Topographic wetness indices (TWIs) calculated from digital elevation
models (DEMs) are meant to predict relative landscape wetness and
should have predictive power for soil and vegetation
attributes. While previous researchers have shown cumulative TWI
distributions shift to larger values as DEM resolution decreases,
there has been little work assessing how DEM scales affect TWI
spatial distributions and correlations with soil and vegetation
properties. We explored how various DEM resolutions (2, 5, 10, 20,
30, and 50
The TOPMODEL topographic wetness index and its derivatives (TWIs) are intuitively attractive to hydrologists because they elegantly encapsulate our beliefs about variable source area (VSA) behavior (Dunne and Black, 1970; Dunne et al., 1975; Hewlett and Hibbert, 1967) and the spatial variation of soil moisture. In turn, landscape position and soil moisture exert first order control on soil properties (e.g. texture, organic matter, chemistry) and vegetation characteristics, so we expect these characteristics to correlate with TWI values (e.g. Florinsky et al., 2002, 2004; Sariyildiz et al., 2005; Seibert et al., 2007). Under the assumption that relative wetness should be a major determinant of soil characteristics, we used both quantitative and qualitative techniques to investigate the correspondence of soil and vegetation characteristics to relative wetness as predicted by the TOPMODEL TWI calculated at fine to coarse DEM resolutions.
Previous researchers have shown that distributions of TWIs and their components are highly sensitive to scale (e.g. Zhang and Montgomery, 1994; Quinn et al., 1995; Kienzle, 2004), consistently demonstrating that TWI distributions shift to higher values as the DEM grid size increases. However, the question of how DEM resolution affects the spatial distribution of TWI values has been largely ignored. We investigated how the spatial arrangement of TWI values varied with DEM resolution. The question of appropriate DEM scale has become more critical since the advent of Light Detection and Ranging (LiDAR) data now makes high-resolution DEMs an easy choice when LiDAR is available. Our interest was in seeing how high-resolution DEMs changed our perceptions of relative landscape wetness predicted by TWI values.
The application of TWIs to explain the spatial distribution of moisture and
soil and vegetation characteristics has encountered problems of several
types. Efforts to determine optimal DEM resolution have found the answer
depended on the terrain attribute investigated, but was generally within the
range of 5–20
The TOPMODEL topographic wetness index is defined as the natural log of the
ratio of upslope contributing area per unit width to local slope (Beven and
Kirkby, 1979). We derived TWI values from a high resolution LiDAR dataset of
the Savannah River Site (SRS) subsampled to 2, 5, 10, 20, 30, and
50
Because of these characteristics, this is an unusual landscape for evaluating the TWI concept. The TOPMODEL TWI intrinsically assumes lateral slope-parallel flow is an important process distributing water in the landscape. If interflow is not a dominant hydrologic process, uncertainty exists whether TWIs should be predictive of hydrologic responsiveness, soil characteristics, and vegetation characteristics. For example, in the highly infiltrative chalk geology of southern England, Blyth et al. (2004) found very little correlation between TWI values and soil wetness. This study is both an evaluation of DEM scale on the correspondence of TWI values with soil and landscape characteristics and also an evaluation of the applicability of these indices to a landscape in which interflow is not a dominant process.
The specific objectives of this study were to investigate the effects of DEM scale upon (1) spatial arrangements of TWI values, (2) the correlation of TWI values with point measurements of soil carbon and nitrogen content, (3) and the relationship between TWI values and distributions of landscape characteristics associated with relative wetness, including: Natural Resources Conservation Service (NRCS) soil maps, depths to groundwater, and vegetative cover. We randomly sampled hillslope and riparian soils on hillslope grids and riparian transects and analyzed the samples for carbon (C) and nitrogen (N) content. We then evaluated these soil characteristics against TWI values calculated at each scale to determine the degree of correlation between soil characteristics and TWI. We compared the distribution of TWI maps at various resolutions to hydric soil map units of NRCS soil series, riparian vegetation polygons, and depths to groundwater. We also evaluated the variation of TWI values along the ephemeral and intermittent stream network. These evaluations and visual inspection of the spatial arrangement of TWI values through maps reveals DEM resolution effects not shown by analyzing the aggregate distributions of TWI values.
The forested hillslopes in the study area are managed for timber production, so in this case the vegetative cover distribution does not represent climax vegetated conditions partly controlled by soil wetness, but rather incorporates past planting and management reflecting land manager perceptions of relative wetness and soil productivity as well as best management practice guidelines for leaving undisturbed vegetation around streams and wetlands. For example, pines are planted and managed on hillslopes and plateaus that are expected to be drier while unmanaged or lightly managed mixed hardwood forests are left on the wetter valley floors. Wetlands are left unmanaged. Thus, our tests of relative wetness are quantitative (correlations with point soil characteristics), qualitative (comparisons with mapped soil and landscape attributes), and partly sociological (comparisons with vegetation distributions whose management is dictated by human perceptions of wetness).
The study area is a first-order intermittent stream basin measuring
approximately 1
Topsoils in the study area are primarily sand to sandy loam in texture, and
they are underlain by an argillic sandy clay loam that occurs usually between
0.5 and 1.5
A 1
Distributions of the TWI values for the entire basin were computed
using JMP 9.0.0 (SAS Institute Inc., Cary, NC, USA). Quantile metrics
were used to quantify key differences among the distributions. There
were over 200 000 cells in the basin at the 2
We visually compared the maps of calculated TWI values of various resolutions to the maps of NRCS-mapped soil units especially with the hydric soil types, USFS-delineated vegetation types, and also depth to groundwater using kriged well data. Soil pedology should vary across landscapes partly in response to relative wetness. For example, the NRCS soil map units in the riparian zone and depressions (wetlands) show poorly drained, frequently flooded Pickney sand, Rembert sand and Ogeechee sand units (Fig. 2a). These map units are identified as hydric sandy soil with very high hydraulic conductivity value that occupies a narrow stretch along the river section and the depressions (wetlands). The growth and management of vegetation is also partly dictated by relative wetness, distance from mapped drainage ways and wetlands, and site productivity. For example, according to the vegetation distribution, the valley floors are dominated by sweet gum and yellow poplar hardwoods (Fig. 2b).
A depth-to-groundwater map was created in several steps. First
a snapshot of groundwater level was taken on 18 October 2012 from
measurements of 16 wells located in and around the study basin (
In order to compare the distribution of various TWI resolutions with
depth to groundwater maps, we created depth to groundwater maps having
corresponding horizontal grid sizes that matched with the resolutions
of TWIs. The effects of resolution on TWI were evaluated based on how
the distributions in TWI maps were correlated to the corresponding
depth to groundwater maps. Within the hydric and streamside polygons,
the distributions of each TWI resolutions were classified into five
classes. The classifications of the TWI values were based on their
quantile distribution:
To quantify the effect of DEM resolution on the characterization of
valley pixels, we defined the stream channel pixels and valley thalweg
pixels (above the point of channel initiation) and assessed the
variation of valley TWI values among the 2, 10, and 20
Soil samples were collected from two
In July 2010, a soil sample was extracted from the A horizon sampling
point to a depth of 7.5
Elemental concentrations for each sample unit were transformed to
content (
DEM resolution strongly affected the distribution of TWI values and
quantiles of TWIs calculated over the study basin (Table 1). Minimum,
and median TWI values were smallest at the highest resolution and
increased as DEM resolution decreased. There was nearly a three-fold
difference in minimums between the 2 and 50
The spatial distribution of relative wetness values (Fig. 3) were
strongly affected by DEM resolution. Most striking, at the finer
resolutions, valleys disappeared as the TWI values of most valley
pixels were defined by the local microtopography and were insensitive
to the larger drainage area flowing to the valley at that point. Thus,
2 and 5
Depths to groundwater within the basin vary from 15
Figure 6 shows the effect of TWI resolution based on the distribution of the
various wetness classes within the hydric soil and streamside management
polygons. Across all resolutions, the proportion of relatively wet cells is
dominant in hydric soil polygons. This is consistent with the fact that soils
vary across landscape in response to relative wetness. The wet cell
proportion significantly increased as the resolution coarsened (Fig. 6a). The
proportion of wet cells increased from about 44 % at 2
The streamside vegetation polygon reflects the legacy of past forester
decisions about where soil conditions are too wet for growing pine
trees or the minimum allowable distance between aquatic habitat
(defined streams and wetlands) and timber harvest allowed by USDA
Forest Service management guidelines. The land managers at this site
plant pine trees on the hillslopes where soils are drier and favorable
to pine growth, and they leave hardwoods growing in the bottomlands
and near streams and wetlands. Because of the riparian buffer
requirements, the streamside vegetation polygons are generally wider
than the hydric soil polygons and contain regions of relatively drier
conditions. Figure 6b shows coarsening TWI resolution increases the
proportion of wet cells and decreased the proportion of medium wetness
cells. This was similar to the change observed within the hydric soil
polygons, but the effect was not as strong. The proportion of wet
cells increased from 24 % at 2
Variability of TWI values along the stream network decreased with
decreasing DEM resolution (Fig. 7). For 2
Graphing point-scale mineral soil C and N contents against TWI values
interpolated from the resident and nearest pixels, as well as against pixel
TWI values, indicated that the form and direction of the relationship
depended on DEM scale (Figs. 8 and 9). At the finest DEM scale of
2
The finer scale sampling used in parts of the variably-spaced sampling grid
resulted in multiple points located in the same pixels at the larger DEM
scales (Fig. 8b). These points also indicated substantial and not surprising
variability in soil chemistry over the scale of even 5
The TOPMODEL TWI assumes that water accumulation and routing is driven by
shallow slope-parallel subsurface flow. The evaluation of landscape
attributes against the spatial arrangement of TWI values at different scales
suggests that the scales at which shallow subsurface processes drive soil
wetness are not as fine as the scale of surface microtopography, and that the
optimal scales for estimating relative wetness range from about 20 to
30
We did not select this watershed specifically to test TWI values against other landscape metrics associated with soil moisture. We were already using LiDAR data to study this watershed's hydrology, and we knew from previous work that TWI distributions varied with DEM resolution, but we were surprised to find that not only did distributions change, but spatial patterns changed substantially. These spatial patterns reveal that fine scale topography may poorly predict site conditions when soil moisture is driven by much larger scale groundwater processes. Unfortunately, this watershed is managed, so vegetation distributions reflect not just landscape controls but also management decisions. Furthermore, the soils have been previously disturbed, so topsoil characteristics may partially reflect legacy effects of management. Ideally, a future study would examine the relationships between TWI distributions, vegetation distributions, and soil characteristics in relatively undisturbed watersheds varying in dominant hillslope flow pathways. Despite these limitations, this study demonstrates that DEM scale strongly affects inference generated from TWI distributions.
The spatial distribution of TWI values across the landscape, and
the inference generated therefrom, is highly sensitive to DEM scale. We
documented a migration of the bulk of the wettest cells from higher to lower
elevation as resolution decreased and a change in the thickness and number of
high TWI threads leading from the ridges to the valleys. Higher resolution
DEMs mapped thinner and more numerous drainage threads while coarser
resolutions mapped wider and fewer drainage threads. Essentially, valleys
disappeared at the finest resolutions while the drainage network became very
abstract at the coarsest 50
Consistent with the findings of previous researchers, cumulative distributions of TWI values consistently shifted as DEM resolution changed. As DEM resolution coarsened, the median value and minimum TWI values increased and the range of predicted TWI values decreased. This was also observed by Zhang and Montgomery (1994) and Kienzle (2004). However, each of these studies observed different relative sensitivities of the TWI distributions to DEM scale. When evaluated in concert with previous studies, the investigation indicates the sensitivity of TWI distributions to DEM resolution depends on terrain characteristics.
The correspondence between TWIs and depths to groundwater is very
sensitive to DEM resolution. The study showed that TWI and depth to
groundwater are more correlated at lower resolution DEMs. Visual
observation of the distribution of TWI map and its correlation with
depth to groundwater map showed that the 20 and 30
The distribution of TWI values in the hydric soil showed a significant
increase of wetness with decreasing resolution. Even at high
resolution (2
Prior to the advent of LiDAR, the choice of DEM resolution was largely a moot question, as the effort to create high resolution DEMs was prohibitive for most projects. Previous assessments of the effects of DEM scale on topographic indices relied on laborious photogrammatic interpretation. With available high resolution LiDAR, the creation of high resolution DEMs for this study was easy, but not necessarily beneficial, depending on the desired analysis. The results indicate that LiDAR is not a panacea for hydrologic assessments but rather raises questions about integration of different scales relevant to different watershed characteristics and processes.
These results illustrate that considerations of scale effects must be made when choosing a DEM resolution for characterizing relative soil wetness and hydrologic sensitivity. The spatial and elevational distributions of high TWI values (expected to be variable source areas) were highly sensitive to the choice of resolution, an issue that must be considered when applying the TOPMODEL TWI in modeling and land-use planning applications.
This project was funded by the U.S. Department of Energy's Bioenergy Technology Office and the Department of Energy-Savannah River Operations Office through the U.S. Forest Service Savannah River under Interagency Agreement DE-AI09-00SR22188. John Blake of the USFS provided substantial assistance with site logistics and data layer acquisition. Also, we appreciate the support of our research associates Louise Jacques, Bob Bahn, Meg Williamson, Ben Morris, and Erin Harris.
Quantile distribution characteristics across resolutions. Sample sizes
of each resolution were: 2
Above: Fourmile Branch of Savannah River in South Carolina, with the general study area shaded in inset. Below: hillshaded map of the study area. The basin boundary as delineated by hand is depicted as a solid black line. Road F runs N–S on the right side of the hillshade figure.
Spatial distribution of TWI values for
Depth to groundwater table
Correlation between spatial distribution of TWI values and depth to
groundwater across the study watershed for 2, 5, 10, 20, 30, and 50
Fractional distribution of TWI classes in the hydric soil
polygon
Variation of TWI values along the ground-truthed stream
network for 2, 10, and 20
Natural log of carbon contents from point samples of mineral soils
(0–7.5