Thermal remote sensing of water under flooded vegetation: New observations of inundation patterns for the ‘Small’ Lake Chad

Summary

Lake Chad at the border of the Sahara desert in central Africa, is well known for its high sensitivity to hydroclimatic events. Gaps in in situ data have so far prevented a full assessment of the response of Lake Chad to the ongoing prolonged drought that started in the second half of the 20th century. Like many other wetlands and shallow lakes, the ‘Small’ Lake Chad includes large areas of water under aquatic vegetation which needs to be accounted for to obtain the total inundated area. In this paper, a methodology is proposed that uses Meteosat thermal maximum composite data (Tmax) to account for water covered by aquatic vegetation and provide a consistent monthly time series of total inundated area estimates for Lake Chad. Total inundation patterns in Lake Chad were reconstructed for a 15-yr period (1986–2001) which includes the peak of the drought (86–91) and therefore provides new observations on the hydrological functioning of the ‘Small’ Lake Chad. During the study period, Lake Chad remained below 16,400 km² (third quartile ∼8800 km²). The variability of the inundated area observed in the northern pool (standard deviation σ_{northern pool} = 1980 km²) is about 60% greater than that of the southern pool (σ_{southern pool} = 1250 km²). The same methodology could be applied to other large wetlands and shallow lakes in semi-arid or arid regions elsewehere using Meteosat (e.g. Niger Inland Delta, Sudd in Sudan, Okavango Delta) and other weather satellites (e.g., floodplains of the Lake Eyre Basin in Australia and Andean Altiplano Lakes in South America).

Introduction

Across the world most shallow lakes and wetlands are largely or in part covered by aquatic vegetation (Burgis and Symoens, 1988, Hughes and Hughes, 1992, Talling and Lemoalle, 1998, Fraser and Keddy, 2005). In central Africa, the transition from an ‘Average’ to a ‘Small’ Lake Chad in 1973 (Fig. 1, Fig. 2) was accompanied by the development of aquatic vegetation which now dominates the inundated area of Lake Chad (Carmouze et al., 1983, Lemoalle, 2004). Despite the critical need for information during stages of low levels, the hydrological functioning of the ‘Small’ Lake Chad during drought periods is not well documented (Olivry et al., 1996, Lemoalle, 2004). While the gauging station in Bol has been operating almost continuously since 1953 in the southern pool (Fig. 2), no in situ level data are available from the northern pool of the lake since 1975, when the lake was split into different water bodies. Quantitative long-term time series of observations of the northern pool can only originate from satellite data.

Using the earliest data available from satellite radar altimetry Topex/Poseidon (T/P), it is possible to estimate water height fluctuations in the southern pool as far back as 1992 (Birkett, 1995, Birkett, 2000, Cretaux and Birkett, 2006). T/P radar altimeter has no track above the northern pool but ERS-2 and Envisat satellites could provide some water levels from 1995 and 2002 respectively. Radar altimetry in the northern pool remains difficult given the complex nature of this region with the presence of islands, small channels, and changing vegetation.

Satellite data have been used extensively to measure inundated areas, which is arguably a more pertinent parameter than water level to capture the dynamics of large shallow water bodies (e.g., Ozesmi and Bauer, 2002, Jones et al., 2009). However, main limitations still remain; especially for the detection of water under flooded vegetation. Optical sensors, such as Landsat TM and MODIS, can easily detect open water using the strong absorption of solar energy by water in the near and middle infrared. Shallow depths and turbid waters, are better detected at greater wavelengths (>1 μm; mid-infrared/short-wave infrared) where the illumination of the suspended materials or of the shallow bottom of a water column is considerably reduced (Li et al., 2003, Bukata, 2005). The mapping of water underneath vegetation is problematic for optical sensors as they only observe the skin surface of the vegetation. This limitation may be overcome by using classification techniques with multispectral data to selectively map hydrophyte plants (e.g. Zomer et al., 2009). However hydrophyte plants may survive in wet soils when the inundation receeds and other types of vegetation (e.g. forest) may occasionally get flooded. In water limited environments, a high photosynthetic activity will differentiate flooded vegetation from the rest of the dry land. Satellite based vegetation indices, which are derived from optical data, give a measure of the photosynthetic activity. Vegetation indices from the NOAH/AVHRR and MODIS sensors have been used in semi-arid and arid regions to map water under aquatic vegetation across large wetlands outside the rainy season (e.g., Leblanc et al., 2003b, Mariko, 2003, Sakamoto et al., 2007). This technique is hindered by the fact that vegetation activity may often remain high for several weeks after water has receded. Normalised Difference Wetness Index (Gao, 1996, Xiao et al., 2002), or the Tasseled Cap wetness index (Ordoyne and Friedl, 2008) have also been applied. However the main limitation of all these optical-based approaches is the inability to map beneath clouds; and this is particularly a problem for tropical regions which are characterised by frequent cloud coverage.

Radar imaging is less affected by cloud cover and can penetrate vegetation at a depth which depends on the wavelength used and the structure (density and height) of the vegetation. Synthetic Aperture Radar (SAR) sensors are active instruments that measure the backscattering coefficient of target surfaces (e.g., ERS1 and 2, Envisat ASAR, JERS, PALSAR, RADARSAT). Under low wind conditions, open water reflects specularly the radar signal resulting in a characteristic low backscatter (return) to the sensor, while water underneath vegetation may often be identifiable by a double bounce effect (Lewis, 1998). A main limitation of radar data is that wind-induced waves (Braggs waves) increase scattering from open water surfaces (Alpers, 1985, Smith and Alsdorf, 1998). As a result the radar signal over water surfaces continually changes with waves. This phenomenon is particularly strong in C band but is often insufficient to affect L band data (Smith and Alsdorf, 1998, Alsdorf et al., 2007). L band is also capable of greater penetration of the vegetation and has been used successfully to map flooded areas under thick forest (Hess et al., 2003, Frappart et al., 2005, Martinez and Le Toan, 2007, Rosenqvist et al., 2007, Alsdorf et al., 2007). The use of L band data from the JERS and PALSAR sensors for inundation monitoring is mainly restricted by acquisition times and limited archives rather than by weather or vegetation condition. In the future L band archives may improve with the recent launch of a second PALSAR satellite and more systematic acquisitions, but, in general, frequent acquisition of radar data over large areas may remain expensive and impractical. Passive microwave data (e.g. SSM/I and ISCCP) are helpful for delineating inundated areas (e.g., Sippel et al., 1998, Hamilton et al., 2002), in particular when used in conjunction with other sensors to limit confounding factors such as atmospheric condition and vegetation (Prigent et al., 2007), but their use is limited by their low spatial resolution (tens of km).

To date, the remote sensing of inundation patterns in the specific case of Lake Chad can be summarised as follow. Mohler et al. (1989) used registered photographs from manned space flights to estimate the open water area of Lake Chad for 8 dates between 1966 and 1986. Citeau et al. (1989) and Rigal (1989) used Meteosat channel 1 in the visible domain (2.5 km resolution) to monitor the dynamics of open water in 88/89. Wald (1990) compiled a time series showing the variation of the area of open water in Lake Chad between 1966 and 1989 (23 dates with 83% of the data between 1984 and 1989) derived from space photography (Mohler et al., 1989, Wood et al., 1989) and Meteosat visible channel (Anonymous, 1989, Citeau et al., 1989, Rigal, 1989, Wald, 1990). The area of the annual maximum inundation in the northern pool observed in January was derived using Landsat MSS band 7 (0.8–1.1 μm) data from 1973 to 1976 and Meteosat data in the visible channel from 1977 to 1990 (Lemoalle, 1978, Lemoalle, 1991, Olivry et al., 1996). More frequent inundation maps capturing the seasonal variability of Lake Chad from 1995 to 1998 were obtained by Birkett (2000) using a time series of NOAA/AVHRR LAC images. Birkett (2000) used a threshold technique in AVHRR band 2 (0.7–1 μm) to detect open water. Rigal (1989) first demonstrated the use of thermal infrared data to map inundated areas of Lake Chad under aquatic vegetation. The author used nine AVHRR images between November 1988 and April 1989 to estimate the variations of the total inundated area of Lake Chad, using a normalised water index based on the contrast of water (open and covered by vegetation) in the thermal infrared (band 5) and near infrared (band 2) channel. Rosema and Fiselier (1990) used the principle of thermal inertia to detect the total inundated area using Meteosat noon and midnight thermal images in the Yaeres flood plain, south of Lake Chad which are typically covered by tall grasses. They captured both open water and water covered by aquatic vegetation but processed only 11 dates between 1978 and 1986. Travaglia et al. (1995) also applied this thermal inertia method using AVHRR/LAC thermal data (band 4) to map the extent of water bodies in the Sudd wetlands in Sudan. This region has a similar environment to that of Lake Chad with wetIands covered by dense mats of floating vegetation within which areas of open water occur. Travaglia et al. (1995) report that AVHRR/LAC thermal data (band 4) can achieve a clear distinction of both open water and water under vegetation from the bare soil and soil with active vegetation.

In this paper, a methodology is proposed that uses Meteosat maximum thermal composite data (Tmax) to account for water under flooded vegetation and provide a consistent time series of monthly estimates of the total inundated area of Lake Chad. The long archive of Meteosat thermal composite data also allows us to reconstruct monthly inundation patterns of the lake from 1986 to 2001 and therefore provides new observations on the hydrological functioning of the ‘Small’ Lake Chad.