Empirical correlations between plate tectonic setting and sand/sandstone composition have been the basis for large scale petrological models. These models do not explicitly treat sampling scale. Four areas from the western USA with diverse tectonic settings and rock types provide a natural laboratory for sampling sand at three different scales: talus piles to small drainages (first order), streams and rivers draining mountain ranges (second order), and large rivers and marine environments (third order). Existing plate tectonic petrofacies models should only be applied to third order settings because the data were derived from studies of such settings. This is especially true in tectonic settings with diverse source rocks (e.g. continental rifts and transform settings). On the other hand, some settings, such as active magmatic arcs and foreland fold-thrust belts, provide uniform results at any sampling scale because of homogeneity of source rocks.
The Rio Grande drainage area is especially complex, with diverse igneous, metamorphic and sedimentary source areas. Some components (e.g. basalt) are destroyed with minimal transport, whereas others (e.g. quartz) are relatively enriched with greater transport. In this complex continental rift setting, first and second order sand is diverse and heterogeneous due to input from tributaries. The Santa Clara River of southern California also has heterogeneous sand due to diverse source rocks in this transform setting. It is only after considerable homogenization and stabilization due to weathering and mixing with more stable components, and/or considerable transport, that homogeneous compositions are produced in these two settings.
In contrast, the Cascade magmatic arc and the Canadian Rocky Mountain fold-thrust belt have uniform source rocks (dominantly volcanic in the former and dominantly sedimentary in the latter). Uniform sand composition that is unique to each of these tectonic settings results at any sampling scale in these two cases.
Uniformity of data collection and analysis is essential for reproducible results. Use of the Gazzi-Dickinson point counting method allows direct comparison among source rocks (zero order samples), modern sand of any order and ancient sandstone of unknown provenance.
Lack of recognition of the effect of sampling scale in the development of actualistic petrofacies models has led to incorrect rejection of many existing models. Third order sands are excellent predictors of plate tectonic setting, but first and second order sands can provide ambiguous plate tectonic interpretations in many settings. More complex actualistic petrofacies models based on diverse sampling scales are needed.