Despite the importance of channel avulsion in constructing fluvial stratigraphy, it is unclear how contrasting avulsion processes are reflected in stratigraphic-stacking patterns of channelized fluvial sand bodies, as a proxy for how river depocenters shifted in time and space. Using an integrated, geospatially referenced, three-dimensional data set that includes outcrop, core, and lidar data, we identify, for the first time in an outcrop study, a predictive relationship between channelized sand body architecture, paleochannel mobility, and stratigraphic-stacking pattern. Single-story sand bodies tend to occur in vertically stacked clusters that are capped by a multilateral sand body, indicating an upward change from a fixed-channel system to a mobile-channel system in each cluster. Vertical sand body stacking in the clusters implies reoccupation of abandoned channels after "local" avulsion. Reoccupational avulsion may reflect channel confinement, location downstream of a nodal avulsion point that maintained its position during development of the sand body cluster, and/or aggradation and progradation of a backwater-mediated channel downstream of a nodal avulsion point. Sand body clusters and additional multilateral sand bodies are laterally offset or isolated from each other, implying compensational stacking due to "regional" switching of a nodal avulsion point to a new, topographically lower site on the floodplain. The predictive links between avulsion mechanisms, channel mobility, and resultant sand body distributions and stacking patterns shown in our findings have important implications for exploring and interpreting spatiotemporal patterns of stratigraphic organization in alluvial basins.

Reservoir models can be improved through the incorporation of process-based models as training images for simulation with MPS. Process-based models in this study are reported to be excellent sources of training images. When combined with knowledge of depositional stratigraphic information, delta complex architectural trends can be accurately reproduced. This approach is applicable to systems that are well constrained by knowledge of depositional processes that can be efficiently synthetized to generate an appropriate process model-derived training image.