Uncertainty in Estimated Trends Using Gridded Rainfall Data: A Case Study of Bangladesh

Round 1

Reviewer 1 Report

General Comments

In summary, the manuscript proposes to compare different gridded rainfall data and their trend over the period 1979-2010 in comparison with the grid of observations (34 stations) obtained by interpolation. The purpose of this paper is to evaluate the grids and the ability of each to represent the spatial rainfall in terms of annual and seasonal trends.

For this, the trends are calculated using two tests: the Mann Kendall test and the modified Mann Kendall test. Several indices are calculated to evaluate the grids and their performance to reproduce observed rainfall. In addition, the spatial similarity between the observations and the different grids are evaluated, as is the probability of detecting trends. The findings reveal that the grids show a high spatial variability of precipitation between products, however the GPCC product shows the best results in estimating monthly precipitation and in the ability to assess observed trends.

The introduction is well structured with a state of the art of the different studies already conducted in Bangladesh on rainfall. The data is sufficiently detailed. The methodology part needs to be improved especially the Mann Kendall trend subpart. There is a consistency in the presentation of the results however this paper can’t be submitted with these figures, in particular the legends (cf major comments) and some results require a better analysis. The last paragraph of the discussion is to be reviewed. In my opinion, your conclusions are too precocious and require more bibliographies.

I propose a Major Revision of the manuscript, and encourage the authors to improve the figures, methods, results, discussion and conclusion parts.

Major Comments:

·         3.1. Observed data

You perform a homogeneity test (double mass curved method) and a variation test (Student’s test) on your observation data. However your introduction details the different studies and their divergent conclusions following the missing data. You say lines 102-103 "However, data before 1979 contains a large number of missing records, while missing data after 1979 is very few" The term "very few" suggests that there are missing data in your time series. My question is: how do you process your missing data knowing that you are working on monthly accumulation.
You can’t say in your introduction that the results of the different studies are contradictory in part due to the quality of the data, without you explain how you deal with the missing data, even if they are few, especially for a rainfall study.

·         Figures : The weakness of this paper are the figures. I have several remarks :

Ø  Figures 2 to 9: I advise you to put on the legends, the corresponding values between the colors.

Ø  Figures 4 to 9 : you must keep a similar color scale. Because in your figures the value 0.0 shows an orange color (Fig 4) sometimes yellow (Fig 5) or green (Fig 8). The comparison between the seasons becomes impossible.

Ø  I don’t understand your values on legends; at time your values do not follow any logic. Example in Figure 6, the value of 0.0 is orange. On both sides of 0.0 we have -81 (orange) and +81 (yellow) and if we add 81 we get 161 (green) but on the other side we have -197 (dark red) ; or the size of the colored bar does not change. I strongly suggest that you put the corresponding values in the color change and not in the center of the color. Taking the example of Figure 6, I guess the difference between orange and yellow is 81/2 is 40.5 mm / decade? Same problem figure 4 (216 + 108 = 324 and not 378), Figure 7 with -22 and +34 on both sides of 0 ...

I therefore advise for each figure to set the value 0 between two colors and thus make a gradient of green (or other color) for the positive trends and a gradient of red (or other color) for the negative trends and always keep this legend for all figures with intensity values (mm/decade) between the colors.

·         Lines 344-345: You said, "The POD of APHRODITE was the highest in detecting trends in pre-monsoon rainfall using both MK and mMK tests. "

If I refer to Figure 5, the pre-monsoon trends, comparing APHRODITE and OBS the trends are opposite especially in the NW.

How do you evaluate the sign criterion in the value of the POD. Trends are detected, POD = 0.85, however they are in opposite directions. Can we really say that the POD is good?

·         Figure 9

Here again the colorbar are unreadable and they must be constructed like the other figures with values between each color and keeping a consistency on the values of the both sides of the 0. Moreover I will do the same scale for the 5 maps. In this way we will be able to see the differences between seasons. Indeed, currently, in PreR center in red refers to trends of about -15mm/decade while the same red in AnnR is -68 mm/decade.

·         Paragraph 4 .2 + Figure 4

When annual trends are calculated, are they from annual accumulation or monthly accumulation ? Same for seasonal, are these trends calculated from seasonal accumulation or monthly accumulation ? Because :

1)      If it is "mm/yr" or "mm/season" I don’t understand why using the mMK knowing that there is no seasonal autocorrelation and that the natural variability as multi-decadal variability is very low over 31 years of data and does not lead to sufficient autocorrelation detection. If this multi-decadal variability is strong and leads to a significant autocorrelation then it must be mentioned.

2)      If it is "mm/month", then the MK test is completely biased by the presence of the seasonal cycle (strong autocorrelation). In fact, the Mann Kendall test is a rank test that calculates concordances and discrepancies between each pair of points. However, the presence of autocorrelation renders the results obtained in Figure 9 in AnnR (black signs) uninterpretable

The strengths and weaknesses of these two tests need to be further explained in paragraph 4 .2. And especially why do you use these tests with your data, rather than others like Spearman test for example. In addition it is necessary to explain a minimum how they are constructed rather than to refer to the bibliography.

·         Discussion/conclusion

You conclude that GPCC is the best gridded rainfall data for representing spatial precipitation in Bangladesh. However with your tests the winter season shows the best scores for GPCC but these scores are very bad (Pbias -552.6%, md = 0.05, SS = -0.29). I think, these winter results need further discussion.

·         Lines 393-394, 399-401: Is this a conclusion or a hypothesis?

I think it's very premature if it's a conclusion. Especially if this conclusion is based only on the different results obtained between the MK and mMK tests, which in my opinion can’t be applied to the same time series. Moreover we know that the links between precipitation and climate change are much more complex.

Minor Comments:

Abstract: Add the study period and the type of data used (monthly or annual accumulation, monthly mean  ...?)

4.1: I didn’t understand if your evaluation is made between interpolated observed rainfall versus gridded rainfall or times series stations versus gridded rainfall.

Line 177 : « x_sim,i is the interpolated time series from gridded data » I will add at the end of this sentence « which included station i »

Line 229 : « 1971-2007 for Aphrodite » would it not be 1979-2007 ?

Lines 231-232 : « The Sen’s slope estimated at the observed locations were interpolated to a résolution of 0.5°×0.5° for fair comparison with gridded data. »

I'm not sure to understand how Sen slope is calculated. 1) the Sen slope is calculated at each station and then interpolated on the same interpolation scheme of the OBS grid or 2) the slope is calculated from the grid 0.5 ° x0.5 °, so on the OBS already interpolated ?

In terms of comparison with other grids 2) is more consistent.

Lines 232-236 : The reference to Figure 4 must be indicated.

Figure 4 to 9 : Never, do you observe a significant trend with both MK and mMK? That is, a grid point with the sign in both white and black?

Figure 4 : The legend must be expanded. You say more in the text (lines 232-236) than in the legend. In particular, add the meaning of the positive and negative signs. Also to say that for the Obs the points represent the stations because one thinks that it is a third "sign". We must be able to understand the figure without having to look in the text.

Lines 241-242 : To place before, in addition I will put Figure 4 before Table 3. Because you already refer to Figure 4 in paragraph lines 232-236. This part is to reorganize.

Lines 257-258 : «A nearly similar to the observed spatial distribution of Sen’s slope was obtained using Aphrodite, GPCC, and PGF ». I don’t have the impression that APHRODITE, GPCC and PGF are similar to OBS, especially in the NW. The term "nearly similar" bothers me.

Lines 289-290 : repetition with lines 286-287

Line 304 : winter is not in figure 6 but in figure 8

Lines 303-304 : « Compared to other seasons, a better consistency in rainfall changes among the gridded data was observed for winter ». How can you assert this with Pbias values in winter strongly negative (Table 5)?

Table 5 : There is one point that I don’t understand. You said: "The results of statistical indices used to assess the spacial similarity in changes (Sen's Slope) ». The formulas of the indices don’t involve the slopes, but simply the time series and therefore the differences between the two series (Gridded and Obs)? So why talk about "spatial similarity in changes (Sen Slope)" ?

Line 391 : The reference refers to Figure 9 instead? 

Figure 9 : Is the intensity of the "-" an indication to be taken into account?

Figure 4 to 8 : We agree that in the figures between APHRODITES and OBS the period over which the MK and mMK are performed are different?

So your statistical tests between APHRODOTE and Obs are made over the period 1979-2007?

Author Response

General Comments

In summary, the manuscript proposes to compare different gridded rainfall data and their trend over the period 1979-2010 in comparison with the grid of observations (34 stations) obtained by interpolation. The purpose of this paper is to evaluate the grids and the ability of each to represent the spatial rainfall in terms of annual and seasonal trends.

For this, the trends are calculated using two tests: the Mann Kendall test and the modified Mann Kendall test. Several indices are calculated to evaluate the grids and their performance to reproduce observed rainfall. In addition, the spatial similarity between the observations and the different grids are evaluated, as is the probability of detecting trends. The findings reveal that the grids show a high spatial variability of precipitation between products, however the GPCC product shows the best results in estimating monthly precipitation and in the ability to assess observed trends.

The introduction is well structured with a state of the art of the different studies already conducted in Bangladesh on rainfall. The data is sufficiently detailed. The methodology part needs to be improved especially the Mann Kendall trend subpart. There is a consistency in the presentation of the results however this paper can’t be submitted with these figures, in particular the legends (cf major comments) and some results require a better analysis. The last paragraph of the discussion is to be reviewed. In my opinion, your conclusions are too precocious and require more bibliographies.

I propose a Major Revision of the manuscript, and encourage the authors to improve the figures, methods, results, discussion and conclusion parts.

Answer:

Thank you very much for your careful revision and comments on our paper. We revised the paper based on your and other reviewer’s comments. The comments of the reviewers helped us to improve the quality of the manuscript. Details of the revisions made based on your comments are given in point-to-point answers to your comments. All the revisions made in the manuscript are marked with red color.  

 Major Comments:

3.1. Observed data

You perform a homogeneity test (double mass curved method) and a variation test (Student’s test) on your observation data. However, your introduction details the different studies and their divergent conclusions following the missing data. You say lines 102-103 "However, data before 1979 contains a large number of missing records, while missing data after 1979 is very few" The term "very few" suggests that there are missing data in your time series. My question is: how do you process your missing data knowing that you are working on monthly accumulation.

You can’t say in your introduction that the results of the different studies are contradictory in part due to the quality of the data, without you explain how you deal with the missing data, even if they are few, especially for a rainfall study.

Answer:

The filling up the missing data are discussed in the revised manuscript as below:

“Complete daily rainfall data after 1979 was available at 10 stations. Missing data at other stations was less than 1%. The missing data was mostly random. Continuous missing data for 2 to 3 months a year were found at 4 stations. Data for the whole year was discarded when it was found missing continuously for a month. randomly missing data were filled up using an artificial neural network (ANN) model developed by Shahid [27]. The ANN model was used to estimate the missing value at the station of interest from the rainfall of six neighbouring stations.

The complete daily rainfall data generated for the period 1979-2010 (after filling the missing data) was used in this study”

Besides, we revised the Introduction, showed the gap in other studies in revised table (Table 1) and try to justify the cause of the disparity in obtained trends in different studies.

Comment
Figures : The weakness of this paper are the figures. I have several remarks :

Ø  Figures 2 to 9: I advise you to put on the legends, the corresponding values between the colors.

Ø  Figures 4 to 9 : you must keep a similar color scale. Because in your figures the value 0.0 shows an orange color (Fig 4) sometimes yellow (Fig 5) or green (Fig 8). The comparison between the seasons becomes impossible.

Ø  I don’t understand your values on legends; at time your values do not follow any logic. Example in Figure 6, the value of 0.0 is orange. On both sides of 0.0 we have -81 (orange) and +81 (yellow) and if we add 81 we get 161 (green) but on the other side we have -197 (dark red) ; or the size of the colored bar does not change. I strongly suggest that you put the corresponding values in the color change and not in the center of the color. Taking the example of Figure 6, I guess the difference between orange and yellow is 81/2 is 40.5 mm / decade? Same problem figure 4 (216 + 108 = 324 and not 378), Figure 7 with -22 and +34 on both sides of 0 ...

I therefore advise for each figure to set the value 0 between two colors and thus make a gradient of green (or other color) for the positive trends and a gradient of red (or other color) for the negative trends and always keep this legend for all figures with intensity values (mm/decade) between the colors.

Answer:

Thank you for your comment. The figures of the paper are revised based on your comments. The interpretability of the results improved after revision of the figures based on your comments. Please see the Figures 6 to 11.

Comment:

Lines 344-345: You said, "The POD of APHRODITE was the highest in detecting trends in pre-monsoon rainfall using both MK and mMK tests. "

If I refer to Figure 5, the pre-monsoon trends, comparing APHRODITE and OBS the trends are opposite especially in the NW.

How do you evaluate the sign criterion in the value of the POD. Trends are detected, POD = 0.85, however they are in opposite directions. Can we really say that the POD is good?

Answer:

POD is a simple measure of accurate detection of trend. In detecting similar signs in trends, it considers the ability of detecting positive, negative and no change. We have discussed the method clearly in the revised manuscript as below to omit any confusion:

“…. the Probability of Detection (POD) index was used to assess the reliability of gridded datasets in detecting the spatial pattern of trends. POD (equation 6) measures how many significant trend sign (positive, negative, no trend) obtained using observed data were correctly estimated by gridded data. For example, if the sign of observed trend is found same as the sign of the trend in the corresponding grid point, the POD counts it as a correct detection. The sign means positive, negative and no trend. This means that if no significant trend at an observed location is also detected at the corresponding grid point, the POD counts it as a correct detection.”

The limitation of POD is mentioned during discussion of POD results, as below:

“Here, it should be noted that POD was estimated only based on the ability of gridded data to detect the sign of significant trend (positive and negative) and no trend in observed data. Changes in other grid points where no station data is available were not taken into consideration during the computation of POD.”

Comment:

Figure 9

Here again the colorbar are unreadable and they must be constructed like the other figures with values between each color and keeping a consistency on the values of the both sides of the 0. Moreover I will do the same scale for the 5 maps. In this way we will be able to see the differences between seasons. Indeed, currently, in PreR center in red refers to trends of about -15mm/decade while the same red in AnnR is -68 mm/decade.

Answer:

Thank you for your comment. We revised the figure to keep consistency on the values of the both sides of 0. However, it was not possible to provide same scale for all the 5 maps as it would cause no color variation for some maps. The values of rainfall change for different seasons are very different. For annual rainfall, the range is -121 to 155 mm/decade, while for winter it is only between -19.4 and 11.3 mm/decade. If same scale is used for presentation of both maps, there will be no color variation in winter map. Therefore, separate scales are used for presentation of rainfall changes for separate seasons.

Comment:

Paragraph 4 .2 + Figure 4

When annual trends are calculated, are they from annual accumulation or monthly accumulation ? Same for seasonal, are these trends calculated from seasonal accumulation or monthly accumulation ? Because :

1)      If it is "mm/yr" or "mm/season" I don’t understand why using the mMK knowing that there is no seasonal autocorrelation and that the natural variability as multi-decadal variability is very low over 31 years of data and does not lead to sufficient autocorrelation detection. If this multi-decadal variability is strong and leads to a significant autocorrelation then it must be mentioned.

2)      If it is "mm/month", then the MK test is completely biased by the presence of the seasonal cycle (strong autocorrelation). In fact, the Mann Kendall test is a rank test that calculates concordances and discrepancies between each pair of points. However, the presence of autocorrelation renders the results obtained in Figure 9 in AnnR (black signs) uninterpretable

Answer:

Thank you very much for your comment. Annual accumulation (mm/year) or seasonal accumulation (mm/season) values were used for the trend analysis. This has been mentioned in the beginning of trend analysis (Section 5.2) of the revised manuscript as below:

“The rainfall data for the period 1979-2010 was converted to annual and seasonal total rainfall to assess the trends in annual and seasonal rainfall.”

We added two paragraphs to show autocorrelation and multi-decadal variability in annual rainfall. Two figures (Figures 4 and 5 are also added).

“The characteristics of rainfall data were analyzed before trend analysis to reveal the presence of autocorrelations and multi-decadal variability in time series. The autocorrelation function (AFC) was used in this study to find a significant correlation for various time lags while the presence of decadal and multi-decadal variability in the time series of climate variable was assessed through wavelet decomposition of time series data [46]. The AFC plot of annual rainfall data at two locations is shown in Figure 4. The vertical lines in the plot that exceed the blue confidence band (horizontal lines) indicate significant correlation. The figure clearly shows positive autocorrelation up to 7-lag years in the time series.

Different levels of decompositions reveal the presence of different cycles in time series. Obtained results for annual rainfall at two stations are shown in Figure 5. The fourth level decomposition of data revealed the presence of a cycle of nearly 20 years in both stations. The x-axis of the graph shows the number of years and the y-axis shows the decomposed precipitation anomaly. Similar results were obtained at other stations. The results indicate the presence of short- and long-term autocorrelations in annual rainfall of Bangladesh. The presence of such multi-decadal variations in annual rainfall time series can significantly affect the trend in rainfall if it is not taken into consideration during trend analysis. Therefore, mMK test along with MK test was also used in the present study.

Comment:

The strengths and weaknesses of these two tests need to be further explained in paragraph 4 .2. And especially why do you use these tests with your data, rather than others like Spearman test for example. In addition it is necessary to explain a minimum how they are constructed rather than to refer to the bibliography.

Answer:

Thank you for your comment. The details of the methodology of MK and MMK tests are provided in revised manuscript. The strengths and weaknesses of the tests are also mentioned.

“Sen’s slope [34] was used to calculate the magnitude of change in observed and gridded monthly rainfall data, while MK [35,36] and mMK [37,38] tests were used to assess the significance in change. Non-parametric MK method is widely used for trend test because it needs only the assumption of data independence as serial autocorrelation in data can increase the chance of significance in trend [17,39]. However, recent studies also showed that the significance trends over time was also sensitive to the assumptions of whether the underlying data have short term or long term autocorrelation. Koutsoyiannis and Montanari [40] stated that MK trend test statistic is heavily affected by long-term autocorrelation due to multi-decadal variability of climate. Thus, MK test overestimates the significance of trend due to long-term fluctuation in time series caused by natural variability in climate. Hamed [38] proposed a mMK trend test that takes scaling of the data into account to discriminating the multi-scale variability from unidirectional trends. Several recent studies, in different regions, have concluded that significant trends in hydro-climatic data obtained using MK test resulted from ignoring of natural variability of climate [3,6,19,41,42]. Therefore, mMK test was used in this study to confirm the trend detected using the MK test. In mMK test, significant trend found in time series was first removed. The equivalent normal variants of the rank of the de-trended series were then obtained to derive the Hurst coefficient and its significance. If the Hurst coefficient is found to be significant, the significance of MMK trend is estimated using a function proposed by Hamed [38]. The full description of the Sen’s slope, MK and mMK methodologies can be found in [3,35-37].

Comment:

Discussion/conclusion

You conclude that GPCC is the best gridded rainfall data for representing spatial precipitation in Bangladesh. However with your tests the winter season shows the best scores for GPCC but these scores are very bad (Pbias -552.6%, md = 0.05, SS = -0.29). I think, these winter results need further discussion.

Answer:

Thank you very much for your comment. Winter rainfall is very low in Bangladesh. It shares only 3% of total annual rainfall. Therefore, small deviation in winter rainfall between observed and GPCC caused a large variation in bias and other statistics. This has been explained in details in the revised manuscript as below:

“Though GPCC was found most suitable in estimating rainfall changes and trends in Bangladesh, the statistical scores of GPCC for winter was very low (Pbias -552.6%, md = 0.05 and SS = -0.26). The winter rainfall in Bangladesh shares only 3% of the total annual rainfall. In some years, it is 0 at some stations. Therefore, a small deviation in winter rainfall between observed and GPCC caused a large variation in bias and other statistics. The mean winter rainfall in Bangladesh is 27 mm while the GPCC estimated the mean winter rainfall as 4.89 mm and thus the Pbias is -552.6%.”

Comment:

Lines 393-394, 399-401: Is this a conclusion or a hypothesis?

I think it's very premature if it's a conclusion. Especially if this conclusion is based only on the different results obtained between the MK and mMK tests, which in my opinion can’t be applied to the same time series. Moreover we know that the links between precipitation and climate change are much more complex.

Answer:

Thank you very much for your comment. The sentence (lines 393-394) has been revised as:

“Thus, it can be concluded that annual rainfall trend estimated at the three grid points in Bangladesh by MK test may be due to its ignorance to natural variability of climate.”

The sentence (lines 399-401) has been deleted.  

Minor Comments:

Comment:

Abstract: Add the study period and the type of data used (monthly or annual accumulation, monthly mean  ...?)

Answer:

Study period and scale of data is mentioned in abstract. The revised sentence is as below:

“A study has been conducted to assess uncertainty in the spatial pattern of rainfall trends in six widely used monthly gridded rainfall data for the period 1979-2010.”

Comment:

4.1: I didn’t understand if your evaluation is made between interpolated observed rainfall versus gridded rainfall or times series stations versus gridded rainfall.

Answer:

The issue has been made clear in the beginning of result section as below:

“There are two general ways to compare gridded data with station observation: (i) areal average precipitation for each grid box is computed from available station data and then the grid-to-grid comparison is conducted; (ii) gridded data is interpolated to station location and then compared with observed data [1,10]. For the assessment of the ability of gridded rainfall to estimate the observed rainfall, the monthly rainfall series of the four nearest grid points surrounding a station were interpolated at the station location using the inverse distance weighting method. These interpolated series were compared with the observed gauge series.”

Comment:

Line 177 : « xsim,i is the interpolated time series from gridded data » I will add at the end of this sentence « which included station i »

Answer:

The sentence has been revised as below:

“ is the interpolated time series from the gridded data at station i,..”

Comment:

Line 229 : « 1971-2007 for Aphrodite » would it not be 1979-2007 ?

Answer:

Sorry for the mistake. It is corrected as: 1979-2010.

Comment:

Lines 231-232 : « The Sen’s slope estimated at the observed locations were interpolated to a résolution of 0.5°×0.5° for fair comparison with gridded data. »

I'm not sure to understand how Sen slope is calculated. 1) the Sen slope is calculated at each station and then interpolated on the same interpolation scheme of the OBS grid or 2) the slope is calculated from the grid 0.5 ° x0.5 °, so on the OBS already interpolated ?

In terms of comparison with other grids 2) is more consistent.

Answer:

Thank you for your comment. The issue has been made clear in the text as below (line 249):

“For the comparison of slopes in observed and gridded data, observed data were gridded to the resolution of gridded data (0.5°×0.5°) and the areal average rainfall for each grid box was computed. The grid-to-grid comparison of slopes was conducted by comparing the Sen’s slope estimated for the areal average of observed rainfall at each grid box with the Sen’s slope estimated for gridded data.”

Comment:

Lines 232-236 : The reference to Figure 4 must be indicated.

Answer:

Thank you for your suggestion. Figure 4 is mentioned in text as reference.

Comment:

Figure 4 to 9 : Never, do you observe a significant trend with both MK and mMK? That is, a grid point with the sign in both white and black?

Answer:

The mMK test is conducted only if MK test is found significant. This has been mentioned clearly in description of mMK test in the revised manuscript as below:

“In mMK test, significant trend found in time series was first removed. The equivalent normal variants of the rank of the de-trended series were then obtained to derive the Hurst coefficient and its significance. If the Hurst coefficient is found to be significant, the significance of MMK trend is estimated using a function proposed by Hamed [38].”

Therefore, the point showing significant change in mMK test means MK test also showed significant change at the point.

Comment:

Figure 4 : The legend must be expanded. You say more in the text (lines 232-236) than in the legend. In particular, add the meaning of the positive and negative signs. Also to say that for the Obs the points represent the stations because one thinks that it is a third "sign". We must be able to understand the figure without having to look in the text.

Answer:

Thank you for your valuable suggestion. The figure legend has been revised to make the figures self-explanatory.  

Comment:

Lines 241-242 : To place before, in addition I will put Figure 4 before Table 3. Because you already refer to Figure 4 in paragraph lines 232-236. This part is to reorganize.

Answer:

Thank you for your suggestion. We placed Figure 4 before Table 3 in the revised manuscript.

Comment:

Lines 257-258 : «A nearly similar to the observed spatial distribution of Sen’s slope was obtained using Aphrodite, GPCC, and PGF ». I don’t have the impression that APHRODITE, GPCC and PGF are similar to OBS, especially in the NW. The term "nearly similar" bothers me.

Answer:

Thank you for your comment. The sentence is removed.

Comment:

Lines 289-290 : repetition with lines 286-287

Answer:

Thank you for your careful revision. The sentence is deleted.

Comment:

Line 304 : winter is not in figure 6 but in figure 8

Answer:

Sorry for the mistake. It has been corrected.

Comment:

Lines 303-304 : « Compared to other seasons, a better consistency in rainfall changes among the gridded data was observed for winter ». How can you assert this with Pbias values in winter strongly negative (Table 5)?

Answer:

We just mentioned consistency among the gridded data, not with observed data. It has large difference with observed data which has been mentioned in subsequent texts.

Comment:

Table 5 : There is one point that I don’t understand. You said: "The results of statistical indices used to assess the spacial similarity in changes (Sen's Slope) ». The formulas of the indices don’t involve the slopes, but simply the time series and therefore the differences between the two series (Gridded and Obs)? So why talk about "spatial similarity in changes (Sen Slope)" ?

Answer:

This issue has been made clear in the revised manuscript as below:

“The statistical indices were calculated between the slopes estimated from interpolated observed data and gridded data at all the grid points to assess the spatial similarity in slopes.”

Comment:

Line 391 : The reference refers to Figure 9 instead?

Answer:

Sorry for the mistake. It has been corrected.

Comment:

Figure 9 : Is the intensity of the "-" an indication to be taken into account?

Answer:

In the present study, we consider the significance of trends for fair comparison.

Comment:

Figure 4 to 8 : We agree that in the figures between APHRODITES and OBS the period over which the MK and mMK are performed are different?

So your statistical tests between APHRODOTE and Obs are made over the period 1979-2007?

Answer: Thank you for your comment. The study was conducted when APHRODITE data was available only unit 2007. Recently, APHRODITE released up to date data until 2015. Those data are used to assess the trends in APHRODITE rainfall until 2010 in the revised manuscript. In the revised manuscript the trends in observed and all gridded rainfall are assessed for the period 1979-2010.

Reviewer 2 Report

The manuscript discusses precipitation trends in Bangladesh and uses several gridded databases covering various periods but analysis were performed for 1979-2010 period.  Trends were calculated with non-parametric Mann-Kendall method and related Sen’s slope, which are commonly used for precipitation series. The manuscript makes use of 4 statistics to assess the performance of gridded data.

It is difficult to recognise whether the manuscript focuses on assessment of gridded databased for trend analysis in Bangladesh (methodological approach) or it seeks to deliver information (up to date) on precipitation trends. It seems that the second is valid but why datasets finish in in 2010 or 2007 are analysed. They are not useful for assessing to date trends. Now we have 2019. Comparison and assessment of the quality of gridded data can be performed on the series finishing in 2010 for the purposes of filling gaps in data.

In the introduction, there is very strong criticism of the quality of data used in the already published papers on precipitation trends in Bangladesh. It seems that all these papers are based on inhomogeneous data ... . I am wondering how it is possible to publish a paper based on a data of such a low quality and delivering unreal trends. This is quite serious accusation needing strong evidence. Please, deliver (in Table 1) information on gaps in the data used in these papers and on the methods used to fulfil the gaps and to calculate trends. Another issue is that the authors claim that various results of trend analysis presented in these studies result from incorrect data and various methods of trend analysis. As I can see from the table 1, the studies used different time windows. Trend analysis strongly depends on the values at the beginning and at the end of the series – this can be a reason for various trends.    

Description of reanalysis would be justified if the paper is of methodological character otherwise such a description is not necessary (subsection 3.2 gridded datasets). Moreover, information in Table 2 are repeated in the description below the table. I suggest including to table 2 information on the methods used to create the gridded dataset and kind of data used for gridding and skip description.

In fact, the analysis were performed on modified gridded data – the authors once again interpolated the gridded data to obtain precipitation for meteorological stations. The question rises how this additional modification of the data influence the results of trend analysis.  The results of trend analysis based on gridded data are surprisingly different from trends calculated from rain-gauge data.

Jaccard similarity index is not fully explained. KGF, SS, md – it should be explained how to interpret the values of this statistics (in the methods section). It is not clear what t-test was used for. Confidence level does not inform on trend significance.

It is unclear how maps based on rain-gauge data were created – they consist of grids and some grids with no station inside exhibit significant trends.  

In subchapter “Comparison of trends” trends for data from Aphrodite cover quite different period, therefore they are not comparable to other datasets. The differences in trends may come from different periods.

In subchapter 5.2 Comparison of trends, it is difficult to find which maps the description refers to. Include clear information that first you describe the map created on the base of observation data.  In this subsection the expression “replicate” is used – how do you assess it – no measure of replicability is included into this chapter? I my opinion subchapter 5.2, 5.3 and 5.4 should be put together. The assessment of maps similarity in 5.2 is subjective – no measure is used while such a measures are included in chapters 5.3 and 5.4, which, by the way are very short.

Some parts of discussion unnecessarily repeat information from the introduction. Main conclusion of this paper says, “GPCC is the most suitable rainfall dataset for assessment of the spatial pattern of the trends in rainfall of Bangladesh”. I suggest modifying this conclusion. The GPCC occurred to be the best out of the gridded data used in this study, however the differences in trend distribution between the GPCC maps and rain gauge maps are still surprisingly big even in case annual rain. This should be stressed in the conclusions. Discussion includes up-to-date map of precipitation trends but only for GPCC dataset. Similar maps should be created using observation data, which, regardless of possible gaps, deliver reliable information on precipitation trends.       

Some basic information included to Introduction is not necessary, e.g. what trend analysis is used for (lines 33-35).

Language should be corrected and text of the manuscript clod be more consistent.

Detailed comments.

lines 25-26: what do the numbers mean? If trends, please deliver units – change per what time-window?

line 29: These Key Words are not the best choice: “evaluation; uncertainty in trends”

line 36: there is no need to adapt to climate variability

line 40-41: “Reliable analysis of climatic trends requires data for at least 30 years which is often difficult for many regions of the world.” – I do not agree that 30-yer series are not available for many regions of the world. It is not true.

line 45: Please deliver citation for some of this “many studies”

lines 69-70: “To overcome the shortcomings of observed 69 data, gridded rainfall and temperature data are generally used for hydro-climatic studies.” - the gridded data are useful but  station data usually deliver more reliable results - this statement seems like recommendation of gridded data while in the further part of this manuscript the performance of gridded datasets is assessed by comparison with station data. This is kind of inconsequence.

equation 6 – there is no information in numerator that only significant trends are counted.

Figures – Information on time-window should be included to the captions. It also concerns tables.

Author Response

Comment:

The manuscript discusses precipitation trends in Bangladesh and uses several gridded databases covering various periods but analysis were performed for 1979-2010 period.  Trends were calculated with non-parametric Mann-Kendall method and related Sen’s slope, which are commonly used for precipitation series. The manuscript makes use of 4 statistics to assess the performance of gridded data.

It is difficult to recognise whether the manuscript focuses on assessment of gridded databased for trend analysis in Bangladesh (methodological approach) or it seeks to deliver information (up to date) on precipitation trends. It seems that the second is valid but why datasets finish in in 2010 or 2007 are analysed. They are not useful for assessing to date trends. Now we have 2019. Comparison and assessment of the quality of gridded data can be performed on the series finishing in 2010 for the purposes of filling gaps in data.

Answer:

Thank you for your comments. Both are the objectives of the paper. We tried to find the best gridded data which can be used for assessment of rainfall trends in Bangladesh. Second one is to inform the long-term trend in rainfall of Bangladesh using best gridded data and modified MK test which can consider both short- and long-term autocorrelation in data to omit the natural variability in climate to assess unidirectional trends in rainfall in Bangladesh.

Considering your comment, we have extended our analysis until 2017 as GPCC data are available until that period. The figure and texts in discussion have been changed accordingly.

Comment:

In the introduction, there is very strong criticism of the quality of data used in the already published papers on precipitation trends in Bangladesh. It seems that all these papers are based on inhomogeneous data ... . I am wondering how it is possible to publish a paper based on a data of such a low quality and delivering unreal trends. This is quite serious accusation needing strong evidence. Please, deliver (in Table 1) information on gaps in the data used in these papers and on the methods used to fulfil the gaps and to calculate trends. Another issue is that the authors claim that various results of trend analysis presented in these studies result from incorrect data and various methods of trend analysis. As I can see from the table 1, the studies used different time windows. Trend analysis strongly depends on the values at the beginning and at the end of the series – this can be a reason for various trends.    

Answer:

Thank you for your comment. We completely agree with you. We have revised the introduction section. We showed the gap in other studies and try to justify the cause of the disparity in obtained trends in different studies. All the revisions are marked with red color. In revised manuscript we justified the issue as below:

“Contradictory results, particularly in trends of rainfall have been reported in different studies even when the same method is used. This is mainly due to different periods used for trend analysis, data used without necessary quality control and the methods used for filling the missing data. Table 1 shows that different results were obtained when different periods were used for trend analysis. The table also shows that proper quality control of data was not conducted before trend analysis in most of the studies. Besides, different approaches were used to handle missing data in trend analysis. The missing data were attempted to fill before trend analysis in some of the studies. Some other cases those were ignored following different rules, for example, a record of the whole month was discarded if rainfall for consecutive days was found missing. These missing data filling techniques and the rule adopted for the consideration of missing data in discarding monthly, seasonal and annual rainfall records also resulted in different trends in climate.”

Comment:

Description of reanalysis would be justified if the paper is of methodological character otherwise such a description is not necessary (subsection 3.2 gridded datasets). Moreover, information in Table 2 are repeated in the description below the table. I suggest including to table 2 information on the methods used to create the gridded dataset and kind of data used for gridding and skip description.

Answer:

We remove the description of data from the text following your comment. We just keep Table 2 for information as suggested by you. We also included sources of data used for the development of the gridded data in Table 2.

Comment:

In fact, the analysis were performed on modified gridded data – the authors once again interpolated the gridded data to obtain precipitation for meteorological stations. The question rises how this additional modification of the data influence the results of trend analysis.  The results of trend analysis based on gridded data are surprisingly different from trends calculated from rain-gauge data.

Answer:

Thank you for your comment. The performance of gridded data assessed following two general ways. It has been discussed in details in revised manuscript as below:

“There are two general ways to compare gridded data with station observation: (i) areal average precipitation for each grid box is computed from available station data and then the grid-to-grid comparison is conducted; (ii) gridded data is interpolated to station location and then compared with observed data [1,10]. For the assessment of the ability of gridded rainfall to estimate the observed rainfall, the monthly rainfall series of the four nearest grid points surrounding a station were interpolated at the station location using the inverse distance weighting method. These interpolated series were compared with the observed gauge series.”

While for assessment of the slopes:

“For the comparison of slopes in observed and gridded data, observed data were gridded to the resolution of gridded data (0.5°×0.5°) and the areal average rainfall for each grid box was computed. The grid-to-grid comparison of slopes was conducted by comparing the Sen’s slope estimated for the areal average of observed rainfall at each grid box with the Sen’s slope estimated for gridded data.”

Comment:

Jaccard similarity index is not fully explained. KGF, SS, md – it should be explained how to interpret the values of this statistics (in the methods section). It is not clear what t-test was used for. Confidence level does not inform on trend significance.

Answer:

Details of the Jaccard similarity index are provided in revised manuscript. Other methods are also discussed elaborately in the revised manuscript. The revised text is as below:

“Jaccard similarity index (JSI) [43,44] is a statistical measurement of the similarity between two sets of data using the concept of intersection over union. It can be mathematically calculated as,

where  is the JSI between X and Y data sets based on a similarity threshold value.

The JSI computes the number of data shared in two sets and represents it as a percentage of the total number of data in both sets. Thus, it can have a value between 0 and 100% where the higher percentage represents more similarity between the datasets. Yin and Yasuda [45] compared JSI with other 19 well-known similarity assessment indices and found JSI as the best in providing stable and discriminable results.”

Interpretation of other indices are also mentioned as below:

“Therefore, the KGE, md and SS near to 1 indicate a better match between observed and gridded data.”

Comment:

It is unclear how maps based on rain-gauge data were created – they consist of grids and some grids with no station inside exhibit significant trends.  

Answer:

Thank you very much for your comment. This is the standard way to assess the performance of gridded data. It has been mentioned clearly in revised manuscript as below:

In section 4.1:

“There are two general ways to compare gridded data with station observation: (i) areal average precipitation for each grid box is computed from available station data and then the grid-to-grid comparison is conducted; (ii) gridded data is interpolated to station location and then compared with observed data [1,10].”

And in results,

“For the comparison of slopes in observed and gridded data, observed data were gridded to the resolution of gridded data (0.5°×0.5°) and the areal average rainfall for each grid box was computed. The grid-to-grid comparison of slopes was conducted by comparing the Sen’s slope estimated for the areal average of observed rainfall at each grid box with the Sen’s slope estimated for gridded data.”

.

Comment:

In subchapter “Comparison of trends” trends for data from Aphrodite cover quite different period, therefore they are not comparable to other datasets. The differences in trends may come from different periods.

Answer:

Thank you for your comment. The study was conducted when APHRODITE data was available only unit 2007. Recently, APHRODITE released up to date data until 2015. Those data are used to assess the trends in APHRODITE rainfall until 2010 in the revised manuscript.

Comment:

In subchapter 5.2 Comparison of trends, it is difficult to find which maps the description refers to. Include clear information that first you describe the map created on the base of observation data.  In this subsection the expression “replicate” is used – how do you assess it – no measure of replicability is included into this chapter? I my opinion subchapter 5.2, 5.3 and 5.4 should be put together. The assessment of maps similarity in 5.2 is subjective – no measure is used while such a measures are included in chapters 5.3 and 5.4, which, by the way are very short.

Answer:

Thank you for your comment. Subsections 5.2, 5.3 and 5.4 are put together in the same section. Texts have been modified where necessary for clarity. Statistical measures are also used to show the similarity in slopes maps presented in Section 5.2. This has been clearly mentioned in revised manuscript as below:

“The JSI was calculated for Sen’s slope maps prepared for each gridded data against the observed data at 0.5° resolution (Figures 4 to 8).”

Comment:

Some parts of discussion unnecessarily repeat information from the introduction. Main conclusion of this paper says, “GPCC is the most suitable rainfall dataset for assessment of the spatial pattern of the trends in rainfall of Bangladesh”. I suggest modifying this conclusion. The GPCC occurred to be the best out of the gridded data used in this study, however the differences in trend distribution between the GPCC maps and rain gauge maps are still surprisingly big even in case annual rain. This should be stressed in the conclusions. Discussion includes up-to-date map of precipitation trends but only for GPCC dataset. Similar maps should be created using observation data, which, regardless of possible gaps, deliver reliable information on precipitation trends.      

Answer:

Thank you for your comment. We revised discussion section based on your comment. All the repetitions are removed. Up to date maps of trend (1901-2017) have been provided in discussion section. The conclusion section is also modified based on your suggestions. Observed data is available is only for recent years (1979-2014). Therefore, such long-term trend maps cannot be created. Trend maps created with observed data for recent years are presented in different sections during comparison of gridded data.

Comment:

Some basic information included to Introduction is not necessary, e.g. what trend analysis is used for (lines 33-35).

Answer:

Thank you for your comment. We have revised the Introduction section. All the less important texts have been removed including the sentence you mentioned.

Comment:

Language should be corrected and text of the manuscript clod be more consistent.

Answer:

The manuscript has been proof read by an experienced scientific writer. All the grammatical mistakes have been corrected. The English language of the manuscript has been improved to make it more readable.

Detailed comments.

Comment:

lines 25-26: what do the numbers mean? If trends, please deliver units – change per what time-window?

Answer:

Those are different statistical indices. Only Jaccard similarity index (JSI) has a unit of %, others do not have unit. The unit for JSI value is provided

Comment:

line 29: These Key Words are not the best choice: “evaluation; uncertainty in trends”

Answer:

The keywords are changes as below:

“Trend analysis; gridded rainfall data; Mann-Kendall test; Jaccard similarity index”

Comment:

line 36: there is no need to adapt to climate variability

Answer:

The sentence has been revised as below:

“Assessment of hydro-climatic trend is one of the most important fields of research in climate change. Trend analysis provides important information required for planning adaptation and mitigation to climate changes [1-3].”

Comment:

line 40-41: “Reliable analysis of climatic trends requires data for at least 30 years which is often difficult for many regions of the world.” – I do not agree that 30-yer series are not available for many regions of the world. It is not true.

Answer:

The sentence has been revised as below:

“Reliable analysis of climatic trends requires quality assured data for at least 30 years [6] which is often difficult for many regions of the world.”

Comment:

line 45: Please deliver citation for some of this “many studies”

Answer:

Thank you for your suggestion. A number of references have been added to support the statement

Comment:

lines 69-70: “To overcome the shortcomings of observed 69 data, gridded rainfall and temperature data are generally used for hydro-climatic studies.” - the gridded data are useful but  station data usually deliver more reliable results - this statement seems like recommendation of gridded data while in the further part of this manuscript the performance of gridded datasets is assessed by comparison with station data. This is kind of inconsequence.

Answer:

Thank you for your comment. We have revised the sentence as below:

“To overcome the problem of unavailability of long-term quality assured climate data, gridded rainfall and temperature data are generally used for hydro-climatic studies.”

Comment:

equation 6 – there is no information in numerator that only significant trends are counted.

Answer:

Signs are provided only for significant trend. Therefore, “trend sign” in equation indicates significant trend. This is also mentioned in text.

Comments:

Figures – Information on time-window should be included to the captions. It also concerns tables.

Answer:

Time periods are mentioned in all figures and table caption

Round 2

Reviewer 1 Report

The part concerning the description of the MK and mMK tests (4.2) as well as the 5.2 part have been very clearly improved and expanded notably with the addition of figures 4 and 5.

An effort has been made on the legends of the figures but in my opinion there is still a problem in Figures 6-7-8-9-10-11 in the second version.
I don’t understand why the scale of legends varies between the negative and the positive part. It is necessary to keep a proportionality in each change of colors, which for me is major for a scientific papers, even if one goes beyond the value reached.
Example Figure 6: -183 and +216
Example Figure 7: -121 and +124
Example Figure 8: -197 and +242
Example Figure 9: -22 and +34
Example figure 10: -2.4 and +4.4

For the figure 11, agree to keep a different colorbar for the 5 maps however the addition of black bars is not enough to better read the figure. For example for "PreR" it would be more legible to apply a colorbar with a color change for proportional integers of type : -20; -10; 0; 10; 20; 30 ; 40. And do this same type of colorbar for the 5 maps of this figure.

Author Response

After revision, the manuscript has been greatly improved by taking into account the considerations of the reviewers.

The part concerning the description of the MK and mMK tests (4.2) as well as the 5.2 part have been very clearly improved and expanded notably with the addition of figures 4 and 5.

An effort has been made on the legends of the figures but in my opinion there is still a problem in Figures 6-7-8-9-10-11 in the second version.
I don’t understand why the scale of legends varies between the negative and the positive part. It is necessary to keep a proportionality in each change of colors, which for me is major for a scientific papers, even if one goes beyond the value reached.
Example Figure 6: -183 and +216
Example Figure 7: -121 and +124
Example Figure 8: -197 and +242
Example Figure 9: -22 and +34
Example figure 10: -2.4 and +4.4

Answer: Thank you for your careful revision. The legends color ramp and intervals values have been corrected for Figures 6 to 11 based on your comment. Now, both sides (+ve and -ve) of the color ramp have the same proportionality.

For the figure 11, agree to keep a different colorbar for the 5 maps however the addition of black bars is not enough to better read the figure. For example for "PreR" it would be more legible to apply a colorbar with a color change for proportional integers of type : -20; -10; 0; 10; 20; 30 ; 40. And do this same type of colorbar for the 5 maps of this figure.

Answer: The color bars have been changed in Figure 11 based on your comment. Now, all the maps have a proportional integer. 

Reviewer 2 Report

I suggest including the followin minor corrections:

line 35-36: “high quality” instead “quality assured”

line 150-151:  Please delete “ and for the others is one” from the sentence: “The optimal value of RMSE is zero, and for the others is one.”

line 224: “climate indices” instead of “climate variable”

Figure 5 – please include description for vertical axis

line 246: “ to assess trends” is enough – delete please the further part of the sentence

264: Saying: “exact spatial distribution”, do you mean: “exactly the same spatial distribution” (?). If yes, please change the sentence.

Author Response

I suggest including the following minor corrections:

line 35-36: “high quality” instead “quality assured”

Answer: Thank you very much for careful revision of the manuscript and the comments. The text has been changed according to your comment.

line 150-151:  Please delete “ and for the others is one” from the sentence: “The optimal value of RMSE is zero, and for the others is one.”

Answer: The text has been changed according to your comment

line 224: “climate indices” instead of “climate variable”

Answer: The text has been changed according to your comment.

Figure 5 – please include description for vertical axis

Answer: Thank you for your comment. “Decomposed Precipitation Anomaly” was added as a label of the vertical axis in Figure 5.  

line 246: “ to assess trends” is enough – delete please the further part of the sentence

Answer: Thank you for your careful revision. The mentioned texts have been deleted.

264: Saying: “exact spatial distribution”, do you mean: “exactly the same spatial distribution” (?). If yes, please change the sentence.

Answer: The text has been changed according to your comment. It is changed to “exactly the same spatial distribution”.

The author would like to thank you for the reviewers for their valuable comments which indeed helped us to improve the quality of the paper.