1. Introduction
Terrestrial surface water covers only about 5% of the Earth’s ice-free land surface [
1,
2], but plays a key role in global biogeochemistry, hydrology and wildlife diversity [
3,
4]. Consequently, it is critical to monitor the distribution of terrestrial water at large spatial and high temporal scales [
5,
6,
7,
8]. The work in [
9] estimates that nearly two-thirds of all terrestrial freshwater wetlands disappeared between 1997 and 2011. A more recent study used three million Landsat images to provide a high resolution map of surface water extent [
10]. The authors of this study estimated that 90,000 km
of permanent surface water had disappeared between 1984 and 2015.
Several different methods based on mapping water bodies from remote sensing datasets were used since the development of the Earth’s space observations: (1) visible; (2) infrared; (3) active microwave; (2) passive microwave and; (4) hybrid approach (passive and active microwave). Each method offers varying degrees of success in providing quantitative estimates of wetlands and inundation extents.
Water surface can be sensed by optical remote sensing methods. These methods typically exploit the absorption of longer wavelengths of light in water, especially the near and shortwave infrared parts of the electromagnetic spectrum [
11,
12]. Optical remote sensing provides very accurate mapping of water bodies. For example, [
13] senses lakes with a spatial resolution of 15 m, whereas [
14] sensed global water bodies at 30-m resolution using the Landsat data. The majority of the studies using optical remote sensing for water bodies’ detection provided only one snapshot of the hydrology stage. Due to the low revisit time of the optical sensors, few maps of a large area are available, and the minimum and/or the maximum of the flooded area are not always observed. The detection of sudden changes impacting the hydrologic cycle [
10] is also not sensed with accuracy. These limitations are crucial issues for hydrology application. However, some studies [
15,
16] managed to follow the temporal dynamics of the water surface in specific places. The most important limitation of the optical sensors is their inability to penetrate clouds and dense vegetation cover, which is essential during tropical wet seasons over the Amazon Basin.
Active microwave (scatterometers and Synthetic Aperture Radar (SAR)) is also sensitive to the water surface and has the ability to penetrate clouds and, to a certain extent, vegetation. Open water surfaces are generally characterized by low backscattering coefficients. Contrary to passive microwave, the signal is more contaminated by the vegetation. The spatial resolution of scatterometers is about 25–50 km, whereas the SAR provides higher resolution, typically around 10–150 m. Several studies have shown the ability of active microwave to map surface water at regional scales, such as over the Amazon region [
17] and over the Arctic region [
18]. Satellite altimeters are radars that observe at nadir to measure surface topography. They provide accurate measurements of water heights in rivers, lakes and wetlands [
19,
20,
21]. Due to their high spatial resolution, altimeters do not provide sufficient spatial coverage to analyze the water bodies’ temporal dynamics, except in polar regions [
22]. The future Surface Water Ocean Topography (SWOT) mission [
23] intended to be launched in 2021 is expected to provide K-band SAR interferometry, enabling continental altimetry.
Passive microwaves are sensitive to the distribution of liquid water in the landscape; they can operate day and night for all weather conditions. However, they are limited by a low spatial resolution (approximately 30 km). They can sense only large wetlands or regions where the cumulative area of small wetlands comprises a significant portion of the field of view. Consequently, they provide the capability to map the temporal evolution of surface water over the land surface due to their high temporal resolution. In previous studies, passive microwave measurements have shown the capability to sense the dynamics of terrestrial surface water at coarse resolution [
24,
25,
26,
27,
28,
29,
30,
31]. The basic principle of the surface water measurements based on passive microwave is explained by the difference of the emissivity between the water and the soil. Flooding surfaces decrease the emissivity in both vertical (V) and horizontal (H) polarization and increase the difference between the two polarizations, especially at low frequencies. This approach produces ambiguous estimation of surface water over regions with mixtures of open water and other complex surfaces (topography effects). The work in [
26,
27] has extensively studied the inundation area over the Amazon Basin with the Scanning Multichannel Microwave Radiometer (SMMR). However, their studies focus essentially on a restricted area close to Manaus town from 1979 to 1987.
Hybrid approaches combine the strengths of different types of sensors. For example, altimetry data are characterized by a high spatial resolution and a low temporal resolution and can be combined with passive microwave data having low spatial resolution and a high temporal resolution to obtain a product with both high temporal and spatial resolution. The Global Inundation Extent from Multiple-Satellites (GIEMS) products are based on merged data from passive (Special Sensor Microwave/Imager (SMM/I)), active microwave (European Remote Sensing satellite (ERS)) and data from an optical sensor (Advanced Very High Resolution Radiometer (AVHRR)) [
32].
Table 1 presents the major studies related to the observation and detection of water bodies from space by using the techniques presented above. Visible and infrared remote sensing methods were extensively used, but provided static maps of water bodies at the global scale or a dynamic map at the regional scale [
15,
33]. A lack of studies concerning dynamic water surface extent from 2013 to the present is clearly identified in this table.
The floodplains and wetlands of the Amazon River are important in terms of water volume and in terms of fluxes between the land and the atmosphere. Mapping water fraction under the Amazon tropical dense forest is challenging, but sensing water under dense vegetation remains a key issue in the remote sensing scientific community.
In this study, we developed a method to map the temporal evolution of the water bodies at coarse spatial resolution and weekly temporal resolution by using a microwave sensor at L-band (1.4 GHz) called Soil Moisture Ocean Salinity (SMOS) over the Amazon Basin. The SMOS satellite operates at L-band, and it was shown that this frequency is the most suitable, being less impacted by vegetation than higher frequencies [
34,
35,
36]. Originally, the SMOS satellite was dedicated to sense soil moisture over land surfaces and the ocean salinity. The SMOS physical signal (brightness temperature) is highly impacted by the presence of standing water over the ground.
Our motivation is to use a contextual radiative transfer model and a single dataset to estimate the water fraction over the tropical basin. The area of study and the datasets used in this work are presented in the
Section 2 and
Section 3, respectively.
Section 4 presents the algorithm permitting retrieving the water fraction extent from the SMOS data, and
Section 5 contains the results and the validation. The discussion and conclusions are presented in
Section 6 and
Section 7.
2. Study Areas
This study focuses on the Amazon Basin, which is the largest tropical basin with an area of approximately 6,000,000 km
and contributes up to 15% of the global river discharge to the ocean (approximately 200,000 m
s
discharge). With a sediment load of three million tons near its mouth [
48] and drainage area covers about 6,200,000 km
, almost 5% of all of the continental masses, the Amazon Basin is one of the most impressive hydrological basins of the world. The Amazon is highly interconnected by floodplain channels, resulting in complex flow patterns.
Figure 1 presents the Amazon Basin with the main rivers and floodplains. Covering more than 300,000 km
, the Amazon extensive floodplains play a crucial role for global climate and biodiversity, but they are still poorly monitored at a large scale, limiting our understanding of their role in flood hazard, carbon production, sediment transport, nutriment exchange and air-land interactions. Surface water stored in floodplains represents about half of the terrestrial water storage and 15–20% of the water that flowed out of the Amazon floodplains [
49,
50,
51,
52,
53]. Because it extends over two hemispheres, the Amazon region is characterized by several rainfall regimes. Rainfall shows opposing phases between the Northern and the Southern Hemisphere with a rainy season in austral winter in the Northern Hemisphere and summer in the Southern Hemisphere. The rainfall shows a gradient from northwest to southeast with decreasing rainfall amount and increasing length in the dry season. For the eastern part of the basin, the rainy season occurs from March–May, and the dry season prevails from September–November. For the northern regions, low rainfall seasonality is observed with wet conditions throughout the year. For more information on the Amazon hydrological regime, see [
54,
55,
56].
7. Conclusions and Prospect
This study presents the validation and the link to other hydrological components of regional (Amazon Basin) daily and multi-year (2010–2015) water surface extent maps from the SMOS mission at coarse resolution (25 km × 25 km). The SWAF product is based on L-band acquisitions. At such a frequency, the signal is highly sensitive to the standing water above the ground, and it is expected to penetrate deeper in the vegetation than at higher frequencies, such as visible and infrared or microwave at higher frequencies. As the L-band signal is more sensitive to open water under dense vegetation, the SWAF product provides surface water extent estimates (percentage of inundation in a pixel of 25 × 25 km) with a high temporal resolution (<3 days) based on the accumulation of daily surface water extent in the Amazon Basin between 2010 and 2015. The SWAF product is computed from the L-band, and it can be computed easily and quickly without any ancillary data. Over this basin, the water surface extent showed a strong seasonal and interannual variability with two marked droughts in 2010 and 2015.
The SWAF data were compared to three sets of static land cover maps provided from visible sensors (IGBP, GlobeCover and ESA CCI) and the average inundation extent from GIEMS over 1993–2007. It was found that the SWAF products are close to the IGBP and ESA CCI maps. On average and during the 2010–2015 period, 270,000 km were inundated over the Amazon Basin. A slight overestimation of the flooded areas could be noticed. Over the Amazon Basin, the SWAF products were highly correlated with water levels measured by Jason-2 (r> 0.8) for the significant stations. The temporal dynamics of the SWAF products were also validated against precipitation (TRMM data) and in situ discharge at the mouth of each river. It was found that over the Amazon Basin, the precipitations often precede the inundation by three months, and the water surface extent impacts the discharge at the mouth of the Amazon after one month. As expected by the microwave theory, the mall water fraction could not be detected by the large footprint of SMOS. This implied that low water fraction extent (<4%) could not be mapped by the SWAF products. The mountainous areas were also a limitation of the SWAF products. The topography-modified local incidence angles implied significant impact on the microwave signal and, consequently, on the water surface estimation. The effects led to overestimation of the water fraction. To avoid this effect, the areas with high topography slopes were flagged in the SWAF products.
Based on the SMOS product, the SWAF products declined with several incidence angles at two polarizations (H and V). It was clear that high incidence angles (>47± 5) were not suitable to sense the water surface from the L-band microwave signal. The H-polarization tended to increase the lower value of the water fraction extent with respect to the V-polarization. The SWAF products computed with different angles and polarizations led to similar results with very slight differences over the Amazon Basin. For future use, the authors advise the use of SWAF computed with low incidence angles (32± 5 and/or 37± 5) at V for the Amazon Basin.
The methodology permitting retrieval of the water fraction applied in this study does not require much computation time and can be easily be applied to another L- band microwave dataset, such as the new Soil Moisture Active and Passive (SMAP) data or an older dataset (SSM/I…). The method had been validated over the Amazon Basin by taking advantage of the numerous data and research performed over this area.
In the near future, this recent water surface fraction product can be easily extended with the future SMOS data and the Soil Moisture Active and Passive (SMAP) data to obtain a long record of inundation products under dense vegetation. These data will be useful to better understand the water, carbon and methane cycles over the tropical areas. By adding a third component (saturated soil) on the first-order radiative transfer, this method is likely to be applied in other regions in the world.