Forty-seven 50 m × 50 m quadrats were sampled systematically for vertebrates at Litchfield National Park, northern Australia, in both 1995–96 and 2001–02. A total of 184 vertebrate species was recorded from this sampling, of which 92 species were recorded from five or more quadrats. There was substantial change in the reported species composition of these quadrats between these two periods: the mean Bray–Curtis index for similarity in species composition from the baseline to subsequent sampling of a quadrat was only 22.1 (for an index that varies from 0 for complete turnover in species to 100 for unchanged composition). For individual species, correlations across quadrats in the abundance scores from baseline to resampling varied from –0.12 to 0.85. Matched-pairs testing showed that there was significant change in abundance for 18 species from the baseline to repeat sampling, and significant increase in total bird species richness and total native mammal abundance, but significant decrease in reptile species richness.

Fire history was recorded biannually for 40 of the 47 quadrats. Fire was very frequent, with quadrats being burnt in an average of 3.65 years of the six years between fauna samples. Three aspects of this fire history (total number of years of fires, number of fires in the late dry season, and interval from the last fire to the date of resampling) were related to change in the fauna composition of quadrats. Neither the similarity in species composition, nor change in richness or total abundance of all vertebrates or of the four taxonomic classes considered (frogs, reptiles, birds and mammals) were significantly correlated with these components of the fire history of sampled quadrats. This lack of association was possibly because the monitoring period was too short to show pronounced directional change, because the system was responding to many factors other than fire, because the variations in abundance were too large and the number of samples too small to detect true associations, or because fire histories preceding baseline sampling were not considered.

The apparent instability of fauna species and communities in this system provides a considerable challenge for broad-brush (that is, vertebrate community–wide) monitoring. Power analysis demonstrated that, for most species, more than 1000 sample sites are needed to be 90% certain of detecting a 20% change in abundance, and with a 10% chance of accepting a Type I error. This level of sampling effort is commensurate with the current level of vertebrate sampling in this region. Broad-brush monitoring approaches such as described here are valuable, but need also to be complemented by more targetted monitoring for individual threatened species or species of particular management interest.