This manuscript presents an approach to analyzing harvest data, shows that simulations recover parameters adequately when the fitting and generating model are identical and there are relatively large sample sizes, and applies the model to data from four species and compares predictions to external abundance estimates (or an index in the case of one species). I found the manuscript to be well written and straightforward to follow and I liked the presentation of external validation data, however, I have a few reservations regarding the suggested approach (many of which are shared with similar approaches that have been presented in the literature in recent years and are by no means unique to this manuscript) and one broad comment about the structure of the manuscript:
1) I would be worried about applying this approach in a management context with the limited set of simulations presented. The bare minimum for a simulation study is to show an approach recovers parameters when the generating and fitting model are identical. A more convincing approach is to show that a model can recover parameters when there is unmodelled heterogeneity (either from a missing covariate that acts like a random effect or because of factors that lead to overdispersion in the underlying abundance) and/or that some model selection approach can recover the generating model from a set of models under reasonable sample sizes. In most real-world situations covariates are imperfect proxies, there are unknown/unmodelled sources of heterogeneity, and a savvy manager wants an approach that has been tested in these situations.
2) There is no justification provided as to why the max harvest was used as the response variable (Line 111-113) in the application. This approach was not taken in the simulations. Choosing the maximum is curious and suggest the approach would be highly dependent on the number of years available. I worry that agreement between external data and model output may be a result of this choice and that otherwise the model would underpredict abundance. Why not assume abundance is relatively constant and use variation in harvest among years to better understand heterogeneity the observation process? Or perhaps just chose one year of data and use other years as out of sample data?
3) Since this is supposed to be a general approach it makes more sense to me to start the methods with a description of the model, followed by your simulation, with the applications described last. In the results, I would start with simulation results and then present the application as the simulations need to be examined first before a reader can critically assess the application output.
4) Part of my wonders whether integrating the external data (instead of using it as out of sample) in the model is a better approach. In this case, the manuscript might argue that limited small scale intensive studies can help inferences from large scale data. I imagine the harvest data would be useful for getting at environmental drivers, whereas intensive studies would help to peg the absolute abundance and also describe the underlying heterogeneity in abundance. We did something similar in this manuscript (https://esajournals.onlinelibrary.wiley.com/doi/full/10.1002/ecs2.4240) to model population dynamics when we had a long time series of catch (count) data and 4 estimates of total population size. In your case it would just require linking the latent abundances for the pixels with external data to the estimate abundances.
A stated above, the authors have written a nice paper and I think the simulations check the bare minimum box, however I would like to see some examination of the model before I would recommend its application in an important management scenario.
charles yackulic

Nice work! I appreciate the effort the authors put into addressing my initial set of comments and have no additional concerns.
- charles yackulic