the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Integrating satellite-based soil moisture within a rainfall-runoff database for operational flood forecasting in Mediterranean catchments
Abstract. Flood forecasting in Mediterranean catchments is challenging due to strong rainfall variability and rapidly evolving antecedent wetness conditions. These challenges are further amplified in data-scarce basins, where limited hydrometric observations constrain the application of conventional real-time forecasting approaches. This study develops a probabilistic Rainfall-Runoff Database (RRDB) for the Crati River basin (2440 km2) in Calabria, southern Italy. The framework combines stochastic rainfall and air-temperature generation with a continuous semi-distributed hydrological model to generate, offline, a large ensemble of precomputed hydrographs spanning different event durations, precipitation amounts, and antecedent soil moisture conditions. The RRDB is evaluated across the three gauging sections in the study basin: Cosenza-Ponte Europa, Castiglione Cosentino, and Sibari using observations of precipitation, satellite soil moisture and streamflow. The stochastic rainfall generator reproduces the main statistical properties and extremes of observed precipitation, while the hydrological model performs well across the evaluated gauging stations, with Nash-Sutcliffe efficiency (NSE) = 0.85 and Kling-Gupta efficiency (KGE) = 0.93 at the Sibari calibration section. Independent comparison of the modelled soil moisture with Advanced SCATterometer (ASCAT) satellite observations yields correlation coefficients of 0.50–0.72 across the three sub-catchments, supporting the use of satellite observations to characterize basin antecedent wetness conditions. Event-based validation shows that most observed hydrographs, peak discharges, and runoff volumes fall within or close to the corresponding RRDB probabilistic envelopes, although some extreme peaks are underestimated. The RRDB is evaluated in operational mode during the February 2026 flood event through its integration into HYPES (Hydro Prediction Engine System), an operational decision-support and flood forecasting platform. In this setup, forecast rainfall and satellite-derived surface soil moisture are used to identify representative precomputed scenarios within the RRDB and retrieve probabilistic discharge forecasts in real time. At Sibari, the observed peak discharge of 663 m3 s-1 falls within the interquartile range of the retrieved probabilistic forecast (497–840 m3 s-1), with an expected value of 674 m3 s-1.
- Preprint
(3977 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 04 Dec 2026)
- RC1: 'Comment on egusphere-2026-5965', Anonymous Referee #1, 07 Nov 2026 reply
-
CC1: 'Comment on egusphere-2026-5965', Mulugeta Tekleyohannes Hailekiros, 12 Nov 2026
reply
Introduction
This manuscript develops a probabilistic Rainfall-Runoff Database (RRDB) for the Crati River basin in northern Calabria (southern Italy) for the purpose of operational flood forecasting. It integrates stochastic rainfall and temperature generation with the semi-distributed MISDc hydrological model. A satellite derived ASCAT surface soil moisture is used to describe antecedent moisture conditions, and an operational demonstration is provided for the February 2026 flood via the HYPES platform.
The manuscript addresses an important and timely challenge: flood forecasting in rapidly responding, data-scarce Mediterranean catchments. In addition, the concept of relying on a precomputed probabilistic database (RRDB), conditioned on satellite-derived soil moisture, is vital as it reduces the need for real-time hydrological model execution while still retaining uncertainty information. Overall, the manuscript is well organized, clearly presented, and supported by an extensive set of simulations and event-based validation datasets.
Furthermore, I consider the study a valuable contribution because it brings together three elements, i.e., synthetic forcing generation, hydrological modelling, and satellite soil-moisture conditioning into a single operational workflow. The February 2026 case study is particularly interesting since the observed peak discharge at Sibari lies within the RRDB interquartile range. I also appreciate how the authors explain their rationale for selecting the Crati River basin as a representative basin for the Mediterranean region.
Major comments
- Abstract and operational application: The manuscript acknowledges in conclusion that the operational assessment is based on a single event and that additional testing is required. However, the abstract states that the RRDB was “evaluated in operational mode” and used to retrieve probabilistic discharge forecasts “in real time” without explicitly indicating that this evaluation is limited to the February 2026 flood event. I suggest adding a phrase in the abstract, such as “for a single flood event,” to ensure that the operational demonstration is not interpreted as evidence of established operational forecast skill.
- The independent comparison between ASCAT and MISDc soil moisture requires more careful interpretation. The correlation values are 0.72 at Sibari but only 0.50 - 0.54 at the two internal sub-catchments. The authors interpret these results as supporting the use of ASCAT to define antecedent conditions, but the reason for the weaker agreements are less described and needs a more detailed discussion.
- The selection of “representative” RRDB scenarios should be more precisely defined. Section 3.4 states that the system extracts hydrographs associated with “similar” rainfall and soil-moisture conditions, but the exact matching rules are not given in this manuscript. What degree of tolerance is used around the observed cumulative rainfall and soil-moisture value? How many simulations are retrieved? Are the 24, 48, and 72 h duration classes treated separately? These details are vital for the reproducibility of the method used and to judge whether the probabilistic envelope is meaningful.
- The treatment of rainfall forecasts should be clarified. In the February 2026 application, the RRDB appears to be queried using observed rainfall over the 24 h before the peak, whereas a true operational forecast would require a quantitative precipitation forecast. The authors mention this in the conclusions, but the operational section should state more explicitly whether the demonstration is retrospective or genuinely real time. If it is retrospective, the claimed “24 h lead time” should be phrased carefully.
- The performance at validation stations should be discussed more critically. Like correlation values, the NSE values at Cosenza-Ponte Europa (-0.06) and Castiglione Cosentino (0.24) are weak. Even though the authors attribute this partly to uncertain rating curve, the implications of these weak discharge validations on the reliability of the upstream RRDB ensembles should be discussed since the RRDB is evaluated at all three stations. This will help to show whether the poor performance is mainly due to discharge observation uncertainty, model structural uncertainty, spatial rainfall errors or a combination of these factors.
- The independence of the RRDB validation could be more strengthened. The RRDB is built using MISDc, calibrated at Sibari, and then evaluated against observed events also using MISDc-derived initial soil-moisture. Although ASCAT provides an independent soil-moisture comparison, the event validation remains closely tied to the same model used to construct the database. Please clarify which parts of the validation are truly independent and which are internal consistency checks.
- Extreme-event performance deserves further analysis. The authors highlighted that some observed peak discharges exceed the upper envelope of RRDB. Because flood warning is most important precisely during extreme events, please quantify how often this occurs across all events, rather than describing it qualitatively. Including a table or figure showing the percentage of observed peaks and runoff volumes that falls within the 50% and 90% envelopes, by station and duration class, would greatly strengthen the validation.
- Moderate claims of transferability of the framework developed. The conclusion states the transferability of the framework to data-scare regions as it only requires rainfall and temperature observations and satellite soil moisture. This is an important claim, but the study uses a calibrated hydrological model, local discharge data to validate the model, and a basin with at least some hydrometric observations. Could you please clarify what minimum data requirements are actually needed and do not imply that the approach is ready for ungauged basins without further testing.
Conclusion of the review
This is an interesting and potentially useful manuscript presenting a well-constructed probabilistic rainfall-runoff database and a promising integration with satellite soil moisture for operational flood forecasting. Generally, the methods are appropriate, the case study is relevant, and the February 2026 event flood application shows the practical potential of the approach.
However, the manuscript could be improved by a more careful interpretation of the single-event operational test, a clearer description of the real-time selection of RRDB scenarios, and a more critical discussion of the moderate ASCAT–MISDc agreement and weak validation performance at the internal gauging sections. I therefore suggest a major revision. The underlying work is promising, but the operational claims and validation evidence need to be more transparent and better quantified before the manuscript can be accepted.
Minor comments
- Abstract, lines 19–21: “Independent comparison” is used for the ASCAT evaluation; please specify that independence refers to the soil-moisture variable, not to the hydrological model calibration.
- Abstract, lines 25–26: Clarify whether “forecast rainfall” refers to observed rainfall used retrospectively or an actual operational quantitative precipitation forecast (QPF).
- Introduction, lines 65–67: The statement that no previous study has integrated satellite soil moisture into a precomputed RRDB is strong; consider softening it to “to the best of our knowledge” if not already sufficient, and ensure that this can be fully supported by literature review.
- Section 2.1.1, line 103: The hydrological model is introduced only as “MISDc,” but its full name is not provided anywhere in the manuscript. Please define the acronym at first mention, for example: “the continuous semi-distributed MISDc (Modello Idrologico Semi-Distribuito in continuo) model.”
- Section 2.1.1, line 158: Similarly, the statistical method is introduced only as “ARIMA,” but its full name is not provided anywhere in the manuscript. Better to define it at first mention.
- Section 2.1.1, lines 100–103: Since rainfall and temperature are generated independently, consider adding a brief justification earlier in the methods for why this assumption is acceptable, or explicitly refer the reader to the sensitivity analysis in Sect. 3.2.1, where the implications of this assumption are quantified.
- Figure 2: In panel (b), the caption refers to three sub-catchments, but only two appear to be clearly highlighted; all three should be distinctly coloured and identified in a legend. Please revise this figure for clarity and consistency. In addition, please also align the three map panels more closely, reduce the size of the north arrow, and use a consistent font size across all panels, particularly panel (a).
- Section 2.3, lines 240–245: Please provide more information on how discharge uncertainty from the rating curves was assessed or cite the specific evidence supporting the statement that Sibari is the most reliable station.
- Section 2.3, lines 247: The acronym EUMETSAT should be defined at first use.
- Figures 3 and 4: These multi-panel figures are difficult to read at their current size. Please consider simplifying them, increasing the size of individual panels, or moving the full versions to supplementary material. Figure 4 could be moved to the Supplementary material and replaced in the main text with a concise summary of the agreement between observed and simulated air-temperature distributions.
- Section 3.2, lines 345–354: Consider adding a short explanation of why NSE can be negative in Cosenza-Ponte Europa while R and KGE remain positive; this would help readers interpret the metrics.
- Figure 6: Panels (b) and (c) appear to be in the reverse order compared with the caption. Please rearrange the panels or correct the figure caption to ensure that each panel corresponds to the correct sub-catchment.
- Figure 7: Please ensure that the font type and size of the axis labels are consistent with the manuscript and with the other figures.
- Figure 8 caption, line 435: Specify whether “soil moisture = 0.5” is a relative soil-moisture value and explain why this particular value was chosen in the main body of text before the figure.
- Section 3.4, lines 505–506: What does the term ‘expected value” mean, is it the ensemble mean, median, or another statistic? It is better to define it precisely
- Figure 11: Since the interface is in Italian, consider adding English translations directly in the caption for all key labels. Moreover, in Figure 11b, the text at the top of the figure appears larger than the text in the rest of the figure.
Citation: https://doi.org/10.5194/egusphere-2026-5965-CC1
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 289 | 97 | 28 | 414 | 33 | 34 |
- HTML: 289
- PDF: 97
- XML: 28
- Total: 414
- BibTeX: 33
- EndNote: 34
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The manuscript describes a framework that combines stochastic rainfall and air-temperature generation with a continuous semi-distributed hydrological model to generate a large ensemble of precomputed hydrographs spanning different event durations, precipitation amounts, and antecedent soil moisture conditions. The generated hydrograph scenarios are used to identify plausible hydrological responses under different rainfall and soil moisture conditions without real-time model running. The scientific significance and applicability of the method are of concern. Therefore, I rejected this paper for now, but would encourage resubmission if my comments below can be well addressed.
One motivation of the study is flood forecasting in poorly monitored catchments. However, the selected study region has well-monitored rainfall, air temperature, and discharge to be used for the methods. I understand the author would argue about the reliance on satellite-based soil moisture, but such a statement is misleading. I would suggest the author rephrase the relevant sentence.
Many rainfall-runoff models are computationally efficient. I did not see the value of this method compared to directly running the rainfall-runoff model in real-time with forecasted precipitation, as the latter is usually far more accurate. I may miss something here, but I would suggest the author include some evidence to support the necessity of a pre-calculated hydrograph.
A limitation of the method is that no rainfall spatial variability is considered when identifying the hydrograph ensemble. This limitation may be negligible for small watersheds, but for large watersheds, rainfall spatial variability can cause large differences in generated hydrographs.
The manuscript has a lot of figures plotting the simulated hydrograph and observed hydrograph. In the supplementary, in many cases, the simulated hydrograph can not capture the observations. I would recommend the authors use some metrics to quantify the ensemble performance, such as reliability diagrams, rank histograms, and skill-spread plots. These evaluations can show more clearly the skill of the proposed method in flood forecasting.
The objective of the study is to validate the use of satellite-based soil moisture in a rainfall-runoff database-based flood forecasting. Besides the lack of quantitative evaluation mentioned above, the sensitivity of soil moisture is not addressed. A comparison with in-situ soil moisture data may be needed.