SURFEX, the land-surface modeling platform originally developed and currently maintained by Météo-France (Masson et al., 2013; Le Moigne et al., 2020), has been chosen to perform the LSM simulation used in this study. Simulations for the IP and Balearic Islands developed within the HUMID project have 5×5 km spatial resolution and use the forcing provided by the Iberian application (Quintana-Seguí et al., 2008, 2016, 2017) of the SAFRAN meteorological analysis system (Durand et al., 1999). This is a modeling chain whose offline mode operates using the atmospheric forcing of SAFRAN to feed the LSM SURFEX-ISBA and simulate even the hydrology with Eaudyssée–RAPID (David et al., 2011).

ISBA (Noilhan and Mahouf, 1996) is the SURFEX module in charge of simulating natural surfaces. There are different versions of ISBA. In this study, we have used the diffusion version (ISBA-DIF; Boone, 1999; Decharme et al., 2011), which performs better in the study area than the simpler three-layer force restore version (Quintana-Seguí et al., 2020). In this version of ISBA, the leaf area index (LAI) has a prescribed annual cycle (constant every year), which may limit the ability of the model to reproduce the long-term effects of drought on vegetation. The model simulates the soil column, but it is unable to simulate groundwater, which despite its impact on soil moisture memory, is not very relevant in the Ebro basin. SURFEX-ISBA requires additional physiographic information that is incorporated from the ECOCLIMAP-II land cover database (Faroux et al., 2013), which includes topographic, soil and land cover information at high resolution.

The available SAFRAN forcing data allow us to simulate the period 1979–2017, but the period used for this study is restricted by the relatively short length of the RS SMOS data coverage (2010–present) compared to the model. To ensure the comparability of RS-based and LSM-based drought indices, the study period is 2010–2017, for which both RS and LSM data are available. To ensure that the RS soil moisture and the LSM soil moisture are comparable, we have averaged the first three soil layers of the model according to their discretization in the first 5 cm of the soil (1, 3 and 10 cm of depth, respectively). The simulation is performed using a regular 5×5 km resolution grid based on a custom Lambert conformal conic projection.

Drought indices allow quantifying several aspects of drought, like the magnitude and duration, and may also focus on particular variables depending on the scope of interest (i.e., precipitation, soil moisture, aridity, etc.). In view of the convenience to combine several single-variable indicators to describe most of the drought mechanisms and our focus on rainfed environments, we adopt the use of the standardized precipitation index (SPI) (McKee et al., 1993), the evapotranspiration deficit index (ETDI) and the soil moisture deficit index (SMDI) (Narasimhan and Srinivasan, 2005). Using these three indices, the study aims to investigate the interaction between the two main water fluxes (rainfall and evapotranspiration) and the main storage (soil moisture) involved in the water balance of the land–atmosphere system. The aggregation periods of the SPI inform us about the different responding times of rainfall, soil moisture, streamflow and groundwater anomalies. By evaluating the evolution of these indices along with their interactions, this study aims to characterize drought mechanisms in the Ebro basin. In this study, the indices have been computed using gridded datasets, thus generating a time series for each grid point.

The first two aims of the study are to evaluate the suitability of the SPI, ETDI and SMDI to characterize the main anomalies in water exchanges of the land–atmosphere system and to evaluate the suitability of adopting the weekly scale for the analysis of drought indices compared to the use of the monthly scale. Regarding the first, results shown in Fig. 2 indicate a general agreement of the SPI, ETDI and SMDI (computed at either a monthly or a weekly scale) on the major events of dry and wet anomalies of the period 2010–2017. The dry period of 2011–2012 and the wet period from the end of 2012 to 2015 were properly depicted. However, we identified differences, especially in the case of the SMDI. This index tends to show a generally lower variability than other indices when calculated with the LSM. RS results of the SMDI differ from the other indices during the start of 2010 due to the uncertainties during the test period of the SMOS mission. The left column of Fig. 2 shows the monthly SPIm-i and the monthly averaged ETDIm and SMDIm. Correlations between the SPI and the ETDI and SMDI are calculated on the monthly scale (Table 1, top row) and on the weekly scale (bottom row) for both RS and LSM data (where i= 4, 13, 26 and 52 weeks of aggregation, equivalent to 1, 3, 6 and 12 months). We test the significance of the correlations (p value <0.05) between indices for two subsets: the entire period (2010–2017) and a subset of dry periods (i.e., when SPI <0, ETDI <0, SMDI <0). Then, we compare the observed (RS) and simulated (LSM) estimates of the indices to explore differences between data sources. RS and LSM ETDIm and SMDIm indices are moderately correlated (barely over 0.5, significant). Table 1 reports a value of r=0.58 (significant) between ETDIm RS and SMDIm RS, which is significantly higher than the r=0.32 between the ETDIm LSM and SMDIm LSM. In general, despite the resemblance of RS and LSM series of the ETDI and SMDI shown in Fig. 2, these two series differ. There is a higher agreement between the RS and the LSM estimates of ETDIm (r=0.77, significant) than between the RS and LSM ones of SMDIm (r=0.27).

Table 1 also reveals differences in the moderate correlation of SPIm-i with the ETDI and the SMDI of different temporal aggregations (i= 1, 3, 6, 12 months of SPIm-i accumulation) as well as between their RS and LSM versions. Correlation between SPIm and ETDIm RS increases with the aggregation period of SPIm (r= 0.39, 0.62, 0.72 and 0.81 for SPIm-1, SPIm-3, SPIm-6 and SPIm-12, respectively) while correlations of SPIm with the ETDI LSM peak at SPIm-3 (r=0.8) and remain high for SPIm-6 (r=0.78) and SPIm-12 (r=0.71). The correlations between SPIm-i and SMDI RS show moderate correlation ranging from SPIm-1 (r=0.42) to the maximum value of SPIm-12 (r=0.63). The SMDIm LSM exhibits a decreasing correlation pattern with the increasing aggregation of SPIm from SPIm-3 to SPIm-12 (r= 0.45, 0.31, 0.3 and 0.24 for SPIm-1, SPIm-3, SPIm-6 and SPIm-12, respectively). Remarkably, correlations of the LSM SPIm–SMDIm are lower than their RS pairs, which suggests data uncertainties in RS and the LSM, as reported by Barella-Ortiz and Quintana-Seguí (2019) and by Quintana-Seguí et al. (2020).

The monthly correlation analysis is additionally conducted for the subsets of dry periods (those with negative signs of SPIm, ETDIm and SMDIm). Using the dry subset instead of the entire period primarily decreases all correlations. Compared to the correlations of the entire period, the RS ETDIm–SMDIm values decrease a little while the LSM ones increase a little. Also the SPIm-i–ETDIm and the SPIm-i–SMDIm correlations decline noticeably. Despite the loss of correlation and significance with the dry subset, the increase in the period of aggregation causes an increase in correlations, equal to the case of the entire period series.

Figure 2 provides an overview of the effect of adopting the weekly scale (right column) instead of the monthly scale (left column). The weekly scale substantially improves the temporal resolution of the plots. Subplots of SPI-i (a–d), the ETDI (e) and the SMDI (f) show how the weekly scale accurately reproduces the magnitude, tendency and duration of the monthly-scale anomalies, while the increase in temporal resolution additionally captures quick changes. Graphically, we noticed that the aggregation period applied to the SPI counteracts the gain in the resolution of the weekly scale, which often prevents users from adopting the weekly scale instead of the monthly scale. Conversely, Fig. 2e2 and f2 illustrate the strong increase in the resolution of the ETDI and SMDI resolution on the weekly scale.

The right columns of Table 1 indicate the correlations between indices on a weekly scale compared to those on a monthly scale (the “w” sub-index of ETDIw, SMDIw and SPIw denotes the weekly scale). There is an overall decrease in correlations at the weekly scale compared to the monthly scale, accentuated by the increasing period of aggregation (from SPIw-3 to SPIw-12). Correlations increase compared to the monthly scale at the lowest period of aggregation of SPIw-i, especially for SPIw-1–ETDIw (from r=0.39 to r=0.57) compared to SPIw-1–SMDIw (from r=0.51 to r=0.68). For both RS and LSM data, the weekly scale lowers the correlation values of SPIw-i with SMDIw more than those with ETDIw. Within the dry-period subset, all the correlations decrease too. The weekly scale lowers correlations due to the increase in variability in the weekly time series compared to the monthly ones but enables capturing the increasing complexity of interactions at shorter timescales. Therefore, since quick shifts (Fig. 2e1 and f1) are of great interest for drought analysis, we consider the weekly scale the most convenient for the lag analysis of the next section. This decision benefits comparing drought indices at their highest temporal definition, which for the ETDI and SMDI is natively the weekly scale and for the SPI can be easily adopted.

The analysis of correlations between pairs of temporally lagged time series of drought indices provides valuable insights into the interactions between indices (e.g., in terms of reciprocity) as indicators of synchronicity of one variable with respect to the other. The analysis addresses the characterization of the interactions between rainfall, evapotranspiration and soil moisture while also aiming to examine whether the land–atmosphere exchange of semi-arid areas under drought events is a quickly evolving or rather inertial system. Plots of the fraction of the area affected by each correlation level (from -1 to 1 in steps of r=0.2) for each lag period are shown in Figs. 3–7. It is anticipated that the three evaluated indices agree the most in a narrow range around lag =0 (no lag) and that their correlation fades away progressively for the increasing lag periods beyond the scale of the propagation of drought. Figures 3 to 6 in which SPIw-i interacts with the ETDI and SMDI illustrate the range of aggregation periods of SPIw (i.e., from 4 weeks of aggregation SPIw-4 to 52 weeks SPIw-52) along which the interpretation of the interactions between drought indices becomes most altered. The range of aggregation periods aims to open discussion about the optimal timescale to analyze the interacting drought processes. In general, the upper two subplots of Figs. 3 to 6 identify the admissible aggregation periods to interpret the clusters of interaction (SPIw-4 and SPIw-13, equivalent to SPI-1 and SPI-3 on a monthly scale), while the bottom two subplots (SPIw-26 and SPIw-52, equivalent to SPI-6 and SPI-12 at the monthly scale) depict the increasingly merged clusters of interaction due to excessively long aggregation periods of the SPI.

Analyzing the differences in correlations between indices at monthly and weekly scales (Fig. 2), we support the necessity of adopting the weekly scale to study lags. The monthly scale preferred for drought assessment from a hydrological perspective may overlook the quick response of the land–atmosphere interactions. The clusters of moderate to high correlation between indices mostly occur within the first month preceding or following an anomaly (Figs. 3–7), particularly in the short to very short term. High correlations tend to peak and plunge in an interval of a few weeks. This short-term response recommends the use of the weekly scale and aggregation periods below the seasonal scale, such as SPIw-13 (equivalent to SPI3), to evaluate the delay or precedence between indices.

Our results showing soil moisture response to rainfall (-5 to 5 weeks) and evapotranspiration (-10 to 5 weeks) anomalies in a matter of weeks are consistent with previous works showing soil moisture echoes rainfall anomalies in a range from days to weeks (Scaini et al., 2015; Martínez-Fernández et al., 2016) and also when driven by evapotranspiration (Otkin et al., 2013). Therefore, in a basin exposed to the high-energy characteristics of semi-arid climates, the weekly scale allows us to diagnose disregarded short-term interactions, such as the ones of the ETDI and SMDI on the SPI, without losing resolution in the identification of mid-term to long-term interactions. The need to apply a weekly scale to the SPI, which is barely used below the monthly scale, is recommended not only for the analysis of rainfall–evapotranspiration interactions but also for soil moisture ones, which are often assessed over the monthly scale. Using the weekly scale for drought assessment demands an increase in spatial resolution. This is the reason why LSMs using coarse inputs, such as the semi-aggregated physical characteristics of the basin, may generate results with scale effects. In fact, scale effects due to the coarse resolution of the spatial characteristics can increase the temporal scale at which the processes are effectively correlated (Rodriguez-Iturbe et al., 2001). Therefore, lag analysis needs to compile datasets of appropriate spatial and temporal scales for their aims, like the high-resolution RS datasets evaluated in this study or LSM ones based on completely distributed data.

Adopting specific drought indices for rainfall, evapotranspiration and soil moisture allows exploring the interactions between variables of different levels of the land–atmosphere system. The pertinence of using the SPI, ETDI and SMDI to evaluate the interactions between the variables' anomalies can be the subject of discussion depending on the specific advantages, drawbacks and applicability of each index. There can be alternative drought indices even for soil moisture and evapotranspiration of better stability and statistical characteristics worth exploring. Furthermore, assessing the interactions of anomalies may be possible without using drought indices. However, since the commonplace in the analysis of the anomalies behind drought has been widely based on drought indices for comparable interpretation, we consider the SPI, ETDI and SMDI to represent a set of comprehensive indices to flexibly evaluate interactions at different timescales.

Figure 8a and graphically Fig. 8b summarize the annual mode of interactions between the SPI, ETDI and SMDI. In the short to middle term, both the ETDI and the SMDI interactions with the SPI concur on having moderate significance, with only a few negative and positive low interactions in the middle to long term. Positive correlations around lag 0 of the ETDI and SMDI with the SPI indicate direct precedent dependence of the indices, which means changes on the ETDI and SMDI correlate positively (negatively) to positive (negative) changes on the SPI. The short-term correlations after rainfall for both the ETDI and the SMDI (lagged response of these indices to the SPI) are straightforward and have been reported before in similar Iberian regions (Martínez-Fernandez et al., 2016). Sustained dry or wet anomalies in both variables favored by a positive correlation between indices are primarily restricted to a length of three seasons. Correlations beyond the year may represent the multi-annual persistence of anomalies common in Mediterranean climates.

The strong aggregation impact occurring when adopting SPIw-26 and SPIw-52 indicates the analysis of interactions may be uncertain when indices are aggregated beyond the seasonal scale. It is evaluated if the magnifying effect of clusters with long aggregation periods (SPIw-26 and SPIw-52) is the autocorrelation of the indices. Significant autocorrelated values always extend for less than the period of aggregation of SPIw-i (2 weeks on SPIw-1, 3 weeks on SPIw-4, 10 weeks on SPIw-13, 18 weeks on SPIw-26, 35 weeks on SPIw-52, 40 weeks on SPI-78, 45 weeks on SPIw-104). Hence, we presume autocorrelation values of the SPI series are only partly caused by the period of aggregation. The partial autocorrelation of both the ETDI and the SMDI shows mostly two significant lags, which indicate the existence of an autoregressive model of type AR(2) (Fig. S2). An AR(2) model may articulate the combination of growing and decaying factors behind the shifting balance of positive and negative correlations between indices. In this way, the clusters of interactions identified in Figs. 3 to 7 can be attributed to interactions between rainfall, evapotranspiration and soil moisture anomalies. Conversely, clusters identified in series with a period of aggregation over the seasonal scale may be misleading for the analysis of drought interactions in Mediterranean environments. Thus, drought evolution in semi-arid environments can be considered an expression of quick exchanges between land–atmosphere variables which develop within the seasonal scale. This implies the temporal resolution of drought indices must adopt short timescales (i.e., weekly scale) to capture the onset of drought.

The existence of the precedent influence of the ETDI and SMDI on the SPI (clusters of the negative range of lags) implies some unequal reciprocity (feedback) between evapotranspiration and soil moisture with rainfall. This precedent influence is weaker than the influence of the SPI on subsequent ETDI and SMDI anomalies (lagged response) but still remarkable. It is reasonable that the lagged response of evapotranspiration and soil moisture to rainfall is stronger and longer-lasting than the precedent influence (Fig. 8a). Furthermore, the precedent influence period between the ETDI and SPI is stronger and of longer duration than the one of the SMDI on the SPI. This asymmetry suggests that the ETDI, more than the SMDI, has a weekly to seasonal precedent influence on rainfall (Fig. 8b). We expected a longer period of positive correlations of the SMDI influencing rainfall, given the multiple reports of soil moisture inducing memory at the near-surface atmosphere (Manning et al., 2018).

One reason why the ETDI shows a longer influence on the SPI than the SMDI may be that the ETDI from MOD16A2 is fed by the whole depth of soil moisture, while the SMDI based on SMOS1km is limited to the top 5 cm of soil moisture, a very exposed soil level in semi-arid climates. The complexity of soil moisture dynamics, which barely follow a cyclic interaction (Rodriguez-Iturbe et al., 1991), can also explain a weaker relationship between the SMDI and SPI compared to the ETDI. Other reasons for this may lie in the prevalence of maritime advection as the main contributor to evapotranspiration in the IP (Gimeno et al., 2010) compared to the prevalence of local soil moisture recycling common in more continental areas of Europe (Bisselink and Dolman, 2008). The advective explanation is supported by the contrast between the few weeks of precedent influence of soil moisture on rainfall we observe in the Ebro basin and the up to 250 d of precedent influence of continental areas prone to soil moisture recycling (Rowntree and Bolton, 1983; Bisselink and Dolman, 2008). Some studies focused on continental climates of relevant summer rainfall have described the implications of the alteration of the recycling due to soil moisture depletion during heat waves and drought which can eventually alter the atmosphere (Rasmijn et al., 2018; Miralles et al., 2019). In the Mediterranean climate of the Iberian Peninsula characterized by a lack of summer rainfall, soil moisture annually reaches such low levels that we can expect annual summer alterations in the near atmosphere. Differences between areas where soil moisture plays a role, like central Europe, and areas where soil moisture is unable to control the evolution of the system under high-energy conditions, like the Iberian Peninsula, have been reported before in Mediterranean-like Western Australia (Herold et al., 2016).

In consequence, our results at the Ebro basin seem compatible with the frequent activation of a reinforcing or self-intensification loop (Brubaker and Entekhabi, 1996), by which the precedent influence of negative (eventually positive) anomalies of evapotranspiration reducing (increasing) rainfall cascades into a depletion (rise) in soil moisture that further limits (enhances) the response of evapotranspiration and restarts the cycle (Fig. 8c, right column). The weak precedence of soil moisture on rainfall compared to that of evapotranspiration expresses the limited duration of the control capacity of the soil moisture over evapotranspiration in semi-arid climates of the Mediterranean type (left column of Fig. 8c). Negative correlations when indices differ in sign (r<0 conditions, Fig. 8a) can be indicative of transitional periods of mid-seasons. The sharp shift from the cluster of short-term positive correlations to the cluster of mid-term to long-term negative correlations suggests a limit in the persistence of the self-intensification mechanism. A physical interpretation of the shift may be related to the change in the dominance of the sequence from the one under high-energy conditions (Fig. 8c, right column) to the one under low-energy conditions (Fig. 8c, left column). A low-energy inhibiting mechanism of negative correlations between soil moisture and surface temperature under low temperatures has already been described by Brubaker and Entekhabi (1996). Given the direct link between evapotranspiration and temperature, the shift from positive to negative interactions happening at timescales over the semester but below the yearly scale suggests that the arrival of winter low-energy conditions terminates the dominance of the self-intensifying loop of evapotranspiration.

In this way, the annual cycle can be modeled as the seasonal succession of two sequences: one under the low-energy conditions of winter when evapotranspiration no longer outweighs the inhibiting of soil moisture due to rainfall (left column of Fig. 8c) and the other under high-energy conditions driven by evapotranspiration (right column of Fig. 8c). The shift between the long period of interactions dominated by evapotranspiration (right column of Fig. 8c) and a short period of interactions controlled by rainfall and soil moisture (lower sequence of Fig. 8c) is generally driven by an energy threshold. However, certain levels of rainfall and soil moisture anomalies may temporally advance or delay the shift. This reason explains why under high-energy conditions drought may terminate due to heavy rainfall, while soil moisture deficits under low-energy conditions may cause an anticipated onset of the self-intensification loop of evapotranspiration.

The conceptualization of the interactions illustrated in Fig. 8 aims to raise awareness about the power of evapotranspiration anomalies to alter the land–atmosphere system, year-round, beyond hydrometeorological extremes (Seneviratne et al., 2006; Otkin et al., 2013; Teuling, 2018; Miralles et al., 2019). Another reason for supporting the year-round implications of the dominance of evapotranspiration over soil moisture is that rainfall mostly transfers to evapotranspiration in semi-arid climates (Rodriguez-Iturbe et al., 2001), where the often underestimated interception (Savenije et al., 2004) further increases evaporation at the expense of soil moisture. Additionally, as the semi-arid Mediterranean climate likely presents thresholds of rainfall, evapotranspiration and soil moisture anomalies different from those triggering hydrometeorological extremes in other areas (Tramblay et al., 2021), the evapotranspiration-dominated sequence may initiate not only more often but also more abruptly than in regions of lower-energy inputs. All these aspects, together with the increasing chance of extremes in the Mediterranean area due to climate change (Samaniego et al., 2018), recommend assessing changes in the balance of land–atmosphere interactions from the basis of this study.

However, we bear in mind that our results may oversimplify the causality, since processes not analyzed in this study may also play a role. The multiple periods showing neither prevalently positive nor prevalently negative correlations between indices indicate a loss of linear interaction. A source of non-linearity is vegetation due to its mediating role in water exchanges of the land–atmosphere system. Plants can control evapotranspiration and soil moisture in adaptation to water stress in non-linear manners that depend more on the type of vegetation (Katul et al., 2012), particularly within the Mediterranean floras (Boulet et al., 2020), than on the evapotranspiration or soil moisture status. Vegetation can also modulate the partitioning of energy governing evapotranspiration (Lansu et al., 2020) but similar processes are reported with soil moisture (Barbeta et al., 2015). In consequence, the quick response to drought of rainfed crops and sclerophyllous vegetation (Vicente-Serrano et al., 2019) may obscure the interpretation of the links between rainfall, evapotranspiration and soil moisture. This is of concern regarding the ETDI and SMDI results because interactions of vegetation integrate the status of the atmospheric and the land-surface variables (Peters et al., 1991). Nonetheless, additional factors of uncertainty may arise from teleconnections such as the well-known North Atlantic Oscillation (NAO) or Western Mediterranean Oscillation (WeMO; Barnston and Livezey, 1987; Conte et al., 1989) or the oceanic ones like the Atlantic Multidecadal Oscillation (AMO; Kerr, 2000) altering the land–atmosphere system at large scales.

The RS and LSM results of the lag analysis of the ETDI–SPI interactions show consistently comparable results in contrast to the remarkable disagreement between RS and LSM results for the SMDI–SPI interaction. Results of the SMDI obtained with the LSM show substantially lower correlations than the ones of RS while also differing in the timing of the clusters of correlation. We expected the opposite, i.e., that the LSM, being simpler than reality, has stronger SPI–ETDI–SMDI correlations than the RS dataset. We assume the implicit accumulation of uncertainties in modeling (Rodriguez-Iturbe et al., 1991), partly inherited from inputs but also from the LSM structure, is the cause of the decrease in correlations. This is particularly true for soil moisture, a variable integrating exchanges between climate, soil and vegetation (Rodriguez-Iturbe et al., 2001). Secondly, this is an offline simulation, where the atmosphere (SAFRAN) is forcing the land surface (SURFEX-ISBA) without explicit feedback of SURFEX-ISBA influencing SAFRAN back. SAFRAN estimates real conditions by ingesting observations, so the feedback is implicit in results, which may be insufficient to represent reality. Thirdly, the model itself does not consider important processes like the interactive response of vegetation. ISBA has an interactive vegetation module (ISBA-A-gs), but Mediterranean vegetation can be particularly challenging for it. We expect to test the capabilities of interactive modeling vegetation in a follow-up study. Uncertainties in ISBA with vegetation also have roots in the use of the ECOCLIMAP-II database, which shows inaccuracies in cover type and the LAI. ECOCLIMAP assumes the maximum/minimum LAI occurs in June/February in contrast with the early spring and autumn LAI maximums characteristic of the Mediterranean environment (Queguiner et al., 2011). All in all, the differences between the LSM and RS datasets are already an important result to improve the LSM and comprise a useful insight into the use of offline LSM drought simulations.

Our results positively verify that RS represents an effective tool to overcome the problem of sparsely observed soil moisture or evapotranspiration, whose crucial role in drought evolution requires high-resolution data similarly to precipitation (AghaKouchak and Nakhjiri, 2012). Including high-resolution evapotranspiration products from MODIS (MOD16A2) and soil moisture from SMOS missions (SMOS1km) together with the distributed rainfall reanalysis data allows dedicated interpretation of the interactions between these two drought-relevant variables and rainfall and their role in the water balance of the land–atmosphere interface (Dai, 2011). Especially for evapotranspiration, the maps and series of the LSM SURFEX-ISBA are comparable to those of RS, which supports the reliability of LSMs despite their limited capability in arid regions (De Kauwe et al., 2015).

The temporal and spatial patterns of the anomalies are overly identified by both RS and the LSM. The RS data seem able to capture a more complex scheme of interactions than the LSM, despite the intrinsic data issues of the RS sensors and the performance of the algorithms used to generate the products. Conversely, the LSM seems sensitive to uncertainties from input data, especially surface properties, and the offline forcing. The parametrization of the model assumes a semi-distributed approach by sub-basins of the catchment on which each sub-basin is defined based on average values of land cover and soil characteristics of the ECOCLIMAP database, which may induce some patchiness of LSM results compared to the RS results. The offline run means that the meteorological data force the LSM, but the feedbacks are lost beyond the meteorological observations included as observation in the model. Additional aspects can be, for instance, the impact of groundwater redistributing soil moisture depending on topography, which is underrepresented in the LSM. Limitations of LSMs have been reported in multiple works before and are the subject of improvement efforts (Teuling et al., 2006; Samaniego et al., 2018). Either way, uncertainties are causes of major concern in the lag analysis where they can alter the correlations between indices and obscure the interpretation of the interactions.

These aspects together can be critical to improving evapotranspiration and soil moisture estimates (Ukkola et al., 2016). Solving these inaccuracies would increase the value of RS and LSM estimates. This study exemplifies the potential of high-resolution RS and LSM products for a wide range of applications, such as drought analysis.