Analysis of the relationship between yield in cereals and remotely sensed fAPAR in the framework of monitoring drought impacts in Europe
- European Commission, Joint Research Centre (JRC), 21027 Ispra (VA), Italy
- European Commission, Joint Research Centre (JRC), 21027 Ispra (VA), Italy
Abstract. This study analyses the relationship between satellite-measured fAPAR (Fraction of Absorbed Photosynthetically Active Radiation), which are continuously monitored by the European Drought Observatory (EDO) of the EU’s Copernicus Emergency Management Service, and crop yield data for cereals, which are collected by Eurostat. Different features of the relationship between annual yield and multiple time series of fAPAR, collected during different periods of the year, were investigated. Two key outcomes of the analysis were the identification of the period from March to October as that when the highest positive correlation between fAPAR and yield is detected in Europe on average, and February to May as the period when most of the negative correlation are observed. While both periods align well with the commonly assumed dynamic of the growing season, spatial differences are also observed across Europe. On the one hand, the Mediterranean regions report the highest correlation values (r > 0.8) and the longest continuous periods with positive statistically significant results (up to 7 months), covering most of the growing season. On the other hand, the central European region is characterized by the most limited positive correlation values, with only 2 months or less showing statistically significant results. While marked differences on the overall capability to capture the full dynamic of yield are observed across Europe, fAPAR anomalies seem capable to distinguish between drought and no/drought years in most of the cases if negative yield anomalies are used as a proxy variable for drought impacts.
Carmelo Cammalleri et al.
Status: open (until 18 Aug 2022)
-
RC1: 'Comment on nhess-2022-178', Anonymous Referee #1, 16 Jul 2022
reply
General comments:
The paper evaluates the relationship between cereal yields and the Fraction of Absorbed Photosynthetically Active Radiation (fAPAR, a satellite derived product) in Europe with the aim of capturing the effects of droughts on crop production. The anomalies of fAPAR were plotted against the yearly yield deviations. The correlation between fAPAR and yield anomalies is positive between the period from March to October, which corresponds to the cereals growing season in Europe. Some negative correlations were observed between February and May; they are limited in length but their analysis could be interesting to assess if they can be considered valuable early warning information. The average growing period in Europe is characterized by a marked south to north gradient; the season has an early start in central Europe and in southern Mediterranean, while it has a late start in southern and Western Europe. An interesting result of the study is the good correlation between fAPAR anomalies and crop yield anomalies over the Mediterranean, while the correlation is limited in most of regions of Central Europe. fAPAR anomalies are found to be useful to distinguish between drought and no/drought years in the majority of situations in which yield anomalies are used as proxies for drought impacts on agriculture.
Overall, the paper represents a good contribution to the understanding of drought and its effects on crop yield. The method applied is valid and the results are appropriately discussed. The work is well structured, the figures quality is good and the number of figures and tables is appropriate.
Specific comments:
Line 32-33: Together with the two FAO reports of the 2015 and 2018 I suggest citing a most recent one: “The impact of disasters and crises on agriculture and food security: 2021” (FAO, 2021).
Line 48: I suggest citing a recent study (Monteleone et al., 2022) evaluating the effect of drought on different phenological stages of maize in Italy.
Line 94: Could you specify which crops Eurostat includes in the definition of “cereals”? Do you believe that the same results discussed in your study could have been obtained considering crops different from cereals?
Discussion: I suggest comparing the obtained results with the ones reported in (López-Lozano et al., 2015), who found a significant correlation between fAPAR and official yields (R2>0.6) in water-limited yield agro-climatic conditions (e.g. the Black Sea region and the Mediterranean basin) for wheat, barley and grain maize.
Table 1: It could be useful to add a column where the drought impacts on agriculture described in the cited studies are briefly summarized.
Fig. 2: NUTS2 regions are quite difficult to visualize; I suggest to increase the size of the various panels and eventually having a sort of table with three columns and six rows instead of the actual figure in which there are six columns and three rows.
Technical corrections:
Line 177-179: The latter may occur when a strong vegetative growth IS observed early in the season during drought years, especially in 179 energy-limited conditions.
References
FAO. (2021). The impact of disasters and crises on agriculture and food security: 2021. Rome: FAO. doi:10.4060/cb3673en
López-Lozano, R., Duveiller, G., Seguini, L., Meroni, M., García-Condado, S., Hooker, J., … Baruth, B. (2015). Towards regional grain yield forecasting with 1km-resolution EO biophysical products: Strengths and limitations at pan-European level. Agricultural and Forest Meteorology, 206, 12–32. doi:10.1016/j.agrformet.2015.02.021
Monteleone, B., Borzí, I., Bonaccorso, B., and Martina, M. (2022). Developing stage-specific drought vulnerability curves for maizeâ¯: The case study of the Po River basin. Agricultural Water Management, 69(May), 107713. doi:10.1016/j.agwat.2022.107713
-
AC1: 'Reply on RC1', Carmelo Cammalleri, 18 Jul 2022
reply
We would like to thank the reviewer for his/her comments and appreciation for the submitted manuscript. We are carefully evaluation all the suggestions, and we are planning to incorporate the following adjustments to the paper:
- We will integrate the referenced literature by adding the reviewer’s suggestions. We will integrate the text to discuss this additional literature.
- We will expand the details on the Eurostat data included in the analysis, and clarify the crops included under cereals.
- We will revise Table 1 by adding some details on the reported impacts for each study.
- We will rearrange the panels in Figure 2 in order to improve the visualization.
-
AC1: 'Reply on RC1', Carmelo Cammalleri, 18 Jul 2022
reply
-
RC2: 'Comment on nhess-2022-178', Anonymous Referee #2, 29 Jul 2022
reply
The manuscripts highlights the important correlation between the satellite-derived fAPAR product and observed cereal yield in Europe, clearly positioning the study and its results in drought impact monitoring efforts. Therefore, this work fits the journal’s scope and can be seen as a timely contribution to the growing early warning and climate services research. In particular the unique scope of the paper, looking at longer, homogeneous periods with robust, significant correlations makes for an interesting, novel article that is written well.
While I enjoyed reading it, some part could benefit for extra clarifications. Besides, I would like to propose a few comments/ideas which could maybe be considered as additional discussion points to strengthen and broaden the manuscript.
- I wonder if the Corine land cover map arable land area is similar to the Eurostat cereal production area data for each NUTS2. It would strengthen the method if this is the case
- Could you please better justify why you would take insignificant (but “at least different than zero”) values into account (doesn’t this reduce precision of the analysis?) and why the particular threshold of 0.15 was chosen?
- I do not fully understand the following part in the method (L182): “Starting with a minimum length of 2 dekads, up to 990 periods of various length (L, from 2 up to 45 dekads) can be analyzed for each region, and for each of these periods four main metrics are computed”. In the rest of the manuscript, it seems the analysis is done on a 1 dekad level (then only combining dekads based on their correlation, not for calculating the correlation)
- It would strengthen the full analysis to check for sensitivity regarding the chosen -1 thresholds for the drought conditions (L202 and onwards). While not a necessary addition, the paper mainly focusses on its use in drought monitoring systems hence it would be interesting if a similar result would be obtained with other standard deviations as thresholds.
Another sensitivity that could be considered to be evaluated is the detrending method. - Besides, in this regard (L205), I wonder why drought years are defined as yield anomaly years (these are agricultural impact years, maybe caused by droughts but potentially by other shocks). The authors could consider reversing the analysis, looking at the average yield anomaly during years with a low average fAPAR during the optimal period. This would not guarantee excluding other shocks (that might impact fAPAR too) but would be more interesting in terms of its capacity to be used as an impact monitoring of prediction product (also give insight on the FA rate for example). (Similar remark could be made for Figure 2: it shows that indeed, during large drought episodes in Europe, the fAPAR is low, but does not show anything about potentially low fAPAR values during years not considered droughts)
- I wonder what happened if two periods with Fp+=1 are equally long? Could you please explain how this is handled in the analysis?
- In the method, multiple exclusion criteria (related to fAPAR and EUROSTAT data) are stated, however the results show a full map without data gaps. Does that mean no NUTS2 were excluded based on these criteria?
- In figure 4; it is a pity no spatial signal could be visualized. It would be a great addition to show which region of Europe contributes to what here.
- Technical remark: “It is possible to observed two “flexing points” …” à OBSERVE
- I feel the method behind figure 7 could be explained better. Could the authors please reflect on this choice for the bounded length of 6 to 8 months (rather than only the optimal, correlated period)?
- I wonder if the overall limited correlation in central Europe might be caused by changing crop types over the years / fAPAR is calculated based on the full period thus assuming homogeneity in land cover over this period.
- I would like for the authors to better explain figure 8: what is meant with performance? Here, how are Fp+/ F+ calculated? Based on the average fAPAR over the optimised period? I think this is missing in the method.
- In the discussion (eg L363), it seem the authors equate growing season with the season where fAPAR relates to the yield; but can this be supported by agronomic observations? Is this correlation not a sign of a crop growing period vulnerable to droughts rather than a representation of the full season?
- In the discussion (L396-400); the interpretation is not extremely clear. So the inverse relationship between fapar and yield is a result of hot-dry summer months (dry spell months or all months?) and their lagged effect: how? Is this correct(ly interpreted?)
With kind regards
-
AC2: 'Reply on RC2', Carmelo Cammalleri, 03 Aug 2022
reply
The manuscripts highlights the important correlation between the satellite-derived fAPAR product and observed cereal yield in Europe, clearly positioning the study and its results in drought impact monitoring efforts. Therefore, this work fits the journal’s scope and can be seen as a timely contribution to the growing early warning and climate services research. In particular the unique scope of the paper, looking at longer, homogeneous periods with robust, significant correlations makes for an interesting, novel article that is written well.
We would like to thank the reviewer for her/his positive stance on our research.
While I enjoyed reading it, some part could benefit for extra clarifications. Besides, I would like to propose a few comments/ideas which could maybe be considered as additional discussion points to strengthen and broaden the manuscript.
- I wonder if the Corine land cover map arable land area is similar to the Eurostat cereal production area data for each NUTS2. It would strengthen the method if this is the case
- Could you please better justify why you would take insignificant (but “at least different than zero”) values into account (doesn’t this reduce precision of the analysis?) and why the particular threshold of 0.15 was chosen?
- I do not fully understand the following part in the method (L182): “Starting with a minimum length of 2 dekads, up to 990 periods of various length (L, from 2 up to 45 dekads) can be analyzed for each region, and for each of these periods four main metrics are computed”. In the rest of the manuscript, it seems the analysis is done on a 1 dekad level (then only combining dekads based on their correlation, not for calculating the correlation)
- It would strengthen the full analysis to check for sensitivity regarding the chosen -1 thresholds for the drought conditions (L202 and onwards). While not a necessary addition, the paper mainly focusses on its use in drought monitoring systems hence it would be interesting if a similar result would be obtained with other standard deviations as thresholds.
We will clarify those choices in the revised version of the manuscript, and consider further analyses on different thresholds if necessary to improve the analysis.
Another sensitivity that could be considered to be evaluated is the detrending method.
We indeed explored different detrending techniques, and we will expand on this topic in the revised version.
- Besides, in this regard (L205), I wonder why drought years are defined as yield anomaly years (these are agricultural impact years, maybe caused by droughts but potentially by other shocks). The authors could consider reversing the analysis, looking at the average yield anomaly during years with a low average fAPAR during the optimal period. This would not guarantee excluding other shocks (that might impact fAPAR too) but would be more interesting in terms of its capacity to be used as an impact monitoring of prediction product (also give insight on the FA rate for example). (Similar remark could be made for Figure 2: it shows that indeed, during large drought episodes in Europe, the fAPAR is low, but does not show anything about potentially low fAPAR values during years not considered droughts)
We would like to clarify Figure 2 reports yield anomalies, not fAPAR, precisely to verify if yield anomaly patters resemble the expected drought patters. Beside this, we agree that the last part of our study can be performed also in reverse, and we will expand this section with extra analyses to test what is suggested by the reviewer.
- I wonder what happened if two periods with Fp+=1 are equally long? Could you please explain how this is handled in the analysis?
This did not occurred very often, since data are at relatively high temporal resolution (10 days). However, in those instances we explored the neighbor NUTS2 and chosen the period closer to the surrounding. This also helped to obtain smoother spatial patterns in the outcomes. We will clarify this in the revised version of the manuscript.
- In the method, multiple exclusion criteria (related to fAPAR and EUROSTAT data) are stated, however the results show a full map without data gaps. Does that mean no NUTS2 were excluded based on these criteria?
Even if the Eurostat data used in this study are at NUTS2 level, they are often provide by national authorities. Hence, the availability usually shows national patterns (i.e. rarely a nation provide data only for some NUTS2). In the final outputs, it is possible to see many NUTS2 masked in full countries (e.g. Norway, Switzerland, among other). We will better clarify that in the revised version on the manuscript.
- In figure 4; it is a pity no spatial signal could be visualized. It would be a great addition to show which region of Europe contributes to what here.
This limitation is overcome by the next analysis (and in particular by Figure 7) where some spatial patterns in the optimal periods can be deducted.
- Technical remark: “It is possible to observed two “flexing points” …” à OBSERVE
- I feel the method behind figure 7 could be explained better. Could the authors please reflect on this choice for the bounded length of 6 to 8 months (rather than only the optimal, correlated period)?
We had to introduce a minimum length, otherwise the optimal average correlation will always be achieved, by definition, by a single dek period corresponding to rmax. Similarly, an upper boundary was introduce to the optimal length to avoid very long optimal periods over regions with flat signals in the correlogram. We will explain better these constrains in the revised version of the text.
- I wonder if the overall limited correlation in central Europe might be caused by changing crop types over the years / fAPAR is calculated based on the full period thus assuming homogeneity in land cover over this period.
- I would like for the authors to better explain figure 8: what is meant with performance? Here, how are Fp+/ F+ calculated? Based on the average fAPAR over the optimised period? I think this is missing in the method.
- In the discussion (eg L363), it seem the authors equate growing season with the season where fAPAR relates to the yield; but can this be supported by agronomic observations? Is this correlation not a sign of a crop growing period vulnerable to droughts rather than a representation of the full season?
- In the discussion (L396-400); the interpretation is not extremely clear. So the inverse relationship between fapar and yield is a result of hot-dry summer months (dry spell months or all months?) and their lagged effect: how? Is this correct(ly interpreted?)
We will improve the discussion section to better clarify the key outcomes of our research concerning spatial differences among regions of Europe and the relationship between optimal periods and growing seasons.
Carmelo Cammalleri et al.
Carmelo Cammalleri et al.
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
175 | 39 | 15 | 229 | 3 | 3 |
- HTML: 175
- PDF: 39
- XML: 15
- Total: 229
- BibTeX: 3
- EndNote: 3
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1