Seafloor data from multibeam sonar and sidescan sonar can cover an entire area (sampling frame) and is a real boon for efficient survey designs. In particular, multibeam data can facilitate the production of base maps covering tens or hundreds of square kilometres, with accurate geo-location. Multibeam data enables the survey design team to produce a design that picks out the major sources of variation in the ecosystem (typically depth and hard substrate), which can then be used to alter inclusion probabilities. To use these data one must consider how the multibeam data might be related to the variance in the target biota being sampled; – under certain circumstances it is reasonable to spend greater survey effort on hard substrate to reduce variation in parameter estimates. For example, sponge abundance will have higher variance on hard bottom than on soft bottom and so a sponge survey should disproportionally target hard bottom. Once these areas have been (accurately!) identified, then the inclusion probabilities for those regions can be increased, which will increase the chance of sampling hard substrate but maintaining the ability to infer to the sampling frame. This is the intuition in the approach that was used in Lawrence et al. (2015).

Although our recommendation is to map the survey area using multibeam prior to designing biological surveys, it is not always possible. One alternative approach, which tries to leverage as much multibeam information as possible, is to stage the sampling: perform a limited amount of multibeam mapping and work within those limited areas. Done smartly, like in Lawrence et al. (2015) this approach can still offer good estimates of biota. However, it is not without difficulties (principally in the analysis stage) and these complications could be, in some cases, overly limiting. Another possible alternative is to predict the presence and location of important habitat features, such as reefs, using statistical models (!Ref) and use the resulting habitat map to guide inclusion probabilities. We would caution, however, that such an approach will only be advantageous with a well-validated model.