Surface freshwater storage and dynamics in the Amazon basin during the 2005 exceptional drought

The Amazon river basin has been recently affected by extreme climatic events, such as the exceptional drought of 2005, with significant impacts on human activities and ecosystems. In spite of the importance of monitoring freshwater stored and moving in such large river basins, only scarce measurements of river stages and discharges are available and the signatures of extreme drought conditions on surface freshwater dynamics at the basin scale are still poorly known. Here we use continuous multisatellite observations of inundation extent and water levels between 2003 and 2007 to monitor monthly variations of surface water storage at the basin scale. During the 2005 drought, the amount of water stored in the river and floodplains of the Amazon basin was ∼130 km3 (∼70%) below its 2003–7 average. This represents almost a half of the anomaly of minimum terrestrial water stored in the basin as estimated using the Gravity Recovery and Climate Experiment (GRACE) data.


Introduction
The amount of water stored and moving through the floodplains and wetlands of large river basins plays a major role in the global water cycle and is a critical parameter for water resources management. Covering more than 300 000 km 2 (i.e., 5% of the surface of the entire basin) Content from this work may be used under the terms of the Creative Commons Attribution-NonCommercial-ShareAlike 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. (Diegues 1994, Junk 1997, the Amazon extensive floodplains are particularly crucial to global climate and biodiversity, but they remain still poorly monitored at the large scale, limiting our understanding of their role in flood hazard, carbon production, sediment transport, nutrient exchange and air-land interactions. The droughts that affected large areas of this basin in recent years are among the most severe in the past centuries (Marengo et al 2008a), with the 2005 and 2010 events still considered as the most exceptional in the past 40 years. Mostly located in the Solimões, the Madeira, the Amazon rivers (figure 1(a)) and its southwestern tributaries  (Marengo et al 2008b, Tomasella et al 2011, the 2005 drought indeed affected an extensive area of 1.9 × 10 6 km 2 for the dry season, to 2.5 × 10 6 km −2 considering the maximum climatological water deficit (MCWD), based on satellite-derived rainfall anomalies (Lewis et al 2011). The impact on the Amazon rainforest was strong, with several studies reporting an increase in tree mortality and loss of biomass (Philips et al 2009), peaks of forest fires and burning of biomass (Aragão et al 2007, Koren et al 2007, Bevan et al 2009, and highlighting its vulnerability to extreme drought conditions, with large potential impacts on regional biogeochemical and carbon cycles (Philips et al 2009). During the low water stage season of 2005, in situ observations reported historic minima of river water levels, up to several meters below their mean (Marengo et al 2008a, Zeng et al 2008, Tomasella et al 2011 with important consequences also on human activities and economy. Despite the advent of hydrology-oriented Earth observation satellite missions, the spatial and temporal dynamics of surface freshwater storage are still poorly known Lettenmaier 2003, Alsdorf et al 2007). So the signatures of extreme climatic events such the drought of 2005 on the dynamics of surface freshwater volumes can only be inferred indirectly from satellite-based estimates of rainfall (Zeng et al 2008), from gridded measurements of rainfall (Marengo et al 2008a(Marengo et al , 2008b or from observations of integrated terrestrial water storage (TWS) variations as measured by the Gravity Recovery and Climate Experiment (GRACE) mission (Chen et al 2009). In spite of being the largest component of freshwater in the watershed on the seasonal time-scale, but also one of the major factors controlling surface processes and basin-wide hydrology, the surface freshwater stored in the Amazon is still not measured at proper space and time-scales, leaving major questions open: what is the seasonal amount of water in and out of the Amazon floodplain, its interannual variability and its behavior during exceptional drought events?

Maps of surface water levels
Maps of water levels over the floodplains of the Amazon Basin were obtained by combining observations from a multisatellite inundation dataset and altimetry-based water levels at monthly time-scales over 2003-7, where all the datasets overlap. Water levels, derived from ranges processed with the Ice-1 algorithm to obtain more accurate estimates (Frappart et al 2006), for 534 ENVISAT RA-2 altimetry stations (Santos da Silva et al 2012) were bilinearly interpolated over inundated surfaces estimated using multisatellite observations , Prigent et al 2007. Each monthly map of surface water levels has a spatial resolution of 0.25 • and is referenced to the EGM2008 geoid. The error on these estimates is lower than 10% (Frappart et al , 2011a. A map of minimum water levels was estimated for the entire observation period using a hypsometric approach to take into account the difference of altitude between the river and the floodplain (see the supplementary information available at stacks.iop.org/ERL/7/ 044010/mmedia).

Time series of water volume variations
At the basin scale, the time variations of surface water volume are simply computed as (Frappart et al 2011a): where V SW is the volume of surface water, R e the radius of the Earth (6378 km); P(λ j , ϕ j , t), h(λ j , ϕ j , t) and h min (λ j , ϕ j ) are respectively the percentage of inundation, the water level at time t and the minimum of water level of the pixel of coordinates (λ j , ϕ j ); λ and ϕ are the grid steps in longitude and latitude, respectively. This minimum of water level is estimated through a hypsometric approach relating the percentage of inundation of a pixel to its elevation (see the supplementary information available at stacks.iop.org/ERL/7/ 044010/mmedia for more details).
Accordingly, the time variations of volume of TWS anomalies from Level-2 GRACE solutions filtered using an independent component analysis (ICA) approach (Frappart et al 2010b(Frappart et al , 2011b are computed following Ramillien et al (2005): where h tot (λ j , ϕ j , t) is the anomaly of TWS at time t of the pixel of coordinates (λ j , ϕ j ).

Results
For the very first time, a continuous mapping of surface water levels and surface water volumes, as well as their temporal dynamics at interannual time-scale, are presented for the Amazon river, the largest drainage basin on Earth. Second, surface water volume variations for the Amazon river are also estimated using surface water level maps (see figure 1(b), figure 2 and figure S2 (available at stacks.iop.org/ERL/7/044010/mmedia), or Frappart et al (2008Frappart et al ( , 2010aFrappart et al ( , 2011a for more details). The time series of surface water volume over 2003-7 for the Amazon basin was decomposed into interannual (figure 1(c)) and annual (represented for 2005 in figure 1(d)) terms using a 13-month sliding average and compared to river discharge for the whole Amazon basin. The surface water volume leads the interannual variations of the river discharge in  (Zeng et al 2008). The annual cycle of surface water storage for 2005 was close to or above the mean from February to June 2005, peaking in May with a value around +σ (one standard deviation or STD). Then, it dropped significantly below the mean (values lower than −σ ) from July to December ( figure 1(d)). These results are also in good agreement with what was observed on river discharge in Obidos (Tomasella et al 2011).
This unique opportunity to monitor the changes of water level all along the hydrological cycle at monthly time-scales is illustrated in figure 2 (along with figure S2 available at stacks.iop.org/ERL/7/044010/mmedia) for the drought of 2005. The anomalies of surface water levels averaged over two consecutive months during 2005 are compared with bi-monthly anomalies of rainfall from the Tropical Rainfall Measuring Mission (TRMM, see the supplementary information available at stacks.iop.org/ERL/7/ 044010/mmedia) (with an advance of two months) and TWS from GRACE (figure 2 for the dry season, from July to December, and figure S2 (available at stacks.iop.org/ERL/7/ 044010/mmedia) for the wet season, from January to July). Rain deficits (upper panel) in the northern and western part of the basin in the heart of the rainy season (May-June), are responsible for anomalously low levels in the wetlands of the central corridor of the Amazon two months later (September-October), in good accordance with the TWS observations (lower panel). The spatial and temporal patterns in the anomalies of surface water (center panel) are consistent with both in situ measurements of water levels and discharges and satellite-derived observations of TWS (lower panel). For instance, in the central part of the Amazon (from Manacapuru (60.61 • W, 3.31 • S) to Obidos), the surface water maps present levels close to or above the mean until May-June 2005 (figure S2 available at stacks.iop.org/ERL/7/044010/mmedia) that then started to drop until a minimum in September-October 2005 (figure 2) is reached, similarly to what was recorded by gauges (Tomasella et al 2011). In the Madeira basin, the water levels between 10 • S and 5 • S were close to the mean until March-April 2005, and then below, with a minimum in September close to 5 • S as observed in Fazenda Vista Alegre (60.03 • W, −4.90 • S). In the Negro basin, an important contrast is observed between the upper (above the mean over the whole period) and the lower ( Time variations of surface water volume over 2003-7 were analyzed in the major western and southern tributaries of the Amazon. The most important contributions come from the Solimões and the Madeira basins (∼30% and ∼25% respectively) whereas the contribution from the Tapajos represents less than 6% of the water stored in the surface reservoir of the Amazon basin. The interannual variations of surface water generally precede the interannual variations of discharge by one month in the Solimões (R = 0.94) and Madeira (R = 0.84) basins (figures 3(a) and (c)). Good, but lower agreement can be observed between interannual variations of surface water storage and discharge for the Tapajos (R = 0.71, t = 0, where t is the time shift between the two time series to be compared, figure 3(e)). The discharge values for the four stations were obtained from ORE HYBAM (see the supplementary information available at stacks.iop. org/ERL/7/044010/mmedia). These differences in time shift are consistent with what we know about the dynamics of surface water in these sub-basins. The white waters (turbid with large amount of dissolved organic carbon) originating from the Andes loaded with sediments during their stay in the extensive floodplains distributed along the Solimões and most of the tributaries forming the Madeira have a longer residence time in the basin than the clear waters (transparent and containing low content of dissolved organic carbon) of the Tapajos descending from the Brazilian shield through numerous waterfalls and rapids. The analysis of the 2005 annual cycle also reveals differences among these sub-basins. The volume of surface water in the Solimões basin was close to the mean or above during the rising period, peaking at a value greater than +σ in May, then declined rapidly, with a minimum reached below +σ in October ( figure 3(b)). Similar behavior is found in the Tapajos (with a peak reached in April, one month earlier than usual, figure 3(f)). Most of surface waters in the Tapajos are located in the large estuary formed by its encounter with the Amazon. At its mouth, its level is controlled by the stage of the Amazon. This can account for the similar temporal pattern found in the Tapajos and Solimões 2005 annual cycle for surface waters. The lower agreement with discharge (R = 0.71) at the interannual time-scale is more likely caused by differences of the hydrological regime between the upper and lower parts of the Tapajos (figure 3(f)). In contrast, the volume of surface water in the Madeira basin was below the mean until May, and then close to the mean (figure 3(d)). These results are consistent with what was observed at in situ gauges (Marengo et al 2008a, Tomasella et al 2011. The impact of the 2005 drought was quantified for the surface water storage and the TWS for the whole Amazon basin (respectively 129 and 245 km 3 below the 2003-7 average), and for the three sub-basins mentioned above for which different hydrological behaviors are observed during the 2005 drought (table 1). The minimum volume of water stored in the Amazon was by 71% lower for the surface reservoir, under the assumption that the storage below the minimum water level can be neglected, compared to the average during 2003-7, and by 29% for the total hydrological reservoirs. If the 2005 drought strongly affected the four different western and southern tributaries, its impact on TWS also differs from one to another, giving us information on the importance of the surface reservoir in the Amazon basin. Notice that surface water storage and TWS were much more affected by drought in the Solimões basin than in the three other tributaries. This coincides with the areas of largest anomalies of MCWD and increase in tree mortality (Aragão et al 2007, Lewis et al 2011, and with regions with important fire activity in 2005 (Koren et al 2007).

Discussion and conclusion
Our results provide the first interannual estimates of the variations of surface water storage in a large basin at monthly time-scale. They reveal that, during 2003-7, the variations of surface water reservoir vary from 800 to 1000 km 3 per year, which represents 15%-20% of the water volume that flowed out of the Amazon basin and about half of the variations of the total amount of water in the Amazon basin as detected using GRACE data. This result is 3-4 times greater than what was found by a previous study solving the water balance equation with gravimetric and imaging satellite methods (i.e., GRACE, SRTM, GPCP and JERS-1) for six GRACE gridcells  of 330 km of spatial resolution encompassing the floodplains along the Amazon mainstem (Alsdorf et al 2010). The major reason for this discrepancy must come from the leakage from other regions, due to the spherical harmonics representation of the GRACE data, which contaminate the signal at the GRACE gridcell resolution. Our estimates agree well with (i) analysis of GRACE data and GLDAS/NOAH outputs which show that the TWS is equally partitioned between surface and sub-surface reservoirs and soil water (Han et al 2009), and (ii) modeling results from ensemble hydrological simulations with river routing which found that surface water and shallow groundwater represents 73% of the TWS in the Amazon basin . In addition, the method presented here to derive water levels from multisatellite datasets over rivers and floodplains offers the first opportunity to continuously monitor mass transport in the surface water reservoir before the launch of the NASA-CNES Surface Water and Ocean Topography (SWOT) mission in 2019. It makes it possible to study the changes affecting the hydrological cycle in large river basins covered with floodplains. It also helps better understand the complex dynamics of surface water in large drainage basins (i.e., backwater effects, the Amazon flood-pulse linked to the strong seasonality of the rainfall or time residence of water in the floodplains).
The surface water level maps give unique and valuable spatial information on the time evolution of floodplain reservoirs during the hydrological cycle in response to rainfall forcing caused by interannual and longterm variability of both the tropical Pacific and northern Atlantic tropical oceans.
They permit one to directly identify the regions most severely affected by exceptionally low stages during the extreme drought of 2005 (the volume of surface water in the Amazon basin during the 2005 low stage period was 71% below its 2003-7 average according to our results). The estimated spatial and temporal patterns of surface water storage are in good agreement with in situ gauge records, satellite-derived hydrological variables, and ecological parameters. Removed from GRACE-derived TWS, they will permit a direct estimate of the soil water and groundwater storages in the Amazon basin.