This paper presents our new results on lucky ranging utilizing a towed array in environments subject to unknown fluctuating spatial coherence losses. We derive a lucky maximum likelihood range estimator based on the probabilistic assumption that each collected data snapshot is either coherent or purely incoherent with some probability. Our lucky range estimator can be interpreted as first ranking the coherence quality of each data snapshot according to an array gain-like quantity during the parameter search, followed by accumulation of likelihood surfaces out of data snapshots of high spatial-coherence. This effectively avoids the wash-out or the smearing results encountered in the traditional processing procedures of utilizing a long integration time without a prior screening for the data spatial-coherence. An important advantage of the lucky approach is that it makes no prior assumptions about the signal spatial coherence loss model. This estimator has greatly improved robustness over the conventional estimator when coherence is low and time-varying.