Sociotechnological Sustainability in Pasture Management: Labor Input and Optimization Potential of Smart Tools to Measure Herbage Mass and Quality

Comment 1: This paper evaluates various tools for measuring herbage mass to identify labor and workflows for improved human input efficiency through automation. This paper is well written and has a novel and under evaluated foci. Although some major concerns are included directly below.

Response: We appreciate your constructive feedback and are glad that we get the chance to improve our manuscript based on all the comments. Your comments were very helpful in further improving the manuscript and in making the study repeatable. We agree that we did not elaborate enough on the “smartness” of the used technologies. We tried our best in responding to your comments. Please find our responses below.

Comment 2: Barriers to entry and use are not addressed. Typically, the RPM is used by producers but it unclear how NIRS and UAV could be adopted given the library/database development requirement, postprocessing, and technology use requirements for each system. Unclear scenario(s) in which these 'smart' farming practices could be implemented.

Response: We fully agree, that we elaborated too little on the implementation of the tools, as well as on the term “smart farming tools”.

We changed the manuscript in the way that we defined the term “smart farming tools” more clearly. In fact, all of the smart farming tools we investigated have an integrated prediction algorithm (RPM) or more than one (UAV and NIRS). In addition, they are all operated in combination with a smart device (phone or tablet) or a computer. This user interface is crucial to use and understand the application fast, to visualize the measurement data and interpret it. This way farmers get measurements in nearly real-time and on-farm. Therefore, we think the term “smart farming tool” is well suited. We incorporated the mentioned aspects in the Introduction and the Materials and Methods Section.

In addition, we elaborate more on the development stage (in case of the UAV) and the commercialization and re-calibration of the NIRS in the Introduction, so that the reader gets an idea of the stage of implementation of these tools in practice. Although, satellite imagery is already used to predict herbage growth in Australia, using UAVs is not yet commercialized (as far as we know), but has the advantage of having a higher image resolution compared to satellite imagery and being more independent of weather conditions (particularly cloudy skies). Even though the UAV approach might be a technique of the future, it deserves to be under study when it comes to labor input and work flows. Studying such topics in early development stages, may help to better commercialize the techniques and thus also increase adoption in farming practice.

Postprocessing of multispectral images is detailed in Section 2.1.2 Unmanned Aerial Vehicle. However, we extended the section a bit more when it comes to how the technology is applied by explaining that two separate apps had to be used to operate the UAV and to operate the multispectral camera. This will be easier in the next generation of the quadcopter.

Changes made: Lines 73-74, 78-79, 208-210; lines 81-82, 90-95; lines 170-172.

Comment 3: In reality, NIRS and RPM and not smart farming practices. The former is an analytical tool for identifying, in this case, forage composition and quality, the latter a way of estimating mass in situ. The spectral imaging is the only smart farming practice that was actual tested. None of these methods are actually measuring the same things and are not comparable.

Response: A part of this comment is already addressed in our response above. Based on your comment, we assume it was not clearly enough written that the NIRS we investigated was a mobile tool that could be used on farm or in the boot of a car and allowed for near real-time analysis of un-dried herbage. We integrated this information in section “2.1.4 Cut Samples and Near-Infrared Reflectance Spectrometer”. The “smart” thing about the RPM is also included now (i.e. it automatically converts compressed sward height into herbage mass and visualizes the grass wedge), as mentioned in the response above.

We disagree with the forth sentence of your comment: the tools have in common that they all determine herbage mass. However, the NIRS does not weigh the cut samples but it “only” determines dry matter concentration which is necessary to estimate herbage mass as kg DM per ha. We found this not clearly explained in the manuscript and improved the description in section “2.1.4 Cut Samples and Near-Infrared Reflectance Spectrometer”.

Changes made: Lines 209-211, 78-79, 214-240

Comment 4: After reading the paper, I am largely concerned that actual herbage mass, composition, or quality was not measured. It is unclear how one can identify a superior technique without properly or adequately measuring forage composition and quality via the status quo.

Response: We can assure you that actual herbage mass and composition was measured during our work observations. The work observations were performed during an evaluation study, where the accuracy and precision of the tools was studied. The results are published by Hart et al. [27] (we now included this information in the text). However, your assumption was right, that some of the tools are not yet accurate enough for practical use by farmers, and are constantly developed further. Not only the prediction models for multispectral imagery are evolving but also the calibrations for the NIRS are constantly updated with broader training data from all over the world. We have made adjustments in the text and communicate these aspects in a transparent manner.

Changes made: Lines 81-83, 92-95

Comment 5: This experiment had an extremely low sample size, with no replication. Further, methodological details were lacking (time of sampling, forage type, fertilization, sampling frequency, experimental design, data analysis, etc.). Thus, making this experiment non-repeatable.

Response: For the specific method used, the work element method, a sample size of 3-5 users with 2-5 repetitions each, is a valuable sample size (for the sup-operation of “Image processing” we measured only one user because the work procedure was mainly determined by machine time, here more specific computer processes. As we have an extensive database of work elements, our measurement of image processing matched the work element times already stored in the database. Therefore, there was no need to measure another user). We set out to communicate the data set as openly and transparent as possible (i.e., in Table 2), and show in the appendix the sample size of each work element used (Table A3). In addition, we were able to use work elements that were well validated by work experiments for modeling (Table A2). The limited number of work observations is unfortunately related to the fact that the investigated tools are not yet applied in broad agricultural practice. We tried to point this out in lines 120-122 “No work diaries or survey data were available for two tools investigated, because they were still relatively new and have not yet been widely applied in practical agriculture.”

Regarding the methodological details, we agree that some information was missing to be able to reproduce the experiment. Therefore, we added information on location of the farms, the users, and the grasslands in section “2.2.3 Time Measurements”. The data analysis is well described in the same section, but we are happy to include further information if you could provide us with the missing points. The sampling frequency (regarding work observations) is given in Table 2, but if you refer to the sampling frequency of herbage measurements the answer is that we observed the users on occasion. It means that even if users measured grasslands once every week throughout the vegetation season, we only joined them 3-5 times dependent on if a meeting could be scheduled.

Changes made: Lines 317-325

Comment 6: Authors hypothesized that one sample per paddock was sufficient to make a statement about herbage mass and quality. This is a terribly erroneous assumption. Given that these are not issues that can be augmented or altered in this study, it was suggested that this paper be rejected.

Response: We are aware of the accuracy and precision problem when measuring only one spot per paddock. For scientific purposes it would be unaccountable to do so. For practical application, we based our assumption on the recommendations of Teagasc, the state agency providing research, advisory and education in agriculture, horticulture, food and rural development in Ireland, who propose one sample as a minimum for farmers to measure. However, we fully agree with you and therefore adapted the discussion in the way that a minimum of two samples is required in order to get a more valid herbage mass estimate. This new assumption is based on the papers of Nakagami who proposed the use two samples per paddock, the one with the lowest herbage mass and the highest herbage mass for on-farm purposes: (1) Nakagami, K. 2015. A method for approximate on-farm estimation of herbage mass by using two assessments per pasture. Grass and Forage Science, 71, 490-496. (2) Nakagami, K. 2019. A simple variance estimation of herbage mass based on two assessments per pasture, JARQ 53 (3) 207-214. doi: 10.1111/gfs.12195.

We adapted the labor time model, created new figures and changed labor time estimates in the text. The labor time requirement for cutting samples is now 152.1 instead of 101.4 manpower minutes (under the conditions of the standardized farm). We included Nakagami [45] as a reference.

Regarding your comment that a change from one to two samples cannot be done, we use a modelling approach based on work elements. It is therefore easy to adapt the model to any number of cut samples.

Changes made: Figures: 2, 3, and 5; Material and Methods: lines 214-224, 303, 312-313, 399-402; Results: lines 427, 430, 435-438, 450-451,486, 528, 535-536; Discussion: 621-639, Conclusions: lines 691-715; References: expanded and new numeration.

Comment 1: How long does it take to learn and apply the described methods?

Response: Thank you for your review, the positive feedback and constructive advice. Below we answer the comments point by point.

Regarding your first comment: The use of the RPM is very self-explanatory. There is a video tutorial for the tool on Youtube and a user manual is included. A user who can operate a smartphone and establish a Bluetooth connection can use the tool directly after a short installation phase (max. 30 min). 

For the HarvestLab (NIRS) and the UAV, training by an experienced person is advisable. For the HarvestLab, the distributor of the tool provided us the service that a product manager presented the functions of it and we analyzed sample feeds during a hands-on training session with a duration of 3 hours. For the UAV, we used a consumer drone. Its use can also be learned on the basis of user manuals and tutorials. However, we had a separate multispectral camera from another manufacturer that had to be attached to the UAV and operated separately with the smartphone. This has already changed by now. There are drones from DJI, for example, that carry a multispectral camera for use in agriculture: https://www.dji.com/ch/p4-multispectral. We have added this to the paper in the introduction section, and we mention the trainings in the discussion section.

Learning to use and apply a tool is not included in the calculation in labor time models because it is a one-time activity and depends strongly on the skill and interest of the tool user.

Changes made: Lines 171-173, lines 646-647

Comment 2: How many kilos of milk are needed to keep each sampling method operational.

Response: Unfortunately, we did not do a full cost analysis to calculate the economics of a precision grazing management using the tools we studied. The smart farming tools cost around 32.000 Euros (NIRS), 1500 Euros (RPM) and 5000 Euros (UAV with camera).

There is a study from New Zealand that found 15% increase in profitability when performing herbage measurement regularly (Beukes, P. C., et al. (2018). "Regular estimates of herbage mass can improve profitability of pasture-based dairy systems." Animal Production Science 59: 359.) but they did not investigate different methods of measuring herbage and herbage measurements focused on determining herbage mass and not herbage quality (which in our case makes the NIRS and UAV methods so expensive).

However, we would like to point out that development of smart farming tools is an ongoing and fast process. Therefore, it is not helpful to run an economic analysis for tools which are in development.

Changes made: None

Comment 3: The NIRS device was not planned to predict forage mass, but it has several prediction equations implemented for chemical components of the forage plant that can be used to predict forage mass.

Response: This comment was very helpful, because we realized that it was not clearly described that the algorithm of the NIRS for determining DM% was necessary to measure herbage mass. You were absolutely right with your assumption that the NIRS was needed to determine herbage mass and not only nutrient concentration. Therefore, we changed paragraph “2.1.4 Cut Samples and Near-Infrared Reflectance Spectrometer” to explain that DM% was determined before the grass wedge was created.

Changes made: Lines 227-240

Comment 4: Data in Figure 4 deserves publication as it clearly shows the relationship between the number of cut samples and labor time.

Response: Yes, we hope we are allowed to publish in “Sustainability”. Regarding this relationship, there were no such figures published before, to the best of our knowledge.

Changes made: None

Comments 5 and 6: I suggest the authors read the following papers:

- Nakagami, K. 2015. A method for approximate on-farm estimation of herbage mass by using two assessments per pasture. Grass and Forage Science, 71, 490-496.

- Nakagami, K. 2019. A simple variance estimation of herbage mass based on two assessments per pasture, JARQ 53 (3) 207-214. doi: 10.1111/gfs.12195.

A single sample per paddock must create precision and accuracy issues. You must use at least two, even though it'll increase sampling times.

Response: Thank you for this piece of advice in Comments 5 and 6. We have read the papers, and they were very valuable. Of course, we are aware of the accuracy and precision problem when measuring only one spot of a paddock. For scientific purposes it would be unaccountable to do so. For practical application, we based our assumption on the recommendations of Teagasc, the state agency providing research, advisory and education in agriculture, horticulture, food and rural development in Ireland, who propose one sample as a minimum for farmers to measure. Similar as to what Nakagami explains, it is common knowledge in grazing research that there is the need to measure herbage but for farmers it has to be time efficient or they don’t start implementing the techniques. However, we fully agree with you and therefore adapted the discussion in the way that a minimum of two samples is required in order to get a more valid herbage mass estimate. Therefore, we adapted the labor time model, created new figures and changes labor time estimates in the text. The labor time requirement for cutting samples is now 152.1 manpower minutes instead of 101.4. under the conditions of the standardized farm. We included Nakagami [45] as a reference.

Changes made: Figures: 2, 3, and 5; Material and Methods: lines 214-224, 303, 312-313, 399-402; Results: lines 427, 430, 435-438, 450-451,486, 528, 535-536; Discussion: 621-639, Conclusions: lines 691-715; References: expanded and new numeration.

Comment 7: Data in Figure 5 deserves publication.

Response: Thank you very much for pointing this out.

Changes made: None

Comment 8: The authors show the sampling procedures plainly, with their advantages and disadvantages focused on the use of working time. For me, there is no bias towards any.

Response: Our sincere thanks!

Changes made: None