The recent surge of machine learning models for wireless sensor networks brings new opportunities for environmental acoustics. Yet, these models are prone to statistical deviations, e.g., due to unforeseen changes in recording hardware or atmospheric conditions. In a supervised learning context, mitigating such deviations is all the more difficult that the area of coverage is vast. I propose to mitigate this problem by applying a form of adaptive gain control in the time-frequency domain, known as Per-Channel Energy Normalization (PCEN). While PCEN has recently been introduced for keyword spotting in the smart home, i show that it is also beneficial for outdoor sensing applications. Specifically, i discuss the deployment of PCEN for terrestrial bio-acoustics, marine bio-acoustics, and urban acoustics. Finally, i formulate three unsolved problems regarding PCEN, approached from the different perspectives of signal processing, real-time systems, and deep learning.