddy covariance sites across northern Wisconsin and Michigan’s upper peninsula with up to two decades of observations per site. Two wetlands (Allequash Creek (US-ALQ) and Lost Creek (US-Los)) and an unmanaged and managed forest (Sylvania Wilderness Area (US-Syv) and Willow Creek (US-WCr), respectively) were considered. To consider the fuller carbon budget for wetlands, we also incorporated stream discharge data from the United States Geological Survey. In most of the measured years, on average, NEE in both types of ecosystems was negative (carbon uptake by the ecosystem). US-ALQ and US-Los had a yearly averaged standard deviation of ~4.3 µmol CO2 m^-2 s^-1, while for US-Syv and US-WCr it was ~5.5 and ~6.3 respectively, implying greater variability for the forests than wetlands. Interannual water availability (precipitation and discharge) was the main driver for wetland carbon variation while radiation was the best predictor of carbon dynamics in the forests. Our results demonstrate that for this region, wetlands are a more reliable biome for carbon storage on a decadal scale than forests. In addition, this capacity may be enhanced through restoration efforts focusing more on water availability rather than afforestation/reforestation.