Mapping Dambo Wetlands in Central Uganda Using Remote Sensing and Topographic Data



Yüklə 64,18 Kb.
Pdf görüntüsü
tarix18.06.2018
ölçüsü64,18 Kb.
#49319


UGANDAN DAMBO WETLAND CLASSIFICATION USING MULTISPECTRAL AND 

TOPOGRAPHIC REMOTE SENSING DATA

 

 

Matthew K. Hansen, Research Assistant 

Philip E. Dennison, Assistant Professor 

Scott A. Graves, Research Assistant 

Department of Geography 

University of Utah 

260 S. Central Campus Dr., Rm. 270 

Salt Lake City, UT 84112 

matt.hansen@geog.utah.edu

dennison@geog.utah.edu

scott.graves@geog.utah.edu

 

David J. Brown, Assistant Professor 

Department of Crop and Soil Sciences 

Washington State University 

PO Box 646420 

Pullman, WA 99164 

david_brown@wsu.edu



 

 

ABSTRACT

 

 

The seasonally saturated dambo wetlands of Central and Southeastern Africa are likely a substantial source of 



methane (CH

4

), an important greenhouse gas.  Dambo soil, vegetative, and hydrologic characteristics vary along a 



gradual topographic gradient; from relatively dry uplands, to occasionally inundated margins, to more frequently 

inundated floors, to perennially inundated bottoms.  CH

4

 production presumably also varies along the gradient from 



uplands to bottoms, making mapping dambo topography an important first step in determining regional dambo 

methane production.  Multispectral remote sensing data and topographic attributes were used to map upland, margin, 

floor, and bottom classes of dambo wetlands in a 2,200 square kilometer study area located in central Uganda.  Two 

Système Pour l’Observacion de la Terre (SPOT) 4 multispectral scenes were acquired to coincide with the beginning 

and end of the January-to-March dry season in the study area.  Spectral indices and spectral mixture modeling 

fractions were calculated for both SPOT scenes.  Phenological changes occurring within each dambo class were 

quantified by comparing remote sensing metrics in each SPOT scene.  Multispectral metrics were combined with 

topographic data from the Shuttle Radar Topography Mission (SRTM) using a binary decision tree (BDT) to 

classify the study area into upland, margin, floor, and bottom classes.  Field data were used for training the classifier 

and assessing the accuracy of the final classification.  This classification will be used to locate methane emission 

sampling sites within the study area as part of the first region-scale dambo methane emissions modeling project. 

 

 

BACKGROUND

 

 

 

The anaerobic microbial decomposition of biomass that occurs in wetlands makes them the largest natural 



source of atmospheric CH

4

.  Wetlands are responsible for 20-45% of the total annual global emissions of CH



4

 

(Shindell et al., 2004), and CH



4

 is approximately 20 times as effective a greenhouse gas as carbon dioxide (CO

2

) on 


a molecular level (Rodhe, 1990; Faulkner, 2004).  CH

emissions from wetlands in the tropics are among the least 



understood (Bartlett and Harriss, 1993), adding to the significance of this research. 

 

Dambo” is one of several dialectal terms used to describe the seasonally saturated, grassy, narrow depressions 

covering as much as 20% of the plateau regions of Central and Southern Africa (Bullock, 1992; Mäckel, 1974).  

While most dambo wetlands are waterlogged for at least a portion of the year, many of them dry out at the surface 

during the four- to six-month dry season typically found in semi-arid Africa (Acres et al., 1985).  However, the 

“sponge-like” center of most dambos stays moist even during the dry season, sustaining the higher-density 

herbaceous vegetation in this zone (Mäckel, 1985).  The relatively planar topography (½ – 2º slope) typical of 

ASPRS 2008 Annual Conference 

Portland, Oregon   April 28 - May 2, 2008 

 



dambo wetlands produces little hydraulic energy, which in turn facilitates soil saturation and inhibits channel 

formation (Acres et al., 1985; Von der Heyden, 2004). 

 

 

margin



margin

floor


floor

bottom


upland

upland


 

Figure 1. Dambo cross-section. 

 

Dambo vegetation and soil types vary, but researchers have identified certain patterns characteristic of the 



landforms.  Acres et al. (1985) used vegetation, soil type, and topography to identify three general zones within a 

dambo: the margin, floor, and bottom (Figure 1).  The margin, at the perimeter of a dambo, is the transition zone 

from higher, often wooded terrain to grasslands that gently slope toward the center of the wetland (Von der Heyden, 

2004).  The dambo floor is a continuation of the grassland that transitions to sedges and other herbaceous plants 

near the bottom.  At the center of the idealized dambo is the bottom, an area usually only suitable as habitat for 

those plants adapted to saturated soils.  Because inundation varies according to these dambo zones, becoming 

progressively wetter from margin to bottom, distinguishing between them is essential to modeling methane 

emissions. The purpose of this project was to use remotely sensed information to classify dambo wetlands within a 

study area in Uganda, resulting in an accurate map for use in subsequent region-scale methane emissions 

measurement and modeling. 



 

 

METHODS

 

 

 

This landscape classification designated each 20 meter pixel within a study area as bottom, floor, or margin, 



with the addition of an upland class for the surrounding terrain.  The 2,200 sq. km. study area (Figure 2) was 

selected from a remote region of central Uganda bounded by the Mayanja and Lugogo Rivers.  The study area 

demonstrated the topography, climate, and vegetation typical of African dambos.  Additionally, the site offered a 

range of elevation and precipitation patterns in a relatively compact area, hopefully allowing for broader application 

of the methane model that will result from future research. 

Field sampling of the dambo classes occurred in January and February, 2007, during one of the annual dry 

seasons typically experienced in the study area.  The 236 sites sampled (Figure 2) during the field campaign were 

chosen in one of two ways.  First, to capture dambo wetlands on multiple scales, a stratified random sample set was 

selected according to stream order.  Second, samples were taken opportunely as they were observed within the study 

area.  Two sampling techniques were also employed; one recorded the location of relatively homogeneous areas 

representing each of the dambo classes, and one focused on variation by measuring the extent of each class along 

transects running perpendicular to hydrologic flow. 

Dambo classes were identified in the field according to a combination of soil, topography, and vegetation 

characteristics (Table 1).  Bottoms were readily identified by their distinctive hummocky micro-topography and 

frequent inundation, margins were distinguished by sandy surface soils and slope measurements in excess of 2%, 

and uplands were categorized by their reddish soils according to Munsell chroma and hue.  Because floors exhibited 

fewer unique soil or slope characteristics, they were identified by not belonging to any other class.  In practice, flat, 

grassy terrain and the absence of hummocks were generally indicative of dambo floors. 

Several sources of remotely sensed topographic and multispectral image data served as inputs to the 

classification (Table 2). Two four-band Système Pour l’Observacion de la Terre (SPOT) 4 satellite images, captured 

December 10, 2006, and February 21, 2007 (Figure 2), provided the primary inputs and set a standard spatial 

resolution for the classification of 20 meters.  The SPOT images were also used to calculate two vegetation indices, 

the Normalized Difference Vegetation Index (NDVI; Rouse et al., 1973) and the Normalized Difference Infrared 

Index (NDII; Hunt and Rock, 1989).  Endmember fractions resulting from a spectral mixture analysis (SMA; Adams 

et al., 1993), specifically, the percentages of photosynthetic vegetation (PV), non-photosynthetic vegetation 

(NPV)/soil, and shade for each SPOT pixel, provided additional vegetation-related inputs.  The differences between 

the vegetation index and SMA values for the two SPOT scenes were also included in the classification to observe 

phenological changes that occurred during the dry season.  To measure the spatial heterogeneity of a given variable, 

the classification also included the standard deviation within nine-pixel neighborhoods for each input.   

ASPRS 2008 Annual Conference 

Portland, Oregon   April 28 - May 2, 2008 

 



 

 

Figure 2. Study area with sample sites overlaid on the February SPOT 4 image.  

(includes material © CNES 2007, Distribution Spot Image S.A., France SICORP, USA, all rights reserved) 

 

 



Table 1. Field identification criteria 

 

 



Dambo Class 

Identification Criteria 

Upland 


Surface soil Chroma ≥ 4 and Hue ≤ 7.5 YR,  

or Chroma ≥ 3 and Hue ≤ 5 YR 

Margin Slope 

2% 



Floor 

NOT Upland, Margin, or Bottom 

Bottom Hummocky 

microtopography 



ASPRS 2008 Annual Conference 

Portland, Oregon   April 28 - May 2, 2008 

 


Digital elevation model (DEM) data obtained from the Shuttle Remote Topography Mission (SRTM) provided 

the base topographic input for the classification.  Prior to resampling, these 90-meter resolution DEM data were also 

used to calculate percent slope and a series of ranked relative elevation layers.  Relative elevation was calculated as 

the rank of a given pixel within a moving window, with window sizes from 11 to 201 DEM pixels (roughly 1 km to 

18 km).   Inclusion of these relative elevation layers was intended to improve dambo classification by distinguishing 

topographic variation at multiple spatial scales.



 

 

Finally, northing and easting values from the projected coordinate system, Universal Transverse Mercator 

(UTM), were inputs to the classification.  These coordinates were chosen to serve as a proxy for climatic gradients, 

particularly precipitation, which vary spatially across the study area. 

The field samples for each dambo class were then randomly subdivided into training and accuracy assessment 

sets for the classification.  As is typical for a supervised classification, the training set provided the locations for 

which pixel values were extracted from corresponding areas of each input variable.  These extracted values were 

evaluated using the “tree” package (Ripley, 2006) developed for the statistical software R to construct a binary 

decision tree (BDT) classifier (Figure 3). 

BDTs consist of a series of “nodes” and “branches,” composed of classification input variables and connections, 

respectively.  As pixels from the input layers progress through the tree, they are sorted at the nodes by comparing 

their value for a given variable with a threshold value established by a decision rule.   They eventually arrive at a 

terminal node and are thereby assigned to a specific class.  In the case of the dambo wetland classifier, uplands were 

identified exclusively by relative elevation.  Slope, SPOT bands, vegetation indices, and UTM northing values were 

used to classify bottoms, floors, and margins. 

 

Table 2. Classification input variables 

(*denotes decision tree node) 

 

Multispectral Inputs 



Topographic Inputs 

SPOT 10 Dec 06 

SRTM Elevation Data 

Band 1: green 

DEM 

Band 2: red 



Slope* 

Band 3: near infrared 

Relative elevation 11 

Band 4: short-wave infrared 

Relative elevation 21* 

SPOT 21 Feb 07 

Relative elevation 31 

Band 1: green 

Relative elevation 41 

Band 2: red 

Relative elevation 51 

Band 3: near infrared 

Relative elevation 61 

Band 4: short-wave infrared* 

Relative elevation 71 



Vegetation Indices 

Relative elevation 81 

NDVI Dec 

Relative elevation 91 

NDVI Feb 

Relative elevation 101 

NDVI difference* 

Relative elevation 111 

NDII Dec 

Relative elevation 121 

NDII Feb* 

Relative elevation 131 

NDII difference 

Relative elevation 141 



Spectral Mixture Analysis 

Relative elevation 151 

PV Dec 

Relative elevation 161 



NPV Dec 

Relative elevation 171 

Shade Dec 

Relative elevation 181* 

PV Feb 

Relative elevation 191 



NPV Feb 

Relative elevation 201 

Shade Feb 

UTM Coordinates 

PV difference 

Northing* 

NPV difference 

Easting  

Shade difference 

  

ASPRS 2008 Annual Conference 

Portland, Oregon   April 28 - May 2, 2008 

 



 

 

 



 

Figure 3. Binary decision tree classifier. 

 

 



RESULTS

 

 

 

The BDT classification resulted in a reasonably accurate dambo class map from a qualitative perspective 



(Figure 4).  In general, the mapped dambo classes corresponded with the spatial patterns described in the literature 

and observed in the field.  Bottoms were identified in the central, lowest portions of larger drainages, with floors, 

margins, and uplands surrounding them in succession.  The classification was less accurate beyond the boundaries of 

the study area where field samples were not obtained.  In these areas, floors and margins tended to be overclassified.  

Within the wider Lugogo, Towa (center left), and Mayanja River channels, bottoms were occasionally misclassified 

as upland or margin.  In quantitative terms, the classification resulted in an overall accuracy of 75.5% and a Kappa 

coefficient of 0.67.  Subdividing the accuracy assessment set in a confusion matrix (Table 3) revealed that uplands 

were the most readily identifiable class.  Confusion with adjacent classes was the most common error among all of 

the classes.   

 

  



Table 3.  Pixel-level confusion matrix for BDT classification 

 

 



 

Ground Truth 

 

 



Bottom Floor Margin Upland 

 

 

Bottom 



520 

142 88  22 

67.36% 

Floor 

130 


1002 

232 0 


73.46% 

Margin 

56 258 1159 

294 65.59% 

Upland  

0 0 102 


1396 

93.19% 


Classifie

d

 

 

73.65% 71.47% 73.31% 81.54% 



Producer’s 

Accuracy 

User’s Accuracy 

 

 



ASPRS 2008 Annual Conference 

Portland, Oregon   April 28 - May 2, 2008 

 


 

 

Figure 4. Dambo class map of study area.  



 

 

DISCUSSION

 

 

 

Given the relationship between topography and hydrology, it was not surprising to learn that elevation data 



would play a major role in this landscape classification.  Uplands were identified only by elevation, with both fine- 

and coarse-scale relative elevation being selected for the decision tree.  DEM-derived slope was used to distinguish 

the relatively steep margin class from the flatter bottoms and floors.  Future work could benefit from the use of 

higher-resolution DEM data.  Were these data available, the accuracy of the classification would likely increase, 

while the need for additional optical remote sensing inputs might decrease.   

ASPRS 2008 Annual Conference 

Portland, Oregon   April 28 - May 2, 2008 

 



ASPRS 2008 Annual Conference 

Portland, Oregon   April 28 - May 2, 2008 

 

With the relatively coarse elevation data used in this study, however, vegetation-related variables served as 

important inputs to the classification.  Drier margins were distinguished from wetter bottoms by the phenological 

change detected in the NDVI between December and February.  Shortwave Infrared (SWIR) reflectance, the fourth 

band provided by SPOT 4 imagery, appears to be important for discriminating floors from bottoms, with two SWIR 

variables selected for the decision tree.  NDII, also a measure of water content that was used to distinguish floors 

from bottoms, is calculated using the Near Infrared (NIR) and SWIR bands.  The SWIR band may have been 

important because it experiences large changes in reflectance due to grass senescence.   The UTM northing input’s 

significance may be due to climatic variation within the study area.  A marked latitudinal difference in green 

vegetation and standing water was observed while conducting field work.  This variation may be linked to 

precipitation gradients and a general decrease in elevation moving from south to north in the study area. 

The results of this first-of-its-kind classification of dambo wetlands were promising.  However, the classifier 

did not appear to perform as well outside of the study area where no sampling occurred.  Further refinement of the 

classifier will likely be required if it is to be applied in other areas.  Due to the range of vegetation and topography 

found in the areas of dambo occurrence, it may not be possible to develop a universal dambo wetland classification 

technique.  In spite of this fact, this classification delineated dambo extents within the study area accurately enough 

to proceed with the intended methane emissions measurement and modeling.  These efforts will provide an estimate 

of dambo contributions to global methane emissions.  In addition, this work will add to the understanding of the 

interplay between dambo methane production, hydroclimate, and anthropogenic land cover change.  

 

 

ACKNOWLEDGEMENT 

 

This material is based upon work supported by the National Science Foundation under Grant No. 0620206. 



 

 

REFERENCES

 

 

Acres, B.D., Rains, A.B., King, R.B., Lawton, R.M., Mitchell, A.J.B. and Rackham, L.J., 1985. African dambos: 

their distribution, characteristics and use, In Dambos: small channelless valleys in the tropics (M.F. 

Thomas and A. S. Goudie, Eds.), Zeitscrift für Geomorphologie, 52: 63-86. 

Adams, J.B., Smith, M.O., and Gillespie, A.R., 1993. Imaging spectroscopy: Interpretations based on spectral 

mixture analysis, In Remote Geochemical Analysis: Elemental and Mineralogical Composition (C.M. 

Pieters and P.A. Englert, Eds.), Cambridge University Press: Cambridge, UK, pp. 145-166. 

Bartlett, K.B. and Harriss, R.C., 1993. Review and assessment of methane emissions from wetlands, Chemosphere

26: 261-320. 

Bullock, A., 1992. Dambo hydrology in Southern Africa - Review and reassessment, Journal of Hydrology, 134: 

373-396. 

Hunt, E.R. and Rock, B.N., 1989. Detection of changes in leaf water content using near- and middle-infrared 

reflectances, Remote Sensing of Environment, 30: 43-54. 

Faulkner, S., 2004. Soils and sediment: understanding wetland biogeochemistry, In Wetlands (Spray, S.L. and 

McGlothlin, K.L., Eds), Rowman & Littlefield: Lanham, MD, pp. 30-54. 

Mäckel, R., 1974. Dambos: a study in morphodynamic activity on the plateau regions of Zambia, Catena, 1: 327-

365. 

Mäckel, R., 1985. Dambos and related landforms in Africa – an example for the ecological approach to tropical 



geomorphology, In Dambos: small channelless valleys in the tropics (M.F. Thomas and A. S. Goudie, 

Eds.), Zeitscrift für Geomorphologie, 52: 63-86. 

Ripley, B.D., 2006. The tree package (Version 1.0-24) [Computer software]. The R Project for Statistical 

Computing. Retrieved September 20, 2006. Available from 

http://www.r-project.org/

.  


Rodhe, H., 1990. A comparison of the contribution of various gases to the greenhouse effect, Science, 248: 1217-

1219. 


Rouse, J.W., Haas, R.H., Schell, J.A., and Deering, D.W., 1973. Monitoring vegetation systems in the Great Plains 

with ERTS, Third ERTS Symposium, NASA SP-351, 1, 309-317. 

Shindell, D.T., Walter, B.P. and Faluvegi, G., 2004. Impacts of climate change on methane emissions from 

wetlands, Geophysical Research Letters, 31: L21202. 



Von der Heyden, C.J., 2004. The hydrology and hydrogeology of dambos: A review, Progress in Physical 

Geography, 28: 544-564. 

Document Outline

  • Next Page
  • Previous Page
  • ========================
  • Table of Contents
  • Author Index
  • Exhibitors
  • Copyright
  • ============================
  • Print

Yüklə 64,18 Kb.

Dostları ilə paylaş:




Verilənlər bazası müəlliflik hüququ ilə müdafiə olunur ©genderi.org 2024
rəhbərliyinə müraciət

    Ana səhifə