Thomas [1985; 1991] and Thomas and Lewis [1993; 1995] develop an improvement to stratified sampling. One can obtain unbiased estimates of the load by randomizing the sampling (computer-controlled pumping samplers facilitate this method). Furthermore, one can choose sampling probabilities for each stratum so as to maximize an objective--generally to reduce the mean square error of the estimated transport. In practice, a crude model of sediment dynamics can be used to design a nearly-optimal sampling program for estimating long-term sediment load.
Thomas [1985] presents a method based on variable probability sampling, called Sampling At List Time (SALT), for which the probability of sampling is proportional to its estimated contribution to total sediment transport. Thomas and Lewis [1993] describes a stratified random sampling method where stratification is by time. Time-stratified sampling allows efficient estimation of loads when one wants estimates for multiple constituents (e.g. for both suspended sediment and for nutrients).
For estimating sediment transport from small watersheds, where the validity of regression relations may be in doubt, Thomas and Lewis's [1993; 1995] and Thomas's [1985; 1991] methods provide one of the few alternatives to nearly-continuous sampling. These methods are expected to see increased use in the future.