The Impact of SMOS Soil Moisture Data Assimilation within the Operational Global Flood Awareness System (GloFAS)

Round 1

Reviewer 1 Report

This is a great paper. I suggest acceptance for publication after some minor revision. My reasons are as follows:

Soil moisture is very important for hydrological studies. In the past, our forecast relies heavily on observations in the air than in the soil, mainly because of the limited availability of soil moisture data. With the increasing availability of soil moisture datasets from multi-sources, it provides better opportunities to examine the role of soil moisture in hydrological models. It can be assumed that the soil moisture wet/dry status is of great importance to the accuracy of flooding prediction models. Therefore, the manuscript presented by the authors is important and in time.

I have to point out, that linking soil moisture to flooding models are not linear and transparent work, due to the complex procedures that the whole process may involve. Soil types, vegetation cover, flooding models, spatial resolution of the datasets, etc., all contribute to the sensitivity of the results. Therefore, showing results not strictly in line with the assumption is sure within my expectation. Although the paper concluded that the soil moisture did not show constantly huge importance to the flooding model at all high flow/low flow scenarios, the concept of using data denial analysis to test the importance of soil moisture data in terms of flooding model accuracy is meaningful.

I suggest the authors keep working on this topic in the future using some alternative, high-resolution, spatially consistent datasets, such as SMAP, NLDAS, etc. with the consideration of downscaled soil moisture products. Cheers!

Specific revision suggestions for the authors:

Line 90 Please define LISFLOOD as this is the first appearance of it.

Please add a paragraph of how you process the difference of the temporal resolution (Daily, hourly, etc.), and how you align the UTC with SMOS’s local solar time.

Datasets section: As this study included so many datasets (or models) and equations, I strongly recommend that all the datasets, including their spatial and temporal resolutions, should be listed in a table to show a more organized structure.

Author Response

Thank you very much to the reviewer for your kind comments on our manuscript. We agree that it is a challenging subject especially given the non-linearities and uncertainties within hydrological modelling. We will endeavour to continue with this work when new higher resolution datasets become available.

In response to your specific comments:

Line 90 Please define LISFLOOD as this is the first appearance of it.

LISFLOOD is referred to in previous literature as a proper-noun rather than a defined acronym, see van der Knijff et al., 2012 [https://www.tandfonline.com/doi/abs/10.1080/13658810802549154]. The model was based on the LISEM model which stands for LImburg Soil Erosion Model [de Roo & Offermans, 1995: http://hydrologie.org/redbooks/a231/iahs_231_0399.pdf], but was adapted for flood forecasting hence the ‘FLOOD’ component. However it is not correct to define LISFLOOD as the LImburg Soil FLOOD model. This is why we have not defined the word in the text.

Please add a paragraph of how you process the difference of the temporal resolution (Daily, hourly, etc.), and how you align the UTC with SMOS’s local solar time.

Lines 239-254 and Table 2 explain how the 6 hourly data from the IFS experiment were aggregated to the 24 hour resolution of the GloFAS experiment. However we have made some edits to the table to improve its readability.

Regarding solar time, ECMWF receives the level 1c SMOS data from ESA who convert the local solar time to UTC. Therefore when ECMWF receive the data from ESA it is already in UTC and no further conversion is required. We have added this sentence into line 131 of the manuscript “ECMWF receives the SMOS level 1c data from ESA where it has already been converted from local solar time to UTC, this enables it to be incorporated into the LDAS which is based on UTC.”

Datasets section: As this study included so many datasets (or models) and equations, I strongly recommend that all the datasets, including their spatial and temporal resolutions, should be listed in a table to show a more organized structure

Thank you for this suggestion, we have added the following table to line 269 of the manuscript:

Reviewer 2 Report

I would suggest integrating the Materials and Methods section with a flow chart in order to make easier and faster the understanding for the reader.

Results:

Even if it is clearly explained that it wasn’t possible to point out a clear spatial trend for the impact of SMOS data assimilation upon GloFAS at a global scale, I would suggest a more detailed investigation to point out whether certain land uses and land cover types are more frequently represented among areas where the assimilation of SMOS soil moisture improves (or degrades) the GloFAS prediction skill. Is it possible to identify those land cover types more frequently occurring where the GloFAS simulation including the assimilation of SMOS soil moisture data outperforms the simulation without the assimilation of SMOS (maybe through an overlay between KGE_mod skill scores and a land cover map in GIS environment)? How could be classified (i.e. wooded grassland, bare soil…) those areas showing the best/poorest results as to KGE_mod skill scores (if possible)? Similarly, I would more deeply discuss the Figure 8, showing the difference in the 95^th percentile of specific discharge from the GloFAS simulations with and without SMOS data assimilation, as regards land cover types and environmental conditions (it was shortly addressed: lines 479-480).

Specific comments:

Line 46-to-49, Page 2: “This is because the hydrological prediction chain starts with the IHC’s to initialise a hydrological model, forcings from Numerical Weather Prediction (NWP) forecasts are then used to produce a streamflow forecast”. It is not very clear to me, I would write: “[..] IHCs (being used) to initialise a hydrological model [..]”.

Line 111, Page 3: “top few centimeters” instead of “top few centimetres”.

Author Response

I would suggest integrating the Materials and Methods section with a flow chart in order to make easier and faster the understanding for the reader.

Thank you for this suggestion, it is similar to the comment made by reviewer 1 who suggested adding a summary table of all the datasets. We have added this table (table 2 on line 269 in the manuscript) to the manuscript and believe that this will make this section easier to understand for a reader.

I would suggest a more detailed investigation to point out whether certain land uses and land cover types are more frequently represented among areas where the assimilation of SMOS soil moisture improves (or degrades) the GloFAS prediction skill. … Similarly, I would more deeply discuss the Figure 8, showing the difference in the 95th percentile of specific discharge from the GloFAS simulations with and without SMOS data assimilation, as regards land cover types and environmental conditions (it was shortly addressed: lines 479-480)

Thank you for this suggestion, in response at each observation station location in the United States and Australia we extracted the land cover classification from the ESA Climate Change Initiative (CCI) data for 2018 (https://www.esa-landcover-cci.org/; https://cds.climate.copernicus.eu/cdsapp#!/dataset/satellite-land-cover?tab=overview). We then identified all stations where the modified Kling-Gupta Efficiency skill score (KGE_modSS) was <=-0.05 (indicating a degradation with SMOS data assimilation) and all the stations where KGE_modSS was >=0.05 (indicating an improvement with SMOS data assimilation). Within each of the degradation and improvement categories, these were further broken down into the landcover classes from ESA CCI. Results in table 1 below show that for both degradation and improvement most stations belonged to the grass, tree, water and shrub landcover classes. Therefore it appears that the landcover status does not explain the spatial pattern of degradations or improvements in the GloFAS prediction skill.

Table 1. Modified Kling-Gupta efficiency skill score (KGE_modSS) values at station locations in the United States and Australia calculated from the GloFAS experiments with and without SMOS data assimilation broken down by ESA CCI landcover class. The second column shows stations where KGE_modSS degraded with SMOS data assimilation, the third column shows where it improved.

At the global scale we have compared the ESA CCI landcover map against the changes in the GloFAS 95^th percentile of specific discharge. The greatest changes in the 95^th percentile occurred in northern Canada, central United States, the Sahel and Australia. These areas all coincided with open landcover classes such as sparse vegetation, herbaceous, grassland, cropland and grassland. SMOS is better able to capture soil moisture conditions in these settings as there is less interference from the forest canopy or urban infrastructure for example. In other landcover classes such as urban or forested areas the greater interference to the SMOS signal will result in a higher measurement error which will likely lead to the removal of such data points prior to assimilation within the ECMWF IFS. This explains why weaker or no changes are detected in the GloFAS 95^th percentiles in these areas.

There was already a paragraph in the discussion section (line 505) which discussed the relationship between the changes in GloFAS streamflows and landcover type, but this has now been expanded to read include the station analysis and the global scale discussion:

“The areas of the world which showed the greatest impact upon high flows in this study appeared to coincide with areas which have open land covers (Figure 8). Comparing the results of this study against landcover data from the ESA Climate Change Initiative (CCI) dataset for 2018 [57] confirms that the greatest changes occur in sparsely vegetated, herbaceous, grassland, cropland and shrubland classes. Forested and urban areas showed little impact of SMOS soil moisture data assimilation upon GloFAS streamflow predictions. This is likely because SMOS measurements in these areas are subject to interference which increases the measurement error, meaning they are filtered out and are not assimilated into the model. It may be possible that certain land cover types are associated with either an improvement or a degradation of GloFAS streamflow skill with the assimilation of SMOS soil moisture data. To investigate this further, at each observation station location in the United States and Australia the land cover classification from the ESA CCI data for 2018 [57] were extracted. Then all stations where the modified Kling-Gupta Efficiency skill score (KGE_modSS) was <=-0.05 (indicating a degradation with SMOS data assimilation) and all the stations where KGE_modSS was >=0.05 (indicating an improvement with SMOS data assimilation) were identified. Within each of the degradation and improvement categories, these were further broken down into the landcover classes from ESA CCI. Results show that for both degradation and improvement most stations belonged to the grass, tree, water and shrub landcover classes (Table 5). Therefore it appears that the landcover status does not explain the spatial pattern of degradations or improvements in the GloFAS prediction skill.”

Line 46-to-49, Page 2: “This is because the hydrological prediction chain starts with the IHC’s to initialise a hydrological model, forcings from Numerical Weather Prediction (NWP) forecasts are then used to produce a streamflow forecast”. It is not very clear to me, I would write: “[..] IHCs (being used) to initialise a hydrological model [..]”.

Thank you for this suggestion, we agree that the original wording is unclear. We have rephrased the sentence to the following: “This is because the hydrological prediction chain starts with the IHC’s which are used to initialise a hydrological model, then forcings from Numerical Weather Prediction (NWP) forecasts are used to produce a streamflow forecast.”

Line 111, Page 3: “top few centimeters” instead of “top few centimetres”

Done, all other instances of this spelling have also been corrected.