Reply to 1st comment on the manuscript of title: 

"Soil Moisture consistency for operational drought monitoring" 

We thank the reviewer Mathias Zink for his thorough revision of the text and his positive impression of the quality of the manuscript and research conducted. His insightful comments on the convenience of emphasizing the novelty and target research gap addressed by the study as well as about the minor corrections have been of great help to improve the manuscript. 

In the following lines, we incorporate detailed answers to both the major items requiring attention and the minor corrections demanded. Reviewer demands in bold, author's response without bold and when referring to long chunks of introduced text in italics)

Major items 

• The authors should make clearer what is the novelty and the research gap addressed by their study. 

There is no single soil moisture (SM) product able to address the monitoring of SM over all environmental conditions given the specific limitations of the different SM data types (i.e. either model-based or from remote sensing active and passive missions experience uncertainties over land covers dominated by vegetation, in climates where snow has a relevant role on the SM regime or over areas of difficult soil characteristics). Furthermore, some combined soil moisture products may overlook the consideration that the products combined should be independent, particularly regarding any participating modelling schemes, in order to maximize the informative value of the combined product or at least to avoid inherited uncertainties. Therefore, there is need for products able to monitor SM evolution over a wider range of conditions which encourages the exploration of combined SM products. All these aspects have been clarified along different sections of the manuscript (abstract (L19-20), last part of introduction where objectives of the study are described (e.g. L111-115), and results (e.g. L395-397, L400-402) and conclusion (e.g. L595-598). 

• I suggest to change the title by removing the term drought as the paper is not referring to droughts at all. 

While it is true that most of the article refers to a great extent to the SM data of diverse sources instead of to drought characteristics derived from it, there are results using the anomalies of SM (e.g. trends) which indirectly define thresholds, magnitude and duration of the drought events. Nonetheless, we clarify this aspect in multiple sections (e.g. L116) to prevent readers from wondering which is the commented relation of SM to drought. 

• How are quality assurance scores (ESA CCI, ISMN) treated within this study? 

Quality assurance of ESA CCIp: The main source of quality uncertainty in the case of the application of ESA CCIp over the European continent is derived from the seasonal presence of snow and related processes in the northernmost areas and mountain ranges. Results of the linear and TCA analysis display how snow-capped regions stand out as the most uncertain areas, independently of their vegetation cover (i.e. the other primary factor of uncertainty), primarily due to the altered conditions for signal retrieval under snow cover or frozen soils. Flags of ESA CCIp dataset for snow and frozen soil not only consider snow cover detection based on temperature but also on freeze and thaw conditions based on (Ku-, K- and Ka-band) retrieval (CCIp ATDB guide). In consequence, we apply ESA CCIp snow flags to remove the seasonally affected areas, and as the most restrictive flag of these conditions among the three products, we apply these flag restrictions in all analyses to the rest of the SM products over the same boreal areas.  

Quality assurance of ISMN: Flags of ISMN data are present for multiple factors of influence beyond the features of SM series themselves (e.g. temperature and rainfall threshold beyond for instance SM series characteristics such as saturation). In spite of the worth of these multiple flags, all the stations that are included in the ISMN database after harmonization and quality control across Europe with available data in the period were included. This means that despite some flags in certain stations indicating dubious values of SM or about associated variables (such as BIEBRZA network which is known for providing very high SM values) all data available was included, since beyond the reliability of key ISMNs (according to the Quality Assurance for Soil Moisture initiative identifying the most consistent networks) there is no abundance of stations over all the climate types of Europe. Therefore, it was preferred to deal with uncertainty due to intermediate quality scores of some stations’ data than losing the power of valuable yet not optimal SM data from non-key ISMN stations for validating the remote-sensing and model-based products across such a wide range of conditions from boreal to semi-arid conditions. 

We have included mention to these relevant aspects at the end of section 3.4.1.: 

L226-230: “The different product additionally provides metrics of the error characteristics to identify the areas and periods affected by relevant impactful factors such as snow cover. The flag scheme of ESA CCIp exemplifies the detailed procedures devoted to distinguishing when the data is subject to further filtering. Consequently, considering the notable importance of the snow cover factor in our analysis over the boreal and mountainous areas of Europe, we have applied the flags of ESA CCIp as mask to the data coverage of the other products of the analysis over the snow-prone regions. “

In L232-237: “Conversely to the case of the distributed datasets, the point data from the ISMN database has not been significantly restricted with flags due to the scarcity of this type of data. Multiple areas where snow and icing processes are frequent are barely observed in situ, and consequently, even despite the seasonal uncertainties, RS products provide much more coverage of these areas than the few ISMN stations over these areas. Therefore, including all ISMN networks and data available was the decision adopted to ensure sufficient amount of data for validation over every climatic type and ensure the representativity of a wide range of observed soil moisture conditions.” 

• L212 How do the datasets used fulfil these assumptions (of TCA)?:  
  
  This study focused on soil moisture data types originating from distinct model-based, active and passive remote sensing missions precisely to avoid overlooking the interdependence existing between multiple available soil moisture products. In numerous cases, remote sensing products of enhanced characteristics such as downscaled or filtered data do introduce further uncertainty by ingesting or using models or equivalent products in their validation. Therefore, our application of only the passive CCIp dataset instead of the combined one aimed to avoid dependence between the active ASCAT H120/H121 dataset and CCI combined also including this MetOp datasets. In this way, we proposed the use of ASCAT H120/H121 as a consistent product known for the quality of its curation as representative of the active RS SM products while CCIp as representative of a well-documented passive product generated without assimilation of models or active remote sensing data. For the same reason we choose the LISFLOOD model inside EDO instead of popular land-surface models participating in reanalysis and the curation of multiple L3 and L4 remote sensing SM products as the model-based product. This choice ensured none of the other selected SM products for the study has utilized LISFLOOD in its generation. Additionally, given the primary operational use of LISFLOOD, it emphasizes our aim of evaluating soil moisture products for monitoring. Nonetheless, clarification on these aspects has been extended in Section 3.4.2. (L242-257): 

“ The TCA model assumes linearity of SM retrievals, stationarity of signal and independence of errors from signal or in between product errors (Gruber et al., 2016; Massari et al., 2017; Filippucci et al., 2021). However, since the purpose of the study is not to evaluate the sensitivity of the assumptions of the TCA model we focus on the adequacy of the selected SM products for triplets based on their reported independence and consistency. The LISFLOOD model is not used for the processing or validation of neither the passive and active RS products that complete the product neither on the curation of the ISMN data. The ASCAT active RS SM datasets upgraded from H120 to H121 are supposed to resolve the increasing trend and consequently can be considered only marginally non-stationary. CCIp uncertainties are to a great extent already resolved in the dedicated processing methods developed for the blending of data from different passive SM missions. The same assumption can be applied to the residual trends of EDO and CCIp SM datasets, that are of a magnitude likely more relevant for long-term analysis rather than for its impact on the TCA assumptions.“  

Regarding the assumptions of the TCA model, for the assumption of the independence of residuals, overall, the number of triplets available in the period of study is far bigger (up to 536) than the 100-threshold estimated by Scipal et al, (2008) to ensure the independence and no bias of the residual errors.  Even the regions prone to be snow-covered present enough coverage (far beyond the 100 values of the threshold) to guarantee that no downgrade in the reliability of the TCA model originates from a lack of data representativity. 

Fig. 1: Coverage in number of valid triplets for the analysis of the three products ASCAT H120/H121, CCIp and EDO across Europe in the period 2007-2022. 

 Regarding the assumed linearity between the signal and the errors, values of the logaritmic signal-to-noise ratio (logSNR) in decibels stay within ranges of linearity of between logSNR and R2 (Mean around 4-7 dB). Furthermore, debates on the linearity matter as reported in (Gruber et al., 2015) indicate that despite the bias that might be introduced by this assumption there is no clear alternative to the consideration of linear models to the relationship between the model and the data because other methods also introduce bias through matching, scaling and application of polynomials. Therefore, the major risk of non-linearity may arise from stationarity that is the following assumption discussed here. 

While some of the datasets included here display some non-stationarity (it is the second major goal of the study of evaluating such trends), the assumption refers to both signal and errors also being of constant variance over time. In the first case, since we already indicated in the study the notable correction of the trends from h120 to h121 product, the comparison between the values of error variance, sensitivity and logarithmic signal-to-noise ratio do not vary remarkably in between both cases. These results indicate that despite the importance of the trends for the analysis of changes in soil moisture anomalies, particularly if the analysis focuses on the identification of long-term variations due to climate change, the values of non-stationarity revealed in the Section 4.4. of the manuscript do not seem to compromise the validity of the TCA results in the study. 

 For the second case of the influence of non-stationarity, we checked the error variance and the other auxiliary metrics between different periods within the study period of known contrasted stationarity (2007-2015 (overall not trendy) and 2015-2022 (overall very trendy)). The results indicate that both the improved versions of the products (e.g. h121) and the slightly trended previous version of products (e.g. ASCAT h120) do not show notable differences in correlation (R_TCA), variance of the errors (VAR), sensitivity (SENS) and signal to noise ratio (logSNR)(metrics now specified at lines 269-274) when considering the different periods. Therefore, despite the inevitable existence of residual non-stationarity, given that the purpose of the study was to highlight the increasing temporal stability of products regarding trends, it has no noticeable consequences in the provided results. 

All in all, these considerations referring to the compliance of TCA with its assumptions, favor the independence of the results provided here, but if included in the manuscript might excessively lengthen the text or obscure the interpretation of the main results. They can be included as supplementary information, if necessary. 

Fig. 2a: Triplets of ASCAT H121, CCIp and EDO across Europe in the period 2007-2022 (leftmost column), 2007-2015 (central column) and 2015-2022 (right column) showing the results for H120 regarding correlation (R_TCA), variance (VAR), sensitivity (SENS) and logaritmic signal to noise ratio (logSNR).

Fig. 2b: Triplets of ASCAT H121, CCIp and EDO across Europe in the period 2007-2022 (leftmost column), 2007-2015 (central column) and 2015-2022 (right column) showing the results for H121 regarding correlation (R_TCA), variance (VAR), sensitivity (SENS) and logaritmic signal to noise ratio (logSNR).  

• L297ff The low scores of EDO in Northern Europe may arise from low performance but good agreement of the other 2 RS products. As you mentioned before they are known to be worse in forested and icy regions. Please elaborate a little in the discussion and add in situ observations as reference to get a better picture. 

 There are results in Figure 3 that illustrate the case and that can help clarify the origin of the low scores. The detailed description has been added to the section 4.1.1 (L342-349) and included here:  

 “The negative R-Pearson scores shown in Fig. 3 in between EDO and H120/H121, between CCIp and EDO, and to a lesser extent between CCIp and H120/H121 show even negative R-Pearson correlations in snow-dominated areas and thin soils in mountains and in arid climates. These negative values refer to disagreement between products beyond low performance. The combination express that EDO is the greatest source of such disagreement but differently for northern and southern latitudes: the negative scores induces by the participation of EDO seem to be mostly restricted to snow-dominated regions (compare Fig. 3a and 3b with 3c) while H120/H121 seems to be the primary source of low scores in the arid areas. These results are consistent with the known difficulties of EDO-Lisflood in snow-dominated hydrological regimes and of ASCAT H120/H121 over thin soils of arid areas.”

Figures 4 and 5 also explain differences between the same products in terms of performance over challenging regions, but in this case with triple collocation scores.  The additional comments have been added to the section 4.1.2 (L372-378) in between Fig. 4 and 5: 

“The low values of triple collocation correlations concurring over the snow-prone areas of Scandinavia and snow-caped mountain ranges (e.g. Alps) affect particularly EDO and CCIp, and experienced a significant improvement over snow-prone lowlands with the update of ASCAT data from H120 and H121. Beyond the evidence that EDO model is limited over snow-dominated regimes, CCIp seems to display a bimodal downgrade of scores with low values over Scandinavia and intermediate ones over the also seasonally snow-dominated but less densely vegetated East Europe suggesting that uncertainties originated from high-latitude and dense canopies may differ from those originated from the snow regime alone (Blyverket et al., 2019)”. 

Additional comments regarding this distinct limitation of each data product have been included in the Section 4.3 against in-situ data in the paragraph after Figure 7 (L454-L468): 

 "“However, the low performance of EDO compared to the other products over boreal areas is in fact better at the specific locations of the ISMN networks of Scandinavia (FMI, GTK, NVE) compared to the results when R Pearson or R_TCA scores are considered. This disagreement indicates more the need of additional networks in such a vast boreal area than actual better performance of model-based estimates over the remote sensing ones.”   

Minor items 

• L76 please rephrase numerical model into process based or conceptual 

Thanks for noticing the need of clarification 

• Fig. 1 Labels a and b missing. 

Corrected, thanks for noticing the need for clarification. 

• L171 Web address of ISMN is outdated. Please update with https://ismn.earth. 

Thanks for finding the outdated web address 

• Fig. 2 A distinction into further classes like forest, urban etc would be useful. 

Fig 2 primarily aims to illustrate the location of the ISMN networks used for the study while providing secondary information about the dominant land use in such location. Our purpose for the land cover colouring was to differentiate if the ISMN locations correspond to areas with (green) or without canopy (yellows) given that vegetation can be the most influential factor over the SM retrievals with remote sensing. Therefore, yellows comprise all land use classes without vegetation effects over the SM retrievals and green those land use classes with vegetation susceptible of effects. Urban areas are also currently colored in their own hue (pink) and in fig. 2 are also visible for great urban areas such as Paris, London, Berlin, Milan... All these aspects will be clarified in the caption of Fig. 2.: 

“Yellow/green colours comprise the multiple land cover classes without/with significant vegetation effects on the retrieval of soil moisture using remote sensing. Urban areas are coloured in pink. “ 

• L190 Please rephrase (The daily values ...) 

To clarify the meaning and improve the readiness of the manuscript at L190, the confusing text has been rephrased as "All



“All initial values within each of third of the month at daily time step are aggregated to the reference date of the corresponding interval.”

• L197 I guess SMsat shall be ms

Yes, it refers to surface soil moisture saturation, thanks for identifying the clarification needed. 

• I suggest to add a table detailing the different spatial and temporal representation/resolution of the data source

While the table can facilitate the understanding of the details of all datasets simultaneously, it barely will include many more details than those included in the data description. Therefore, while the table can still be added, we meanwhile provided clarified description of the spatial and temporal sampling, resolution and their resampling in their sections 3.1 and 3.2 

• Please review the use of hyphens instead of a comma when you refer to two panels in a figure, e.g, L260 (Fig. 3b-3e), L261... 

Thanks for realizing the misuse of it and the recommendation to revise them. 

• L229 R TCAHSAF should be SMHSAF ___ correct the equation 5 

Thanks for identifying the mistaken term. Corrected 

• L270 caption of Figure 3: Map of temporal R-Pearson correlation instead of spatial. 

Yes, corrected, thanks for noticing. 

•  L318 English grammar, please rephrase (Hence, ...). 

Rephrased and clarified in meaning. 

• Fig. 7: Please add an axis label to all x axes. 

While some of the axis may seem to lack label, as all products a-f) rephrased and clarified in meaning. 

• L372 Please rephrase, sentence is hard to understand. 

Rephrased and clarified in meaning. 

• L399 I am not quite sure about which dashed lines you are talking about. 

It was a mistake, referring to previous versions of the figure. 

• L400 Dfb instead Db 

Yes, corrected, thanks for noticing.