The capacity of coral reefs to provide ecosystem services, to keep their diversity and their productivity are related to their three-dimensional structural complexity. This parameter is also correlated to total fish biomass, to the general reef resilience to external stresses and to their ability to dissipate wave energy. However, information on structural complexity (also defined as reef rugosity) has been uncommonly assessed in historical monitoring programs, with the result that the long-term trend of this variable is generally unavailable. In this study, we show that it is possible to predict and hindcast the three-dimensional complexity of coral reefs by combining photogrammetry, statistical modeling and historical benthic community data. We calibrated a lasso generalized linear model to predict structural complexity obtained by 57 photogrammetry transects recorded at 13 sites around the island of Moorea (French Polynesia). Our model was able to predict structural complexity with high accuracy (cross-validated R2 = 0.81 ± 0.12). We then used our model to hindcast historical trends in 3D structural complexity using community composition data collected in Moorea from 2004 to 2017. The temporal analysis highlighted the dramatic effect of a crown-of-thorns outbreak in 2006-2009 and Cyclone Oli in 2010. These two events together reduce coral cover from ~50% to almost zero. Our model captured these effects, confirming the possibility to predict structural complexity on the basis of assemblage composition.