Challenges in flood modelling over data scarce regions : how to exploit 1 globally available soil moisture products to estimate antecedent soil wetness 2 conditions in Morocco

Abstract. The Mediterranean region is characterized by intense rainfall events giving rise to devastating floods. In Maghreb countries such as Morocco, there is a strong need for forecasting systems to reduce the impacts of floods. The development of such a system in the case of ungauged catchments is complicated but remote sensing products could overcome the lack of in-situ measurements. The soil moisture content can strongly modulate the magnitude of flood events and consequently is a crucial parameter to take into account for flood modeling. In this study, different soil moisture products (ESA-CCI, SMOS, SMOS-IC, ASCAT satellite products and ERA5 reanalysis) are compared to in-situ measurements and one continuous soil moisture accounting (SMA) model for basins located in the High-Atlas Mountains, upstream of the city of Marrakech. The results show that the SMOS-IC satellite product and the ERA5 reanalysis are best correlated with observed soil moisture and with the SMA model outputs. The different soil moisture datasets were also compared to estimate the initial soil moisture condition for an event-based hydrological model based on the Soil Conservation Service Curve Number (SCS-CN). The ASCAT, SMOS-IC and ERA5 products performed equally well in validation to simulate floods, outperforming daily in situ soil moisture measurements that may not be representative of the whole catchment soil moisture conditions. The results also indicated that the daily time step may not fully represent the saturation state before a flood event, due to the rapid decay of soil moisture after rainfall in these semi-arid environments. Indeed, at the hourly time step, ERA5 and in-situ measurements were found to better represent the initial soil moisture conditions of the SCS-CN model by comparison with the daily time step. The results of this work could be used to implement efficient flood modelling and forecasting systems in semi-arid regions where soil moisture measurements are lacking.



38
The Mediterranean region is characterized by intense rainfall events generating floods with a very 39 short response time (Gaume et al., 2004;Merheb et al., 2016;Tramblay et al., 2011). The socio-40 economic consequences of these floods are very important in terms of fatalities or damages to the 41 infrastructures in particular for Southern countries (Vinet et al., 2016). This highlights the need for 42 forecasting systems to reduce the impacts of floods. Unfortunately, the development of such systems is 43 very complicated in the case of ungauged catchments (Creutin and Borga, 2003) such as in North 44 Africa and requires remote sensing products to overcome the lack of in situ measurements. The Moroccan catchments are exposed to intense flash floods, such as the event of August 17, 1995 in 50 the Ourika river where the max discharge reached in 45 minutes a peak discharge of 1030 m3/s 51 causing extensive damages and more than 200 casualties (Saidi et al., 2003). Few studies have been 52 carried out in Morocco to minimize the impact of floods by improving the forecasting systems, either 53 by event-based modeling of floods (El Alaoui El Fels et al., 2017;Boumenni et al., 2017;El Khalki et 54 al., 2018) or by hydro-geomorphological approaches (Bennani et al., 2019) to identify the areas at risk 55 of flooding. The severity of floods in these semi-arid regions is controlled by several factors including 56 precipitation intensity, soil permeability, steep slopes and soil moisture content at the beginning of 57 event (El Khalki et al., 2018;Tramblay et al., 2012). In Mediterranean regions, the soil moisture 58 content varies between events and is known to strongly modulate the magnitude of floods (Brocca et 59 al., 2017;Tuttle and Salvucci, 2014) and particularly to be useful for flood modeling and forecasting 60 systems (Brocca et al., 2011;El Khalki et al., 2018;Koster et al., 2009;Marchandise and Viel, 2010;61 Tramblay et al., 2012). However, studies in North African basins are lacking to document the rainfall-62 runoff relationship with soil moisture during floods (Merheb et al., 2016).

64
In most Mediterranean regions and particularly in North Africa, only a few measurements of soil (Wagner et al., 2013). 2) The Soil Moisture and Ocean Salinity Mission (SMOS) product, which 74 begins in January 2010 with a spatial resolution of 50km (Kerr et al., 2012). The improvement of the 75 robustness of satellite soil moisture products can be achieved by merging passive and active 76 microwave sensors as initiated and distributed by ESA-CCI (European Space Agency Climate Change 77 Initiative) (Liu et al., 2011) providing data from 1978 to 2018. However, remote sensing products 78 might suffer from several problems in complex topography or very dense vegetation and snow cover 79 (Brocca et al., 2017). For this reason and before any use the data, it is necessary to validate them (Al-80 Yaari et al., 2014;Van doninck et al., 2012;Ochsner et al., 2013), either by in-situ measurements, if 81 they exist, or by using Soil Moisture Accounting models (Javelle et al., 2010;Tramblay et al., 2012) to 82 simulate soil moisture in the ungauged basins.

84
In this context, with an increasing number of satellite products becoming available to estimate soil 85 moisture, clear guidelines and recommendations about the most suitable products to estimate the initial 86 soil moisture content prior to floods are lacking for the semi-arid basins of North Africa. The purpose 87 of this study is to compare different satellite soil moisture products with in-situ soil moisture 88 measurements and the recently developed ERA5 reanalysis to estimate the initial soil moisture before 89 flood events. The goal is to identify the best products to be used for flood modelling that could 90 improve forecasting systems. This comparison is performed for two basins representative of medium-91 size catchments of North Africa that are the most sensitive to flash flood events. The validation of the 92 different soil moisture products is made with a Soil Moisture Accounting (SMA) model, to test the 93 capabilities of the different soil moisture products for the sake of estimating the initial conditions for 94 an event-based hydrological model for floods. The paper is organized as follow: In section 2, an 95 overview of the study area and all used data (hydro-meteorological and soil moisture products).

96
Section 3 explains the methods adopted in this paper. Section 4 presents the results. The conclusion 97 and perspectives are given in the last section. consists of clays and calcareous marl. The basin area includes agricultural activities that are irrigated 120 in the downstream part of the basin. The irrigation comes from seguias, earthen-made channels that 121 traditionally draw their water supply from the river itself, by building small diverting dams on the side 122 of the river (Pérennès, 1994). The seguias channels are usually filled up during floods, and water is 123 distributed to the neighboring agricultural parcels. The map on the seguias in the Issyl basin can be 124 seen in Figure 1, covering the northern part of the basin. The system is unmonitored and in a context 125 of high evaporation rates the portion of runoff diverted from the stream is not quantified. Due to the 126 temporary nature of seguias, they can be partially destroyed during large floods and consequently their 127 hydraulic properties and the amount of water collected can be modified over time.  (Jarlan et al., 2015;Khabba et al., 2013) (Chaponnière et al., 2008). The 137 hydrometric data was provided by radar installed in each basin's outlet. The data is covering only the for each selected events are ranging from 13.1 to 34.1% for Rheraya and from 1.2 to 7.2% for Issyl.

148
This indicates the important role of initial conditions in both basins, with a much higher infiltration 149 capacity in the Issyl basin in addition to potential water loss due to irrigation. We used 5 temperature Soil moisture measurements are available at one location with three Thetaprobes at two different 172 depths (5cm and 30cm). In this study we used Thetaprobes with 5cm depth, which is comparable with 173 the depths of satellite products (Massari et al., 2014). The site is located in Rheraya basin, with an Hargreaves-Samani equation (Hargreaves and Samani, 1982) are used as inputs to the SMA model.   4. The ESA-CCI soil moisture product (http://www.esa-soilmoisture-cci.org/) regroups 211 active and passive microwave sensors to measure soil moisture, giving three type of 212 products: Active, Passive and Combined (Active + Passive). In this paper, the ESA-CCI 213 V4.5 -Combined product is used (Dorigo et al., 2017;Gruber et al., 2017Gruber et al., , 2019. The 214 product has been validated to be useful by 600 ground-based measurement points around 215 the globe (Dorigo et al., 2015), as well as it was compared with ERA-Interim products 216 (Albergel et al., 2013). In the field of hydrological modeling, several global studies have 217 used the ESA-CCI product to initiate the hydrological model (Dorigo et al., 2012(Dorigo et al., , 2015 218 Massari et al., 2014) at the scale of Morocco (El Khalki et al., 2018). We extracted for 219 each basin the pixel that corresponds to it. In-situ data preparation consists of averaging the 5cm depth probes in order to get a single value to 239 work with and take into account the plot-scale variability of the measurements. This data is considered 240 as a reference for soil moisture data in the Rheraya basin, so that all the other soil moisture products 241 are compared to it. The different soil moisture products are compared to the observed soil moisture 242 over the entire period and also on a seasonal basis.

244
The SMA model is used to represent the soil moisture aggregated at the catchment scale. The rationale 245 behind the use of such model here is that continuous rainfall and temperature series are often available 246 in monitored catchments, unlike soil moisture, and a calibrated SMA model can sometimes palliate the 247 lack of soil moisture measurements (Tramblay et al., 2012). For the SMA model, the A parameter, 248 representing the soil water holding capacity, is calibrated to obtain the best correlation between 249 observed and simulated soil moisture (S/A). The calibration with observed data can only be performed  (Brocca et al., 2009b(Brocca et al., , 2010Loew and Mauser, 2008;Loew and 269 Schlenz, 2011;Martínez-Fernández and Ceballos, 2005;Miralles et al., 2010;Wagner et al., 2008).

270
According to (Massari et al., 2015), the coarse satellite observations can be beneficial for small basins, 271 in the case if the in-situ observation falls in the satellite product pixel. This means that the in-situ 272 measurements can represent a good benchmark (Liu et al., 2011). In this study we considered the in-273 situ measurement as a benchmark to validate different soil moisture products.

274
275 Where and are the limits of the confidence interval (the upper and the lower 95%) 286 where X is the soil moisture estimate, S is the true soil moisture, α and β are additive and 287 multiplicative biases, respectively. Eventually, ε is the zero-mean random error.

308
where σ_T^2 is the true soil moisture variance, σ_ε^2 is the variance of the random error, and σ_(ε_X 309 ε_Y ) is the error covariance between X and Y.

311
And the least squares solution for the parameters in S is given as: with i, j in [X, Y, Z] and i ≠ j.