Journal of Environmental Management
Journal of Environmental Management
Journal of Environmental Management
Quantifying changes in flooding and habitats in the Tonle Sap Lake (Cambodia)
caused by water infrastructure development and climate change in the Mekong
Basin
Mauricio E. Arias a, Thomas A. Cochrane a, *, Thanapon Piman a, Matti Kummu b, Brian S. Caruso a,
Timothy J. Killeen c
a
Department of Civil and Natural Resources Engineering, University of Canterbury, Private Bag 4800, Christchurch 8140, New Zealand
b
Water & Development Research Group, Aalto University, P.O. Box 15200, 00076 Aalto, Finland
c
Carbon and Commodities Program, World Wildlife Fund, 1250 24th Street, N.W. P.O. Box 97180, Washington, DC 20090-7180, USA
a r t i c l e i n f o a b s t r a c t
Article history: The economic value of the Tonle Sap Lake Floodplain to Cambodia is arguably among the highest
Received 24 February 2012 provided to a nation by a single ecosystem around the world. Nonetheless, the Mekong River Basin is
Received in revised form changing rapidly due to accelerating water infrastructure development (hydropower, irrigation, flood
28 June 2012
control, and water supply) and climate change, bringing considerable modifications to the flood pulse of
Accepted 5 July 2012
Available online 9 August 2012
the Tonle Sap Lake in the foreseeable future. This paper presents research conducted to determine how
the historical flooding regime, together with human action, influenced landscape patterns of habitats in
the Tonle Sap Lake, and how these habitats might shift as a result of hydrological changes. Maps of water
Keywords:
Habitat modeling
depth, annual flood duration, and flood frequency were created for recent historical hydrological
Hydro-ecology conditions and for simulated future scenarios of water infrastructure development and climate change.
Tonle Sap Lake Relationships were then established between the historical flood maps and land cover, and these were
Mekong River Basin subsequently applied to assess potential changes to habitat cover in future decades. Five habitat groups
Flood pulse were clearly distinguishable based on flood regime, physiognomic patterns, and human activity: (1)
Tropical floodplain Open water, flooded for 12 months in an average hydrological year; (2) Gallery forest, with flood duration
GIS of 9 months annually; (3) Seasonally flooded habitats, flooded 5e8 months and dominated by shrublands
Natural resources management
and grasslands; (4) transitional habitats, flooded 1e5 months and dominated by abandoned agricultural
fields, receding rice/floating rice, and lowland grasslands; and (5) Rainfed habitats, flooded up to 1
month and consisting mainly of wet season rice fields and village crops. It was found that water infra-
structure development could increase the area of open water (þ18 to þ21%) and the area of rainfed
habitats (þ10 to þ14%), while reducing the area covered with seasonally flooded habitats (13 to 22%)
and gallery forest (75 to 83%). Habitat cover shifts as a result of climate change include a net increase
of open water (2e21%), as well as a reduction of rainfed habitats by 2e5% and seasonally flooded habitats
by 5e11%. Findings from this study will help guide on-going and future conservation and restoration
efforts throughout this unique and critical ecosystem.
Ó 2012 Elsevier Ltd. All rights reserved.
1. Introduction River Basin (MRB), with a floodplain that extends over 15,000 km2
and stores 50e80 km3 of water from the Mekong River that enters
The Tonle Sap Lake has been traditionally regarded as “the Heart the ecosystem during the wet monsoon season (MRC, 2005). The
of Cambodia”, an expression that goes far beyond its location. The annual flood pulse into the Tonle Sap Lake Floodplain ecosystem
Tonle Sap is the largest lake and wetland ecosystem in the Mekong (herein referred only as Tonle Sap) has shaped a unique mosaic of
natural and agricultural habitats. These habitats also support the
largest fishery complex in Cambodia, which provides for the
* Corresponding author. Tel.: þ64 3 364 2378; fax: þ64 3 364 2758. majority of the protein consumed in the country (Bonheur and
E-mail addresses: mauricio.arias@pg.canterbury.ac.nz (M.E. Arias),
tom.cochrane@canterbury.ac.nz (T.A. Cochrane), thanapon.piman@
Lane, 2002; Baran and Myschowoda, 2009). The ecological and
canterbury.ac.nz (T. Piman), matti.kummu@iki.fi (M. Kummu), brian.caruso@ social importance of the Tonle Sap led to the establishment of
canterbury.ac.nz (B.S. Caruso), Tim.Killeen@wwfus.org (T.J. Killeen). a UNESCO Biosphere Reserve e the only one of its kind in the
0301-4797/$ e see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jenvman.2012.07.003
54 M.E. Arias et al. / Journal of Environmental Management 112 (2012) 53e66
country (UNESCO, 2010). The importance of the Tonle Sap to the Cambodian floodplain during 1999e2003 in order to construct
Cambodian people has prevailed since the Khmer Empire domi- a water balance and to verify a hydraulic dynamic model of the
nated the region and established a complex society based on a large Cambodian floodplain. Milne and Tapley (2004) used Airsat data to
agricultural and irrigation network from the IX to the XV century classify vegetation and flooding of a region in the northwest of the
(Kummu, 2009). The hydrological, ecological, nutritional, and Tonle Sap (Siem Reap Province). Furthermore, Benger (2007, 2009)
cultural value that the Tonle Sap provides to Cambodia is unques- reported on a classification of satellite images to determine vege-
tionable, and it is arguably among the highest provided to a nation tation distributions and hydrological patterns.
by a single ecosystem around the world. Flooding is the primary factor driving the Tonle Sap ecosystem,
Nonetheless, the future of the Tonle Sap and the Mekong is and therefore, structure, quality, and function of the floodplain
uncertain. The MRB is changing at a rapid rate due to investments habitats are influenced by flooding characteristics. Humans are
in water resources infrastructure (hydropower, irrigation, flood secondary drivers of change. Flooding alters ecosystem structure
control, and water supply) and climate change. A large hydropower first by delimiting the area where agriculture (primarily rice) takes
scheme has been envisioned by the countries that share the basin, place in the floodplain; cultivated areas are restricted mainly to the
and if completed, the number of large hydropower projects in the outer part of the floodplain where unregulated flooding only occurs
MRB could triple while the active storage could increase from for a few days in wet years. Within the non-cultivated areas of the
8.6 km3 to over 90 km3 in the next few decades (MRC, 2009; floodplain, plant species and biomass are distributed according to
Kummu et al., 2010). Irrigation is the major consumer of water in the environmental conditions and human use dictated by the
the MRB, and if current trends of economic development and flooding regime.
growth continue, the total irrigated area could increase 43% and One of the main impediments to a full understanding of the
65% by 2030 and 2060 respectively (Young, 2009; MRC, 2011). In hydro-ecology of the Tonle Sap is the absence of published data on
addition to infrastructure development, climate change will alter habitats, primary production, and human interactions (Kummu
temperature and rainfall throughout the region, bringing signifi- et al., 2006; Lamberts, 2006; Lamberts and Koponen, 2008).
cant changes to the hydrology of the basin (Eastham et al., 2008; Comprehensive field surveys of habitat characteristics in relation to
Hoanh et al., 2009; TKK and SEA START RC, 2009; Västilä et al., the flooding regime and human disturbance may ultimately help to
2010; Kingston et al., 2011). Changes in the MRB as a result of fill this knowledge gap, but this will be more effective and mean-
hydropower, irrigation, and climate change are expected to bring ingful if it is guided by a landscape assessment that incorporates
considerable modifications to the flood pulse of the Tonle Sap in the existing hydrological, GIS, and remote sensing datasets.
future (Kummu and Sarkkula, 2008; MRC, 2010; Västilä et al., 2010). Examples of studies of landscape interactions between flood
The surface hydrology of the Tonle Sap is well understood and pulse and floodplain vegetation, similar in nature to the Tonle Sap,
has been explained through numerical models (Sarkkula et al., exist for the Okavango Delta in Botswana and the Amazon. The
2003; MRC, 2005; Inomata and Fukami, 2008). The Tonle Sap Okavango River Delta’s spatial scale, pulsing characteristics, and
River discharges a maximum 10,000 m3/s from the lake toward the basin-wide problematic are similar to the Tonle Sap’s. Murray-
Mekong Delta from June to October (Inomata and Fukami, 2008); at Hudson (2009) studied how changes in the hydrological regime
the end of this period, the lake reaches its minimum monthly depth due to potential future scenarios of climate change and develop-
of 1.5 m and a surface water area of 2600 km2. When the wet ment could impact the vegetation cover of the Okavango Delta.
monsoon reaches the MRB, the level of the Mekong River rises to Results from a hydrological model were used to determine the size
a much higher level than the Tonle Sap system, forcing the Tonle and distribution of flooding area under future scenarios. Flooding
Sap River to reverse its flow toward the lake. This phenomenon frequency and duration were assigned to functional floodplain
causes the lake to fluctuate from 1.5 to 10 m and an expansion of types related to the existing vegetation cover. A baseline vegetation
the surface area subject to flooding from 2600 to 15,000 km2. cover was modeled as a function of flooding probability, and then
Remote sensing and Geographical Information Systems (GIS) this was compared to the resulting vegetation cover types simu-
have become one of the main research tools in environmental lated from the probability maps for future scenarios. In a different
management and impact assessment of the Tonle Sap in recent study, Milzow et al. (2010) investigated how vegetation cover in the
years. Kummu and Sarkkula (2008) assessed how changes in the Okavango River Delta could change as a response to hydrological
Mekong River as a response to hydropower would affect the flood future scenarios by linking the results from a groundwater model to
pulse characteristics of the Tonle Sap. This study found that the area a vegetation cover map. Vegetation classes were found to have
of open water could increase by 17e40% as a response to higher distinct histograms of depth to groundwater level, so predictions of
water levels during the dry season, which would ultimately result vegetation cover were based on the assumption that vegetation
in the permanent inundation of all the area of gallery forest. The classes would maintain the same depth to groundwater histogram
study also found that the area of seasonally inundated floodplain in the long-term. Wittmann et al. (2002, 2004, 2006) classified
could decrease by 7e16% due to the decline in water level during forest types according to the flooding regime in Amazon floodplain
the wet season. forest in Brazil by combining field observations, aerial photography,
A more recent analysis was carried by the Mekong River and satellite imagery. These studies found clear differences in the
Commission (MRC, 2010) in which impacts from eight different canopy structure and species distribution as a result of flood height
future scenarios to the Tonle Sap were considered. The study and duration. Studies in the Okavango Delta and Amazon floodplain
concluded that there could be a reduction of flood depth during the illustrate a direct link between hydrology and floodplain vegeta-
wet season of over 0.5 m and an increase of water level during the tion, and their methodologies are relevant to the Tonle Sap.
dry season of 0.2e0.6 m. These changes could reduce the flooded Remote sensing tools have also been used to study hydrology
area by 400e900 km2. Consequently, flooded areas covered by and vegetation patterns of different types of large tropical
forests could be reduced by 22e100 km2, grasslands by wetlands. For instance, the vegetation communities of the Ever-
50e150 km2, and rice fields by 300e630 km2. glades National Park (USA) were found to be distributed along
Other studies in the Tonle Sap have used remote sensing data to hydrological gradients represented by percent time of inundation
infer historical and present status of flooding and vegetation. Fujii and mean water depth (Todd et al., 2010). The resulting vegetation
et al. (2003) used a series of Radarsat images to estimate inunda- classification rules were then linked to a water depth model driven
tion area and storage volume of the Tonle Sap and the rest of the by rainfall inputs in order to investigate potential changes in
M.E. Arias et al. / Journal of Environmental Management 112 (2012) 53e66 55
vegetation cover due to climate change (Todd et al., 2011). Other A number of digital elevation models (DEMs) have been used to
studies have used remote sensing to map ecosystem coverage and support studies on the Tonle Sap (Kite, 2001; Eloheimo et al., 2002;
structure of large floodplains and wetlands without inferring Milne and Tapley, 2004). Kummu and Sarkkula (2008) prepared
a relationship between vegetation and flooding (Simard et al., a DEM derived primarily from an actual topographic survey carried
2006; Odunuga and Oyebande, 2007; MacAlister and Mahaxay, by Certeza Surveing Co. in 1964 and from bathymetry measure-
2009). ments from the 1999 MRC Hydrographic Atlas. For areas not
Although similarities can be elucidated between other flood- covered by either dataset, the DEM includes information from
plains around the world and the Tonle Sap, the uniqueness of this Shuttle Rapid Topography Mission (SRTM). The original DEM
ecosystem requires an in-depth analysis of the interactions covered an area of 27,232 km2, an elevation range of 0e112 m amsl,
between hydrology, human use, and dominant habitats. The and had a horizontal raster grid resolution of 100 m by 100 m. For
objectives of this paper are thus to establish how the historical purposes of this study, the DEM was modified to exclude terrain
flooding regime has shaped landscape patterns of habitats in the elevations higher than 15 m amsl, decreasing the study area to
Tonle Sap, examine how they might shift as a result of hydrological 21,123 km2. The threshold elevation of 15 amsl was chosen because
changes, and discuss implications on future natural resource it is higher than the historical lake’s highest daily record (10.36 m)
management and conservation efforts. and inclusive of all future water level predictions reviewed. The
DEM was reprojected to WGS 1984 UTM Zone 48N, which was the
2. Methods default coordinate system used for all GIS datasets in this study.
Fig. 1. Schematic of assessment and mapping approach used in this study. Input data (DEM and water depth records, Section 2.1) are used to create GIS-based flood maps (Section
2.2), which are then combined with a LULC (land use/land cover) map (JICA, 1999) to create relationships between flood duration and habitats (Section 2.5). The flood duration-
habitat rules are then used to predict future habitat as a result of basin-wide hydrological changes (Section 2.6).
M.E. Arias et al. / Journal of Environmental Management 112 (2012) 53e66 57
Table 1
Summary of model future scenarios used in this study.
Arealargest Areaoverlap orchard, paddy fields with villages, swidden agriculture, and village
%difference ¼ 100% (1) garden crop), and open water (composed of the Tonle Sap perma-
Arealargest
nent lake and Boeng Chmar, a smaller lake permanently connected
Where Arealargest is the largest flooded area between each pair of to the larger lake).
MODIS-based and GIS-based maps, and Areaoverlap is the flooded
area identical to each pair. 2.4. Historical (1996e2005) land cover change
The GIS-based flood maps were compared to daily depth maps
with actual water depth measurements taken during 2010 in the Land cover change in the Tonle Sap during post-war times was
early dry season (JanuaryeFebruary) and the wet season (October). investigated using two different LULC maps compiled from aerial
Measurements were taken throughout the floodplain in areas imagery approximately one decade apart. The first dataset is the
accessible by boat using a Dr Lange HT 1 probe and/or a Speedtech JICA (1999) LULC map described in Section 2.3, and the second
Instruments water depth probe. The location of each depth dataset was classified by the Cambodian Fisheries Administration
measurement was recorded with a Garmin Legend GPS handheld based on 1:5000 aerial imagery captured in 2004e2005 (PASCO-
device. Location and depth measurements were then converted FINNMAP CONSORTIUM, 2005). Once geographical projections
into an ArcGIS point feature shape file (Fig. 2) and overlaid on water and extents were coordinated, the classification nomenclature was
depth maps created for the specific dates of field measurements. harmonized according to recent field observations and with the
The difference between field measurements and depth maps values terminology that is commonly used to describe the Tonle Sap
were calculated to further validate the accuracy of the flood habitats and/or LULC classes. Finally, area coverage of each LULC
mapping procedure presented in this paper. class was calculated and compared between the two maps.
2.3. Tonle Sap Lake Floodplain habitats 2.5. Relationship between flooding and habitats
Historical cover of habitats on the Tonle Sap was derived from The historical interaction between flooding and floodplain
a Cambodia’s country-wide land use/land cover (LULC) map habitats was assessed with GIS by overlaying the flood duration
developed by the Japanese International Cooperation Agency (JICA, maps on the LULC map and then calculating the histogram of flood
1999). This dataset was created using SPOT and Landsat images zones on habitats using spatial analyst tools (See relationship
from the late 1990s (Sarkkula et al., 2003; Kummu and Sarkkula, flooding and habitats in Fig. 1). These histograms can be expressed
2008). Although no ground-truthing was done for the dataset due in a matrix form as follows:
to the political instability of Cambodia at the time, it was later 2 3
x0;1 x0;2 . x0;j
shown to provide good estimates of the actual vegetation cover 6 x1;1
6 x2;2 . x2;j 7
7
throughout the floodplain (Hellsten et al., 2003). The LULC map has Yt ¼ 4 (2)
« « « 5
a 30 m by 30 m grid resolution, and includes 27 different classes xk;1 xk;2 . xk;j
within the area studied. Most of the minor classes were excluded or
merged onto larger classes, so that only those classes that covered where Yt is the matrix representing the histogram for the year t
more than 1% of the total area below 15 m amsl were included in (average, dry, or wet), and the element xk,j represents the area
the analysis (Fig. 2). Merging of classes was only done on two covered by the flood zone k on the habitat j. Flood zone k has 13
occasions in which the nomenclature and location implied very distinct categories, which represent annual zones of 0e12 months
little functional difference between the classes. Merged classes of inundation within the Tonle Sap, and habitat j has 10
included village crops (composed of field crops, garden crops, categories representing the habitats from Fig. 2. The resulting
58 M.E. Arias et al. / Journal of Environmental Management 112 (2012) 53e66
Fig. 2. Overview map of the Tonle Sap with major LULC (land use/land cover) classes from JICA (1999) below 15 amsl (above mean sea level) and water depth measurements.
histograms were then normalized in two ways to infer two annual basis; therefore, the classification rules were applied over
different properties of the Tonle Sap habitats in relation to flood the range of flood duration values during the average, dry, and wet
duration: dominance and preference. Dominance refers to the hydrological years considered. Once the range of flood duration
distribution of habitats within each flood zone, and it can be values was determined, each raster pixel was directly assigned
derived from the histogram in Eq. (2) by dividing each matrix a habitat class that best fit the results from Eq. (5). LULC classes that
element by the total area covered by their corresponding flood yielded very similar flood duration values were merged into
zone: a habitat group with unique distribution of flood duration. To assess
the validity of the habitat modeling approach used in this study, the
Yt baseline modeled map was overlaid onto the original JICA (1999)
dominancet ¼ Pj (3)
x map. Areas of habitat overlap and shifts from one habitat to
i¼1 k;j
another were then estimated. Modeled map accuracy was calcu-
Preference refers to the distribution of each specific habitat lated as the percent of overlap area (%overlap) between habitats from
among all flood zones. It was derived by dividing all elements of the JICA (1999; Areaoriginal) and the model baseline map:
matrix Yt by the total area covered by their corresponding habitat:
Areaoverlap
%overlap ¼ 100% (6)
Yt Areaoriginal
preferencet ¼ Pk (4)
i¼0 xk;j Future shifts in habitat cover were assessed by applying the
These two properties were then used to select the habitat of classification rules derived from the baseline data to the flood
maximum likelihood of occurrence within each flood zone by duration maps associated with the different scenarios from
simply applying a scalar multiplication between the matrices and hydrological modeling simulations (Table 1). Future impacts were
finding the greatest element for each row of the resulting matrix: assessed by determining the change in habitat cover relative to the
model habitat map for the baseline scenario.
Max likelihoodt ¼ dominancet $ preferencet (5)
3. Results
2.6. Habitat mapping
3.1. Water levels and flood maps
A model habitat map for the baseline hydrological years was
created by applying classification rules to the flood duration maps The 1986e2010 water depth records at Kampong Loung show
based on the maximum likelihood of habitat occurrence within a clear annual flood pulsing behavior with variable water levels
each flood zone (See habitat mapping step in Fig. 1). It is assumed during the wet season and consistent levels during the dry season
that the spatial distribution of habitats is a response to long-term (Fig. S1). The year 1998 was the driest in the period of record, with
decadal flooding conditions rather than changes occurring on an a maximum monthly average water level reaching only 6.7 m,
M.E. Arias et al. / Journal of Environmental Management 112 (2012) 53e66 59
whereas 2000 was the wettest year with a maximum monthly there is a progression of water level changes from wet year (least
average water level of 10.2 m. Water levels during 1997 were impacted) to dry years (most impacted). During wet years, very
representative of the annual central tendencies for the period of little change is expected as a result of climate change and a slight
record. The minimum and maximum monthly average water levels decrease during the rainy season (July through September) is ex-
in 1997 were 1.5 and 9.1 respectively, compared to 1.6 and 8.9 for pected as a result of water infrastructure development. During
the entire period. average years, mean water levels during October and November
The impact of five future infrastructure development and increase by 0.50 m as a response to climate change, but decrease by
climatic change model scenarios on Tonle Sap flood levels were 0.35 m during the same months due to development. The largest
studied for dry (1997), average (1998), and wet years (2000). changes occur during dry years. As a result of climate change, water
Hydrographs from the two 2030 scenarios are compared to the levels remain the same during the low water months of April and
observed monthly averages in Fig. 3. The hydrographs show that May, but increase by 0.2e0.75 m during the rest of the year. As
Fig. 3. Comparison of mean monthly water level at Kampong Loung for historical observed records and model predictions for an (a) average year, (b) dry year, and (c) wet year.
2030CC ¼ climate change scenario for 2030’s; 2030DEV ¼ water resources infrastructure development scenario for the 2030’s. See Table 1 for description of scenarios.
60 M.E. Arias et al. / Journal of Environmental Management 112 (2012) 53e66
Fig. 5. Flood duration (months) zone dominance (aka., percent of flood zone area
covered by each habitat) during hydrological (a) average, (b) dry, and (c) wet years.
Histograms do not add to 100% since only habitats that covered largest area are shown.
Fig. 6. Habitat preference (aka., percent of habitat extent covered by each flood
duration (months) zone) during hydrological (a) average, (b) dry, and (c) wet years.
Habitat preference for distinct flood zones was quantified by Only habitats that covered greatest areas are shown.
Table 2
Flood duration rules used for modeling habitat cover and validation results. LULC refers to land use/land cover.
Clustered habitat LULC class(es) Months of annual flood duration Cover area (km2) % Overlap
average, dry, and wet years, respectively. Transitional habitats was 4. Discussion
defined as the area flooded for 1e5, 0e1, and 3e6 months during
average, dry, and wet years, respectively, and it included aban- 4.1. Historical changes in water levels, flood extent, and LULC
doned fields, lowland grassland, and floating and receding rice.
Seasonally flooded habitats included flooded shrubland and floo- The changes in water level and flood extent in the Tonle Sap
ded grassland. This group was defined by flood duration of 5e8, calculated in this study are consistent with findings from previous
2e7, and 6e11 months during the average, dry, and wet years, studies (Fujii et al., 2003; MRC, 2005; Inomata and Fukami, 2008;
respectively. Gallery forest was left as a single habitat since it Kummu and Sarkkula, 2008), which revealed little change in
dominated the transitional area between the seasonally flooded recent past decades in water levels and flood extent (hence, flood
habitats and the open water (flood duration of 9 months during duration). Some of these studies have analyzed longer time series
average and dry years and 9e12 months in a wet year). Open water at stations in the Tonle Sap and Mekong Rivers, and their results
was classified as the areas flooded for 10e12, 9e12, and 10e12 show that the years used to represent recent historical conditions
months during average, dry, and wet years, respectively. The for the Tonle Sap (1997, 1998, and 2000) are optimal. Despite the
baseline cover map derived from these flood duration rules was in limited water level time series available at Kampong Loung, it
good agreement with the spatial distribution of habitats from JICA appears that decadal patterns of extreme events are influenced by
(1999; Table 2). ENSO patterns; extreme floods match strong La Niña years
Flood duration rules defining each of the 5 habitat groups whereas droughts match years of strong El Niño. This linkage
(Table 2) were then used to simulate the most likely changes in between regional and local climate patterns could be useful in
habitat cover due to potential future scenarios of water infra- studying long-term historical and future climate change in the
structure development and climate change. Area extent and the absence of local long-term hydrological data. Other approaches to
percent difference with respect to the baseline cover map were overcome the data limitation could be to expand the regression
calculated for each of the habitat groups (Table 3; Fig. 7). Area of relationships with other stations with longer historical records
rainfed habitats increases by 10e14% for all water infrastructure (Inomata and Fukami, 2008), or to use other long-term hydro-
development scenarios and decreases by up to 5% for the climate logical indicators such as sediment or tree cores (Buckley et al.,
change scenarios. Area of transitional habitats decreases up to 6% 2010; Day et al., 2011).
as a result of both infrastructure development and climate change A broad comparison of the two classified floodplain-wide land
scenarios. Area of seasonally flooded habitats decreases by up to cover maps (1996 and 2005) revealed expected changes in vege-
22% as a result of infrastructure development, but increases tation cover that followed documented land use modifications in
slightly as a result of climate change. The area most suitable for the floodplain. The expansion of village crops and conventional
gallery forest decreases by up to 83% as a result of development rice paddies can be explained by the economic expansion and
and up to 69% as a result of the 2040 climate change scenario. population growth following the end of a long period of civil
Maps displaying these results show clear habitat shifts as a result conflict. The abandonment of floating rice fields has occurred
of future scenarios (Fig. 7); changes due to water infrastructure since the late 1970s (Hand, 2002; Sarkkula et al., 2003), and the
development show a large shift from gallery forest to open water, results of our analysis suggest that some of the abandoned fields
from seasonally flooded habitats to transitional habitats, and from have more recently been cropped again for conventional rice
transitional to rainfed habitats. Climate change scenarios also paddies. An unexpected finding was the expansion of the area
showed a large expansion of the open lake for the 2040 scenario. classified as gallery forest. Nearly all the area classified as gallery
Other notable shifts for the climate change scenarios were forest in PASCO-FINMAP CONSORTIUM (2005) was previously
seasonally flooded habitats shift to gallery forest, transitional classified by JICA (1999) as either shrublands or grasslands, sug-
habitats shift to seasonally flooded habitats, and rainfed habitats gesting that the change in land cover of this small area (yet largest
shift to transitional habitats. in terms of standing biomass per unit area) has occurred either as
Table 3
Area change from baseline (modeled) habitat cover as a response to different future scenarios.
Model scenarios Rainfed habitats Transitional habitats Seasonally flooded habitats Gallery forest Open water
Fig. 7. Shifts in habitat cover from the baseline model map in area around Prek Toal, the largest core area of the UNESCO Biosphere Reserve. RF ¼ rainfed habitats, T ¼ transitional
habitats, SF ¼ seasonally flooded habitats, GF ¼ gallery forest, and OW ¼ open water.
an unlikely example of ecological succession occurring over short result in slightly higher water levels during the start of the wet
time scales, or as the result of different classification criteria and season and moderately higher levels during its peak. Overall, more
erroneous interpretation of the remote sensing data. With the pronounced changes are expected during dry years than during wet
evaluation of LULC change we intended to acknowledge the years. Although slightly different trends, both models predicted
changes that might have occurred in the floodplain during water level changes of similar magnitude, implying that their input
a period when no significant changes in flooding regime occurred. databases and modeling frameworks are similar and hence their
Changes in the outer portions of the floodplain e covered results are comparable. This was not the case with previous
primarily with agricultural fields e were expected, as flooding is modeling efforts, which showed large discrepancies and predicted
not as dramatic in this area and it could be more easily regulated much larger changes as a result of climate change (Eastham et al.,
than closer to the lake. On the contrary, a larger portion of the 2008) than water infrastructure development (WB, 2004). In
floodplain, where seasonal flooding is much more prominent, has short, the water level changes predicted by the models used in this
experienced much smaller shifts in LULC. These observations study seem to provide a higher level of confidence than previous
suggest that flooding has played a major role in the habitat predictions, transferring less uncertainty to subsequent habitat
distribution of the floodplain, primarily by preventing agriculture changes.
in areas where flooding is too deep and lasts too long. It is worth mentioning that the MRC model scenarios did not
include detail analysis about impacts from hydropower dams on
4.2. Future changes of water level and flood duration Mekong tributaries such as the Sesan, Sre Pok and Sekong (3S),
which contribute a large fraction of the mean annual Mekong River
Moderate changes in average monthly water levels were pre- flow. Development of the 3S’s full hydropower capacity will have
dicted by the two hydrological models. Water infrastructure a higher impact on water flows than the hydropower scheme of the
development is expected to shrink the magnitude of the flood pulse Upper Mekong Basin in China (Piman et al., 2012). This overlooked
of the Tonle Sap by raising water levels during the dry season while factor will subsequently alter the water levels in the Tonle Sap to
reducing water levels during the wet season. Climate change will a greater extent than current estimates.
64 M.E. Arias et al. / Journal of Environmental Management 112 (2012) 53e66
Changes in annual flood extent and duration are slightly more gradient rather than distinct habitat types; gallery forest and
pronounced than those for monthly water levels. Overall, water flooded shrublands share similar vegetation composition, and the
infrastructure development is expected to reduce the flood extent distinction between these two and seasonally flooded grasslands
by up to 1200 km2, while climate change scenarios are expected to could be related to recent fire and cattle grazing (M.E. Arias et al., in
increase the flood extent by up to 1000 km2. The greatest changes preparation). While ongoing and future work are addressing the
from baseline conditions will occur during average years, whereas limitations summarized above, the methodology and results from
wet years will experience the least changes. The trends and this paper present a simple and comprehensive model that could
magnitude of flood duration estimated in this study are in agree- be used as a research and large scale management tool for the Tonle
ment with the ones from Västilä et al. (2010) and MRC (2011). The Sap.
areas that will be impacted the most are those at the fringe of the
open water, with flood duration of 9e10 months, and half way 4.4. Habitat cover changes
between the open water and the edge of the floodplain, which is
flooded approximately 4 months of the year. Overall, extensive Large shifts in habitat cover could occur as a result of future
changes e in the order of 100e1000 km2 e are expected in the scenarios of water infrastructure development and climate change.
spatial distribution of flood duration as a result of sub-meter Although there have not been other attempts to model habitat
changes in water levels, a magnified effect that would be ex- cover change on the Tonle Sap, the results of this assessment are in
pected in a floodplain as large and as flat as the Tonle Sap. The agreement with trends of potential impacts to the Tonle Sap
magnitude of these spatial changes in flood duration highlights the habitats documented by others (Kummu and Sarkkula, 2008; MRC,
importance of this assessment to the overall understanding of the 2010, 2011): water infrastructure development reduces the spatial
hydrological and ecological changes that the Tonle Sap is being extent of seasonally flooded habitats and gallery forests while
subject too. favoring rainfed/irrigated agricultural areas. These changes in
habitat cover could have large implications to the overall func-
4.3. Modeling approach tioning and productivity of the Tonle Sap ecosystem. Forests and
shrublands have the greatest standing biomass per unit area, hence
A large fraction of landscape variability was explained solely as these habitats provide a major role in sediment deposition, nutrient
a function of flood duration, which is extremely useful to isolate cycling, periphyton growth, primary production, fish food and
and quantify future ecological impacts as a result of hydrological refuge. Grasslands and rice fields provide these ecosystem func-
alterations. A considerable amount of work has been done on the tions to a much lesser extent, and therefore impacts to habitat cover
hydrology of the Tonle Sap and the MRB, and this paper demon- and human use influenced by changes in the hydrological regime
strates how to link hydrological and ecological changes using will most likely reduce the overall ecological integrity and
empirical, in situ observations. Although the model developed in productivity of the Tonle Sap.
this study was based on a classification rule scheme, it used a robust Although there is a progressive increase in the extent of the
methodology that considered the full range of historical hydro- impacts with water infrastructure development scenario inten-
logical conditions as well as the full distribution of habitats through sity, this assessment shows that a large magnitude of the changes
the Tonle Sap landscape. The five hydrological-distinct habitat could happen soon as a result of current hydropower develop-
groups take into account both the adaptation of each habitat to ment in the Upper Mekong in China. Although no significant
flooding (aka., optimal or preferred hydrological conditions for each changes in the LMB hydrology have been reported yet, a land
habitat; Fig. 6), as well as the competition among habitats for cover change monitoring protocol should be implemented in
a single flood zone (aka., dominance of habitats on flood duration order to improve our predictive abilities and develop mitigation/
zones; Fig. 5). adaptation strategies in the near future. Habitat cover shifts solely
On the other hand, the modeling approach has a number of as a result of climate change showed a different picture than water
limitations. First, it does not take into account habitat succession, as infrastructure development scenarios, especially with respect to
there are no documented observations on how this concept applies the rainfed habitats and seasonally flooded habitats. The esti-
to the Tonle Sap ecosystem. Therefore, each of the modeled habitat mated changes for the gallery forest and the open water, however,
cover maps represent a snapshot of what the floodplain may look are similar to the development scenarios, with a net increase in
like after the hydrological conditions from future scenarios persist open water area. Overall, the magnitude of shifts resulting from
for sufficient time for the considered habitats to evolve. Further- climate change was found to be smaller than those from water
more, some of the classes from the JICA (1999) map were too small infrastructure development scenarios. This finding should not
and/or determined by human use, and therefore they were not undermine the potential impacts from climate change, but it
accurately estimated as a function of flood duration alone. For should instead promote more detailed and robust modeling of
instance, the optimal area for gallery forest simulated was cumulative impact scenarios.
considerably larger than what was classified by JICA (1999) or Regardless of which future scenario could bring more signifi-
PASCO-FINMAP CONSORTIUM (2005). One reason for the mismatch cant shifts, natural resources managers and authorities in
could be because this small habitat is surrounded by open water Cambodia have little or no control over the drivers of these
and seasonally flooded habitats, both of which cover extensive changes, and the choices that they are faced with have more to do
areas that strongly influenced the dominance flood duration rela- with the implementation of management strategies that would
tionship. Another reason could be that the gallery forest is the most lead the floodplain toward a smoother transition that minimizes
intact section of the seasonally flooded gradient, and the unique impacts to the ecosystem services historically provided by the
vegetation structure of this habitat is an artifact of restricted human Tonle Sap. By taking into account the potential habitat shifts, Tonle
use in comparison to sections further upland. Furthermore, very Sap managers could guide how and where they will focus present
similar hydrological conditions were found between flooded and future efforts on habitat conservation and/or restoration.
shrublands and grasslands, yet these two habitats have very Based on the results of this assessment, habitat conservation and
distinct vegetation structure. Recent field work elucidated answers restoration should take place in areas most suitable for natural
to these two anomalies, as it appears that gallery forest, flooded habitats (aka., grasslands, shrublands, and gallery forests) and
shrubland, and flooded grassland are part of a compositional least useful for rice cultivation under future hydrological regimes.
M.E. Arias et al. / Journal of Environmental Management 112 (2012) 53e66 65
MRC, 2011. Assessment of Basin-wide Development Scenarios (Main Report). In: Todd, M.J., Muneepeerakul, R., Pumo, D., Azaele, S., Miralles-Wilhelm, F., Rinaldo, A.,
Basin Development Plan Programme, Phase 2. Mekong River Commission. Lao Rodriguez-Iturbe, I., 2010. Hydrological drivers of wetland vegetation
PDR, Vientiane. community distribution within Everglades National Park, Florida. Advances in
Murray-Hudson, M., 2009. Floodplain Vegetation Responses to Flood Regime in the Water Resources 33, 1279e1289.
Seasonal Okavango Delta, Botswana. PhD thesis, University of Florida, Todd, M.J., Muneepeerakul, R., Miralles-Wilhelm, F., Rinaldo, A., Rodriguez-Iturbe, I.,
Gainesville. 2011. Possible climate change impacts on the hydrological and vegetative
Odunuga, S., Oyebande, L., 2007. Change detection and hydrological implications in character of Everglades National Park, Florida. Ecohydrology 5, 326e336.
the Lower Ogun flood plain, SW Nigeria. In: Remote Sensing for Environmental UNESCO, 2010. Biosphere Reserves e World Network [WWW Document]. UNESCO-
Monitoring and Change Detection. Presented at the IAHS Symposium on MAB (Man and the Biosphere) Secretariat, Paris, France. URL. http://www.
Remote Sensing for Environmental Monitoring and Change Detection. IAHS unesco.org/mab/.
Press, Perugia, p. 358. Västilä, K., Kummu, M., Sangmanee, C., Chinvanno, S., 2010. Modelling climate
PASCO-FINNMAP CONSORTIUM, 2005. Purchase of Aerial Photography and Digital change impacts on the flood pulse in the Lower Mekong floodplains. Journal of
Orthophotos for the Tonle Sap Environmental Management Project. Cambodia Water and Climate Change 01, 67e86.
Department of Fisheries. WB, 2004. Modelled Observations on Development Scenarios in the Lower Mekong
Piman, T., Cochrane, T.A., Arias, M.E., Green, A., Dat, N.D., 2012. Assessment of flow Basin. World Bank, Vientiane, Lao PDR.
changes from hydropower development and operations in Sekong, Sesan and Wittmann, F., Anhuf, D., Junk, W.J., 2002. Tree species distribution and community
Srepok Rivers of the Mekong Basin. Journal of Water Resources Planning and structure of central Amazonian várzea forests by remote-sensing techniques.
Management. Journal of Tropical Ecology 18.
Sarkkula, J., Koponen, J., Hellesten, S., Keskinen, M., Kiirikki, M., Lauri, H., Varis, O., Wittmann, F., Junk, W.J., Piedade, M.T.F., 2004. The varzea forests in Amazonia:
Virtanen, M., Dubrorin, T., Eloheimo, K., Haapala, U., Jozsa, J., Jarvenpaa, E., flooding and the highly dynamic geomorphology interact with natural forest
Kummu, M., 2003. Modelling Tonle Sap for Environmental Impact Assessment succession. Forest Ecology and Management 196, 199e212.
and Management Support (Draft Final Report). MRCS/WUP-FIN. Wittmann, F., Schongart, J., Montero, J.C., Motzer, T., Junk, W.J., Piedade, M.T.F.,
Simard, M., Zhang, K., Rivera-Monroy, V.H., Ross, M.S., Ruiz, P.L., Castaneda-Moya, E., Queiroz, H.L., Worbes, M., 2006. Tree species composition and diversity gradi-
Twilley, R.R., Rodriguez, E., 2006. Mapping height and biomass of mangrove ents in white-water forests across the Amazon Basin. Journal of Biogeography
forests in Everglades National Park with SRTM elevation data. Photogrammetric 33, 1334e1347.
Engineering and Remote Sensing 72, 299e311. Young, A., 2009. Regional Irrigation Sector Review for Joint Basin Planning Process,
TKK, SEA START RC, 2009. Water and Climate Change in the Lower Mekong Basin: Regional Irrigation Sector Review for Joint Basin Planning Process. In: Basin
Diagnosis & Recommendations for Adaptation. Water & Development Publica- Development Plan Programme, Phase 2. Mekong River Commission, Vientiane,
tions, Helsinki University of Technology, Espoo, Finland. Lao PDR.