Increased floodplain inundation in the Amazon since 1980

Extensive floodplains throughout the Amazon basin support important ecosystem services and influence global water and carbon cycles. A recent change in the hydroclimatic regime of the region, with increased rainfall in the northern portions of the basin, has produced record-breaking high water levels on the Amazon River mainstem. Yet, the implications for the magnitude and duration of floodplain inundation across the basin remain unknown. Here we leverage state-of-the-art hydrological models, supported by in-situ and remote sensing observations, to show that the maximum annual inundation extent along the central Amazon increased by 26% since 1980. We further reveal increased flood duration and greater connectivity among open water areas in multiple Amazon floodplain regions. These changes in the hydrological regime of the world’s largest river system have major implications for ecology and biogeochemistry, and require rapid adaptation by vulnerable populations living along Amazonian rivers.


Introduction
The seasonal flood pulse of the Amazon River, the Earth's largest river system, dictates the ecological structure and function of its floodplains and the ecosystem services they support (Junk et al 1989, Melack andCoe 2021). The composition, productivity, and life cycles of river and floodplain biota are adapted to seasonal flood regimes (Junk et al 1989), including fisheries underpinning human food and income security. The fertile floodplains provide food and fiber to support human communities across the basin (Sherman et al 2016, Langill and Abizaid 2020, Fleischmann 2021. These vast floodplains influence climate (Paiva et al 2011, Santos et al 2019 and carbon cycling within and beyond the Amazon basin, and are hotspots for greenhouse gas emissions (Richey et al 2002, Melack et al 2004, Abril et al 2014, Pangala et al 2017. Changes to the Amazon basin's flooding regime can therefore produce unprecedented impacts from local to global scales. The Amazon River system is increasingly affected by multiple stressors (Castello et al 2013). Changes in regional climate have been superimposed on human disturbances such as construction of dams (Chaudhari and Pokhrel 2022) and deforestation in both uplands (Costa et al 2003) and floodplains (Renó et al 2011). Based on river gage measurements, recent studies have shown higher maximum water levels in the Amazon since the late 1990s (Gloor et al 2013, Barichivich et al 2018), linked to a ∼17% increase in wet-season rainfall over the northern part of the basin (north of 5 • S) from 1981 to 2017 (Espinoza et al 2019a, Haghtalab et al 2020, Funatsu et al 2021. This change has been hypothesized to be driven by changes in sea surface temperature in both the Pacific and Atlantic oceans (Marengo and Espinoza 2016, Friedman et al 2021 and the consequent strengthening of the Walker and Hadley circulation (Barichivich et al 2018, Espinoza et al 2019a, Friedman et al 2021. Increased rainfall in the northern Amazon basin has resulted in higher river water levels and discharges along the mainstem Amazon River (figure 1(a)) (Heerspink et al 2020). Seven of the ten highest maximum annual water levels recorded in the last 119 years at Manaus (Negro-Amazon confluence) have occurred since 2009 (figure 1), including the highest-ever recorded water level in June 2021, which affected more than 500 000 people in the Amazonas State of Brazil (Espinoza et al 2022). This contrasts with a prevailing perception by the public and many scientists that the Amazon is drying out, influenced by the well-publicized decrease in rainfall and streamflow in southern Amazon sub-basins, a trend that has also been linked to changes in ocean-atmosphere interactions (Espinoza et al 2019a).
While trends in water levels are readily apparent from gages along major rivers, understanding their impacts on ecological and biogeochemical processes and human communities requires quantifying how higher water levels translate into commensurate increases in flooding extent and duration, as well as hydrological connectivity of seasonally inundated floodplain areas, which have not yet been analyzed. Here, we use hydrological-hydrodynamic models (MGB (Siqueira et al 2018) and CaMa-Flood (Yamazaki et al 2011)) and long-term satellite-derived datasets (see section 2), supported by in-situ observations, to analyze for the first time a 40 year record of floodplain inundation patterns in relation to basinwide rainfall changes.

Dynamic inundation from hydrological models
Monthly inundation estimates were obtained from two state-of-the-art hydrological-hydrodynamic models-MGB (Siqueira et al 2018) and CaMa-Flood (Yamazaki et al 2011). MGB has been used to investigate Amazon hydrological processes (Paiva et al 2013, Sorribas et al 2016, Fleischmann et al 2020. It is a rainfall-runoff model coupled to a physically-based hydrodynamic routine developed to represent the river-floodplain interactions in large basins (Pontes et al 2017). The version used here is the same one developed for the entire South American continent (Siqueira et al 2018), and was validated against extensive in-situ and satellite data. Within the model, the river drainage network is divided into 15 km river reaches, for which unit-catchments are defined. While the input rainfall data has a daily time step, the model runs with an adaptive time step to maintain model stability (Neal et al 2012). The model was manually calibrated aiming at river discharge simulation, defining parameters for hydrologically homogeneous areas based on lithology/geology information and the boundaries of large South American basins. Multiple hydrological components (water levels, total water storage and evapotranspiration) were validated based on in-situ and satellite data, showing a satisfactory model performance across the Amazon region (Siqueira et al 2018). In turn, CaMa-Flood is a global hydrodynamic model (Yamazaki et al 2011) that uses equations similar to MGB to simulate river-floodplain processes (Bates et al 2010, Yamazaki et al 2013. Here we use the same version as available in the eartH2Observe platform (www.earth2observe. eu) . The model is forced with HTESSEL model runoff fields, with a 0.25 • spatial resolution and 1 h time step. In both MGB and CaMa-Flood, the river channels are considered as rectangular cross sections, and the floodplains are assumed as storage units, so that no flow occurs along floodplains. Both MGB and CaMa-Flood models use daily MSWEP v 1.1 rainfall (Beck et al 2017) and provide daily inundation extent at 500 m resolution for the period 1980-2014. The adopted model versions simulate the Amazon basin in its natural scenario, i.e. without anthropogenic alterations such as dams. The models' ability to simulate inundation extent along Amazon floodplains was recently assessed by Fleischmann et al (2022). In addition to inundation extent, we also assess the flood storage estimates with both models, given their satisfactory capability to estimate both inundation and water depth across the basin (Yamazaki et al 2011, Paiva et al 2013, Siqueira et al 2018. The long-term maximum inundation area (including open water of river channels) over the Amazon mainstem floodplain between the cities of Iquitos and Gurupá is estimated to be 118 500 km 2 and 115 000 km 2 for CaMa-Flood and MGB, respectively, which are close to the estimate by Hess et al (2015) of 115 800 km 2 , which in turn is widely considered as the benchmark for Amazon wetland mapping (Fleischmann et al 2022). The Amazon mainstem floodplain area was delineated considering the flood mask by Hess et al (2015), and includes the lower reaches of major tributaries (see area in figure 1).

Open water remotely-sensed data
Long-term remote sensing observations of nonvegetated, open water extent are scarce. Here we use the high-resolution (30 m) Global Surface Water Occurrence (GSWO) product (available at <https://global-surface-water.appspot.com/ download>) (Pekel et al 2016), which is based on classification of the entire archive of the Landsat 5 Thematic Mapper, the Landsat 7 Enhanced Thematic Mapper-plus and the Landsat 8 Operational Land Imager orthorectified images, covering the period 1985-2020. Given the inability of optical imagery to sense inundation obscured by dense vegetation (Aires et al 2018, Zhou et al 2021, it is only capable of monitoring open water areas or sparsely vegetated inundated areas. Its applicability for the Amazon floodplain region is restricted to areas with extensive lakes. Thus, the lower river reaches, mainly downstream of the Negro-Amazon river confluence, are suitable for the use of GSWO. In upper reaches, inundated forests are more common, although floodplain lakes bordered with non-forest vegetation are present and are used to assess long-term changes. Besides being in agreement with the other inundation datasets and the increasing trend for water levels across the Amazon River (figure 3), the ability of GSWO to represent open water inundation changes was considered satisfactory given its correlation with the in-situ water levels in the Amazon River atÓbidos gage (figure S5(a)), which shows the high agreement between annual maxima of in-situ water level measurements and annual maxima of inundation area (R = 0.81, P < 0.001). The GSWO version used here is provided at a monthly basis and converted to annual maximum and minimum maps to reduce the uncertainties related to lack of images due to cloud-cover, especially during wet season. , based on the JPL RL06M Mascon solution, were used to investigate the impact of extremely wet years on seasonal water storage in the central Amazon. Given its short-term data availability (since 2003) and coarse resolution (∼300 km), no long-term trend analysis was performed, but the GRACE data were used to infer large-scale water storage patterns. In-situ observations of river water levels were obtained from the Brazil's National Water Agency for all gages in the Brazilian Amazon basin, in addition to in-situ data from the Peru's SENAMHI for the Tamshiyacu gage on the Amazon River close to the city of Iquitos. Because of delayed data availability, at the time of this study some gages did not yet have data available for the 2021 extreme flood event that occurred in central Amazon (Espinoza et al 2022); the gages which data covered the 2021 flood are highlighted in figure 1(a).

Ancillary data
The linear model developed by Hamilton et al (2002) was used as an additional independent estimate of inundation extent. It is based on passive microwave emission data from the Scanning Multichannel Microwave Radiometer (spatial resolution of 0.25 • ) and in-situ data from the Manaus river gage. Here, we used the aggregated time series of inundation extent for the entire mainstem Amazon floodplain from Hamilton et al (2002) (see their figure 1). It is important to notice that the linear model was developed for a water level range slightly smaller than the one assessed here (the maximum water level used was 28.9 m, while the 2021 flood reached 30 m).

Trend analysis
Maximum and minimum annual inundation trends obtained from MGB (1980MGB ( -2014, CaMa-Flood (1980-2014 and GSWO datasets (1984GSWO datasets ( -2020 were analyzed individually without any type of merging, and spatially mapped with the rank-based nonparametric Mann Kendall trend test (Kendall and Gibbons 1975) at the product resolution (500 m for MGB and CaMa-Flood and 30 m for GSWO), i.e. for each pixel the trend was assessed considering the entire inundation time series. The long-term changes in annual maximum and minimum flooded areas, as well as annual amplitudes (i.e. annual maxima minus minima), for the Amazon mainstem floodplain were obtained by applying a linear fit to the annual series, and then computing the change between the inundation in the first and last years of the adjusted line.
Step changes in the slopes of the time series were evaluated using the non-parametric approach to the change-point problem (Pettitt 1979), which yields the year in which the largest step change occurs in the whole series.
The annual maximum inundation extent was averaged for the decades 1985-1996 and 2009-2020 for the entire Amazon River floodplain between the cities of Iquitos and Gurupá. These two decades were chosen as representative of 12 years of recent (2009-2020) and earlier (1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996) inundation regimes, given the first year of GSWO availability (1985). The year 1996 has been identified as the beginning of a change in the rainfall regime, as recently reported (Espinoza et al 2019a). The year 2021 was not considered since it was not available in GSWO or with the hydrological models. In the case of the models, data were only available until 2014 because the model forcings (i.e. rainfall and runoff) did not extend to more recent years. Although updated versions of the models are available until present time with different model forcings than those used in the consolidated versions, the new runs do not satisfactorily represent the water level trends across central Amazon due to unsatisfactory bias correction, and thus are unsuitable to assess inundation trends (Yamazaki, pers. comm.; Brêda, pers. comm.). Given the unsuitability of GSWO for forested floodplains and issues with cloud cover, only MGB and CaMa-Flood were used to estimate the long-term changes in inundation extent for the entire Amazon mainstem floodplain. GSWO was only used to assess flood duration and open water lateral connectivity along floodplain open water areas in the lower Amazon downstream of Manaus. Finally, besides inundation, we assess flood storage trends (including all surface water) with both the MGB and CaMa-Flood models, following the same methodology adopted for the assessment of inundation extent.

Open water lateral connectivity
Open water lateral connectivity and flood duration changes were computed over the 1985-1996 and 2009-2020 decades with the GSWO dataset for the lower Amazon floodplain between Manaus and Gurupá, which is characterized by an abundance of large lakes (Sippel et al 1992). To estimate connectivity changes, we used the methodology presented by Trigg et al (2013). This analysis estimates, for a given floodplain area, the degree of connectivity across a given direction, measured as the number of pixel pairs connected across a given distance. The direction adopted for each analyzed area refers to the longest dimension of the floodplain unit, e.g. west-east for the Curuai floodplain ( figure 3(b)). For the flood duration changes, long-term maps (i.e. estimation of how many days per year a given pixel was, on average, inundated) were computed for the decades 1985-1996 and 2009-2020, and subtracted from each other. The use of a full decade diminishes the uncertainties related to data unavailability in specific months due to cloud cover in GSWO.

Recent increases in rainfall, river water levels and total water storage
Long-term records reveal that annual rainfall has increased in certain regions of the Amazon basin, while in others it has decreased (figure 1(a)) (Barichivich et al 2018, Espinoza et al 2019a. At the scale of the entire basin, the influence of areas with increased rainfall outweighs the counteracting influence of areas with decreased rainfall. Trends in annual maximum river levels show increases in several major tributaries and along the mainstem Amazon River ( figure 1(b)). Positive, statistically significant trends for both annual maximum water levels and water level amplitudes (i.e. annual maxima minus minima) are observed for the mainstem Amazon ( figure 1(c)), Negro and Purus rivers, and in lower reaches of other major tributaries subject to backwater effects from the mainstem Amazon (figures 1(a) and S2). Along the lower mainstem region (at theÓbidos gage station), where the largest increase is observed, the average maximum water level has risen by 87 cm (13% of the mean annual amplitude of ∼6.90 m, computed for the period 2009-2021) between 1985-1996 and 2009-2021. These periods refer to the earliest and latest multi-year periods available from the GSWO dataset. In contrast to maximum water levels, annual minima along the central Amazon have remained unchanged over the past 40 years (figure S2). The decrease in rainfall in parts of the southern basin coincides with a negative trend of annual minimum discharge documented in the upper Madeira River (at the Porto Velho gage station), which includes the extensive Llanos de Moxos wetlands along the Mamoré River (Espinoza et al 2019b).
Additional evidence for temporal changes in floodplain water storage, although over a shorter period, is provided by the total water storage variation along the central Amazon floodplains estimated from gravity field variations by the GRACE and GRACE-FO satellite missions since 2002 ( figure 1(d)). An increase in total water storage for individual flood events has been previously shown for the Amazon (Chen et al 2010), but our analysis for the entire 2003-2021 period supports the role of floodplain inundation dynamics in the seasonal and interannual changes of water storage at the regional scale.

Patterns of increased inundation along the Amazon mainstem
Our analysis shows that the change in rainfall has not only increased water levels, as stressed by previous studies, but also the annual maximum inundation extent in the floodplains along the Amazon River mainstem ( figure 2(a)). In the Amazon mainstem floodplains between the cities of Iquitos (Peru) and Gurupá (Brazil) (figure 1(a)), hydrological models suggest that the annual maximum inundation extent increased by 26% (93 000-117 000 km 2 between 1980 and 2014, with an annual increase of approximately 700 km 2 yr −1 , P-value < 0.001). In contrast, annual minimum inundation extent has not changed over the same period (figure S3). As a result, the inundation amplitude (annual maximum minus minimum inundated area) has increased (trends of 600-700 km 2 yr −1 ; P-value < 0.001; figure S3).
The modeled annual maximum inundation area in Amazon mainstem floodplains has the same step change observed for rainfall in the northern Amazon (Espinoza et al 2019a) in 1998 (P-value = 0.02 for MGB and P-value = 0.001 for CaMa-Flood models), increasing by 12% from a mean of 98 000 km 2 for the period 1980-1998 to a mean of 110 000 km 2 for 1999-2014. Furthermore, the modeled increase in maximum inundation extent and water levels translates into increased surface water storage (figure S4), consistent with trends in total water storage derived from GRACE satellite observations ( figure 1(d)), further supporting the hydrological models.
Increased inundation as revealed by hydrological models is corroborated by independent satellite observations, both by passive microwave-based measurements for the central Amazon ( figure S5(b); Hamilton et al (2002)), and by the Landsat-based GSWO. However, the increased inundation observed by GSWO over 1985-2020 is restricted to floodplain areas dominated by open water, which exist mainly downstream of Manaus (figures 2(c) and S6), because Landsat cannot detect inundation beneath dense canopies of flooded forest or floating herbaceous vegetation, in contrast to the two hydrological models. Additionally, a few areas with decreasing inundation trends are associated with natural sediment accumulations along newly formed islands and crevasse splay formations within floodplains (see detail in figure 3(a)).
In addition to increased inundated area, GSWO data indicate that flood duration (i.e. the average number of days that a pixel is subject to flooding in a given year) increased in 65% between the cities of Manaus and Monte Alegre (13 300 out of 20 400 km 2 ; excluding lake surfaces permanently flooded) between the decades of 1986-1995 and 2009-2020 (figures 3(a) and S7(a)). Some of the increases in duration were large, with 23% of the area subject to inundation changes being flooded for an additional 50 or more days per year across the two decades ( figure S7(a)). The total open water area subject to inundation for more than 180 days per year has increased by 14% for the whole floodplain between Manaus and Monte Alegre (from 16 300 to 18 600 km 2 ; figure S7(b)).
Because of more extensive and longer flooding, connectivity among floodplain waterbodies and with the main channels has increased in many places along the Amazon floodplain ( figure 3(b)). This is revealed by GSWO for open water areas downstream of Manaus. The largest impact on connectivity is observed in the Maracu and Camapu areas, with connected area within a 10 km distance increasing fivefold ( figure 3(b)). A smaller yet notable increase (20%-40%) is observed in Curuai Lake, associated with its shoreline expansion, because most of the lake was already subject to flooding by 1986-1995. This increased connectivity can impact exchanges of water, sediments, nutrients, pollutants and organisms between the river and its floodplain (Junk et al 1989, Park andLatrubesse 2017). It is important to note that our analysis does not measure connections between areas with flooded vegetation due to GSWO limitations, and thus our result should be regarded as a lower bound for the increase in surface water connectivity.

Multiple stressors affecting inundation regimes in a changing Amazon
While evidence of a shift to a novel hydroclimatic regime in the Amazon basin is mounting, climate projections remain uncertain about the net effects of climate change on river discharge, water levels, and floodplain inundation. Projections for the middle and end of this century suggest increased rainfall in the Amazon watersheds upstream from the Amazon-Purus confluence (Sorribas et al 2016, Zulkafli et al 2016, Brêda et al 2020. However, the propagation of these effects downstream to the mainstem Amazon and its floodplain is poorly understood. Changes in rainfall regimes are driven by complex oceanatmosphere-land teleconnections, and it is uncertain the extent to which increased rainfall may be offset by decreases of rainfall and river discharges as projected for southern Amazon tributaries (Boisier et al 2015, Brêda et al 2020. Notably, our estimated 26% increase in the maximum inundation extent over 1980-2020 is considerably larger than late-century projections for the Peruvian Amazon (+18%), and much larger than projections for the central Amazon (+4%) (Sorribas et al 2016). Recent studies have shown that the drying of southern tributaries is leading to less area of open water in that region (Souza et al 2019). Along the Amazon mainstem floodplain, however, there is no evidence of a clear trend of increasing droughts (i.e. no negative trends either for minimum water levels, figure S2(b), or for annual minimum inundation extent, figure S3), even though parts of the Amazon have faced several extreme droughts in recent years (e.g. 2005 and 2010), particularly in the central-southern portion of the Amazon Basin. In some cases, droughts have occurred in the same year of an extreme flood along the mainstem river. This calls for more studies to understand spatial variation in the hydrological processes related to simultaneous occurrences of flood and drought events across this vast drainage basin (Ward et al 2020).
If sustained in the future, the increase in maximum inundation extent observed across the central Amazon floodplains, driven by a change in rainfall regime, is unlikely to be the only change affecting floodplain inundation regimes. Human-driven hydrological changes in the basin that potentially affect river discharge and floodplain inundation include construction of hydropower dams (Chaudhari and Pokhrel 2022), the development of industrial waterways in the western Amazon tributaries that require dredging of river channels (AIDESEP 2019), and deforestation in both the uplands and floodplains (Castello et al 2013). Understanding how these changes may interact either synergistically or antagonistically is important for planning dams and navigation channels as well as regulating deforestation. Furthermore, although our estimates based on hydrologic-hydrodynamic modeling considered a natural basin scenario, without the effects of anthropogenic impacts such as dams (Chaudhari andPokhrel 2022, Flecker et al 2022), the impacts of rainfall changes on inundation extent reported here are well aligned with evidence from other, rainfallindependent remote sensing datasets. It is important to stress that such models have simplifications such as time-constant and rectangular river channel cross sections, which are necessary to be assumed due to scarcity of detailed, in situ data to be used in parameterization. However, their ability to simulate large-scale inundation extent and rainfall-runoff processes has been thoroughly investigated and is considered satisfactory (Yamazaki et al 2012, Paiva et al 2013, Siqueira et al 2018, Wongchuig et al 2019, Fleischmann et al 2022. Future developments in model parameterization, especially with improved information on floodplain topography (Yamazaki et al 2017, Fassoni-Andrade et al 2020, will further improve the accuracy of inundation extent estimation at local scales.

Environmental and social implications of increased flooding
Increases in the extent, duration and connectivity of floodplain inundation in the Amazon Basin have multifarious implications for floodplain ecosystems, geomorphological dynamics, biogeochemical processes, and the people who depend on floodplains. The exchange of sediment between floodplain lakes and the river channel is expected to intensify with increased flooding, as reported for the Curuai floodplain lake (Rudorff et al 2018). In addition to local geomorphological changes, increased floodplain sediment deposition could in turn lead to changes in the total sediment export to the ocean (Anthony et al 2021).
Fluxes of methane and carbon dioxide from aquatic habits to the atmosphere in the Amazon are large and dependent on inundated area and exchanges of nutrients and organic matter between rivers and floodplains (Richey et al 2002, Melack et al 2004, Abril et al 2014. Hence, these fluxes will likely increase with greater inundation. The role of wetlands in the increasing rates of global methane emission has been debated in the literature (Zhang et al 2017, Wilson et al 2020. While airborne observations from 2010 to 2018 suggest no clear trend in Amazon methane emissions (Basso et al 2021), the implications for methane emissions of the increased Amazon floodplain inundation require more study.
Increased inundation extent and duration are expected to affect fishes and fisheries yields, likely increasing fish production via increased feeding opportunities, growth, and recruitment (Welcomme 1985, Castello et al 2015. However, fish production also depends on primary production and the response of forest trees to a changing flood regime (Castello and Macedo 2016). Increased duration of inundation results in shorter dry periods, which can affect many floodplain plant and animal species (Castello and Macedo 2016) as well as the production of livestock and crops on the floodplains (Junk et al 2000), and may exacerbate fishing difficulties during high water levels, which is known to induce food insecurity for riverine human populations (Tregidgo et al 2020).
Although human populations along Amazonian floodplains have historically adapted to cope with and benefit from the annual flood pulse, recent extreme floods are driving migrations and negatively impacting disease risk, sanitation, mobility, and the transport of agricultural goods (Pinho et al 2015, Fonseca et al 2022. Adverse effects of flooding are particularly severe for vulnerable populations that depend on subsistence agriculture and that cannot readily move (Pinho et al 2015). A number of adaptive actions are being implemented to cope with increased floods, such as improving flood risk mapping in remote and urban areas, strengthening of civil defense, social programs for disaster preparedness and post-event response, and development of early warning systems (Marengo et al 2013, Pinho et al 2015). However, communication to remote rural communities regarding flood risk and development of adaptation strategies, such as pre-event food saving, agricultural water-resilient structures (e.g. floating planting beds) and crops, and temporary upland migration, remains challenging (Andrade et al 2017), as does the development of flood-resilient urban centers.

Conclusions
The Amazon River basin has been facing multiple extreme events in the last decade. Here, we used two hydrological models and multiple remote sensing datasets to investigate the changes in the inundation dynamics of the Amazon since 1980. Hydrological models show that since then the maximum inundation extent of the central Amazon floodplain has increased by 26%, or ∼25 000 km 2 -an area equivalent to the size of Belgium. The Landsat-based GSWO dataset revealed increases in flooding duration and river-floodplain connectivity in the lower Amazon reaches, where multiple open water areas are common. Such changes have multiple consequences for floodplain ecosystems and the people that rely on them. Governments from local to national, as well as international organizations, must strengthen mitigation measures, particularly for vulnerable populations across the Amazon river-floodplain system. There is thus an urgent need to understand the changes, build resilience to their impacts, and mitigate their drivers to the extent possible.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).

Data and materials availability
All data used in this research are freely available, and described in the Materials and Methods section. In-situ river water level and discharge data from Brazilian National Water Agency are available at www.snirh.gov.br/hidroweb/, and from Peru's SENAMHI at www.senamhi.gob.pe/?p=pronosticometeorologico. Documentation and outputs for the MGB and CaMa-Flood models are available at www.ufrgs.br/lsh/ and http://hydro.iis.u-tokyo.acjp/ ∼yamadai/cama-flood, respectively, and GSWO data are available at Google Earth Engine (more details at https://global-surface-water.appspot.com). CHIRPS rainfall is available at www.chc.ucsb.edu/data/chirps. GRACE data from JPL were obtained at https:// podaac-tools.jpl.nasa.gov/drive/files/allData/tellus/ L3/mascon/RL06/JPL/v02/CRI/netcdf.