In this work, four different algorithms (fast Fourier transform FFT, short-time Fourier transform STFT, continuous wavelet transform CWT, and instantaneous frequency IF) for calculating median frequency (MDF) from surface EMG signals were investigated for studying muscle fatigue during a on-water rowing training. The study protocol included 5 consecutive parts with increasing stroke rate. Six athletes participated in the study aged 36.6+-14.6 years and a rowing experience of 6 to 35 years. We considered eight muscles: biceps brachii right, biceps brachii left, latissimus dorsi right, latissimus dorsi left, erector spinae right, erector spinae left, rectus femoris and biceps femoris. By applying Friedmann test, we found a significant difference in MDF behavior between algorithms in assessing muscle fatigue (p < 0.05). Correlation analyses showed significant correlations between muscle activity duration t{textit{act}} and MDF, which differs for the four considered algorithms and should be taken into account in further experiments. With CWT showing the smallest correlation to t{textit{act}} it might be more robust against time window variations. Our study provides a basis for the development of improved methods for more robust, non-invasive, and continuous detection of muscle fatigue in experiments with dynamic on-water rowing study designs.