A New 60-Year 1940/1999 Monthly-Gridded Rainfall Data Set for Africa

Round  1

Reviewer 1 Report

Research work on new datasets is always welcomed. Different data sets can bring something new than others. That is good you compared this datasets with other datasets.  Some of my comments are as follows:

If this data was that much important, Please explain the reasons, why did this data not extended up to the present?

Can put some of the caveats of this datasets?

Author Response

If this data was that much important, Please explain the reasons, why did this data not extended up to the present?

Concerning an update of these grids, since 1980, the ownership and the management of rainfall data was transferred to the countries and the access to those became very difficult. In the article (lines 501 to 507 of the manuscript) we mentioned this problem and the impossibility to update our work for the beginning of the 21th century. New data acquisition methods will have to be used (use of the attenuation of the GSM signal or of radar waves).

Can put some of the caveats of this datasets?

The comparison with the CRU grids contains the differences in the process and the areas where the results are different. The major differences apprear in the boreal hemisphere, area where the HSM_SIEREM database have more observations due to the presence of hydrologists of IRD in West and Central Africa.

Reviewer 2 Report

The paper present an intresting topic for the journal readers. However some correction must be done.

Firstly, at the indoduction the novelty of the paper must be cleary defined and the state of the art in this topic must be extensively describted and add literature.

Moreover, at the discussion a more professional work must be done. You have to compare your results with other similar studies and give differences and new points of view.

Finally, the englishmust be improved so as to be easier to read and undrestand. 

Author Response

Firstly, at the indoduction the novelty of the paper must be cleary defined and the state of the art in this topic must be extensively describted and add literature.

The introduction has been widely rewritten and this work is  better  placed in its context and references have been added.

Moreover, at the discussion a more professional work must be done. You have to compare your results with other similar studies and give differences and new points of view.

First, asked by another reviewer, we added a section where the consitution of the CRU grids, this section answers partly your request. The first comparison between the two mean annual grids over the period 1940-1999 of the CRU and  HSM_SIEREM is discussed.  It comes mainly from the way the grids are built, one is more a statistical (CRU) method due to an unsufficient amount of data, the other is built using only observed data (due to a more consistant database). The comprison stands only on the precipitation grids of the CRU, their aim was more ambitious than ours, they calculated grids worldwide for 7 climatic parameters.

Finally, the englishmust be improved so as to be easier to read and undrestand. 

This has been done by an american speaker.

Reviewer 3 Report

This review is for the manuscript entitled A new 60-year 1940-99 monthly gridded rainfall data set for Africa and First climatological analysis and authored by Claudine Dieulin, Gil Mahe, Jean-Emmanuel Paturel , Soundouss Ejjiyar , Nathalie Rouche , Bouabid El Mansouri. With this manuscript, the authors’ aim to elaborate a reference precipitation dataset of monthly rainfall grids over African continent using historical using a historical record of rain gauge data.

The topic of the manuscript is of high importance from water resources perspectives. Overall, the manuscript is informative but requires to be organized and improved in some areas before it can be considered for publication. Based on these suggestions, I believe the manuscript requires major revisions in several areas.

General comments:

1.     In the Introduction section, need more elaboration and references on similar reference precipitation datasets and different techniques used for interpolation/resampling. Is there other precipitation datasets for this time period (1940-1999) currently available? If yes, what are their limitations? Why there is a need for novel precipitation dataset?

2.     In the result part, I would like to see  a cross validation experiment that discusses the performance of the interpolation. Specifically, a chart that discusses the skill, MAE, NSE etc. between model gridded precipitation and observed. 

3.     In the discussion, the limitations of the study should be focused more.

Author Response

1.     In the Introduction section, need more elaboration and references on similar reference precipitation datasets and different techniques used for interpolation/resampling. Is there other precipitation datasets for this time period (1940-1999) currently available? If yes, what are their limitations? Why there is a need for novel precipitation dataset?

The introduction has been widely rewritten and this work is  better  placed in its context and references have been added.

A new section was added that describe the way CRU built the grids that they provide. The constitution and the mean annual grid of CRU and HSM_SIEREM are compared and this comparison give the limitations of the two products.

In matter of precipitation dataset, Africa is poorly represented. Some databases exist but very few for the whole continent, preventing a global analysis on a long term.

2.     In the result part, I would like to see  a cross validation experiment that discusses the performance of the interpolation. Specifically, a chart that discusses the skill, MAE, NSE etc. between model gridded precipitation and observed.

Yves Tramblay, a specialist of data processing in Montpellier, who has become co-author of the paper ran a cross-validation. In section 2.2.4 we present the results of it;  it compares IDW without taking elevation into account and Ordinary Kriging taking elevation into account. The average correlation coefficient is 0.86 for IDW and 0.85 for OR, i.e. very close. This cross-validation was done with a random use of the 95 measurement points having data over the whole 1940-99 period.

Section 2.2.4 with figure 6 displays this part.

3.     In the discussion, the limitations of the study should be focused more.

Section 2.2.5 compares the grids from CRU with HSM_SIEREM and in which area the grids differ.

These two sections show in what the two products are different. The section dealing with the fitting of the statistical test results to the known evolution of the African climate in the literature confirms the quality of the HSM_SIEREM grids.

Reviewer 4 Report

This study represents an extensive and painstaking effort to construct a long-term gridded rainfall product for Africa that improves upon existing products (such as CRU). This has the potential to be of significant value to the community. However, I have some concerns about the methodology and analysis presented that I think should be addressed prior to publication.

The gridding procedure appears to be limited to choices available in ArcInfo, as opposed to the best procedure possible. For example, the fact that ArcInfo does not include a straightforward method for handling elevation is not a scientifically appropriate rationale for ignoring its importance. There is also no discussion of the sensitivity of the analysis to the choice of parameters in section 2.2.2.

The observed and gridded products identify the 'rupture' in the timeseries at different years if I understand Table 2 correctly. This suggests that either the gridding procedure is faulty in some fashion or the timeseries analysis is very sensitive to the details of the data, which should be addressed in the discussion.

The identified breaks in the the timeseries also coincide with years where there are sharp changes in the number of available stations. It would be very good to see some sensitivity tests where the number of stations is held constant or only stations that are available for the entire record are used. 

Minor comments:

Figure 7 - use the same ranges for each figure

Lines 287-288: Is this always true locally? Are there no regions or periods where CRU includes more stations?

Author Response

The gridding procedure appears to be limited to choices available in ArcInfo, as opposed to the best procedure possible. For example, the fact that ArcInfo does not include a straightforward method for handling elevation is not a scientifically appropriate rationale for ignoring its importance. There is also no discussion of the sensitivity of the analysis to the choice of parameters in section 2.2.2.

The use of ArcInfo was done due to our expertise with the tool. Nevertheless, taking into account your remark, we added a section 2.2.1 dealing with the way the CRU built their precipitation grids. 

Their process is mainly statistical based on anomaly grids on a standard reference period 1961-1990, period during which they had the most complete data, but those do noe exceed 2000 measurement stations in between 1960 and 1980. When the distance between 2 stations was more than 450 km, they created "synthetic stations".

They used thin-plate spline interpolation method on large tiles (5 for the African continent), these tiles overlapping of 5X5 degrees. This method takes into account longitude, latitude, elevation. The overlapping areas were interpolated with a linear function of the inverse-distance.  

The anomaly grids (in percentage) are used to create the precipitation grids, these grids are calculated usint ADW, and this interpolation does not (cannot) take into account elevation of the measurement points.

We also added a section 2.2.4 for the validation of the IDW interpolation, top evaluate the robustness of the interpolation method. This validation is run on the whole base, using a set of 95 stations having complete data for the whole period and Ordinary Kriging (taking into account elevation of the station) and IDW (without taking into account elevation). The validation result give similar performances between IDW and OK.

The observed and gridded products identify the 'rupture' in the timeseries at different years if I understand Table 2 correctly. This suggests that either the gridding procedure is faulty in some fashion or the timeseries analysis is very sensitive to the details of the data, which should be addressed in the discussion.

In section 3.3, we give a first clue to this difference in the date of rupture. One clue can be that the rupture in 1969 occurs in West Africa, region where the database had the most complete number of stations, even in the 1990s. During this decade, the number of stations is really sparce even inexistant in the Southern hemisphere (Angola, Democratic Republic of Congo, Namibia). This hetherogeneity in the quantity of data is probably the reason why gridded data have a rupture data different from observed data.

The identified breaks in the the timeseries also coincide with years where there are sharp changes in the number of available stations. It would be very good to see some sensitivity tests where the number of stations is held constant or only stations that are available for the entire record are used. 

We agree with you and this has been added in the discussion part of the article. This part has very deeply been rewritten.

Minor comments:

Figure 7 - use the same ranges for each figure

This has been done in the new version of the article.

Lines 287-288: Is this always true locally? Are there no regions or periods where CRU includes more stations?

CRU include always less stations than the ones we had. Hereunder is the graph displayed in reference [26] that shows that even during their “standard reference period” the maximum of stations is less than 2000 when ours is over 3000 for the same period.

Round  2

Reviewer 3 Report

I recommend the paper can be published in the current form.

Reviewer 4 Report

The authors have addressed the concerns raised in my review and I feel the manuscript will be ready for publication after some minor additional proof-reading.