Field heterogeneity of soil texture controls leaf water potential spatial distribution in non-irrigated vineyards

By : Louis Delval, Jordan Bates, François Jonard, Mathieu Javaux

Review for EGUsphere

General comments

This article addresses the issue of estimating vegetation leaf water potential (LWP) values from vegetation indexes (VIs) and followed by spatializing LWP at intraparcel scale in vineyards. This topic is relevant, because although LWP is the reference method for assessing vine water status, it is constraining to measure, which makes spatialization (mapping) LWP with the pressure chamber time consuming. The topic has been investigated before, but never with such detail and accuracy as in this article. The article has many strong points. The first one is that the authors combined VIs based on LIDAR measurements, multispectral images and thermal IR images. By doing so, they show great skills in data acquisition with complexe devices and management of large datasets. In this way they contstructed 48 ( !!) VIs. The second strong point is that the authors tested combinations of VIs. One difficulty of working with VIs in vineyards (or other row crops) is that the images contain information from the row and the interrow. In some articles, pixels contained mixed information of vine vegetation and interrow vegetation (e.g., in Laroche-Pinel et al., 2021), which obviouly introduces a lot of noise in a possible relationship between VIs and LWP. This brings me to the third strong point of this manuscript : the authors developped a very sophisticated workflow to process the segmentation of vine pixels from interrow pixels to generate a pure grapevine mask (Figure 2). The fourth strong point of the manuscript is that the authors always separated their dataset in a calibration and validation subset. The fifth strong point is that the authors show that soil hydraulic conductivity (in relation to soil texture) better explains predicted LWP than slope, elevation or soil water holding capacity (p. 29, lines 738-739). They conclude that variations in soil texture drive intra-parcel variability of predicted LWP.

The authors show great modeling skills, e.g. p. 26, lines 651-655. In the article, the model to predict LWP from VIs is based on stepwise regression. However, they also tested random forest, which gave good results in terms of R² and RMSE, but a great loss in predictive power when the random forest model was applied to the validation dataset.

The results are consistent and the authors conclude that regardless of the model tested, VIs with the greatest predictive power are CLRedEdge and/or CWSIb (p.19, lines 509-511). This is valuable information for all researchers working with VIs as a proxy for vine water status.

A limitation of the study, is that the method was developped on only a few parcels of 2 winegrowing estates in Belgium. As one can expect in the cool and relatively humid climatic conditions of Belgium, vine water deficit was not very strong. The lowest water potentiel recorded was -1.15 MPa, which corresponds to only weak-to-moderate water deficit, according to van Leeuwen et al. (2009). It is likely that the method would have yielded better results if the study was carried out in conditions with a wider range of vine water status conditions. The authors were lucky to have two contrasting vintages (2022 which was dry and 2023 which was wet), but it is surprising that the LWP readings were not so much more negative in 2022 compared to 2023.

Finally, the best relationships the authors found between VIs and LWP were reasonably good (Fig. 7a), but not exceptionnal. Because the authors used a highly sophisticated approach and tested many VIs (and combinations of VIs), it is unlikely that someone will do much better in the future (unless based on a calibration dataset with a wider range of LWP). That also shows the limitation of the approach of predicting LWP from VIs. The method was developped to map vine water status with VIs in order to save time compared to mapping vine water potentials with a pressure chamber. But in the end, the sophisticated workflow presented in this article may be more time-consuming than taking 40 or 50 LWP measurements across a parcel with a pressure chamber and creating a map with oridinary kriging.

I suggest the title to be modified as « Field heterogeneity of soil texture controls leaf water potential spatial distribution predicted from vegetation indexes in non-irrigated vineyards ». As it is now, a substantial part of the content of the article (predicting LWP from VIs) is not covered by the title.

Specific comments

P5, line 120    Delete « the interface between »

P13, line 366  Delete « and »

P17, line 455  In Figure 4, replace « CI » by « CWSIb »

P17, line 460  Add « s » to model (« multple linear regression models »)

P20, line 528  Replace « eastern » by « western »

P26, lines 651 and 653           Two following senences start with « However », please reword.

P42, line 1077 Replace « OENO One » by Journal International des Sciences de la Vigne et du Vin (all articles initially published in the J. Int. Sci. Vigne Vin are available on the OENO One website, but in 2009 OENO One did not yet existe).

Conclusion

This very nice article can be published with very minor corrections in EGUsphere.

References

Laroche-Pinel, E., Duthoit, S., Costard, A. D., Rousseau, J., Hourdel, J., Vidal-Vigneron, M., ... & Clenet, H. (2021). Monitoring vineyard water status using Sentinel-2 images: qualitative survey on five wine estates in the south of France. OENO One, 55(4), 115-127.

van Leeuwen, C., Trégoat, O., Choné, X., Bois, B., Pernet, D., & Gaudillère, J. P. (2009). Vine water status is a key factor in grape ripening and vintage quality for red Bordeaux wine. How can it be assessed for vineyard management purposes? Journal International des Sciences de la Vigne et du Vin, 43(3), 121-134.

EGUsphere – Delval et al. : Field heterogeneity of soil texture controls leaf water potential spatial

distribution in non-irrigated vineyards

GENERAL COMMENTS

The manuscript from Delval and co-workers presents a new approach intended to characterize the heterogeneity of grapevine water status at the field scale and to quantify the influence of environmental factors on this variability. In particular, the authors analysed the heterogeneity of leaf water potential (Ψleaf) in two Belgian vineyards during two seasons. The paper is impressive in the amount of work in present (e.g. 132 in situ measurements of Ψleaf, more than 30 soil water content profiles, several models and vegetation indices tested…). It also presents an important step forward in the field by combining several sensors (i.e. multispectral, thermal and LiDAR) in a single framework. This new approach led to capture the spatial distribution of Ψleaf with high resolution during the growing season. This methodological development is relevant because the actual spatial distribution of Ψleaf is limited to pressure chamber measurements, a time-consuming process. They also highlight an important aspect of grapevine water research: the soil hydraulic heterogeneity has a strong influence on grapevine water status. This finding is consistent with recent results from studies conducted under controlled conditions in other species. The paper is well written and easy to follow. The results are based on robust analysis and evaluation.

MAJOR REMARKS AND QUESTIONS

The authors combined multilinearly vegetation indices from multispectral, thermal and LiDAR sensors to capture the spatial distribution of grapevine water status within vineyards. To discriminate information from the vine canopy pixels to inter-row soil, they developed an interesting analytical pipeline, integrating K-means clustering analysis and different vegetation indices. I'm not familiar with this procedure and whether it has been done in previous studies. If this new pipeline has never been published, I encourage the authors to write a short paragraph on this topic in the discussion, detailing the added value of developing this approach compared to what has been done before. It should again highlight the novelty and interest of this study. Moreover, it should support the advantages of using several vegetation indices from different sensors: performing a better discrimination of the vine pixel and performing a spatial distribution of the vine water status.

The authors mentioned a high spatial resolution of the vine water status captured by their UAS devices. However, I wondered how important this resolution is compared to what has been done in the past. In fact, I did not see any reference and comparison with previous studies. In addition, it would be interesting to know the size of the pixel of the map obtained at the end of the "grapevine pixel extraction". These values could be mentioned in the abstract or in the conclusion, and perhaps compared with the values of previous studies in the discussion, in order to highlight the quality of this work. In the same way, it would be good to add scale bars with the different maps in Figure 2.

The authors efficiently capture the spatial distribution of Ψleaf by testing different combinations of vegetation indices using a stepwise regression method. The results are consistent and the authors find that of the seven data combinations tested, the vegetation indices with the strongest predictive ability are CLRedEdge, NDRE and CWSIb. The authors show a better predictive power of the models with LiDAR data. I'm a bit surprised by this result and wonder what additional information the LiDAR provides to better capture the water status of the vine in this experiment? Indeed, "structural" responses of grapevine to water deficit are often observed at low water potentials (e.g. Ψ stem < -1.6 MPa induces leaf shedding). The lowest water potential recorded was -1.15 MPa (Table 4), corresponding to a moderate water deficit. So it's surprising to observe an added value from the use of LiDAR, and I wonder how you explain the contribution of the LiDAR sensors to the model? It's lower growth induced by stomatal closure? Perhaps you could add a sentence to the discussion explaining the biological process that might explain this result. I also noticed a low contribution of the LiDAR to the R² and RMSE in Figures 5a and 5b (comparison between M+T+L and M+T; M+L and M).

The study presented here focuses on two vineyards in Belgium. The reason for using these two different vineyards is not very well explained and could be better introduced in the manuscript. Moreover, the authors show that the spatial heterogeneity of Ψleaf is less pronounced in the Domaine W vineyard than in the Bousval vineyard and that Ψleaf is lower in Bousval. I wonder if this result could also be linked to the different management systems of the two vineyards? In fact, the Domaine W is characterised by a lower density (higher inter-rang and inter-cep) compared to the Bousval vineyard.

SPECIFIC COMMENTS

In the results part, it might be interesting to provide the values of the results highlighted by the authors. For example, line 394, please give the mean values of Ψleaf measured in 2022 and 2023. The same comment applies to lines 396-398, 533 or 586.

Please consider if "weather" (e.g. line 11 and 18) is the best term here. Perhaps "climatic conditions" is more appropriate.

P.2 L. 35: Perhaps explain why manual measurement is more time-consuming and labour-intensive in a heterogeneous field (e.g. need for more measurements).

P5. L.123: Delete the space in the middle of the word “south”.

P.8 L. 204: Add a space between “m²” and “zones”.

P11. L.294-296: Perhaps give the reason for focusing on these 2x2m² zones (e.g. for leaf water prediction and evaluation).

P16. L.441: Delete the point after “Ψleaf_meas”.

P16. L.443-444: This sentence should be written in the discussion section as it was done in line 649 (the same comment for the lines 446 and 447).

P16. 444: I suggest deleting "It is interesting to note". Let the reader decide if it's interesting or not. So just be factual and present the results without taking a position: "Pearson is greater than Spearman". Same comment on line 496 or for similar formulations (e.g. "This is not surprising to").

P.20 L.528: Replace “eastern part” by “western part” (the same comment for the line 724).

Fig. 6: For ease of reading, the name of the model used to reconstruct the map could be added for each date (e.g. 27/07/22 - Model 1 or 27/07/22 – M.1).

P.24 L.589-590: I suggest adding this sentence to the discussion.

P.27 L.691: Put a dot after "Ψleaf". Same on line 799.

CONCLUSION

The manuscript offers new insights to the scientific community with the development of a new methodological approach based on technological advances in remote sensing and the quantification of environmental factors affecting grapevine water status. The tables and figures are informative and well designed. I have no reservations about publishing it in EGUsphere with minor corrections.