Ground Validation of GPM IMERG Precipitation Products over Iran

Round 1

Reviewer 1 Report

The manuscript titled “Ground validation of GPM IMERG precipitation products over Iran” is a very interesting paper. It presents a validation of GPM IMERG products over Iran based on rain gauge station data. The methodology and resulted extracted are well presented. However, the paper is not yet ready for publication.

I suggest the author to consider the following suggestions prior to its final acceptance.

1. Add more recent published papers on the subject in the Introduction section.

2. Regarding spatiotemporal performance. I could suggest trying to investigate the GPM IMERG performance based on the climatic zones.

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Authors conducted a comprehensive evaluation of different version of GPM/IMERG precipitation estimates over Iran. It is an important step before applying satellite precipitation products in hydrologic and other relevant applications. However, the quality of this manuscript should be further improved to meet the criteria of Remote Sensing. My main concerns are listed as:

comparison of products at different temporal scales; IMERG-Monthly vs observation is evaluated at a monthly scale, while the other three products are compared at a daily scale from Section 3.2. As already pointed out by authors (p9 l299-302), this is not fair because the performance will be always better at a lower time scale. I would suggest to add the error scores of the other three products also at a monthly scale and then compare across all the products.  Impact of algorithm differences on the quality of products; according to the manuscript, I assume that IMERG-Final and -Monthly products are both gauge-corrected versions. However, differences between “IMERG-Early vs. -Late”, and "-Final vs. -Monthly” are not clearly explained. This point should be clearly described, as the algorithm differences could explain the observed performance differences between the products.

Some minor comments are:

p5 l169-170; each IMERG pixel has one exclusive gauge station? I think there could be IMERG pixels that include more than one gauges. If this is the case, how did you conduct comparison between a pixel vs multiple gauges?  Eq1 to Eq3; given that the n is the number of pixel, did you calculate the errors for all pixels at the same time step, and then average them to obtain final error scores? p 9 table 1: isn’t the unit of MAE “mm/month” for IMERG-Monthly…?  P 10 l346-l384; please consider providing synthesis analysis, rather than describing each plot in the Figure 5.  Figure 5. Please consider adding labels to indicate the seasons to the plots directly, instead explaining one by one in the caption. It is a bit hard to follow; even authors made a mistake; i.e., “(j) IMERG-Fall for Winter”.  p14 l443; can you give more details about “rainfall index”? How did you calculated?  p14 l479; did you include zero pairs for the error calculations? e.g., rBias, MAE, to avoid the situation what authors described (i.e. zero rainfall for more than 90% … MAE will not reflect …), when only the pairs for which both IMERG and observant data exceeding a threshold value (e.g., 0.2mm/day) should be investigated. p17-18; the discussions mostly focus on the meaning/characteristics of errors themselves, but the discussion on "the performance differences “between” the IMERG products and possible reasons of the differences are largely missing.  p19 L589; The GPM “constellation” of satellites has a revisit time of approxi. 3h. 

Author Response

Comment 1: comparison of products at different temporal scales; IMERG-Monthly vs observation is evaluated at a monthly scale, while the other three products are compared at a daily scale from Section 3.2. As already pointed out by authors (p9 l299-302), this is not fair because the performance will be always better at a lower time scale. I would suggest to add the error scores of the other three products also at a monthly scale and then compare across all the products.

 

Answer: Thank you for pointing this out. We have now included the comparisons for the daily products (IMERG-Early, -Late, and –Final) at monthly time scale as well. Thus, Table 1 and discussions about that have been updated that now are highlighted in the manuscript.

 

Comment 2: Impact of algorithm differences on the quality of products; according to the manuscript, I assume that IMERG-Final and -Monthly products are both gauge-corrected versions. However, differences between “IMERG-Early vs. -Late”, and "-Final vs. -Monthly” are not clearly explained. This point should be clearly described, as the algorithm differences could explain the observed performance differences between the products

Answer: We have now acknowledged this difference between the gauge-corrected products (final and monthly) and the uncorrected data (early and late), as well as the difference between the uncorrected datasets with latency (late) and the quickly published datasets (early). Thus, we have added more explanations about the differences in section 2.2.1 and mentioned in multiple places that the overall superiority of the IMERG-Final and –Monthly products in relation to the other products is because of their bias correction with regards to the ground-based measurements. More importantly, using Q-Q plot analyses, we found that uncorrected products’ data can, however, be more trustworthy related to extreme events (heavy rainfall and flooding) especially in spring, summer, and fall seasons.

 

Comment 3: p5 l169-170; each IMERG pixel has one exclusive gauge station? I think there could be IMERG pixels that include more than one gauges. If this is the case, how did you conduct comparison between a pixel vs multiple gauges? 

Answer: The grid network of gauge stations is sparsely distributed throughout the whole country which a few of them with many missing data or short period of time excluded. Finally, we had two pixels which had more than one stations (two stations) and as this study was based on the station-based comparison, we considered the both. According to Figure 9, other pixels didn’t have more than one gauge station. Thus, each pixel of GPM IMERG was compared with only one gauge station.

 

Comment 4: Eq1 to Eq3; given that the n is the number of pixel, did you calculate the errors for all pixels at the same time step, and then average them to obtain final error scores?

Answer: The n is the total number of satellite-gauge data pairs, which are being compared. We added this sentence to the manuscript. The n only occasionally differs by location and for a majority of the locations (pixels) the data were historically complete (some 94% of the locations) then n was almost constant.

 

Comment 5: p 9 table 1: isn’t the unit of MAE “mm/month” for IMERG-Monthly…?

Answer: No, it is not. The initial unit for IMERG-Monthly data was in mm/hour that is then multiplied with 24 to report them in mm/day, the same unit as the daily IMERG products.

 

Comment 6: P 10 l346-l384; please consider providing synthesis analysis, rather than describing each plot in the Figure 5.

 

Answer: Although we agree with the reviewer that we described each sub-plot in Figure 5, the descriptions were inevitable to not disregard different results observed for different seasons. However, we have presented several important synthesis conclusions based on these descriptions such as:

The worst distributional fit between IMERG daily products and rain gauge measurements is observed for summer That could be the reason for overestimation of average winter rainfall for the IMERG-Final product observed in Figure 3.

We also added the following sentences supporting our synthesis analyses:

While the bias adjusted GPM IMERG products, which is the case for IMERG-Final datasets, resulted in better match with the gauge measurements for more frequent rainfall events (lower amounts of rainfall), uncorrected datasets of IMERG-Early and –Late products showed to be more trustworthy related to the extreme events (heavy rainfall and flooding) especially in spring, summer, and fall seasons as the bias corrected data from IMERG-Final product deteriorated underestimations observed for extreme rainfalls.

 

Comment 7: Figure 5. Please consider adding labels to indicate the seasons to the plots directly, instead explaining one by one in the caption. It is a bit hard to follow; even authors made a mistake; i.e., “(j) IMERG-Fall for Winter”.

Answer: We apologize for the inconvenience and thank you for your suggestion. We have now added the seasons’ labels to the plots and also corrected the mistake in the caption of Figure 5.

 

Comment 8: p14 l443; can you give more details about “rainfall index”? How did you calculated?

Answer: As it was explained on p7, l246, it was calculated for each rain-gauge station as the ratio of average annual rainfall (in mm) to mean dry period (in days) using the data in the period of study (2014-2017). As we mentioned in the manuscript, we attempted to relate the performance of the GPM IMERG products in estimating rainfall to location specific factors. After investigating different geospatial factors, we found that the rainfall index is more a discriminating factor than other factors.

 

Comment 9: p14 l479; did you include zero pairs for the error calculations? e.g., rBias, MAE, to avoid the situation what authors described (i.e. zero rainfall for more than 90% … MAE will not reflect …), when only the pairs for which both IMERG and observant data exceeding a threshold value (e.g., 0.2mm/day) should be investigated.

Answer: We calculated the criteria indices for both complete pairs of datasets as well as the rainy days of datasets (by removing (0,0) pairs). While the later affected the calculated CC and MAE values, the rBIAS values remained constant. Also, the general trends observed for the criteria indices by location that are presented in Fig. 8, for example, did not change by removing the (0,0) pairs. So, we decided to report the earlier results in the paper (including (0,0) pairs). Thus, all our discussions in the manuscript are on the basis of this assumption. However, we clearly declared to the reader about the data distribution by using q-q plots and depicting the position of percentiles for the plots. Furthermore, related to your next comment, we allocated parts of the text to discuss the way we should evaluate the results based on different combinations of MAE, rBIAS, and etc.

 

Comment 10: p17-18; the discussions mostly focus on the meaning/characteristics of errors themselves, but the discussion on "the performance differences “between” the IMERG products and possible reasons of the differences are largely missing.

Answer: In this part, we only discussed the error itself to argue how we should rely on the products based on the products’ performance. So, the focus was not on the comparison between the IMERG products. However, applying Q-Q plot analysis and on page 15 about Fig. 8, we specifically discuss the differences between the different products and their superiority against each other.

 

Comment 11: p19 L589; The GPM “constellation” of satellites has a revisit time of approxi. 3h. 

Answer: Thank you for spotting this. We corrected this in the manuscript as follows:

“Since the GPM constellation satellites revisit a given spot every approximately three hours, … ”

Round 2

Reviewer 2 Report

Most of my previous comments were taken into account and the authors have made substantial changes to the original manuscript.