SYNERGY OF LOW AND MEDIUM RESOLUTION ENVISAT ASAR AND OPTICAL DATA FOR LAKE WATERSHED MONITORING: CASE STUDY OF POYANG LAKE (JIANGXI, P.R. CHINA)

Andreoli R. (1), Yésou H. (1), Li J (2), Desnos Y-L. (3)

(1) SERTIT, Université Louis Pasteur, Bld Sébastien Brant, BP 10413 F-67412 Illkirch-Graffenstaden,

France, remi@sertit.u-strasbg.fr (2) IWHR, Ministry Water Resource, Beijing, China lijiren@iwhr.com; huangsf@iwhr.com

(3) European Space Agency - ESRIN, Italy Yves-louis.desnos@esa.int

ABSTRACT Poyang Lake, one of the most regularly flooded areas in China, can be considered as a key natural flood control and reduction element within the Changjiang middle basin. Within the Flood DRAGON Project, part of the MOST-ESA DRAGON Programme, Poyang Lake's water extent was monitored based on 64 ENVISAT low and medium resolution ASAR and MERIS Full Resolution data, over a two and half year period. It’s the first time that such an amount of ENVISAT data was exploited in monitoring inland lake water extent variations. This original integration approach permitted: lake-surface variation analysis, yearly submersion-time estimation, and the recognition of three hydrological sub-systems. The results highlight the great potential of ENVISAT and more largely of Earth Observation Medium Resolution data in monitoring and managing large inland water bodies. This approach can be applied worldwide in a global climate change context. Key words: Poyang Lake, floods, submersion time, ENVISAT, ASAR, WSM, MERIS FR, Earth Observation time series, hydrological dynamic. 1. INTRODUCTION Part of the MOST-ESA DRAGON Programme [1, 2], the Flood DRAGON Project's aim is to explore ENVISAT's ASAR and MERIS spatial and temporal characteristics and their potential in rapid flood mapping and monitoring [3, 4]. A second level purpose of the Flood DRAGON project consists of extending and improving methods and tools concerning flood management and hydrological balance analysis in terms of prevention and forecasting using space technologies [5]. Poyang Lake, one of the most regularly flooded areas in China, was selected as the principle test site of the Flood DRAGON Project. Located in Jiangxi Province, Poyang Lake is the largest freshwater lake in China and constitutes a major hydrological subsystem of the middle Changjiang (Yangtze River) basin in central China. Playing a key role in Changjiang basin flood regulation and control, Poyang Lake undergoes very significant seasonal water level variations: the lake's elevation varies between 9 and 18 m and its size fluctuates from less than 1,000 km² during the dry

winter period to more than 4,000 km² during the wet summer season [6]. The Poyang Lake area is one of the most regularly flooded areas in China. Seven major floods have occurred in the past fifty years (1954, 1973, 1977, 1983, 1992, 1995, 1998) and the most severe ever recorded was in 1998 [7]. A previous innovative Flood DRAGON project study demonstrated the potential application of low resolution, ASAR Global Monitoring Mode (GMM) data to regional landscape characterization and water body monitoring [8]. Developing on this, here the integration of ENVISAT optical and ASAR medium resolution data (MERIS and ASAR Wide Swath Mode) and the synergy of multi-source and multi-resolution earth observation time-series data, as applied to flood mapping and monitoring, were explored and assessed. Over the last three years 64 earth observation datasets were analysed and compared both amongst themselves and with high resolution earth observation data. An estimation of the yearly submersion time within the Poyang Lake area based on ENVISAT time-series data analysis is proposed, leading to a preliminary spatial characterization of water dynamics inside the lake. 2. DATABASE AND METHODOLOGY 2.1. The Poyang Lake area database The database (Table 1) over the Poyang Lake area includes a set of optical high resolution reference data (2 Landsat, 6 SPOT4, and 1 CHRIS PROBA data) acquired at different hydrological periods, and a 3 arc second (90 m) SRTM Digital Elevation Model (DEM). Landsat reference data enabled a preliminary hydrodynamic characterization of Poyang Lake’s annual water level variations and a land cover mapping (Fig. 1). Water bodies and wetland areas were extracted from each Landsat image. Lowlands, including lands below 20 m above mean sea level and frequently flooded overbanks, were discriminated using the SRTM DEM and medium scale landscape analysis. The resulting classification shows the annual water level dynamics while differentiating reference low-water levels, areas of seasonal water variations (lakeshore) and wetland soils associated with lowlands (Fig. 1).

_____________________________________________________

Proc. ‘Envisat Symposium 2007’, Montreux, Switzerland 23–27 April 2007 (ESA SP-636, July 2007)

Figure 1: Lowland landcover in the Poyang Lake area

derived from Landsat data

Five multispectral SPOT 4 data acquired within the framework of the Sino-French WARM cooperation programme [5] cover each of the major types of Poyang Lake's dynamics between November 2004 and January 2006 (infilling, high water-level, draw-off, low water-level); the last one was recorded during the dramatic 1998 floods.

Table 1: Reference Earth Observation optical data

Satellite Resolution (m) Date Hydrological

period Landsat 5 TM 28.5 15/07/1989 High water-level

Landsat 7 ETM+ 28.5 10/12/1999 Low water-level

20 08/08/1998 Flooding

20 23/11/2004 Draw-off

20 02/04/2005 Infilling

20 15/07/2005 High water-level

20 16/10/2005 Draw-off

SPOT 4

20 07/01/2006 Low water-level

CHRIS PROBA 18.5 22/06/2005 High water-level

The ENVISAT database (Table 2) includes a total of 64 images acquired between January 2004 and June 2006 with: 14 ASAR Global Monitoring Mode (GMM), 16 ASAR Wide Swath Mode (WSM) and 34 MERIS products (13 Level 1 and 21 Level 2).

Table 2: ENVISAT time-series data

ASAR MERIS FR (300 m)

GMM (500 m)

WSM (75 m) L1 L2

Total

2004 14 5 5 13 37 2005 0 4 6 8 18 2006 0 7 2 0 9

Total 14 16 13 21 64

2.2. Data preparation ENVISAT time-series data analysis is based on a processing chain including three major steps (Fig. 2): a pre-processing step of data calibration and geometric corrections; water body extraction and validation; and finally, submersion time estimation using an average of monthly mean of water presence/absence.

Figure 2: processing chain

Global Monitoring Mode and Wide Swath Mode data were corrected for antenna pattern effects using, respectively, the Rough-range calibration tool and the backscattering image generation tool of the Basic ENVISAT SAR Toolbox (BEST) software [9]. But the Rough-range calibration process does not correspond to a “true” calibration, so GMM data were then standardized to make them comparable. The geometric correction of ENVISAT ASAR GMM data was carried out using a second order polynomial model in ERDAS Imagine V8.7. Wide Swath Mode data were ortho-rectified using the SRTM DEM with the PCI Geomatica v9.1 ortho-rectification tool. Whereas, the ortho-rectification tool of VISAT (BEAM 3.4.1 software) was used on MERIS data, with GETASS DEM as the topographical reference [10], providing good results with respect to data resolution and terrain configuration (relatively flat areas).

2.3. Water extent extraction and analysis The low backscattering characteristic of water in ENVISAT ASAR data, usually appearing in dark tones on SAR images, was exploited to enable an initial water extraction through pixel-based thresholding, This was then manually corrected by an operator. Due to product characteristics, water extraction techniques aren’t similar for MERIS Level 1 and Level 2 products. Two methods were used jointly to extract water bodies from MERIS Level 1 products, according to the hydrological period. During low water-level periods, a single threshold using band 14 (Near Infrared) was performed. During high water-level periods, an NDVImean, using the Eq. 1, was used to extract water bodies through thresholding.

⎪⎪⎩

⎪⎪⎨

⎧

===

=

+

−

=

∑∑

∑∑

==

==

bands red ofnumber bands infrarednear ofnumber

valueereflectancnumber band

:

11

11

9

6

15

10

9

6

15

10

nn

nn

nnNDVI

red

pir

iredipir

iredipirmean

DNi

With

DNDN

DNDN

(1)

A single threshold, using either band 10 or band 13 (Near Infrared), was used to extract water bodies from all MERIS Full Resolution (FR) Level 2 products. 2.4. Submersion time estimation Submersion time P(w) is an average of the monthly mean of water surfaces expressed in percent of year (Eq. 2).

×⎟⎟

⎠

⎞

⎜⎜

⎝

⎛

⎟⎟

⎠

⎞

⎜⎜

⎝

⎛= ∑ ∑

= =1001

121)(

1,

12

1

n

ttm

mw

nwP

m

m

(2)

[ ] { }⎪⎪⎩

⎪⎪⎨

⎧

∈∈ 1 ,0 and ;1with month for the data ENVISAT rom extractedlayer extent water :

month theof layersextent water ofnumber :year theofmonth :

year ofpercent in express timesubmersion :)(

where

,,

tmmtm

m

wntmfw

mnm

wP

Before calculating P(w), the sixty-four ENVISAT water layers were down-sampled to 75 m resolution using a cubic spatial interpolation. The raw P(w) map must be enhanced because of resolution differences in water surfaces extracted from ENVISAT data. The coarser data (GMM and MERIS) lead to an under-estimation of fine and small perennial water bodies but also to distortions of maximal water extent boundaries. Regarding these hydrological elements, a Landsat reference land cover map was used for the manual correction of the P(w) map by visual interpretation.

3. POYANG LAKE MONITORING: RESULTS AND DISCUSSION

The analysis of ENVISAT water extent comparisons between ENVISAT water bodies and SPOT 4 water bodies show that ENVISAT ASAR GMM, WSM and ENVISAT MERIS data are very efficient and that their water recognition concords [11]. 3.1. Poyang Lake surface variation between January

2004 and June 2006 64 water body extents derived from ENVISAT data, acquired between January 2004 and June 2006, corresponding to lake state at the acquisition time, were used to estimate Poyang Lake's surface area (Table 3). The estimation was performed while excluding Junshan Lake (approx. 174.8 km²), it is considered as disconnected with Poyang Lake. Also, rivers and small depressions around the lake were not taken into account.

Table 3: Poyang Lake water surface characteristics derived from ENVISAT data.

GMM WSM MERIS maximal lake surface (km²) 2,939.00 2,725.90 3,103.83

Minimal lake surface (km²) 376.00 913.40 557.91

Mean surface (km²) 1,893.3 Median surface (km²) 1,863.7

The smallest lake surface estimated for the monitoring period, derived from a 20th of February 2004 ENVISAT Global Monitoring Mode image (polarization VV), is about 376 km². The smallest surface recorded by MERIS data (557.91 km²) comes from the same period and tends to illustrate a lower Poyang Lake level than normal. The largest estimated surface is about 3,104 km² from MERIS data acquired the 23rd July 2004. During the 2004 and 2005 wet seasons, the lake's surface exceeded 3,000 km². And, the average Poyang Lake surface was estimated at 1,893.3 km² for the monitoring period (January 2004 to June 2006).

The variation of Poyang Lake's surface derived from ENVISAT data and according to the season appears concordant with Poyang Lake characteristics found in the literature [6, 7] (Fig. 3). The annual maximum water extent is reached during flood season at the end of July for the years 2004 and 2005. The Lake's surface is generally greater than 2,500 km² from June-July to September (3-4 months). The smallest lake surface often occurs during the dry season at the end of December and during January.

Figure 3: Poyang Lake variations observed by low and medium resolution ENVISAT data between January 2004 and

June 2006.

The lake's surface is generally lower than 1,000 km² during months of December and January. The general tendencies, outlined in figure 4, show that the infilling of Poyang Lake starts early – end of February, beginning of March – and takes 4 months (March to June). On the contrary, the draw-off of the lake is rapid (1-2 months). Starting at the end of September, Poyang Lake empties about 1,000 km² in only 1 month (October) to reach 1,500 km² at the beginning of November and less than 1,000 km² in December. The area estimated for February 2005 is anomalous but corresponds to a flood peak [12]. The beginning of 2006 shows a regular infilling of the lake as usual, but this was affected by a drought in the summer, with a very low water level seen. 3.2. Poyang Lake dynamics Submersion time estimation corresponds to a spatial and continuous representation of water presence between January 2004 and June 2006 expressed in percent of a year. It allows a quantitative description (Fig. 4) of the lake's variations but also a spatial characterization of water fluctuation inside the Poyang depression (Fig. 5). The size of Poyang Lake, including perennial water bodies and area of water-level variations, had been estimated at 3,436.2 km² which is 332.4 km² greater than the maximum instantaneous lake surface observed between January 2004 and June 2006. As previously mentioned analysing ENVISAT GMM dataset, flooding events induce not only a dilatation phenomenon but also a translation of flooded area inside the depression [8]. During 3 months a year, Poyang Lake's area exceeds 3,000.0 km² and about 70 days a year the lake's size

goes below 1,000 km² (Fig. 5). Median lake area is about 2,215.1 km² meaning that these areas are under water half a year (183 days). Nearly 88% of the lake's maximum size corresponds to lakeshore (area of water level variations), 3,057.3 km² in other words (Fig. 4).

Figure 4: Yearly submersion time cumulative frequency Therefore, it is necessary to note that the estimated yearly submersion time is only representative of hydrological conditions of the analysed years bearing in mind that 2004 and 2005 were characterized by normal Poyang Lake water levels without pronounced floods, and the year 2006, as previously mentioned, was drier with very low water levels during the summer months. 3.3. Recognition of Poyang Lake's sub-systems Poyang Lake is a major storage basin of the middle Changjiang reach. From a hydrological point of view, the submersion time can distinguish between the

perennial hydrological elements (year-round water surface) and the flood storage areas (temporary water surface). According to the shape of perennial water bodies, water tanks and water transfer channels linking tanks inside the lake system can be identified. The spatial distribution of submersion time also highlights different storage areas according to the duration of water presence and their geographical location within the hydrological system. In this way, Poyang Lake Basin can be divided into 3 major sub-systems (fig. 5): A. The channel linking Poyang Lake to Changjiang

where perennial water bodies form braided streams. Most of the channel is under water nearly 50% of the year. Only border fringes and lateral valley ends diverge from this.

B. Small lateral lakes, such as Kangshan basin, Junshan and Qinglan Lakes lining Poyang Lake to the south-east, are perennial. Only small lateral valley ends present pronounced dynamics. In this sector, the Kangshan basin doesn’t appear totally perennial, but this can be explained by the fact that vegetation usually grows on the edges of the basin and floats on the water. It must also be noticed that Kangshan basin is used as a flood retention basin and it had not been flooded since 1998.

C. The Poyang Lake main basin can be divided in two sub-systems, firstly, the Gan and Fuhe delta and Poyang Lake and, secondly, the Xi He and Dalianzi Hu :

C.1. Gan and Fuhe delta and Poyang Lake, the south-western part of Poyang Lake: Gan and Fuhe delta can be divided into 3 parts drained by 3 different branches of the Gan River. The Northern branch of the Gan River, with Lao River, flow directly into the channel linking Poyang Lake to Changjiang. In this sector of the delta, the lakeshore shows a strong hydrological dynamic. It seems that flows fill the lakeshore basin only when the channel is full. In the central part of the delta, the lakeshore presents a less pronounced dynamic than in its northern and southern parts. Flows from the central branch of the Gan River directly fill the principal tank of the Poyang Lake. The southern part of the delta, drained by the third branch of the Gan River, the Xin River and the Fuhe River, presents the strongest dynamic of the lakeshore. Water flows coming from these rivers appear to be concentrated into a transfer channel which bypasses the main Poyang Lake tank to the north before flowing directly into the channel linking Poyang Lake to Changjiang. This lakeshore part might fill when the transfer channel is blocked by the infilling of the Poyang Lake.

C.2. Xi He and Dalianzi Hu, the north-eastern part of the Poyang Lake: This area is drained by Po River and Xi River. The Xi He water tank, the main perennial water body of this area, collects flows from the two main rivers. Its lakeshores show the strongest dynamics in the north-western part of the Xi He and also in lateral valley ends. The Dalianzi Hu is a lateral lakeshore of the Po River which is not directly supplied by this river but indirectly by overflow of the Xi He and Poyang Lake. The Dalianzi Hu presents the strongest yearly submersion dynamics in this Poyang Lake part.

Figure 5: Estimated Poyang Lake submersion times derived from ENVISAT low and medium resolution

time-series data acquired between January 2004 and June 2006 (lengthy to short from blue to red).

This spatial recognition of Poyang Lake sub-systems based on estimated yearly submersion time is the first step in the development of hydrodynamic models. Over areas where hydrophile flora and fauna species depend on the presence of water, the proposed hydrological sub-systems defined by the submersion time map are key elements in a sensitive ecosystem dynamic characterization. For example, the proposed C.1 area corresponding to the Gan and Fuhe delta provides habitats for rare migratory birds living through winter and it is one of the main national nature reserves of

China. This type of ecosystem is very sensitive to wetland dynamic disturbances such as changes in the spatial distribution of flooded areas and yearly submersion time. 4. CONCLUSION For the first time, a large inland water body was monitored based on 64 ENVISAT images, including low and medium resolution, ASAR and optical data, over a two and a half year period (January 2004 to June 2006). Water bodies were extracted from ENVISAT GMM, WSM and MERIS data, and the comparison of these water bodies both with themselves and with reference, high resolution optical data has shown that ASAR GMM, WSM and MERIS water extents concord. ENVISAT water bodies were combined and integrated based on an original approach in order: 1) to monitor Poyang Lake's surface variations; 2) to estimate the yearly submersion time of the Poyang Lake; 3) and to recognize Poyang Lake's sub-systems. Further steps would be the integration of extreme hydrological years, characterizing drought periods and major floods. The integration of data characterizing these two opposite lake states would allow a complete characterization of Poyang Lake's water variation patterns. These results highlight the great potential of ENVISAT and more generally Earth Observation Medium Resolution data in monitoring and managing large inland water bodies. This approach can be applied worldwide in a global climate change context over large sensitive areas such as Dongting Lake in China or Tonle Sap in Cambodia. In the same way, this approach can be adapted to monitor large river dynamics and floods as on the Lena River or the Ganges and Brahmaputra lower basins. 5. ACKNOWLEDGMENTS The authors are grateful to ESA as this work was realized within the framework of the Sino-European joint research (ESA-NRSCC) DRAGON project 2551 and supported by ESA's “Support Training of Young European Scientists”. The authors would also like to thank Dr Christine KING (BRGM) for supporting additional SPOT coverage and the July 2006 field survey campaign through the Sino-French WARM cooperation programme. 6. REFERENCES 1. Desnos, Y-L., Bergquist, K., Li, Z., 2004. The

Dragon programme: ESA and China cooperate in earth observation. E.S.A bulletin (E.S.A. bull.) ISSN 0376-4265 European Space Agency bulletin 2004, no119: pp22-28.

2. Desnos, Y-L., Li, Z., 2006. EO Science and Applications Development in P.R. China. Proc. 2005 Dragon Symposium “Mid-Term Results”, Santorini, Greece 27 June – 1 July 2005, ESA SP-611: IX-XXV.

3. Yésou, H., Li, J., Li, J., Wang, X., Yida, F., Wang, Y., Huang, S., Xin, J., de Fraipont, P., 2004. Assessment of the Synergistic Exploitation of ENVISAT ASAR and MERIS Data for Plain Flood Rapid Mapping: a Part of the Dragon Flood Project. ENVISAT Symposium Salzburg 6-10 September 2004.

4. Li J., Yésou H., Huang S., Li J., Li X., Xin J., Wang X., Andreoli R., 2006. ENVISAT ASAR medium and high resolution images for Near Real Time flood monitoring in China during the 2005 flood season. Proc. 2005 Dragon Symposium “Mid-Term Results”, Santorini, Greece 27 June – 1 July 2005, ESA SP-611: pp213-226.

5. King, C., Li, J., Costes, M., Yésou, H., Prinet, V., 2004. WARM Water Risk. Management Research Network CEOS meeting Information, Products and Services for Disaster Management, Beijing China, 17-19 November 2004.

6. Shankman, D. and Liang, Q., 2003. Landscape Changes and Increasing Flood Frequency in China’s Poyang Lake Region. The Professional Geographer, 55(4): pp434-445.

7. Shankman, D., Keim, B. D., Song, J., 2006. Flood Frequency in China’s Poyang Lake Region: Trends and Teleconnections. Int. J. Climatol. 26: pp1255-1266.

8. Andreoli, R. and Yésou, H., 2006. Monitoring water level seasonal variations of large natural lake exploiting ENVISAT ASAR low resolution time series: application to Poyang Lake (P.R. China) during the 2004 – 2005 hydrological period. Proc. 2005 Dragon Symposium “Mid-Term Results”, Santorini, Greece 27 June – 1 July 2005, ESA SP-611: pp213-226.

9. ESA, Telespazio, 2005. BEST Basic Envisat SAR Toolbox User Manual Version 4.0.2.: 189p.

10. ESA, BEAM Hand0book: www.brockmann-consult.de/beam/documentation.html

11. Andreoli, R., Yésou, H., Li, J., Desnos, Y-L., Shifeng, H., De Fraipont, P., Poyang Hu (Jiangxi Province, P. R. of China) Area Variations between January 2004 and June 2006 Using ENVISAT Low and Medium Resolution Time Series. Journal of Geographic Information Science, in press.

12. Andreoli, R., Li. J., Yésou, H., 2006. ASAR and MERIS ENVISAT Low and Medium Resolution Product Time Series Exploitation for Large Flood Plain Monitoring (Poyang Lake , P.R. China) Within the Flood DRAGON Project. Proc. 2006 Dragon Symposium, Lijiang, China 10 July – 14 July 2006.