A Multifaceted Approach to Developing an Australian National Map of Protected Cropping Structures

Round 1

Reviewer 1 Report

The authors have used the UNet to map the protected cropping structures in Australia. Despite having limited novelty, the article has a valuable topic, and considering the vastness of its study region, it can be very interesting for the readers. Nevertheless, the authors should address the following issues before the possible publication of the article.

The article needs a better introduction of the used images. Currently, the authors provide a list of image sources. Providing information about how many scenes were acquired for each state and the year for each acquisition is necessary.

Please provide more detail about the implementation of the algorithm, information about the number of scenes used for training and evaluation, the processing environment, and the used parameters should be reported.

The authors claimed that they used a three-stage classification strategy. The UNet was used for the last level of classification. It is unclear which classification algorithm was used for the first two stages.

Author Response

Thank you kindly for reviewing our manuscript. We have addressed all your comments including adding additional details regarding the imagery used. In addition to more information on imagery, extra information about the processing environment and software used has been included. Parameters regarding the U-Net have already been included on lines 275 and 276. 

We have not discussed a three stage classification approach so we were not able to clarify this for you. However, the compilation of the map consisted of remote sensing imagery interpretation, ancillary data, field validation, industry engagement and deep learning. We implemented a three-level hierarchical classification to define PCS. The first level is simply PCS, second level is greenhouse and netting with the third level glasshouse, polyhouse, polytunnel, net and shadehouse. These classes were used to compile the national map of protected cropping structures. The deep learning part of the project only attempted to classify the second level of the classification - greenhouses and nets.

We hope this has satisfied the point you have raised. If not, we are happy to work with you to achieve the appropriate outcome. 

Thank you again for you review of our manuscript.

Reviewer 2 Report

This manuscript reports the results of a multifaceted approach to map protected cropping structures (PCS). Ground data, remotely sensed data and deep learning analytics were used to map the PCS in Australia. In general, the manuscript is well written for all sections with adequate details. However, I have the following specific comments;

L228: add more details about field validation (data collection time and number of observations)

Figure 16: I suggest adding the observation year to the SkySat 1, 2 and KOMPSAT

Discussion section: Please discuss why the 0.5m resolution was the optimal compromise between resolution and area coverage as shown in figure 14. 

Author Response

Thank you kindly for reviewing our manuscript. We have addressed all you comments including providing more information about when the field verification took place and the number of features which were observed in the field. Figure 16 labels now include the year for skysat and KOMSAT satellite imagery and is now consistent with the aerial photography. We have also added a discussion around why the 50cm spatial resolution is the optimal choice. 

Thank you again for you review of our paper. We hope we have satisfied the points you have raised. If not, we are happy to make additional changes.