Methane (CH₄) emissions from the Boreal-Arctic wetlands significantly contribute to the global CH₄ budget and are highly sensitive to climate change. Significant warming has been observed in the high latitude region during the past decades, nearly four times faster than the global average. However, the response of the Boreal-Arctic wetland CH₄ emissions to such a changing environment remains unclear. Limitations in people's understanding of dominant processes of wetland CH₄ emissions and insufficient ground observations, especially in the wetland hotspots, lead to large uncertainties and discrepancies among different process-based bottom-up and top-down models. In this study, we compiled a large in-situ observational dataset, including 139 and 168 site years of eddy covariance and chamber measurements, and generated an upscaling dataset of Boreal Arctic wetland CH₄ emissions from 2002 to 2021 using a data-driven causality-guided machine learning model. The generated upscaling dataset revealed a significant increasing trend, dominated by the CH₄ emission increase during the boreal summer (~76%). Such CH₄ increase was mainly driven by warming and greening, contributing over 90% to the CH₄ changing trend. We also found a peak emission in 2016, surpassing two standard deviations above the 2002-2021 average, triggered by the anomalously high temperature (~2°C) over the Boreal Arctic area. Furthermore, we also discussed the uncertainties caused by model parameters, wetland inundated areas, and input forcing datasets.