Wetlands are the single largest source of methane to the atmosphere and is predominantly considered natural. Although, a portion of wetland methane and nitrous oxide emissions are now considered ‘indirectly anthropogenic’, primarily due to amplified emissions from, e.g., eutrophication, or from the creation/restoration of waterbodies. Rewetting or restoration of previously drained wetland environments is a common policy in the Canadian landscape to mitigate nutrient runoff from agricultural areas, but the potential for greenhouse gas emissions is an understudied tradeoff of this policy. Moreover, aquatic field studies are typically limited in quantity of sampled systems because of logistical constraints, like travel, labor, or funding, but field data is necessary to more accurately upscale and/or validate emissions at regional and continental scales. Here we present results from comprehensive greenhouse gas (CH4, N2O, CO2) sampling at seven restored wetlands of Southern Ontario, but only six times throughout the year. Along with a suite of environmental and water chemistry parameters, we used Random Forest modeling to determine the best predictors of emissions from this small dataset. We found that relatively simple variables such as dissolved oxygen, pH, and certain nutrient species were the best predictors. This machine learning approach helped to create an environmental model from our small dataset that can be tested with future data and allowed us to constrain the most efficient parameters to sample when looking to increase spatial coverage of small systems and upscale emissions.