Advances in sensing and computation have accelerated at unprecedented rates and scales, in turn creating new opportunities for natural resources managers to improve adaptive and predictive management practices by coupling large environmental datasets with machine learning (ML). Yet, to date, ML models often remain inaccessible to managers working outside of academic research. To identify challenges preventing natural resources managers from putting ML into practice more broadly, we convened a group of 23 stakeholders (i.e., applied researchers and practitioners) who model and analyze data collected from environmental and agricultural systems. Workshop participants shared many barriers regarding their perceptions of, and experiences with, ML modeling. These barriers emphasized three main areas of concern: ML model transparency, availability of educational resources, and the role of process-based understanding in ML model development. Informed by workshop participant input, we offer recommendations on how the ecological modelling community can overcome key barriers preventing ML model use in natural resources management and advance the profession towards data-driven decision-making.