Technologies for Forecasting Tree Fruit Load and Harvest Timing—From Ground, Sky and Time

Round 1

Reviewer 1 Report

The authors prepared an excellent structured overview of tree fruit load forecasting and harvest timing methods. The review is extensive, and the examples have been chosen very well.
Comments:
Different methods are required to develop orchards with fruit trees in different parts of the world. An exciting direction would be to analyze the factors specific to a given geographical area and the most popular fruit variety developed for this area and propose a methodology considering local variables conditioning the optimal development of a given variety. It's easy to understand that such analysis is beyond the scope of this review but would be a great start to developing best manufacturing practices.

Rather than describing entirely mechanistic models in predicting tree fruit, this review focuses on using empirical models based on direct counting, manual or machine vision, or correlation with environmental conditions or tree crown attributes. The authors also cite examples with mechanistic models in some parts of the summary. This is a good decision because such an approach gives more comprehensive coverage of the various analyzed fruit varieties and more objective analyses.

An exciting development would be to describe the optimization methods and analyze them over a broader period using data from many seasons within one orchard. Only such long-term analyzes would give a complete picture of the analyzed phenomenon, which is the development of fruit orchards.

In correlation analysis, where research is often based on indicators, such as RENDVI1, it is worth trying to compare several alternative indicators.
The authors rightly point out that several elements are necessary to maintain methodological correctness in growing fruit trees. One such factor is the rigorous use of statistically correct randomization during sampling.
It was also rightly pointed out that after the weaknesses (e.g., occlusion) have been resolved, and computer vision methods could be effectively used for fruit counting and yield mapping.
The paper also addresses several different yield models, climatic variables, and vegetation indices.
As the authors rightly stated in summary, IoT is a trend that has been developing very dynamically in recent years and can help a lot in terms of the discussed topic. It would be worth adding literature studies on this subject; it would also enrich the summary with knowledge on the use of IoT to better control fruit trees' development.

Author Response

Reviewer 1:

The authors prepared an excellent structured overview of tree fruit load forecasting and harvest timing methods. The review is extensive, and the examples have been chosen very well.  <<Thank you!

Comments:
Different methods are required to develop orchards with fruit trees in different parts of the world. An exciting direction would be to analyze the factors specific to a given geographical area and the most popular fruit variety developed for this area and propose a methodology considering local variables conditioning the optimal development of a given variety. It's easy to understand that such analysis is beyond the scope of this review but would be a great start to developing best manufacturing practices.

<< We agree that there is a synergy or inter-twining between technology and agronomy, in this case forecast technologies and orchard structure. Given current length and scope of the review, we have limited ourselves to an addition in the Conclusion of potential for practical advances in this area, as follows:

“Optimal orchard design is a complex interplay of sociological, economic, political and technological factors. This interplay results in need for local optimisations. General trends are, however, evident in the adoption of systems that reduce labour requirements while increasing production and/or facilitate mechanical management, e.g., smaller trees, whether by dwarfing or training. The forecast technologies discussed in this review represent one element in this puzzle. The choice of technology will also require local optimisation on the basis of commodity, production condition, e.g., protected cropping or open field, and market need, e.g., export or local farmers market.”

Rather than describing entirely mechanistic models in predicting tree fruit, this review focuses on using empirical models based on direct counting, manual or machine vision, or correlation with environmental conditions or tree crown attributes. The authors also cite examples with mechanistic models in some parts of the summary. This is a good decision because such an approach gives more comprehensive coverage of the various analyzed fruit varieties and more objective analyses.   <<Again, thank you.

An exciting development would be to describe the optimization methods and analyze them over a broader period using data from many seasons within one orchard. Only such long-term analyzes would give a complete picture of the analyzed phenomenon, which is the development of fruit orchards.

<< It is exciting!....but again in context of current length and scope we have limited ourselves to addition of a ‘pointer’ in this direction, inserted at li 34 (Section 1.1.), as follows:

“Improved monitoring also offers great potential for accumulation of reliable data across time. An experienced agronomist will be able to utilise such a resource, combining this information with knowledge of fruit and tree physiology to provide current season advise and to advise on management practices to optimize orchard development. “

In correlation analysis, where research is often based on indicators, such as RENDVI1, it is worth trying to compare several alternative indicators.

Agreed.  The following text has been added at li 778:

“There are many indices that can be trialled, utilizing visible, far red, or thermal bands.”

Mention is made of the use of other indices is also made in line 804.  Of course, if too may indices are trialled with too few data, then there is risk of overfitting to the data, i.e., finding a spurious correlation, so there is also need for appropriate validation.

The authors rightly point out that several elements are necessary to maintain methodological correctness in growing fruit trees. One such factor is the rigorous use of statistically correct randomization during sampling. <<Agreed. (no change needed)
It was also rightly pointed out that after the weaknesses (e.g., occlusion) have been resolved, and computer vision methods could be effectively used for fruit counting and yield mapping.
The paper also addresses several different yield models, climatic variables, and vegetation indices.
As the authors rightly stated in summary, IoT is a trend that has been developing very dynamically in recent years and can help a lot in terms of the discussed topic. It would be worth adding literature studies on this subject; it would also enrich the summary with knowledge on the use of IoT to better control fruit trees' development.

<< We agree!  But again, for reasons of length and scope we limit our response to an addition to the Conclusion as follows:

“As reviewed by Balafoutis et al. [121], there are many sensors available that can plug into such a network to fine tune estimates of harvest timing and quality, from soil and plant water status sensors to dendrometers, and a developing capacity to use deep learning approaches in the use of this data.”, citing:

Balafoutis, A.T.; Evert, F.K.V.; Fountas, S. Smart Farming Technology Trends: Economic and Environmental Effects, Labor Impact, and Adoption Readiness. Agronomy 2020, 10, 743. https://doi.org/10.3390/agronomy10050743

Reviewer 2 Report

This is a very nice manuscript that presents a very thorough review of the current knowledge in the field of yield forecasting in tree crops. I have only the following minor corrections and suggestions:

Table 4 should be reordered according to commodity or year. Please indicate if images were captured on-the-go or in stop-and-go fashion. Table 5 should also be re-ordered according to commodity

In Section 2.5.1 it would be worthwhile to mention that some trees have a strong natural two-year cycle and in such cases there is no relationship between tree characteristics alone and yield. Also, it should be mentioned that chemical thinning or other operations that do not impact tree size or shape can have a tremendous impact on yield, so that one-to-one relationships between tree characteristics such as volume or shape and yield are unlikely to exist. Finally, since fruits evolve from flowers, a number of studies have attempted to predict yield or at least fruit load from the number of flowers or blooming intensity. Although personally I believe too many factors will affect tree/fruit development between flowering and picking, I recommend mentioning this approach. For instance:

B Braun, D .M Bulanon, J Colwell, A Stutz, J. Stutz, C Nogales, T Hestand, P Verhage, T. Tracht (2018)

A Fruit Yield Prediction Method Using Blossom Detection. 2018 ASABE Annual International Meeting  1801542.(doi:10.13031/aim.201801542)

Trevor Braddock, Duke M Bulanon, Brice Allen and Joseph J Bulanon. Predicting Fruit Yield Using Shallow Neural Networks. https://www.preprints.org/manuscript/202009.0022/v1

At the end of Section 2, in the paragraph starting “Brinkhoff and Robson [87] also”, lasso should be LASSO

The caption Figure 9 is rather confusing. Is this figure presenting the histogram of estimated fruit weight? This figure does not provide information about the accuracy of the relationship between fruit weight, width and length, and I am not sure how it supports the discussion presented in the preceding paragraph.

Section 6.1.2: A reference to the following work should be added:

Dihua Wu, Shuaichao Lv, Mei Jiang, Huaibo Song (2020) Using channel pruning-based YOLO v4 deep learning algorithm for the real-time and accurate detection of apple flowers in natural environments. Computers and Electronics in Agriculture 178, 105742

Author Response

Reviewer 2:

This is a very nice manuscript that presents a very thorough review of the current knowledge in the field of yield forecasting in tree crops. <<Thank you.   I have only the following minor corrections and suggestions:

Table 4 should be reordered according to commodity or year. <<Entries were ordered chronologically, but are now re-ordered chronologically within commodities.  Please indicate if images were captured on-the-go or in stop-and-go fashion. << Done   Table 5 should also be re-ordered according to commodity <<Done 

In Section 2.5.1 it would be worthwhile to mention that some trees have a strong natural two-year cycle and in such cases there is no relationship between tree characteristics alone and yield. Also, it should be mentioned that chemical thinning or other operations that do not impact tree size or shape can have a tremendous impact on yield, so that one-to-one relationships between tree characteristics such as volume or shape and yield are unlikely to exist. Finally, since fruits evolve from flowers, a number of studies have attempted to predict yield or at least fruit load from the number of flowers or blooming intensity. Although personally I believe too many factors will affect tree/fruit development between flowering and picking, I recommend mentioning this approach. For instance: B Braun, D .M Bulanon, J Colwell, A Stutz, J. Stutz, C Nogales, T Hestand, P Verhage, T. Tracht (2018)   A Fruit Yield Prediction Method Using Blossom Detection. 2018 ASABE Annual International Meeting  1801542.(doi:10.13031/aim.201801542)   Trevor Braddock, Duke M Bulanon, Brice Allen and Joseph J Bulanon. Predicting Fruit Yield Using Shallow Neural Networks. https://www.preprints.org/manuscript/202009.0022/v1

<<Suggestion adopted at li 670

“Final tree fruit load, however, reflects a series of events that include flowering, polli-nation, fruit set, fruit retention and fruit growth (sizing). Each step is affected by an inter-play of endogenous and exogenous factors, with limitations not always compensated as in the example of Mizani et al. [62]. Further, as a perennial crop, management and envi-ronmental conditions of previous seasons can impact yield of the current season, and some genotypes have a strong two-year cycle in yield. The yield potential of a given tree is therefore often not reached, with the extent of influencing effects varying between seasons, between orchards, between trees in an orchard and even between branches on a tree. Management actions operations that do not impact tree size or shape can also alter yield,  such as chemical thinning of fruit or timing and extent of irrigation. Thus, the relationship between fruit load and other characters of the tree can vary - in brief, bigger trees do not necessarily have more or larger fruit. The difference between the crop potential and the actual load in a given year has been termed a ‘yield gap’.

A number of studies have attempted to relate fruit load to the number of flowers or blooming intensity, e.g., Braun et al. [63], Bulanon et al. [64] but such a relationships are dependent on a range of conditions."

At the end of Section 2, in the paragraph starting “Brinkhoff and Robson [87] also”, lasso should be LASSO.  << Done   

The caption Figure 9 is rather confusing. Is this figure presenting the histogram of estimated fruit weight? This figure does not provide information about the accuracy of the relationship between fruit weight, width and length, and I am not sure how it supports the discussion presented in the preceding paragraph.   << Preceding text and Figure legend revised to clarify meaning, as follows:

“Wang et al. [97] utilised a Microsoft Kinectv2 ToF camera for the estimation of mango fruit size from a moving farm vehicle mounted platform, reporting a RMSE of 4.9 and 4.3 mm (RRMSE of 4.9 and 5.3%) for length and width, respectively, against a manual measure-ment with calipers. An ellipse was fitted to the detected objects (fruit), and objects were discarded if ellipse parameters did not match those expected for a fruit, thus rejecting partly occluded fruit. An allometric relation was established relating fruit weight to fruit length width (Table 5). Example data from this system is presented in Figure 9.

Figure 9. Estimated weight of mango fruit on-tree, based on lineal dimensions of fruit assessed using a ToF camera for estimation of camera to fruit distance and a RGB camera for estimation of fruit size in pixels.”

Section 6.1.2: A reference to the following work should be added:  Dihua Wu, Shuaichao Lv, Mei Jiang, Huaibo Song (2020) Using channel pruning-based YOLO v4 deep learning algorithm for the real-time and accurate detection of apple flowers in natural environments. Computers and Electronics in Agriculture 178, 105742

>>Reference added at li 1183