Hybrid MSRM-Based Deep Learning and Multitemporal Sentinel 2-Based Machine Learning Algorithm Detects Near 10k Archaeological Tumuli in North-Western Iberia

Round 1

Reviewer 1 Report

It is a very high-quality research. This article shows that the automation processes for the treatment of geographic information open a rich horizon of work to all those disciplines that habitually use georeferenced information.
I add some questions that may help to better understand the results obtained.
In section 2.2 it is not clear to me the total number of Training and Assessment data. You must specify, i think, the number of images used for Training and Evaluation that include the Data Augmentation; It should be clarified what percentage Training-Evaluation of the data has finally been used; and it should be clarified if images are added where there are no burial mounds for training.
The same happens in section 2.3. Lost burial mounds  added are, 278. Are they part of the initial 306 burial mounds? In addition to False Positives, 88. Later, two more DAs are added to retrain the model. Perhaps incorporating information from the data set used in each training attempt will help us better understand the information.

In section 2.4. Two classes are defined for the binary classification of the Sentinel-2 image, class 0 = surfaces not suitable for the presence of burial mounds and class 1 = surfaces suitable for the presence of burial mounds. For class 0 13 polygons are established and for class 1 19 polygons. What exactly do these polygons refer to? If they refer to different types of surface such as water, asphalt, cultivation, could they be listed to know what type of surface they are going to classify? Can you add a figure that can help to better understand the concept?
In section 3.4. In the end, it is estimated that the total number of burial mounds in all of Galicia is 14,626. How has this estimate been made?
These questions perhaps help the comprehension of the paper. 

Author Response

Thank you for your positive comments and for the advice on how to improve the paper. The revised version of the paper follows you advice and adds information on the training data employed in section 2.3 and 2.4. As requested, we have also explained how we calculated the total number of burial mounds in Galicia.

Reviewer 2 Report

It is well-written and factual paper about archaelogical investigations using Artificial Intelligence based strategies (both Machine Learning strategies using the well-known Random Forest classifier and DL strategies for CNN training) for the automatic detection of mounds/tumuli using DTM and multispectral satellite imagery (even if not all the band have been used), focusing also on the pre-process of the primary data and on the development of data augmentation strategies (a well-known deep learning issue, since usually this strategy requires a large amount of data). The topic (exploitation of AI in the field of archaeological and cultural heritage field) is quite new and actual, and very interesting. The goals are well exposed such as the methods and the results. Maybe the introduction/state of art could is  actually too concise and synthtetic actually and it could be increased (maybe adding some references). Even though I'm not native English speaker and I don't feel qualified to judge the grammar of the paper, it seems to me that the article is understandable and easy to read.

Author Response

Thank you very much for your positive review. We have followed your recommendation and added more references to the introduction of the paper.