Remote Sensing of Environment 109 (2007) 295 – 313

www.elsevier.com/locate/rse

Detecting temporal changes in the extent of annual flooding within the

Cambodia and the Vietnamese Mekong Delta from
MODIS time-series imagery
Toshihiro Sakamoto a,⁎, Nhan Van Nguyen b , Akihiko Kotera a ,
Hiroyuki Ohno a , Naoki Ishitsuka a , Masayuki Yokozawa a
a
National Institute for Agro-Environmental Sciences, Tsukuba, Ibaraki, Japan
b
Sub-National Institute of Agricultural Planning And Projection, Ho Chi Minh City, Vietnam
Received 1 September 2006; received in revised form 8 January 2007; accepted 13 January 2007

Abstract

This paper presents the methodology used to detect temporal changes in the extent of annual flooding within the Cambodia and the Vietnamese
Mekong Delta (VMD) based on MODIS time-series imagery (Wavelet-based Filter for detecting spatio-temporal changes in Flood Inundation;
WFFI). This methodology involves the use of a wavelet-based filter to interpolate missing information and reduce the noise component in the
time-series data, as proposed in a previous study. The smoothed time profiles of Enhanced Vegetation Index (EVI), Land Surface Water Index
(LSWI), and the Difference Value between EVI and LSWI (DVEL) are obtained from MOD09 8-day composite time-series data (resolution:
500 m; time period: 2000–2005). The proposed algorithm was applied to produce time-series inundation maps (WFFI products) for the five annual
flood seasons over the period from 2000 to 2004. The WFFI products were validated via comparisons with Landsat-derived results and inundation
maps based on RADARSAT images, hydrological data, and digital elevation model data. Compared with the RADARSAT-derived inundation
maps at the province level, the obtained RMSE range from 364 to 443 km2 and the determination coefficients [R2] range from 0.89 to 0.92.
Compared with Landsat-derived results at the 10-km grid level, the obtained RMSE range from 6.8 to 15.2 km2 and the determination coefficients
[R2] range from 0.77 to 0.97. The inundated area of flooded forests/marsh to the northeast of Tonle Sap Lake were underestimated, probably
because of extensive vegetation cover in this area. The spatial characteristics of the estimated start dates, end dates, and duration of inundation
cycles were also determined for the period from 2000 to 2004. There are clear contrasts in the distribution of the estimated end dates and duration
of inundation cycles between large-scale floods (2000–2002) and medium- and small-scale floods (2003 and 2004). At the regional scale, the
estimated start dates for the southern part of An Giang Province during 2003 and 2004 was distinctly later than that for surrounding areas. The
results indicate that these triple-cropping areas enclosed by dikes increased in extent from 2003 to 2004. In contrast, the estimated end dates of
inundation at the Co Do and Song Hau State Farms were clearly earlier than those for surrounding areas, although the estimated start dates were
similar. Temporal changes in the inundation area of Flood pixels in the Dong Thap and Long An Provinces are in excellent agreement with daily
water-level data recorded at Tan Chau Station. The estimated area of Long-term water body increased in size from 2000 to 2004, especially in
coastal areas of the Ca Mau and Bac Lieu Provinces. Statistical data for Vietnam indicate that this trend may reflect the expansion of shrimp-
farming areas. The WFFI products enable an understanding of seasonal and annual changes in the water distribution and environment of the
Cambodia and the VMD from a global viewpoint.
© 2007 Elsevier Inc. All rights reserved.

Keywords: MODIS; WFCP; Time-series analysis; EVI; LSWI

1. Introduction

The annual flooding of the Cambodia and the Vietnamese

Mekong Delta (VMD) is one of the primary factors that
⁎ Corresponding author. characterizes the local ecosystem and human activity in the
E-mail address: [email protected] (T. Sakamoto). region. For generations, local residents have adapted rice
0034-4257/$ - see front matter © 2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.rse.2007.01.011
296 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313

varieties and cropping systems to the seasonal water cycle under 1991). Using this approach, it is possible to determine the
the influence of the monsoon regime. Riparian species are likely spatial distribution of inundated areas and detect seasonal
to have expanded their habitats and diversified in adapting to changes, but the resolution is insufficient to enable comparisons
environmental variation over many years. Annual flooding links with high-resolution land-use data.
isolated water habitats and supplies various freshwater fauna The useful spatial resolution (ca. 1 km) of NOAA/AVHRR
such as fishes and shrimps to paddy fields and ponds. During and SPOT/VEGETATION data means that they are commonly
the rainy season, Tonle Sap Lake and flooded areas along the employed to detect inundation and temporal changes in the
Mekong River feed a high diversity of aquatic species (Zalinge extent of flooded areas (Harris & Mason, 1989; Liu et al., 2002;
et al., 2003); these aquatic fauna are a valuable source of protein Xiao et al., 2002a). The use of an optical sensor means that the
for local people. Annual flooding also fertilizes agricultural land ground surface is commonly partly obscured by cloud coverage.
by spreading fertile sediment and large amounts of pure water as Bryant and Rainey (2002) used NOAA/AVHRR data with low
a natural irrigation system, although extreme flood events cloud coverage to discriminate the surface area of a lake using a
destroy dwellings and take the lives of many people. The water- threshold value of Channel 2. The authors compared seasonal
flow regime of the Mekong River varies annually depending on changes in the surface area of the lake with time-series
rainfall patterns and typhoons. Therefore, understanding the precipitation data. Sheng and Gong (2001) proposed a ratio of
current status of flood inundation in time and space is important Channel 2 to Channel 1 of NOAA/AVHRR data to enable
in evaluating the relationships between variations in the water discrimination between water and land surfaces. They under-
regime, local agricultural activity, and ecosystem behavior from took a spatial and temporal assessment of flooding in the Huaihe
a global viewpoint. River Basin, China, during the summer of 1991. Xiao et al.
Satellite remote sensing is expected to provide powerful (2002b) evaluated the temporal Normalized Difference Water
techniques for objectively detecting inundated areas. Flood Index (NDWIVGT) calculated from near-infrared and short-
detection is one of the classical themes of remote sensing, and wave infrared data acquired by a SPOT/VEGETATION sensor.
many studies have been undertaken in this field for a variety of The authors concluded that NDWIVGT and NDVI temporal
purposes, including studies of paddy fields as crop-production anomalies are efficient in detecting the inundation and trans-
areas, analyses of wetlands as a source of greenhouse gas and a planting of rice.
habitat for aquatic life and birds, and studies of flooded districts. Since Terra/MODIS was launched in December 1999, it has
The approaches used to detect surface water vary widely become possible to monitor continental-scale inundated areas
between the various sensor types used to generate different with a moderate-resolution optical sensor (ca. 250–500 m).
types of satellite images. MODIS data are also distributed freely throughout the world
Synthetic Aperture Radar (SAR) is the most effective sensor (EOS, 2006). In estimating the extent of paddy fields, Xiao et al.
in detecting flooded areas under cloud cover. Satellite images (2005, 2006) identified flooded pixels from the difference
acquired using RADARSAT, JERS-1, ERS-1/2, and ENVISAT between the Land Surface Water Index (LSWI) and Vegetation
have previously been used to detect inundated areas in a variety Indexes (NDVI or EVI). Zhan et al. (2002) applied the Vegetation
of ways (Haruyama & Shida, 2006; Henry et al., 2003; Cover Conversion Algorithm to two MODIS Level 1B datasets
Heremans et al., 2005; Hirose et al., 2001; Ishitsuka et al., 2003; acquired in July and September 2000 and successfully detected
Laugier et al., 1997; Liew et al., 1998; Nguyen & Bui, 2001; inundated areas during the flooding season in Cambodia. The
Wang, 2002, 2004). When using a pointing device and the Dartmouth Flood Observatory (2006) monitors flood disasters all
selectable sensor mode, RADARSAT and ENVISAT enable the over the world using MODIS data. Up to 2005, the observatory
frequent monitoring of the ground surface at a large scale; published an annual inundation map of the Lower Mekong River
however, this is not a feasible approach for the present study Flood via the Internet (Anderson et al., 2005).
because of the limitations of the recurrence period, the per- The aim of the present study is to create a methodology that
formance of the pointing device, and in particular the high cost can be used to detect spatio-temporal changes in the extent of
of data acquisition. flood inundation. This methodology includes a wavelet-based
With the aim of detecting variations in surface water time- filter for the interpolation of missing data and noise reduction
series data at a regional scale, various case studies have adopted within temporal data. Using this methodology, temporal
an alternative analytical approach using NOAA/AVHRR SPOT/ changes in the extent of the inundated region in the Cambodia
Vegetation, SSM/I, and MOS/SMR. Although the spatial and the VMD are assessed at a resolution of 500 m from 2000 to
resolution is relatively poor (NOAA/AVHRR, SPOT/VEGE- 2004.
TATION: ca. 1 km; SSM/I: greater than 13 km; MOS/SMR:
greater than 23 km), the greatest benefits of using these sensors 2. Study area
is the availability of daily observations because of the wide
swath range and the fact that these data are distributed freely via The study area is located in the southern part of Indochina,
the Internet or at a reasonable cost. In terms of SSM/I and MOS/ including parts of Cambodia and Vietnam [Lat. 8.4–13.8 °N,
SMR, time-series variations in flooded areas are observed using Long. 102.9–107.1 °E] (Fig. 1). The Mekong River is the
a single brightness temperature or various indexes composed of largest international river in Southeast Asia, with an extensive
different polarization/frequency brightness temperatures (Jin, catchment area of 795,000 km2. The river flows across six
1999; Tanaka et al., 2000; Tanaka et al., 2003; Tanaka et al., countries from its headwaters to the river mouth [China,
 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313 297

Fig. 1. Location map of the study area. The 24 provinces delineated by red boundaries are used as base units for area comparisons between WFFI and MRC products.
Seven kinds of Land Use/Land Cover were selected with reference to the land use map of the Vietnamese Mekong Delta in 2002. Blue lines represent major rivers and
canals. Regions I and II, as outlined by black rectangles, are further considered in Figs. 11 and 13, respectively.

Myanmar, Laos, Thailand, Cambodia, and Vietnam]. Cambo- 3. Materials

dia and Vietnam are together home to 28% of the total area of
the Mekong River Basin (Mekong River Commission, 2005). 3.1. MODIS/Terra time-series data
The mainstream from the Thailand–Laos border and the Tonle
Sap River from Tonle Sap Lake converge around Phnom Penh MODIS/Terra data are freely distributed through the Earth
City before branching into the Mekong and Bassac Rivers a Observing System Data Gateway (EOS, 2006). The present
short distance downstream. These rivers then branch off into study involves an analysis of MOD09 8-day composite data
nine different directions within the VMD, each flowing into acquired from 2000 (after DOY 57) to 2005 (before DOY 56).
the South China Sea. It is worth noting that excessive water The product name is “MODIS/TERRA SURFACE REFLEC-
volumes flow backward up the Tonle Sap River into Tonle Sap TANCE 8-DAY L3 GLOBAL 500 M SIN GRID V004”. The
Lake during the wet season (Mekong River Commission, resolution of this product is approximately 500 m, and
2005). The area of Tonle Sap Lake during the rainy season is atmospheric correction has already been carried out (Vermote
3–6 times larger than that during the dry season (Zalinge et al., & Vermeulen, 1999). MOD09 composite products yield the best
2003), as water that was stored during the rainy season is surface spectral-reflectance data for each 8-day period with the
discharged into the lower area during the dry season. Tonle Sap least effect of atmospheric water vapor. The observation date is
Lake therefore plays an important role as a natural regulator in also recorded for each pixel.
the hydrological environment of the Mekong Delta. The lake
moderates flood peaks during the rainy season and provides 3.2. Inundation maps produced by the Mekong River Commission
vital water resources during dry season. Most of study area is
classified as savanna climate, with clear seasonal changes Inundation maps produced by the Mekong River Commis-
in precipitation between the dry and rainy seasons under the sion (herein termed MRC products) are used as a reference to
influence of the Asian monsoon. The concentration of evaluate the estimates derived from MODIS data. The MRC
precipitation during the rainy season (May–October) means products used in this study are based on Digital Elevation
that the water level of the Mekong River rises rapidly from Model (DEM) data, hydrological data, and RADARSAT images
May/June, leading to annual flooding along the river. The acquired on 30th August (DOY 242), 23rd September (DOY
coincidence of typhoons with the flood season causes serious 266), and 17th October (DOY 290) 2001 using the ScanSAR
floods within the Mekong Delta. Narrow B Mode (MRC, 2006). Bearing in mind that C-band
298 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313

Table 1 each grid with a resolution of 500 m to conform with the format
Estimated accuracy of WFFI products relative to the inundation area derived of the MODIS-derived results.
from Landsat images
ID Path/row Date (DOY) RMSE (km2) R2 3.4. Daily-averaged water-level data recorded at Tan Chau
1 127/51 2001/11/23 (327) 9.9 0.95 Station
2 127/51 2002/1/10 (10) 7.4 0.96
3 126/51 2001/7/11 (192) 11.0 0.88
Fig. 2 shows a time-series of daily-averaged water-level data
4 126/52 2001/7/11 (192) 12.2 0.77
5 126/53 2001/1/16 (16) 10.8 0.93 recorded at Tan Chau Station [Lat. 10.8 °N, Long 105.2 °E]
6 125/52 2000/11/6 (311) 6.8 0.97 from 2000 to 2004. According to the definition provided by the
7 125/53 2000/11/6 (311) 15.2 0.92 Hydro-meteorological Center in Vietnam, the scale of flooding
8 125/53 2001/12/11 (345) 13.6 0.85 in the VMD is classified into the following three categories on
Total 10.6 0.94
the basis of the peak water level recorded at Tan Chau Station:
large flood, peak water level higher than 4.5 m; moderate flood,
peak water level of 4.0–4.5 m; and small flood, peak water level
microwaves can penetrate cloud cover and easily discriminate of less than 4.0 m (Gupta et al., 2004; Long, 2003). Based on
open water on the basis of backscatter coefficient data at a high this criterion, large-scale flooding occurred in 2000 (peak water
resolution (50 m), it is assumed that the MRC products based on level at Tan Chau Station of 5.0 m), 2001 (4.8 m), and 2002
the RADARSAT images reveal the details of the flood (4.8 m), while moderate flooding occurred in 2004 (4.4 m). In a
distribution at a satisfactory spatial resolution, even under strict sense, the scale of flooding in 2003 (4.0 m) is also
cloud coverage. The inundated areas in the MRC products were categorized as moderate, although the period for which the
aggregated within each grid at 500 m resolution to enable water level exceeded 4.0 m was only 2 days; however, the
comparisons with results derived from the MODIS data. water-flow pattern recorded in 2003 is clearly different from
that in other years, and on this basis the scale of flooding in
3.3. Water-surface areas identified from Landsat images 2003 is classified as small for descriptive purposes in the
present paper.
Landsat images were also used for a quantitative comparison
with MODIS-derived results. Most of the images listed in Table 1 4. Methods
were downloaded without charge from the Global Land Cover
Facility (GLCF, 2006). The transformation to ground reflectance 4.1. Identification of the water surface using EVI and LSWI
data was performed using ENVI: Remote Sensing Software (ITT
Visual Information Solutions, Boulder, USA). The objective areas Short-wave infrared (SWIR) is highly sensitive to moisture
were then classified into three categories using the simple method content in the soil and the vegetation canopy. Various
described below. approaches have been reported that make use of the
1) Non-analysis areas, including cloud cover and cloud spectroscopic characterization of SWIR against water content
shadow, are determined from the blue reflectance value. The (Gao, 1996; Jackson et al., 2004; McFeeters, 1996; Rogers &
cloud-cover areas are derived from the majority analysis Kearney, 2004; Tong et al., 2004). Xiao et al. (2002a) showed
(window size: 3 × 3) for pixels which whose blue reflectance that NDWIVGT in paddy fields exceeds NDVI derived from
(Landsat Band 1) is greater than 0.2. The cloud-shadow areas SPOT/VEGETATION data for the same period of flooding and
are then masked by the cloud-coverage areas including 10-pixel rice-planting in eastern Jiangsu Province, China. Xiao et al.
buffer zones, where are shifted in the opposite direction from (2005, 2006) used anomalies between the Land Surface Water
the sun orientation.
2) Water-surface pixels are identified where the Normalized
Difference Water Index (NDWI; Rogers & Kearney, 2004) is
greater than or equal to 0.8. The equation for the NDWI used in
this study is as follows:

RED − SWIR
NDWI ¼ ð1Þ
RED þ SWIR

where RED is the surface reflectance value (0–1; 0.63–0.69 μm,

ETM + Band 3) and SWIR is in the short-wave infrared band
(1.55–1.75 μm, ETM + Band 5).
3) ‘Not water-surface’ pixels are identified where the NDWI
is less than 0.8.
After resampling the land-surface map to a grid with a
resolution of 25 m using the nearest neighbor method, the areas Fig. 2. Daily water-level data recorded at Tan Chau Station [Lat. 10.8 °N, Long
of these different land-surface types were aggregated within 105.2 °E] from 2000 to 2004.
 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313 299

Index (LSWI) and Vegetation Indexes (NDVI or EVI) in an BLUE; and (459–479 nm, Band 3). SWIR is in the short-wave
algorithm to estimate the distribution of paddy fields in South infrared (1628–1652 nm, Band 6).
China and South and Southeast Asia. In accordance with their Sakamoto et al. (2005, 2006) devised a wavelet-based filter to
approach, a methodology is proposed in the present study to reduce the noise component and interpolate missing information
detect the spatio-temporal flood distribution in the Cambodia within EVI time-series data. They applied the Wavelet-based
and VMD from the smoothed indexes of these differences. The Filter for detecting Crop Phenology (WFCP) methodology to
equations used to derive EVI and LSWI from MODIS data are successfully detect the heading season in Japanese paddy fields.
as follows: They also used the same wavelet-based filter in the VMD to
classify cropping systems (e.g., double-cropping system in the
NIR−RED rainy/dry season, triple-cropping system) and noted regions
EVI ¼ 2:5  ð2Þ where the number of crops per year was increased from two to
NIR þ 6  RED−7:5  BLUE þ 1
three over the interval 2002 to 2003 (wavelet-based filter for
evaluating the spatial distribution of Cropping Systems, WFCS).
In the present study, the same approach is applied to the EVI
NIR − SWIR
LSWI ¼ ð3Þ and LSWI time-series as well as the differences between the two
NIR þ SWIR to reduce the noise component and interpolate missing
information. This was undertaken to determine spatio-temporal
where NIR is the surface reflectance value: 0–1.0 in the near changes in flood inundation within the Cambodia and the VMD
infrared (841–875 nm, Band 2), RED; (621–670 nm, Band 1), during the period 2000 to 2005.

Fig. 3. Flowchart of the workings of the Wavelet-based Filter for detecting the spatio-temporal changes on Flood Inundation (WFFI).
300 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313

4.2. Wavelet-based filter for detecting spatio-temporal changes less than 32 days) considered to be noise. Fig. 4 shows the
in Flood Inundation (WFFI) domain-averaged smoothed data for EVI, LSWI, and the
Difference Value between EVI and LSWI (DVEL) for the seven
A flow-chart of WFFI is shown in Fig. 3. The previous Land Use/Land Cover types. In accordance with the pioneering
algorithms in the WFCP and WFCS are examined on an annual method described by Xiao et al. (2005, 2006), DVEL was used
basis; however, it is possible that the time component in the in the present study to discriminate between Water-related
interpolated data may be overestimated or underestimated to pixels and Non-flood pixels. If the smoothed DVEL is less than
some degree during the year-end period. To address this 0.05, such pixels are determined to be a Water-related pixel;
problem, the algorithm was improved to enable it to calculate however, as shown in Fig. 4, the smoothed DVEL data for the
the 5-year MODIS data continuously. The beginning data was lake and sea are not always less than 0.05. The high DVEL may
set to 26th February 2000 (DOY 57) and the end date to 25th reflect low LSWI caused by extremely low reflectance (near 0)
February 2005 (DOY 56). The 1st of January was not assigned of open water in the NIR band. To solve this problem, the
as the beginning of the MODIS time-series data because following criterion was empirically added to avoid such
MOD09 8-day composite data are unavailable for those days misclassifications. If the smoothed EVI is less than or equal
prior to 25th February 2000 and overflowing water usually to 0.05 and the smoothed LSWI is less than or equal to 0, the
remains in the upper region of the VMD at this time of year. pixel is determined as a Water-related pixel. If the EVI is
In prescribing the time-series data, cloud-covered pixels are greater than 0.3, the pixel is categorized as a Non-flood pixel.
determined based on the Band 3 value. If the blue reflectance is Even if MODIS is superior to NOAA/AVHRR and SPOT/
equal to or greater than 0.2 (Thenkabail et al., 2005; Xiao et al., VEGETATION in terms of spatial resolution, it is inevitable that
2006), such pixels are treated as missing observation data and pixels with a resolution of 500 m come under the influence of
interpolated linearly. As with WFCS (Sakamoto et al., 2006), mixtures of various types of land surfaces, especially when
the present study did not use the view angle data in the data varying the footprint with different view angles. It is concluded
preprocessing. The coiflet (order = 4) was adopted as the mother that the complicated methodology used to identify sub-pixel
wavelet to reduce the high-frequency component (frequency components requires enormous computation time when treating

Fig. 4. Domain-averaged smoothed Indexes [EVI, LSWI, and DVEL] of seven Land Use/Land Cover types for 2002. The locations of the seven Land Use/Land Cover
areas shown in Fig. 1.
 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313 301

a large dataset. There is also no method available to smooth In terms of land-based aquaculture, shrimp farms are widely
vegetation and water indexes. Thus, Water-related pixels were distributed throughout coastal areas. These areas with an extended
divided into two categories (Flood and Mixture) based on the inundation period (as with rivers, lakes, and the sea) should be
following simple method. EVI in homogeneous open-water discriminated from areas subjected to short-term and seasonal
areas such as large lakes and the sea is generally lower than that inundation, as the present study focuses on temporal changes in
in Mixture pixels (mix of water, vegetation, and soil coverage). annual flooding. Therefore, considering that the shrimp-growing
Thus, it is assumed that the smoothed EVI of the Water-related period lasts for approximately 3–4 months and farmers grow
pixels can be used as a criterion for discriminating between shrimps twice per year in coastal areas, these areas are empirically
Flood and Mixture. If the smoothed EVI was less than or defined as follows. If the total number of days as a Water-related
equal to 0.1, the Water-related pixels were defined as Flood pixel is grater than or equal to 250 days per year (as measured
pixels. If the smoothed EVI was greater than 0.1 and less than or from 26th February to 25th February of the following year), the
equal to 0.3, these pixels were defined as Mixture pixels. pixels are classified as Long-term water bodies.
Temporal changes in the areal ratio of Flood and Mixture pixels Finally, this proposed methodology is termed the Wavelet
in the three types of paddy field and shrimp farm (equivalent to based Filter for detecting spatio-temporal changes in Flood
triple-cropped irrigate rice, double-cropped irrigated rice, double- Inundation (WFFI), and the inundation map estimated using
cropped rainfed rice, and Shrimp farm shown in Figs. 1 and 4) are WFFI is termed the WFFI product.
plotted in Fig. 5. As a result, it would be difficult to detect short-
term irrigation water from the smoothed temporal data. The short 5. Evaluation of the proposed methodology
interval between the continuous cropping seasons of triple- and
double-cropped irrigated rice (Fig. 4) means that EVI values are 5.1. Comparison of the inundated regions assessed using the
higher because of the mixture effects of residual plants and adjacent WFFI and MRC products
pixels. In addition, the cusp of the original data over such a short
interval is smoothed to some degree with the noise component in As in the smoothed temporal data of double-cropped irrigated
the same way as that during the wavelet-based filter. rice in the upper region of the VMD (Figs. 4 and 5), the water

Fig. 5. Temporal changes in the areal ratio of Flood and Mixture pixels in the three types of paddy fields and shrimp farms. The locations of these areas (triple-cropped
irrigated rice, double-cropped irrigated rice, double-cropped rainfed rice, and shrimp farm) are shown in Fig. 1.
302 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313

surface [DOY 197–365] is clearly identified using WFFI. The Compared with daily MOD09 data for the Cambodia and
distribution of inundated regions on the 30th August 2001 (DOY VMD (Fig. 6D1, D2, and D3), the influence of cloud cover is
242) determined using the WFFI and MRC products is shown in considerably reduced in MOD09 8-day composite data (Fig. 6C1,
Fig. 6A1 and B1, respectively. Because the MRC products are C2, and C3); however, the unobserved data assigned to the
based on RADARSAT data, the inundated region is clearly irremovable area of cloud-cover is real, and this poses a
identified in detail without any cloud-cover effects; however, substantive problem for the identification of the inundated region
MRC products were unable to cover the entire objective area using the general methodology.
because of the limited swath range of RADARSAT. False-color In the WFFI method, the missing observation data can be
images of MOD09 8-day composite data and daily MOD09 data interpolated via the wavelet-based filter using available values
for the same date are shown in Fig. 6C1 and D1, respectively. The either side of the missing data. This approach enables an
WFFI products, MRC products, and MODIS false-color images understanding of a wider view of the inundated region without
for 23rd September and 17th October 2001 are also shown in the the influence of spatial noise (Fig. 6A1, A2, and A3); however,
same way (Fig. 6A2 and A3 [WFFI products], 6B2 and 6B3 [MRC it should be noted that the estimated timing of the start and end
products], 6C2 and 6C3 [MODIS 8-day composite false-color of flood inundation might have a margin of error that is
images], and 6D2 and 6D3 [MODIS daily false-color images]). dependent on observation conditions. This situation occurs

Fig. 6. Spatial comparison of WFFI products, MRC products, and MODIS 8-day composite false-color and daily false-color images for the study area, as shown in Fig. 1.
(A1, 2, 3) WFFI products. (B1, 2, 3) MRC products. (C1, 2, 3) MODIS 8-day composite false-color images. (D1, 2, 3) MODIS daily false-color images. R:G:B = Band 6:
Band 2: Band 1.
 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313 303

because the interpolation methodology will never be able to

address the large number of missing values resulting from
conditions of continuous cloud-cover.
As with the three subsequent MRC products (Fig. 6B1, B2,
and B3), the WFFI products (Fig. 6A1, A2, and A3) simul-
taneously revealed the receding and expanding flood areas over
a wide area (e.g., the receding inundated region around the
northern part of Kampong Cham and the expanding region of
inundation in the VMD). The distribution of Flood pixels (blue
region) in WFFI products is consistent in principle with the area
where the inundation ratio is 100% (red region) in MRC
products, with the exception of the northwest of Tonle Sap Lake.
The spatial pattern of the Mixture pixels (green region) is
qualitatively similar to the region where the inundation ratio is
less than 100% and moderate (blue to yellow region) in MRC
products. Small-scale inundated areas along small tributaries in
Fig. 7. Comparison of inundated areas predicted by WFFI and MRC products at
Cambodia were only partly identified in the WFFI products, and
the province level. Target provinces and regions are delineated by red lines in
the WFFI overestimated the extent of inundated regions around Fig. 1.
the Camau Peninsula in Vietnam. These misidentifications may
result from mixtures with adjacent pixels probably due to view
angles. Thus, when the Mixture pixels are used directly for
measurement of the inundated area, the estimation contains MRC products and administrative boundary data for Vietnam
errors related to location. and Cambodia (Fig. 1). When estimating the inundated area in
each region, the result was calculated from the total number of
5.2. Flooded forests to the northwest of Tonle Sap Lake Water-related pixels. Both Flood pixels and Mixture pixels
were counted in equal number without weighing. In terms of the
Areas of flooded forests/marsh are widely distributed to the accuracy of the WFFI products compared to MRC products,
northwest of Tonle Sap Lake (Giri et al., 2000; McKenney & low RMSE and a high determination coefficient (n = 24) were
Tola, 2002; Xiao et al., 2005). Fujii (2004) attempted to detect obtained for the 30th August, 23rd September, and 17th October
the inundated area in this region from RADARSAT data; 2001. The values for these dates are 364 km2 (RMSE) and 0.91
however, it proved difficult to detect the water surface under (R2), 416 km2 (RMSE) and 0.89 (R2), and 443 km2 (RMSE)
flooded forests because of the high backscatter efficiency of C- and 0.92 (R2), respectively. Combining these estimates (n = 72)
band radar. The author was able to conclusively determine for the three different dates, the RMSE and the determination
inundated areas under forests and regions of high vegetation coefficient are 409 km2 (RMSE) and 0.90 (R2), respectively. As
coverage using a combination of water-level data together with shown in Fig. 7, the overall inundated area estimated by WFFI
DEM data. In terms of MRC products, the inundated region to products at the province level tends to be overestimated
the northwest of Tonle Sap Lake is broadly uniform despite the compared to estimates derived from MRC products. There are
presence of vegetation coverage. Thus, these areas may also be many Mixture pixels, especially in the Kien Giang, Siem Reap,
spatially interpolated using hydrological data in combination Ca Mau, and Bac Lieu Provinces. It is clear that the inundated
with DEM data. It is even more difficult to detect these inundated areas in these provinces were markedly overestimated. In
regions under high-vegetation coverage using lower-resolution contrast, the inundated areas for August and September in
optical sensors such as MODIS. Therefore, it would be difficult Battambang Province were considerably underestimated. These
to accurately identify the inundated area of flooded forests/ errors might reflect underestimates in the area around the
marsh in the WFFI products (Fig. 6A and C). The inundated area flooded forests/marsh as explained in the previous section. It is
of flooded forests/marsh was progressively identified as the possible to enhance the estimation accuracy using DEM data or
water level of Tonle Sap Lake increased (Fig. 6A1, A2, and A3) higher-resolution data (Brivio et al., 2002; Wang et al., 2002),
because plant bodies are gradually submerged as the water level and this remains an issue for future studies. Although there is a
rises; this leads to an increase in the area ratio of the water- fundamental overestimation problem resulting from mixed-
surface area to vegetation coverage. For this reason, there may a pixel effects, it can be assumed that temporal WFFI products
time lag involved in identifying inundation in flooded forests/ provide useful criteria for determining annual inundations that
marsh areas. are early or late relative to long-term averages.

5.3. Comparison with MRC products at the province scale 5.4. Comparison with Landsat-derived results at the 10-km
grid level
Fig. 7 shows a comparison of the estimated inundated area at
the province scale obtained from WFFI and MRC products. The The distribution of the water-surface area derived from
24 objective regions were defined based on the coverage area of Landsat images is shown in Fig. 8. In comparing these areas
304 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313

6. Findings obtained from temporal WFFI products

6.1. Spatio-temporal changes in the extent of annual floods in

the Cambodia and VMD from 2000 to 2005

Overlaying multiple images of the inundated region is

effective in detecting the maximum extent of the flooded area
(Sheng & Gong, 2001). Thus, the WFFI products are expected
to be useful in identifying temporal changes in the simultaneous
recession and expansion of flood cycles at a regional scale.
Maximum flood maps estimated from the time-series WFFI
product are shown in Fig. 10. Blue, green, and white colors
represent areas of Flood, Mixture, and Long-term water bodies,
respectively. Classification of the scale of flooding in the VMD
according to the peak water level recorded at Tan Chau Station
(Fig. 2) is in good agreement with the area of Flood pixels
shown in Fig. 10. The area of Flood pixels in 2000 is the largest
among the years 2000 to 2004, while that in 2003 is the
smallest. In additionally, Fig. 11 shows seasonal changes in the
extent of the estimated area of Flood pixels within Dong Thap
and Long An Provinces. The temporal patterns of these
inundated areas from 2000 to 2004 are very similar to the
water level data recorded at Tan Chau Station (Fig. 2). During
the period from July to August [DOY 182–243], the beginning
of the inundation cycle was earliest in 2000, followed by 2002,
2001, 2004, and 2003 in order of the total inundated area. In
addition, the order of the maximum extent of inundated areas is
generally similar to the order of the peak water level, except for
a reverse order between 2001 and 2002. During the period from
December to January [DOY 335–31], the diminution rate of the
Fig. 8. Distribution of the water-surface area derived from Landsat images. The
inundated area is stable, as is the water-reducing ratio. It is clear
colors; blue, green, and red indicate the areas of water-surface, not water-
surface, and non-analysis respectively. The numbers of the Landsat images are that the start dates of the inundation cycles and the scale of
the same as the ID numbers listed in Table 1. flooding in the VMD vary with the amount of inflow water from
the upper regions and the start date of the annual flooding
season.
Fig. 12 shows the estimated flood area for the 1st day of
with the area of Water-related pixels in WFFI products at the every month (June [DOY 152] to January [DOY 1] of the
10-km grid level (Fig. 9), the estimation accuracy of WFFI
products is shown in Table. 1. The RMSE vary from 6.8 to
15.2 km2, and the determination coefficients [R2] vary from
0.77 to 0.97. In grids where the inundated area derived from
Landsat images is greater than 50 km2, the WFFI products
yielded an overestimate (as in the comparison with MRC
products) because of an over-evaluation of Mixture pixels. In
grids where the inundation area is less than 50 km2, the
inundated area derived from WFFI products is underestimated.
This problem occurs because the neighbor-pixel effect makes
it difficult to detect small water-surface areas from low-
resolution images. Specifically, two Landsat images (ID: 3 and
4, path/row: 126/51 and 126/52) were acquired at the
beginning of the rainy season (July 11 2001). These numerous
small water-surface areas led to an overall underestimation of
the WFFI products compared with the Landsat-derived results.
These results, in combination with the results from the
comparison with MRC products, indicate that the WFFI
products can be used to roughly measure the flood extent at Fig. 9. Comparison of the inundated area at a 10-km grid level between the
regional scale. Landsat-derived results and the WFFI products.
 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313 305

Fig. 10. Maximum estimated extent of the floodplain over the period 2000 to 2004.

following year) from 2000 to 2005. Fig. 13 shows the extent of intruded into the southern Svay Rieng Province of Cambodia
inundated regions for every tenth day from 1st June to 17th from the northern Long An Province in Vietnam (Fig. 13).
January of the following year. Because of space limitations, the
analysis focused on the region around the Vietnam–Cambodia 6.2. Temporal characteristics of annual floods: Start, end, and
border, including Dong Thap and Long An Provinces (Region I duration
in Fig. 1). As is generally known, the results show that the flood
started following a rapid increase in water level from July to The continuous flood period is determined according to the
August [DOY 182–243] (Figs. 2 and 12) and reached its largest following procedures. 1) The continuous inundation period
extent across the VMD during October and November [DOY recorded by Water-related pixels (both Flood and Mixture
274–334] (Haruyama & Shida, 2006; Wassmann et al., 2004).
Upon visualization at moderate spatial resolution (500 m) and
high temporal resolution (10 days) (Fig. 13), it is apparent that
the spatio-temporal distribution of the inundated area varies
from year to year. Temporal changes in the area of inundation
within Cambodia in 2002 (Fig. 13) are similar to the estimated
progression of flooding based on RADARSAT data (Fujii et al.,
2003).
The flood that inundates the Dong Thap Muoi (Plain of
Reeds) is mainly caused by floodwater that flows across the
Vietnam–Cambodia border. The proportion of floodwater
sourced from across the border is estimated to be 88%, with
just 12% from the main river and canals (Long, 2003). Fig. 13
shows the same phenomenon, whereby floodwater is directly
sourced not only from the main rivers, but also from the stream
of overflowing water that passes through Cambodia. As
described by Kazama (2001), the results of the present study Fig. 11. Seasonal changes in the extent of the estimated area of Flood pixels
indicate that expanding floodwater within the VMD in 2000 within Dong Thap and Long An Provinces.
306 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313

Fig. 12. Spatial distribution of inundated areas on the first day of every month from June to January of the following year. Areas of blue indicate the distribution of
Flood pixels and areas of green indicate Mixture pixels.
 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313 307

Fig. 13. Temporal changes in flood inundation around the Cambodia–Vietnam border during flood seasons over the period: 2000–2004 (enlargement of Region I in
Fig. 1). Areas of blue indicate Flood pixels, while areas of green indicate Mixture pixels.
308 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313

Fig. 13 (continued ).
 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313 309

Fig. 14. Spatial distribution of the start date, end date, and duration of inundation cycles over the period from 2000 to 2004.

Fig. 15. Start date, end date, and duration of inundation cycles within An Giang and Can Tho Provinces (enlargement of Region II in Fig. 1.). The colors used for the
start date, end date, and duration are the same as those used in Fig. 14.
310 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313

pixels) is calculated from 5-year Water-related/Non-Flood data these triple-cropping regions protected by an embankment
for each pixel. In this case, the continuous inundation periods expanded in area from 2003 to 2004 (intensification and
include irrigation water in paddy fields. 2) If periods of multiplication of paddy crops/upland crops). In contrast, the end
continuous inundation are identified twice or more in a year, the date of the same triple-cropping area is similar to that in
longer period is taken as the period of annual flood inundation. surrounding areas. In the flood-protected areas of southern An
3) The duration of annual flood inundation is defined from the Giang province, at the end of the flooding season (from
start and end dates of annual flood inundation. Fig. 14 shows November) when the flood water is gradually receding and the
estimates of the start dates, end dates, and duration of third of the annual paddy crops (autumn–winter crop) has
inundation cycles from 2000 to 2004. already been harvested, local farmers open their embankment
The spatial distribution of the start dates varies year by year. systems to take the remainder of the flood water into the fields.
The start dates of the largest flood (2000) are mostly earlier than The objectives of this practice are: (1) to flush pollution and
those of the medium-scale flood in 2004 (Fig. 14). For 2001 and pests (e.g., agro-chemicals, insects, and rats) and provide
2002, the start dates along the Bassac and Mekong rivers are natural fertilizer via sedimentation (Long, 2003), and (2) to
similar to those for 2004 (see Fig. 14); however, in the marginal maintain a volume of water for the preparation of land or
region of the floodplain, especially the coastal area near the Gulf sowing of the winter-spring paddy crop (starting from the end of
of Thailand, the start dates for 2001 and 2002 are much earlier November), as this practice reduces the irrigation cost in the dry
than those for 2004 (Figs. 14 and 13 [DOY 242–262]). As for season.
the smallest flood (2003), the start dates along the two main The inundation season on the Co Do and Song Hau State
rivers shows similar patterns to those for 2004 (Fig. 14); Farms located in Can Tho Province (Fig. 1; southeast parts of
however, when focusing on the Flood pixels in Fig. 13 [DOY the windows in Fig. 15) shows distinctive features observed in
202–232], the timing of the appearance of Flood pixels in 2003 the southern part of An Giang Province, as described above.
is obviously later than that in other years. In additional, the start Although the start date for both state farms is similar to that for
dates of flooding around the border between Syey Rieng and surrounding areas, the end dates are clearly earlier. The farming
Long An Provinces in 2003 were clearly later than those in other systems of both sate farms are different from those of sur-
years (Fig. 14). This observation might reflect the fact that rounding areas, which are dominated by mono-paddy cultiva-
sufficient overflowing water failed to arrive in the northern tion. The diversified systems of crops (paddy and upland crops)
marginal region of Long An Province (Vietnam) and that new and the integrated system (paddy and fish/prawn) are used in
embankment and canals were constructed at the province both state farms. The reason for the earlier end dates in both
borders in 2003. state farms is that drainage pumping is performed early for the
These early and late start dates of annual flooding appear to seeding of upland crops (cash crops in the dry season) and
be in good agreement with the dates that the water level preparation for harvesting the fish cultured within paddy fields.
increased (Fig. 2). The water level recorded in 2000 started to
increase earlier in the season than that recorded in other years. 6.3. Shrimp farming
The flood damage during 2000 was the most severe in over
70 years, and the seasonal flood hydrograph began 4–6 weeks The estimated area of Long-term water bodies increased
early (Mekong River Commission, 2005). Although an arbitrary from 2000 to 2004, especially in the Ca Mau and Bac Lieu
measure, the first date that the water level exceeded the given
level of 300 cm was earliest in 2000, followed by 2002, 2001,
2004, and 2003 (Fig. 2).
There is a clear contrast in the estimated end dates of inundation
cycles between large-scale floods (2000, 2001, and 2002) and
small- to medium-scale floods (2003 and 2004) (Fig. 14): the end
dates for large floods are much later than those for the medium- to
small-scale floods. It is also possible to determine the approximate
differences in flood scales from the duration map. It is assumed
that this reflects differences in the natural discharge period that are
related to the annual flood intensity.
A more detailed analysis of the estimates of the timing of
inundation cycles (Fig. 15) reveals that the estimated start dates
for the years 2003 and 2004 in the southern part of An Giang
Province (the central parts of each window in Fig. 15) were
distinctly later than those in surrounding areas. A comparison of
seasonal EVI patterns in 2002 and 2003 suggests that harvesting
in the early part of the flood season was possible for the first
Fig. 16. Total area planted in rice, and Area of water surface used for aquaculture
time in 2003, enabling a change in the cropping system in this and the production of farmed shrimp in Ca Mau and Bac Lieu Provinces from
area from double- to triple-cropping (Sakamoto et al., 2006). 2000 to 2004. Sources: Bac Lieu Statistical Office (2005), Ca Mau Statistical
New findings that arose from this study include the fact that Office (2005), and General Statistics Office (2004).
 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313 311

Provinces (Figs. 1 and 10). This trend may be closely linked to integrated and collaborative approach that considers the river
rapid land-use changes related to growth of the Vietnamese basin as a whole (Trang, 2005). With the aim of changing the
economy. Fig. 16 shows statistical data on the total rice-planted mindsets of policymakers in this regard, the WFFI products may
area, the area of water surface for aquaculture, and the be the most effective information to put forward. The WFFI
production figures for shrimp breeding in both provinces. The products enable us to visualize and understand a spatial and
total rice-planted area tends to decrease over time, while the temporal overview of flood dynamics on the wide floodplain that
area of water surface for aquaculture and shrimp production spans political boundaries. As with satellite remote sensing,
increased from 2000 to 2004. Because the coastal area has thematic maps and animation movies produced from satellite
geographical advantages in terms of brackish-water aquaculture images do not require specialized knowledge to understand
and its good cash income, local farmers are easily able to shift environmental change from a global viewpoint. It is considered
from rice cultivation to aquaculture, especially shrimp farming likely that anybody, even those who are not scientists or
(Binh et al., 2005). There is also a remarkable change in the administrative officers, can understand that the management and
extent of Long-term water bodies estimated from 2000 to 2004 control of water resources is not a private matter to be considered
(Fig. 10). While the estimated extent of Long-term water bodies solely at the province or national level. In considering climate
increased each year, the area of water surface used for change associated with global warming, the focus of interna-
aquaculture and the production of farmed shrimp also increased tional corporations and global attention to changes in the water-
over the same period, especially during 2000 and 2001 (Figs. 10 resource cycle will become increasingly important for the
and 16). According to estimated changes in Land Use/Land coexistence of riparian countries in the near future. It is hoped
Cover within the Cai Nuoc district in Ca Mau Province (Binh that the methodology and WFFI products proposed in the
et al., 2005), large amounts of agricultural land were converted present paper will contribute to consensus-building in terms of
to shrimp farming from 1997 to 2003. This is directly related to a water-resource management and the maintenance of wetland
policy decision of the Vietnamese Government in 2000. ecosystems at an international level.
According to the government No.09/NQ-CP (Binh et al.,
2005), where cultivation is impossible because of environmental Acknowledgements
constraints (e.g., soils, salinity, and acidity), paddy fields can be
converted to other farming systems that provide increased We offer our special thanks to Dr. Takao Masumoto of the
benefits to local farmers. In the case of the delta coast, most of National Institute for Rural Engineering for providing Inunda-
the paddy fields along the coast can only be used for cultivation tion maps produced by the MRC and for his valuable
in the rainy season because of the high risks imposed by physical suggestions. We also thank Miss Trinh Thi Long of the
conditions (e.g., typhoon, drought, salinity intrusion, and soil Southern Institute for Water Resources Research for providing
acidity). This restriction leads to reduced benefits and hardship water level data and conducting the field survey in the Mekong
for the farmers. Shrimp farming is an attractive alternative, Delta. We are grateful to three anonymous reviewers for their
because of the high price commanded by exported shrimp. valuable comments and suggestions. This study was mainly
funded by the Ministry of Agriculture, Forestry and Fisheries,
7. Conclusions Japan. The authors (A.K & M.Y) were partially supported by
the GERF (S-4) of the Ministry of the Environment, Japan.
In this study, the methodology termed the Wavelet-based Filter
for detecting spatio-temporal changes in Flood Inundation (WFFI) References
was devised to use in assessing seasonal and annual changes in
inundation cycles from time-series MODIS data. The WFFI Anderson, E., Brakenridge, G. R., & Caquard, S. (2005). Dartmouth Atlas of
products are in good agreement with the water-surface area Global Flood Hazard: E100N20, Dartmouth Flood Observatory, Hanover,
derived from Landsat images at a 10-km grid level and three USA. Available at. http://www.dartmouth.edu/~floods/hydrography/
inundation maps based on RADARSAT data, hydrological data, E100N20.html
and digital elevation model data in terms of temporal changes and Bac Lieu Statistical Office (2005). Statistical yearbook Bac Lieu Province 2004.
Bac Lieu: Bac Lieu Statistical Office.
spatial distribution. Although the inundated area estimated by Binh, T. N. K. D., Vromant, N., Nguyen, T. H., Hens, L., & Boon, E. K. (2005).
WFFI has a margin of error associated with the low resolution of Land cover changes between 1968 and 2003 in Cai Nuoc, Ca Mau Penisula,
the sensor, the estimates show a strong correlation with the Vietnam. Environment, Development and Sustainability, 7, 519−536.
inundated area derived from MRC products [R2: 0.89–0.92] and Brivio, P. A., Colombo, R., Maggi, M., & Tomasoni, R. (2002). Integration of
Landsat images [R2: 0.77–0.97]. The WFFI products derived from remote sensing data and GIS for accurate mapping of flooded areas. Inter-
national Journal of Remote Sensing, 23, 429−441.
MODIS 500 m imagery shows the ability to study flood dynamics Bryant, R. G., & Rainey, M. P. (2002). Investigation of flood inundation on
and performs similar to Radarsat based flood assessments. playas within the Zone of Chotts, using a time-series of AVHRR. Remote
Considering that MODIS products have a great advantage in the Sensing of Environment, 82, 360−375.
high-frequent observation, we conclude that the WFFI is a useful Ca Mau Statistical Office. (2005). Statistical yearbook Ca Mau Province 2004.
method to clarify the entire extent of the temporal Mekong flood. Ca Mau, Ca Mau Statistical Office.
Dartmouth Flood Observatory. (2006). Dartmouth atlas of global flood hazard.
In Vietnam, collaboration between provinces is sometimes Available at. http://www.dartmouth.edu/~floods/index.html
weak in terms of water-resource management, and such manage- EOS. (2006). NASA earth observing system data gateway. Available at. http://
ment is conducted according to local interests rather than an edcimswww.cr.usgs.gov/pub/imswelcome/
312 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313

Fujii, H. (2004). Hydrological survey and water balance of the Cambodian McKenney, B., & Tola, P. (2002). Natural resources and rural livelihoods in
floodplain in the Mekong River (in Japanese and English Abstract). Cambodia: A baseline assessment — Working paper. Cambodia Develop-
Technical Report of the National Institute For Rural Engineering, 202, ment Resource Institute in Phnom Penh, Cambodia, vol. 23.
127−140. McFeeters, S. K. (1996). The use of the Normalized Difference Water Index
Fujii, H., Garsdal, H., Ward, P., Ishii, M., Morishita, K., & Boivin, T. (2003). (NDWI) in the delineation of open water features. International Journal of
Hydrological roles of the Cambodian floodplain of the Mekong River. Remote Sensing, 17, 1425−1432.
International Journal of River Basin Management, 1, 1−14. Mekong River Commission. (2005). Overview of the hydrology of the Mekong
Gao, B. C. (1996). DWI-A normalized difference water index for remote sensing Basin. Vientiane, November 2005. 73pp.
of vegetation liquid water from space. Remote Sensing of Environment, 58, MRC. (2006). Mekong River commission. Available at. http://www.mrcmekong.org/
257−266. Nguyen, T. L., & Bui, D. T. (2001). Flood monitoring of Mekong River delta,
General Statistics Office (2004). Statistical yearbook 2003. Hanoi, Vietnam: Vietnam using ERS SAR data. Proc. 22nd Asian Conference on Remote
Statistical Publishing House. Sensing, November 5–9 2001, Singapore.
Giri, C., Moe, T. A., Shrestha, S., & Aschbacher, J. (2000). Multi-temporal Rogers, A. S., & Kearney, M. S. (2004). Reducing signature variability in
analysis of the VEGETATION data for land cover assessment monitoring in unmixing coastal marsh Thematic Mapper scenes using spectral indices.
Indochina. Vegetation-2000, Lake Maggiore, Italy, April 3–6 2000. International Journal of Remote Sensing, 20, 2317−2335.
GLCF. (2006). Global land cover facility. Available at. http://www.landcover.org Sakamoto, T., Yokozawa, M., Toritani, H., Shibayama, M., Ishitsuka, N., &
Gupta, A. D., Babel, M. S., & Ngoc, P. (2004). Flood assessment in the Mekong Ohno, H. (2005). A crop phenology detection method using time-series
Delta, Vietnam. Proceedings of the Asia Pacific Association of Hydrology MODIS data. Remote Sensing of Environment, 96, 366−374.
and Water Resources (APHW), 2nd Conference held at the Suntec Sakamoto, T., Nguyen, N. V., Ohno, H., Ishitsuka, N., & Yokozawa, M. (2006).
International Convention and Exhibition Center. Available at: http://www. Spatio-temporal distribution of rice phenology and cropping systems in the
wrrc.dpri.kyoto-u.ac.jp/~aphw/APHW2004/APHW2004.html Mekong Delta with special reference to the seasonal water flow of the
Harris, A. R., & Mason, I. M. (1989). Lake area measurement using AVHRR. Mekong and Bassac rivers. Remote Sensing of Environment, 100, 1−16.
International Journal of Remote Sensing, 10, 885−895. Sheng, Y., & Gong, P. (2001). Quantitative dynamic flood monitoring with
Haruyama, S., & Shida, K. (2006). Flood risk evaluation of the Mekong River NOAA AVHRR. International Journal of Remote Sensing, 22, 1709−1724.
Delta utilizing JERS-1 SAR images (in Japanese with English abstract). Tanaka, S., Sugimura, T., & Tanaka, M. (1991). Observation of flooded area in
Journal of Geography, 115, 72−86. Bangladesh by MOS/MSR on an all-weather condition (in Japanese with
Henry, J. B., Chastanet, P., Fellah, K., & Desons, Y. L. (2003). ENVISAT multi- English Abstract). Remote Sensing Society of Japan, 11, 77−85.
polarised ASAR data for flood mapping. Proceedings of IGRASS'03 France: Tanaka, M., Sugimura, T., & Tanaka, S. (2000). Monitoring water surface ratio
Toulouse. in the Chinese floods of summer 1998 by DMSP-SMM/I. International
Heremans, R., Willekens, A., Borghys, D., Verbeeck, B., Valckenborgh, J., Journal of Remote Sensing, 21, 1561−1569.
Acheroy, M., et al. (2005). Automatic detection of flooded areas on ENVISAT/ Tanaka, M., Sugimura, T., Tanaka, S., & Tamai, N. (2003). Flood-drought cycle
ASAR images using an objected-oriented classification technique and active of Tonle Sap and Mekong Delta area observed by DMSP-SSM/I. Interna-
contour algorithm. Proceedings of the 31st International Symposium on tional Journal of Remote Sensing, 24, 1487−1504.
Remote Sensing of the Environment. June 20–24, 2005 St. Petersburg, Russia. Thenkabail, P. S., Schull, M., & Turral, H. (2005). Ganges and Indus river basin
Hirose, K., Maruyama, Y., Quy, D. V., Tsukada, M., & Shiokawa, Y. (2001). land use/land cover (LULC) and irrigated area mapping using continuous
Visualization of flood monitoring in the lower reaches of the Mekong River. streams of MODIS data. Remote Sensing of Environment, 95, 317−341.
22nd Asian Conference on Remote Sensing. November 5–9 2001, Singapore, Tong, P. H. S., Auda, Y., Populus, J., Aizpuru, M., Habshi, A. A., & Blasco, F.
vol. 1. ( pp. 314−319). (2004). Assessment from space of mangroves evolution in the Mekong
Ishitsuka, N., Saito, G., Murakami, T., Ogawa, S., & Okamoto, K. (2003). Delta, in relation to extensive shrimp farming. International Journal of
Methodology development for area determination of rice planted paddy Remote Sensing, 25, 4795−4812.
using RADARSAT data. Journal of the Remote Sensing Society of Japan, Trang, T. Q. (2005). Water resources management in Viet Nam. A cross-disciplinary
23, 458−472. workshop on the many aspects of Water in Mainland Southeast Asia. Center for
ITT Visual Information Solutions. http://www.ittvis.com/index.asp Khmer Studies, Siem Reap, Cambodia. November 30–December 2, 2005.
Jackson, T. J., Chen, D., Cosh, M., Li, F., Anderson, M., Walthall, C., et al. Available at. http://www.khmerstudies.org/events/Water/water.htm
(2004). Vegetation water content mapping using Landsat data derived Vermote, E. F., & Vermeulen, A. (1999). MODIS algorithm technical
normalized difference water index for corn and soybeans. Remote Sensing of background document, atmospheric correction algorithm: spectral reflec-
Environment, 92, 475−482. tances(MOD09). NASA contract NAS5-96062.
Jin, Y. -Q. (1999). A flooding index and its regional threshold value for Wang, Y. (2002). Mapping the extent of a flood: What we have learned and how
monitoring floods in China from SSM/I data. International Journal of we could do better. Natural Hazard Review, 3, 68−73.
Remote Sensing, 20, 1025−1030. Wang, Y. (2004). Seasonal change in the extent of inundation on floodplains
Kazama, S. (2001). Report on flood and inundation in the Mekong River Basin detected by JERS-1 Synthetic Aperture Radar data. International Journal of
in 2000 (in Japanese with English abstract). Japan Society for Natural Remote Sensing, 25, 2497−2508.
Disaster Science, 19, 493−500. Wang, Y., Colby, J. D., & Mulcahy, K. A. (2002). An efficient method for
Laugier, O., Fellah, K., Tholey, N., Meyer, C., & De Fraipont, P. (1997). High mapping flood extent in a coastal floodplain using Landsat TM and DEM
temporal detection and monitoring of flood zone dynamic using ERS data data. International Journal of Remote Sensing, 23, 3681−3691.
around catastrophic natural events: The 1993 and 1994 Camargue Flood events. Wassmann, R., Nguyen, X. H., Hoanh, C. T., & Tuong, T. P. (2004). Sea level
Proceedings of the third ERS Symposium, ESA SP-414, vol. 1. ( pp. 559−564). rise affecting the Vietnamese Mekong Delta: Water elevation in the flood
Liew, S. C., Chen, P., Kam, S. P., Tuong, T. P., Minh, V. Q., & Lim, H. (1998). season and implications for rice production. Climatic Change, 66, 89−107.
Rice crops monitoring in the Mekong River Delta using combined ERS and Xiao, X., Boles, S., Frolking, S., Salas, W., Moore III, B., & Li, C. (2002a).
RADARSAT synthetic aperture radar. Geoscience and Remote Sensing Observation of flooding and rice transplanting of paddy rice fields at the site
Symposium Proceedings, 1998. IGARSS '98, vol. 5. ( pp. 2746−2748). to landscape scales in China using VEGETATION sensor data. International
Liu, Z., Huang, F., Li, L., & Wan, E. (2002). Dynamic monitoring and damage Journal of Remote Sensing, 23, 3009−3022.
evaluation of flood in north-west Jilin with remote sensing. International Xiao, X., Boles, S., Frolking, S., Salas, W., Moore III, B., & Li, C., et al. (2002b).
Journal of Remote Sensing, 23, 3669−3679. Landscape-scale characterization of cropland in China using vegetation and
Long, T. T. (2003). Impacts of flow regimes on environmental conditions in the Landsat TM images. International Journal of Remote Sensing, 23, 3579−3594.
Mekong Delta of Vietnam. Integrating Environmental Impacts into Water Xiao, X., Boles, S., Liu, J., Zuang, D., Frolking, S., Li, C., et al. (2005).
Allocation Models of the Mekong River Basin, Workshop in Ho Chi Min City, Mapping paddy rice agriculture in southern China using multi-temporal
Vietnam. MODIS images. Remote Sensing of Environment, 95, 480−492.
 T. Sakamoto et al. / Remote Sensing of Environment 109 (2007) 295–313 313

Xiao, X., Boles, S., Frolking, S., Li, C., Bau, J. Y., Salas, W., et al. (2006). Symposium on the Management of Large Rivers for Fisheries. February 11–14
Mapping paddy rice agriculture in South and Southeast Asia using multi- 2003, Phnom Penh Cambodia (pp. 335−357).
temporal MODIS images. Remote Sensing of Environment, 100, 95−113. Zhan, X., Sohlberg, R. A., Townshend, J. R. G., DiMiceli, C., Carroll, M. L.,
Zalinge, N. V., Degen, P., Pongsri, C., Nuov, S., Jensen, J. G., & Nguyen, V. H. Eastman, J. C., et al. (2002). Detection of land cover change using MODIS
(2003). The Mekong River system. Proceedings of the Second International 250 m data. Remote Sensing of Environment, 83, 336−350.