In general, this study presents a novel method for the calibration of a flood simulator using binary flood extent observations. The authors compare their method with the most common method (GLUE) used for calibration and a simpler Bayesian model. The methodology demonstrated using the raster-based Lisflood-fp model. The study area is a short reach on the upper river Thames in Oxfordshire, England. In general, the entire manuscript is well-written, well-referenced, and well-structured. The approaches followed by the authors have novel parts and are very interesting for the hydrological community. Finally, the results are sufficient and support the interpretations and conclusions. 

The manuscript analyzes and compares three broad statistical approaches to learn 2D-flood model parameters from binary observations of flood extent while accounting for various sources of uncertainty. The methodology and the scientific motivation behind this comparative analysis are clearly explained. While there is a large body of scientific literature in hydrology on the effects of using various likelihood functions for time series data, this manuscript fills a critical gap related to our understanding of 2D-flood models and corresponding appropriate likelihood functions/error descriptions for inference using spatially-distributed binary data. Even though simulations are run on only a single flood event, the manuscript provides several valuable and generalizable insights from its results and discussion, which are especially relevant for the researchers working on uncertainty analysis for such hydrologic models. I congratulate the authors for this work and recommend publication after minor revisions (please see the highlighted manuscript for comments).