the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 01 Mar 2023
Indicator-to-impact links to help improve agricultural drought preparedness in Thailand
Michael Eastman
Eugene Magee
Lucy J. Barker
Thomas Chitson
Chaiwat Ekkawatpanit
Daniel Goodwin
Jamie Hannaford
Ian Holman
Liwa Pardthaisong
Simon Parry
Dolores Rey Vicario
Supattra Visessri
Abstract. Droughts in Thailand are becoming more severe due to climate change. Developing a reliable Drought Monitoring and Early Warning System (DMEWS) is essential to strengthen a country’s resilience to droughts. However, for a DMEWS to be valuable, the drought indicators it provides stakeholders must have relevance to tangible impacts on the ground. Here, we analyse drought indicator-to-impact relationships in Thailand, using a combination of correlation analysis and machine learning techniques (random forest). In the correlation analysis, we study the link between meteorological drought indicators and high-resolution remote sensing vegetation indices used as proxies for crop-yield and forest-growth impacts. Our analysis shows that this link varies depending on land use, season, and region. The random forest models built to estimate regional crop productivity allow a more in-depth analysis of the crop-/region-specific importance of different drought indicators. The results highlight seasonal patterns of drought vulnerability for individual crops, usually linked to their growing season, although the effects are somewhat attenuated in irrigated regions. Integration of the approaches provides new detailed knowledge of crop-/region-specific indicator-to-impact links, which can form the basis of targeted mitigation actions in an improved DMEWS in Thailand, and could be applied in other parts of Southeast Asia and beyond.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(1774 KB)
-
Supplement
(1266 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(1774 KB) - Metadata XML
-
Supplement
(1266 KB) - BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Maliko Tanguy et al.
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-308', Samuel Jonson Sutanto, 16 Mar 2023
Title: Indicator-to-impact links to help improve agricultural drought preparedness in Thailand
Authors: Tanguy et al.
Recommendation: minor revision
Summary
This paper correlates meteorological drought indices, represented by SPI and SPEI, and vegetation indices such as VCI, TCI, and VHI with forest growth and crop yield impacts. Two approaches were used in the analysis, which are the Pearson correlation and the Random Forest machine learning model. The authors found that the strength of correlations depends on land use, season, region, and drought duration. Crops are strongly impacted by drought in both wet and dry seasons. The impact of droughts, however, is less apparent for forest growth. The use of the Random Forest technique allows a more in-depth analysis of the importance of different drought and vegetation indicators. The authors also highlighted that the knowledge of linking specific indicators to the drought’s impact on crops will help to improve the DMEWS and perform mitigation actions.
Assessment
This paper analyzes the use of different drought and vegetation indicators to link these indices with the impact of drought on crop yields and forest growth. The manuscript is interesting and well written. I have a few minor comments below and two general comments, but only for clarification. I believe this work is well suited for NHESS.
General Comments
I have two general comments regarding the manuscript but all of them are only for clarification and improvement of the manuscript.
- I am wondering why the authors used the Random Forest (RF) approach only to find the importance of all indicators on crop yields and forest growth. RF can also be used to predict the crop yields by: 1) training the predictor variables, here are e.g., drought and vegetation indices, and the response variables (e.g., crop yields), resulting in crop yield impact model; 2) using the developed model to predict the impact of drought on crop yield by leaving out the predicted year from the training period. Maybe it is interesting to do this since the authors already have the script to develop the RF model and mentioned this in Figure 2. The authors can train the RF model again without the predicted year and in the end forecast the yields and validate the result with the observed crop yield data. Otherwise, it is worth to discuss the use of machine learning to predict crop yield and not only to find the importance of predictor variables.
- I have difficulty to understand figures 8, 9, and 10. I read the caption over and over again but still cannot interpret the figures. Is there any other way to present your results in a simple manner, so thus the readers can understand the results? For example, it is not clear to me why some lines are thick, and some are thin. Also, why VCI N has 6 thin lines and VCI E has only 3 thick lines? How to indicate 24 months accumulation periods in the results? Maybe modify the Y-axis?
Line by line comments
L refers to line and P refers to page.
P1L19: Maybe re-write “…it provides stakeholders…” as “…provided to stakeholders…”?
P2L33: The authors may add a study on extreme high and low flow events in Southeast Asia including Thailand due to climate change (Hariadi et al., 2023).
P2L46: Full stop after the ICID reference.
P2L53: “has” -> “have”
P3L75: Double reference from Stahl et al., 2016.
P3L76: Better to place the EDII and DIR references here. EDII: Stahl et al., 2016 and DIR: Smith et al., 2014?
P4L109: I suggest to mention again the gap (instead of “that” gap) here since it is a new paragraph.
P5L144: Suggestion to rephrase the sentence: “….of Thailand. Although it has suffered….decades, there has been some…”
P8: Figure 2. Here the authors clearly indicated that the RF model can be used to predict the crop yield (my general comment 1). This is one thing that I miss from the result.
P9L8: Please elaborate more how the authors did “detrended”. The authors only said using a simple linear regression.
P10L217: “…both “indices” in our…….
P14L291: I am wondering, it is VI or VCI?
P14L294-296: Maybe elaborate more about the meaning of positive and negative correlations between VCI and meteo indicators. Also, the authors stated that short droughts are beneficial for forest growth. In my opinion, drought is never beneficial for any ecosystem. I suggest to rephrase the word beneficial.
P14L287: The authors can consider to re-write “…the accumulation period best correlated is…” as “…the best correlated accumulation period is…”
P14L298: Here and also in the discussion, the authors conclude that forest is more resistant to short droughts. I believe that this strongly relates to the ability of forest to subtract water from deeper layers, e.g. groundwater. Discuss this.
P18L343-346: How to see the SPI24 from Figure 9 and to see 11 SPEI, 10 VCI, and 6 TCI from Figure 11? See my general comment 2.
P22L397: The authors stated that SPI is more important than SPEI. I am wondering whether the low precipitation in the N region has something to do with the result.
P22L410-411: You may discuss the difference in water consumption by each crop.
P23L437: Rephrase “though this effect is highly variety specific:
P24L453-455: Make two sentences.
P24L459: “…seasons, which suggests… -> “…seasons, suggest…”
P24L468-470: Explain this already in the beginning, thus the readers will not be confused.
P25L498: Here, the authors can link the short drought events with the limitation of using data-driven model, such as machine learning.
P25L505-506: Rephrase “Though powerful tools to produce predictive models from data”
References:
Hariadi et al. (2023). A high-resolution perspective of extreme rainfall and river flow under extreme climate change in Southeast Asia, https://doi.org/10.5194/hess-2023-14.
Smith et al. (2014). Local observers fill in the details on drought impact reporter maps, https://doi.org/10.1175/1520-0477-95.11.1659.
Citation: https://doi.org/10.5194/egusphere-2023-308-RC1 - AC1: 'Reply on RC1', Maliko Tanguy, 06 May 2023
-
RC2: 'Comment on egusphere-2023-308', Veit Blauhut, 28 Apr 2023
Dear authors,
First of all I have to apologise for the strong delay in reviewing your paper. Overall it was a pleasure to review your excellent paper. Very well written and designed in a carefully thought-out way. I believe this study to be essential to support agricultural drought management in the future.
In addition to the comments of SJ Sutantos comments there is only few to add on. All over I recommend this manuscript to be published after minor revisions.
I would appreciate if you could add some introductionary thoughts on the usage of the vegetation indices and their classification it they are rather used a drought index or as a proxy for impacts/ impacts. Also, if VHI can be used without any knowledge on the hazard situation?
In your methods you mention that you spatially aggregated the standardised indices. Please elaborate on your practise applied with a focus on a) the spatial aggregation method of drought indices, were the standardised indices aggregated to province levels or the indicators (temp, precip.) and then the distribution done? And b) how did you aggregate the indices in time.
In my opinion, the discussion on your initial analysis (Fig 4) is a little short and could tolerate a little more discussion on possible drivers (maybe in the discussion section and not in the results). In figure 4b) East inland Thailand, there are three regions neighbouring, having the same major crops paddy rice (and high percentages), but there are either VHI, VCI or negatively correlated.--> why do they perform so different? Irrigation practise (e.g. river fed irrigation?) Elevation?
Some minor points:
Figure 1-4 – Names of neighbouring countries are not readable
Figure 4+ please increase legend size
Furthermore, you might check on the flowing literature. Their results might be usefull for some discussion and or introduction.
Sa-Nguansilp, C., Wijitkosum, S., Sriprachote, A., 2017. Agricultural drought risk assessmentin Lam Ta Kong Watershed, Thailand. International Journal of Geoinformatics 13 (4), 37–43.
Monkolsawat, C., et al., 2001. An. Evaluation of Drought Risk Area in NE Thailand Asian Journal of Geoinformatics 1 (4), 33–44.
Wijitkosum S 2018. Fuzzy AHP for drought risk assessment in lam Ta Kong watershed, the north-eastern region of Thailand. Soil and Water Research, 13(4), 218–225. doi:10. 17221/158/2017-SWR
Citation: https://doi.org/10.5194/egusphere-2023-308-RC2 - AC2: 'Reply on RC2', Maliko Tanguy, 06 May 2023
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-308', Samuel Jonson Sutanto, 16 Mar 2023
Title: Indicator-to-impact links to help improve agricultural drought preparedness in Thailand
Authors: Tanguy et al.
Recommendation: minor revision
Summary
This paper correlates meteorological drought indices, represented by SPI and SPEI, and vegetation indices such as VCI, TCI, and VHI with forest growth and crop yield impacts. Two approaches were used in the analysis, which are the Pearson correlation and the Random Forest machine learning model. The authors found that the strength of correlations depends on land use, season, region, and drought duration. Crops are strongly impacted by drought in both wet and dry seasons. The impact of droughts, however, is less apparent for forest growth. The use of the Random Forest technique allows a more in-depth analysis of the importance of different drought and vegetation indicators. The authors also highlighted that the knowledge of linking specific indicators to the drought’s impact on crops will help to improve the DMEWS and perform mitigation actions.
Assessment
This paper analyzes the use of different drought and vegetation indicators to link these indices with the impact of drought on crop yields and forest growth. The manuscript is interesting and well written. I have a few minor comments below and two general comments, but only for clarification. I believe this work is well suited for NHESS.
General Comments
I have two general comments regarding the manuscript but all of them are only for clarification and improvement of the manuscript.
- I am wondering why the authors used the Random Forest (RF) approach only to find the importance of all indicators on crop yields and forest growth. RF can also be used to predict the crop yields by: 1) training the predictor variables, here are e.g., drought and vegetation indices, and the response variables (e.g., crop yields), resulting in crop yield impact model; 2) using the developed model to predict the impact of drought on crop yield by leaving out the predicted year from the training period. Maybe it is interesting to do this since the authors already have the script to develop the RF model and mentioned this in Figure 2. The authors can train the RF model again without the predicted year and in the end forecast the yields and validate the result with the observed crop yield data. Otherwise, it is worth to discuss the use of machine learning to predict crop yield and not only to find the importance of predictor variables.
- I have difficulty to understand figures 8, 9, and 10. I read the caption over and over again but still cannot interpret the figures. Is there any other way to present your results in a simple manner, so thus the readers can understand the results? For example, it is not clear to me why some lines are thick, and some are thin. Also, why VCI N has 6 thin lines and VCI E has only 3 thick lines? How to indicate 24 months accumulation periods in the results? Maybe modify the Y-axis?
Line by line comments
L refers to line and P refers to page.
P1L19: Maybe re-write “…it provides stakeholders…” as “…provided to stakeholders…”?
P2L33: The authors may add a study on extreme high and low flow events in Southeast Asia including Thailand due to climate change (Hariadi et al., 2023).
P2L46: Full stop after the ICID reference.
P2L53: “has” -> “have”
P3L75: Double reference from Stahl et al., 2016.
P3L76: Better to place the EDII and DIR references here. EDII: Stahl et al., 2016 and DIR: Smith et al., 2014?
P4L109: I suggest to mention again the gap (instead of “that” gap) here since it is a new paragraph.
P5L144: Suggestion to rephrase the sentence: “….of Thailand. Although it has suffered….decades, there has been some…”
P8: Figure 2. Here the authors clearly indicated that the RF model can be used to predict the crop yield (my general comment 1). This is one thing that I miss from the result.
P9L8: Please elaborate more how the authors did “detrended”. The authors only said using a simple linear regression.
P10L217: “…both “indices” in our…….
P14L291: I am wondering, it is VI or VCI?
P14L294-296: Maybe elaborate more about the meaning of positive and negative correlations between VCI and meteo indicators. Also, the authors stated that short droughts are beneficial for forest growth. In my opinion, drought is never beneficial for any ecosystem. I suggest to rephrase the word beneficial.
P14L287: The authors can consider to re-write “…the accumulation period best correlated is…” as “…the best correlated accumulation period is…”
P14L298: Here and also in the discussion, the authors conclude that forest is more resistant to short droughts. I believe that this strongly relates to the ability of forest to subtract water from deeper layers, e.g. groundwater. Discuss this.
P18L343-346: How to see the SPI24 from Figure 9 and to see 11 SPEI, 10 VCI, and 6 TCI from Figure 11? See my general comment 2.
P22L397: The authors stated that SPI is more important than SPEI. I am wondering whether the low precipitation in the N region has something to do with the result.
P22L410-411: You may discuss the difference in water consumption by each crop.
P23L437: Rephrase “though this effect is highly variety specific:
P24L453-455: Make two sentences.
P24L459: “…seasons, which suggests… -> “…seasons, suggest…”
P24L468-470: Explain this already in the beginning, thus the readers will not be confused.
P25L498: Here, the authors can link the short drought events with the limitation of using data-driven model, such as machine learning.
P25L505-506: Rephrase “Though powerful tools to produce predictive models from data”
References:
Hariadi et al. (2023). A high-resolution perspective of extreme rainfall and river flow under extreme climate change in Southeast Asia, https://doi.org/10.5194/hess-2023-14.
Smith et al. (2014). Local observers fill in the details on drought impact reporter maps, https://doi.org/10.1175/1520-0477-95.11.1659.
Citation: https://doi.org/10.5194/egusphere-2023-308-RC1 - AC1: 'Reply on RC1', Maliko Tanguy, 06 May 2023
-
RC2: 'Comment on egusphere-2023-308', Veit Blauhut, 28 Apr 2023
Dear authors,
First of all I have to apologise for the strong delay in reviewing your paper. Overall it was a pleasure to review your excellent paper. Very well written and designed in a carefully thought-out way. I believe this study to be essential to support agricultural drought management in the future.
In addition to the comments of SJ Sutantos comments there is only few to add on. All over I recommend this manuscript to be published after minor revisions.
I would appreciate if you could add some introductionary thoughts on the usage of the vegetation indices and their classification it they are rather used a drought index or as a proxy for impacts/ impacts. Also, if VHI can be used without any knowledge on the hazard situation?
In your methods you mention that you spatially aggregated the standardised indices. Please elaborate on your practise applied with a focus on a) the spatial aggregation method of drought indices, were the standardised indices aggregated to province levels or the indicators (temp, precip.) and then the distribution done? And b) how did you aggregate the indices in time.
In my opinion, the discussion on your initial analysis (Fig 4) is a little short and could tolerate a little more discussion on possible drivers (maybe in the discussion section and not in the results). In figure 4b) East inland Thailand, there are three regions neighbouring, having the same major crops paddy rice (and high percentages), but there are either VHI, VCI or negatively correlated.--> why do they perform so different? Irrigation practise (e.g. river fed irrigation?) Elevation?
Some minor points:
Figure 1-4 – Names of neighbouring countries are not readable
Figure 4+ please increase legend size
Furthermore, you might check on the flowing literature. Their results might be usefull for some discussion and or introduction.
Sa-Nguansilp, C., Wijitkosum, S., Sriprachote, A., 2017. Agricultural drought risk assessmentin Lam Ta Kong Watershed, Thailand. International Journal of Geoinformatics 13 (4), 37–43.
Monkolsawat, C., et al., 2001. An. Evaluation of Drought Risk Area in NE Thailand Asian Journal of Geoinformatics 1 (4), 33–44.
Wijitkosum S 2018. Fuzzy AHP for drought risk assessment in lam Ta Kong watershed, the north-eastern region of Thailand. Soil and Water Research, 13(4), 218–225. doi:10. 17221/158/2017-SWR
Citation: https://doi.org/10.5194/egusphere-2023-308-RC2 - AC2: 'Reply on RC2', Maliko Tanguy, 06 May 2023
Peer review completion
Journal article(s) based on this preprint
Maliko Tanguy et al.
Maliko Tanguy et al.
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 426 | 193 | 15 | 634 | 30 | 3 | 7 |
- HTML: 426
- PDF: 193
- XML: 15
- Total: 634
- Supplement: 30
- BibTeX: 3
- EndNote: 7
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(1774 KB) - Metadata XML
-
Supplement
(1266 KB) - BibTeX
- EndNote
- Final revised paper