RSE-07152; No of Pages 13
ARTICLE IN PRESS
Remote Sensing of Environment xxx (2008) xxx-xxx
Contents lists available at ScienceDirect
Remote Sensing of Environment
j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / r s e
HAND, a new terrain descriptor using SRTM-DEM: Mapping terra-firme rainforest
environments in Amazonia
Camilo Daleles Rennó a,⁎, Antonio Donato Nobre b, Luz Adriana Cuartas a, João Vianei Soares a,
Martin G. Hodnett c, Javier Tomasella a,b, Maarten J. Waterloo c
a
b
c
Instituto Nacional de Pesquisas Espaciais, Av. Astronautas, 1758, São José dos Campos, SP, 12227-010, Brazil
Instituto Nacional de Pesquisas da Amazonia, Escritório Regional do INPA, INPE Sigma, Av. dos Astronautas, 1758, São José dos Campos, SP, 12227-010, Brazil
Vrije Universiteit Amsterdam, De Boelelaan 1105, 1081 HV Amsterdam, The Netherlands
A R T I C L E
I N F O
Article history:
Received 29 May 2007
Received in revised form 24 March 2008
Accepted 29 March 2008
Available online xxxx
A B S T R A C T
Optical imagery can reveal spectral properties of forest canopy, which rarely allows for finding accurate
correspondence of canopy features with soils and hydrology. In Amazonia non-floodable swampy forests can
not be easily distinguished from non-floodable terra-firme forests using just bidimensional spectral data.
Accurate topographic data are required for the understanding of land surface processes at finer scales.
Topographic detail has now become available with the Shuttle Radar Topographic Mission (SRTM) data. This
new digital elevation model (DEM) shows the feature-rich relief of lowland rain forests, adding to the ability
to map rain forest environments through many quantitative terrain descriptors. In this paper we report on
the development of a new quantitative topographic algorithm, called HAND (Height Above the Nearest
Drainage), based on SRTM-DEM data. We tested the HAND descriptor for a groundwater, topographic and
vegetation dataset from central Amazonia. The application of the HAND descriptor in terrain classification
revealed strong correlation between soil water conditions, like classes of water table depth, and topography.
This correlation obeys the physical principle of soil draining potential, or relative vertical distance to
drainage, which can be detected remotely through the topography of the vegetation canopy found in the
SRTM-DEM data.
© 2008 Elsevier Inc. All rights reserved.
1. Introduction
Tropical terrain covered by rainforest presents rich mosaics of very
distinctive environments, often hidden from remote view. The
overwhelming challenge of describing 5.5 million km2 of such
environments and associated dense, tall and closed-canopy vegetation
in Amazonia has made its complete inventory a seemingly impracticable task. Passive optical remote sensing imagery (such as Landsat and
CBERS) can reveal spectral properties of forest canopy with detail (e.g.
Wulder, 1998), but rarely allows for finding accurate correspondence
of canopy features with soils and local hydrology. In Amazonia even
seasonally flooded tropical forests could not be easily spotted and
distinguished from non-floodable terra-firme forests using this type of
data (Novo et al., 1997). The usually flat optical imagery can hint at
relief through either bright or shadow reflection artifacts on the slope
pixels of steeper areas, or where spectral signatures of the vegetation
are distinctive because of local environmental effects (Guyot et al.,
1989; Novo et al., 1997; Nobre et al., 1998), but without stereo images
it lacks the ability to describe relief quantitatively. Optical stereo
⁎ Corresponding author. Tel.: +55 12 39456521; fax: +55 12 39456468.
E-mail address: [email protected] (C.D. Rennó).
images (e.g. obtained from ASTER or SPOT) can be used to produce
digital elevation models (DEM). However, the cost and difficulty of
obtaining cloud-free coverage for many areas of the world, compounded with the requirement of sun angles below 25° (from Nadir)
to avoid long shadows (Jacobsen, 2003), has limited the possibilities
for producing DEMs of large, continuous areas (Hirano et al., 2003). In
addition, because ground truth accessibility to large areas is limited,
many aspects of landscape complexity in those vast tropical surfaces
remain shrouded in mystery.
However, some of these drawbacks are quickly being overcome
with imagery from active space borne sensors, such as synthetic
aperture radars (SAR). Canopy penetrating L-Band SAR imagery from
the Japanese Earth Resources Satellite (Siqueira et al., 2000) revealed
with unprecedented detail patches of flooded vegetation (Barbosa
et al., 2000; Melack and Hess, 2004). Careful mapping of such
previously hidden seasonally flooded biomes has suggested their
occurrence over a far wider area than formerly believed (Siqueira
et al., 2003; Hess et al., 2003). Accurate topographic data are required
for the understanding of land surface processes at finer scales.
Topographic detail has now become available on a global scale
through the C-Band SAR imagery (van Zyl, 2001) of the Shuttle Radar
Topographic Mission (SRTM). The spaceborne SRTM circled the globe
over a wide swath covering all the tropics and more, generating radar
0034-4257/$ – see front matter © 2008 Elsevier Inc. All rights reserved.
doi:10.1016/j.rse.2008.03.018
Please cite this article as: Rennó, C. D., et al., HAND, a new terrain descriptor using SRTM-DEM: Mapping terra-firme rainforest environments
in Amazonia, Remote Sensing of Environment (2008), doi:10.1016/j.rse.2008.03.018
ARTICLE IN PRESS
2
C.D. Rennó et al. / Remote Sensing of Environment xxx (2008) xxx-xxx
data that allowed for the digital reconstruction of the surface relief,
producing the DEM. The SRTM-DEM data, with a horizontal resolution
of 3” (~ 90 m near the equator) and a vertical resolution of 1 m,
constitutes the finest resolution and most accurate topographic data
available for most of the globe. Detailed information on the accuracy
and performance of SRTM can be found in Rodriguez et al. (2006). In
contrast to the passive optical imagery, this new DEM shows the
feature-rich relief of lowland rain forests, adding to the ability to
identify and map rain forest environments through many quantitative
terrain descriptors.
A range of topographic algorithms are available, which allow
various quantitative relief features to be obtained from the DEM. Slope
and aspect (e.g. Jenson and Domingue, 1988), and drainage network
and catchment area (e.g. Curkendall et al., 2003) are a few classical
descriptors. A range of hydrological parameters such as superficial
runoff trajectories, accumulated contributing area and groundwater
related variables (e.g. Tarboton, 2003) add to the suite of relief
descriptors. Relief shape parameters such as curvatures and form
factors can also be calculated (Valeriano et al., 2006). The third
dimension in a DEM, height, is obviously the key parameter, used to
some degree in the derivation of all of the previously mentioned
descriptors. Absolute height (above sea level — ASL) can be used on its
own as a relief descriptor, as large scale geomorphologic features tend
to be associated with altitude relevant geological control (Goudie,
2004). Upon flooding a given catchment for hydro dam development,
for example, the height ASL is the descriptor that will predict the reach
of the impoundment. However, when local environments in the fine
scale relief are considered, height ASL has little, if any, descriptive
power. As a result, local scale environments, although of conspicuous
importance and clearly defined by characteristic terrain topography
that is clearly visible on the SRTM-DEM, have not so far had a good
descriptor.
In this paper we present the development of a new quantitative
topographic algorithm based on SRTM-DEM data. We crafted and
tested the terrain descriptor, applying it for a groundwater, topographic and vegetation dataset from central Amazonia, using ground
calibrated terrain classes for mapping the study area.
Defining G as a set of pairs of Cartesian coordinates of a grid with c
columns and r rows,
G ¼ fhi; jiji ¼ ½1; coN; j ¼ ½1; r oNg;
ð1Þ
we can represent LDD as a function that associates to each grid point
hi,ji the flow direction LDD(hi,ji) which can assume a value according
to its orientation: N, NE, E, SE, S, SW, W, NW or null. The null value is
assigned to all points with undefined flow direction (pits).
The real hydrological meaning of the LDD depends on the quality of
the DEM. The C Band interacts strongly with the vegetation with the
result that the actual topography represented in the SRTM data is
roughly that of the upper canopy (Valeriano et al., 2006). Therefore,
for areas where the soil surface is covered by dense or tall vegetation it
must be expected that a variable degree of relief masking occurs in the
SRTM data (Kellndorfer et al., 2004), producing pits and extensive
unresolved flat areas. Some of these features can be real properties of
the relief, but often they represent artifacts in the data. SRTM data,
perhaps because of radar speckle (noise) or vegetation effects, have
more spurious points than other DEMs (Curkendall et al., 2003).
Besides, forests have characteristic sylvigenetic dynamics with a
relatively high occurrence of tree gaps (Oldeman, 1990), which will
appear in the SRTM-DEM as depressions. Such depressions, if they
occur on a stream, for example, create false interruptions in the
topological continuity of that drainage (apparent impounding).
Another particular feature of SRTM data is related to abrupt
transitions of vegetation types or land uses where vegetation is
suddenly absent, revealing the soil surface in a patchy manner.
According to Lindsay and Creed (2005), the depression artifacts
arising from underestimation of elevation should be filled and the
features caused by elevation overestimation should be corrected by
breaching. In general, filling methods involve raising the inner area of
a depression to the elevation of its outlet point, defined as a point
through which water could leave the depression. The outlet is usually
2. Algorithm development
2.1. Conditioning procedures
The new descriptor algorithm requires a hydrologically coherent
DEM as input, with resolved depressions (sinks), computed flow
directions for each grid point and a defined drainage network. The
procedures to develop these are presented below.
2.1.1. Fixing DEM topology and computing flow directions
Topography is a hydrologic driver since it defines the direction and
speed of flows. Flow directions define hydrological relations between
different points within a basin. Topological continuity for the flow
directions is therefore necessary for a functional drainage to exist.
Hydrological connections by flow direction between two points on a
surface are not the same as those based on Euclidian distances. As seen
in Fig. 1, point A is spatially closer to C, but it is hydrologically connected
to point B because superficial water (runoff) will flow towards the latter.
The flow directions can be represented using different approaches
(Zhou and Liu, 2002). For a DEM represented by a grid, the simplest
and most widely used method for determining flow directions is
designated D8 (eight flow directions) initially proposed by O'Callaghan and Mark (1984). In this method, the flow from each grid point is
assigned to one of its eight neighbors, towards the steepest downward
slope. The result is a grid called LDD (Local Drain Directions), whose
values clearly represent the link to the downhill neighbor. A pit is
defined as a point none of whose neighbors has a lower elevation. For
a pit, the flow direction is undefined.
Fig. 1. Hydrological connection between points A and B. Red line indicates the profile
shown above, blue lines represent the drainage network and the arrows indicate flow
direction. B and C are points on streams. (For interpretation of the references to colour
in this figure legend, the reader is referred to the web version of this article.)
Please cite this article as: Rennó, C. D., et al., HAND, a new terrain descriptor using SRTM-DEM: Mapping terra-firme rainforest environments
in Amazonia, Remote Sensing of Environment (2008), doi:10.1016/j.rse.2008.03.018
ARTICLE IN PRESS
C.D. Rennó et al. / Remote Sensing of Environment xxx (2008) xxx-xxx
defined as the lowest point along the border of the depression basin.
On the other hand, breaching methods create an artificial channel
lowering some points across a divide in order to connect two neighboring depressions.
In this work, both sink and flat areas were eliminated by using a
process similar to depression breaching suggested by O'Callaghan and
Mark (1984), Band (1986) and Jenson and Domingue (1988). Although
depression breaching can create strong anomalies at the edge of the
artificial drainage channels, it produces the least amount of change on
most of the remaining points of the DEM. Fig. 2 exemplifies the
approach used in the DEM fixing process. The first step is to delimit
the sink area (or closed basin) to be corrected based on the LDD
obtained from the original DEM. Note that there are two sink areas in
this example (52 and 50 elevation, the pits in darker hues). Next, the
outlets of these sinks, generally falling on a saddle in the relief, are
3
identified (here, 55 for the first pit). This outlet point will be the
neighbor of the lowest point along the limits of the adjacent sink area
(54). A topological trajectory passing through the outlet and
connecting the two pits is then generated. Taking into account the
distance between the two extremities, the new elevation of every
point along this trajectory for cutting the hill or breaching is calculated
by a linear interpolation. The result is a new functional flow path that
eliminates one sink. Finally, a coherent new LDD is obtained based on
this corrected DEM.
2.1.2. Computing drainage network
Many methods of automatic extraction of drainage networks from
DEMs have been developed. Soille et al. (2003) grouped the drainage
extraction algorithms into two classes, the first based on morphological features and the second on hydrological terrain characteristics.
Fig. 2. DEM fixing process. Red arrows indicate flow directions changed during correction. (For interpretation of the references to colour in this figure legend, the reader is referred to
the web version of this article.)
Please cite this article as: Rennó, C. D., et al., HAND, a new terrain descriptor using SRTM-DEM: Mapping terra-firme rainforest environments
in Amazonia, Remote Sensing of Environment (2008), doi:10.1016/j.rse.2008.03.018
ARTICLE IN PRESS
4
C.D. Rennó et al. / Remote Sensing of Environment xxx (2008) xxx-xxx
where F− 1(hi, ji) represents a set of all neighbors of point hi, ji that flow
to it, that is
Table 1
Δ Function definition
Flow direction
Δ (Flow direction)
N
NE
E
SE
S
SW
W
NW
Null
h0,−1i
h1,− 1i
h1,0i
h1,1i
h0,1i
h−1,1i
h−1,0i
h−1,−1i
h0,0i
F 1 ðhi; jiÞ ¼ fhk; liaN ðhi; jiÞjF ðhk; liÞ ¼ hi; jig
Montgomery and Dietrich (1988) discussed the criteria to predict the
point where channels should begin (headwater). One of the first
procedures for delineating drainage networks (O'Callaghan and Mark,
1984) used a minimum contributing area threshold to identify
channel beginning. The drainage network is thus defined by those
grid points that have a contributing area greater than a given
threshold. The contributing area is computed by counting the number
of points whose flow paths converge to the considered point.
Considering a given point h i,j i, we can determine its neighborhood
as
Nðhi; jiÞ ¼ fhk; liaGjk ¼ fi 1; i; i 1g; l ¼ fj 1; j; j þ 1g; hk; liphi; jig
ð2Þ
The points in N(hi, ji) are called the neighbors of h i,j i. If pointh i,j i is
not on the border grid (i ≠ 1, i ≠c, j ≠ 1 and j ≠r), then it has 8 neighbors,
that is
N ðhi; jiÞ ¼ fhi 1; j 1i; hi 1; ji; hi 1; j þ 1i; hi; j 1i; hi; j þ 1i;
hi þ 1; j 1i; hi þ 1; ji; hi þ 1; j þ 1ig
ð3Þ
otherwise, it will have less than 8 neighbors.
Based on the D8 method, the point hi,ji can be hydrologically
connected to only one of its neighbors. If F(hi,ji) is the point to which hi,ji
flows, it can be defined as:
F ðhi; jiÞ ¼ hi; ji þ DðLDDðhi; jiÞÞ
ð4Þ
where Δ is a function that represents the relative position hΔi,Δji
between two hydrologically connected points. Table 1 shows the
correspondence between the flow directions and the relative positions. Note that if point hi, ji is a pit, then F(hi, ji) = hi, ji.
The contributing area of point hi, ji, defined as A(hi, ji), is computed
through an iterative process.
Let At(hi, ji) be the contributing area of the point hi, ji in the tth
iteration. In the first iteration (t = 1)
A1 ðhi; jiÞ ¼
1
0
ifhi; jigF ðGÞ
otherwise
ð5Þ
where F(G) represents the set of all upward points connected to any
given point, that is
F ðGÞ ¼ fhk; liaGjahi; jiaN ðhk; liÞjF ðhi; jiÞ ¼ hk; lig:
ð6Þ
In this first iteration, the contributing area of all points that initiate
a flow path is equal to one.
For the other iterations (t N 1)
8 t1
ðhi; jiÞ
<A P
At ðhi; jiÞ ¼ 1 þ F −1 ðhi;jiÞ At1 ðhk; liÞ
:
0
if At1 ðhi; jiÞp0
if 0gAt1 F 1 ðhi; jiÞ
otherwise
ð7Þ
ð8Þ
The iteration process stops (t=n) when, for any point hi,ji, At − 1(hi,ji)≠0,
that is, A(hi,ji)=An(hi, ji).
The method based on the contributing area is very easy to
implement and therefore widely used. Tarboton (2003) pointed out
that a significant question with this method is the choice of
contributing area threshold. The basic hypothesis is that channel
heads, where there is a transition from convex to concave profiles, is
also where concentrated fluxes begins to dominate over diffusive
fluxes (Tarboton et al., 1991, 1992). Some authors have called for
objective criteria to define channel heads, such as the relationship
between slope and contributing area (Tarboton et al., 1991). Others use
local curvatures to account for spatially variable drainage densities
(Tarboton and Ames, 2001). Montgomery and Foufoula-Georgiou
(1993) compared the constant contributing area with the slopedependent contributing area. Although these approaches represent
progress towards the automatic determination of channel heads, the
tests we conducted for Central Amazonia showed that the application
of these methodologies to SRTM data did not estimate drainage
density properly. We suspect that vegetation masking ground
topography may be the explanation. For this reason, we decided to
use the contributing area threshold as one of the main criteria to
define the drainage network, validating channel heads with field data.
To define the channel heads we used an additional criteria based on
simplifications of horizontal and vertical geomorphic curvatures.
Firstly, the channel head element should represent a convergent point,
meaning it should have two or more overland flux paths converging to
it (horizontal curvature). Secondly, the channel profile should be
concave, i.e. where the channel head profile point has a smaller
change of elevation than the mean of the elements located uphill and
downhill of it (vertical curvature).
2.2. The height above the nearest drainage terrain descriptor
To quantify relevant parameters that could uniquely identify generalizable spatial properties of hill slopes there was the need to have a local
frame of variable topographic reference that should be more useful than
the all encompassing and generic height ASL, or the third dimension in
the DEM. Horizontal distances of DEM grid points to connected drainage
channels (slope length) have been computed by a number of approaches
(e.g. Tucker et al., 2001). As the initial interest in this study was to predict
potential hydrological properties for each DEM grid point, especially the
depth to the permanently saturated zone, this approach wasn't useful.
The gravitational potential energy difference between any given grid
point and the other extremity of the hill slope flow path, at the functional
stream outlet (explicit in the LDD), defines a unique and permanent
property of that grid point that we call draining potential. The vertical
distance of a given grid point to its drainage outlet (which is
hydrologically important) can in most cases be expressed as a relative
height, or the height difference between those points. Thus there will be
no gravitationally driven water movement between two hydrologically
connected points that share the same height. Classifying all grid points
according to their respective draining potentials allows them to be
grouped into classes of equipotential (equivalent draining gravitational
potential), defining environments or zones with inferred similar
hydrological properties. The linking of every grid point to its outlet on
the drainage system allows for the whole DEM to be normalized for the
drainage network (adjusting point heights in relation to the drainage),
which in this way becomes a distributed frame of topographic reference.
If D is a set of drainage network points, identified by a number
through a bijective function in order that each point of the drainage
Please cite this article as: Rennó, C. D., et al., HAND, a new terrain descriptor using SRTM-DEM: Mapping terra-firme rainforest environments
in Amazonia, Remote Sensing of Environment (2008), doi:10.1016/j.rse.2008.03.018
ARTICLE IN PRESS
C.D. Rennó et al. / Remote Sensing of Environment xxx (2008) xxx-xxx
5
Fig. 3. Procedure to calculate the HAND grid. BBBBlue squares represent grid points belonging to the drainage network. Only black arrows are considered as flow paths.
network be identified by a unique identification number k a N⁎. We
define that
Dðhi; jiÞ ¼
k if hi; ji is a drainage network point
0 otherwise
ð9Þ
and its inverse function as
D1 ðkÞ ¼ fhi; jiaGjDðhi; jiÞ ¼ kg:
ð10Þ
Following respective flow paths, each and every point in the grid is
necessarily connected to a drainage point. Let I(hi, ji) be the function
that identifies the drainage point connected to the point hi, ji. This
function is computed through an iterative process and the result is a
grid that associates the identification number of the drainage point
that each point is connected to. If It(hi, ji) is the drainage identification
number of the point hi, ji in the tth iteration. In the first iteration (t = 1),
1
I ðhi; jiÞ ¼ Dðhi; jiÞ:
ð11Þ
For the other iterations (t N 1)
8
< It1 ðhi; jiÞ
t
I ðhi; jiÞ ¼ It1 ðF ðhi; jiÞÞ
:
0
if I t1 ðhi; jiÞp0
if I t1 ðF ðhi; jiÞÞp0
otherwise:
ð12Þ
The iteration process stops (t=m) when, for any point hi, ji, It(hi,ji)≠0,
that is,
Iðhi; jiÞ ¼ Im ðhi; jiÞ:
ð13Þ
Assuming that all points belong to a flow path and that all flow
paths are associated to respective drainage points, we define the
HAND value of any given point hi, ji as
HANDðhi; jiÞ ¼
H ðhi; jiÞ HD ðhi; jiÞ if HD ðhi; jiÞbHðhi; jiÞ
0
otherwise
ð14Þ
where H(hi, ji) represents the height of the point hi, ji given by the
original DEM and HD(hi, ji) is the height of drainage point hydrologically connected to point hi, ji following the flow path, that is,
HD ðhi; jiÞ ¼ H D1 ðI ðhi; jiÞÞ :
ð15Þ
It is important to note that if the point i,j is a point belonging to
the drainage, then
Iðhi; jiÞ ¼ Dðhi; jiÞ
ð16Þ
D1 ðIhi; jiÞ ¼ hi; ji
ð17Þ
HD ðhi; jiÞ ¼ H ðhi; jiÞ
ð18Þ
and so
HANDðhi; jiÞ ¼ Hðhi; jiÞ H ðhi; jiÞ ¼ 0:
ð19Þ
In other words, all grid points belonging to the drainage network
are zeroed in height, which implies that the draining potential
(according with the HAND definition) along the stream channel is
disregarded. Although the drainage channel is also a flow path, which
effectively drains, the HAND descriptor uses the whole channel as a
flat relative topographic reference, the end of all non channel flow
paths. By definition, the HAND drainage outlet grid point is the nearest
draining point to itself, therefore it can only be subtracted from itself.
It is important to note that the breaching process, required to reach a
topologically sound and accurate drainage network, introduces occasional canyon like artifacts into the DEM, as a result of aberrant height
differences adjacent to the drainage network. These artifacts would be
transferred to the HAND grid if it were computed from the corrected
DEM. When using original SRTM data for the HAND grid computation,
some associations prescribed by the corrected drainage connection grid
result in small negative differences that could be interpreted as points
Please cite this article as: Rennó, C. D., et al., HAND, a new terrain descriptor using SRTM-DEM: Mapping terra-firme rainforest environments
in Amazonia, Remote Sensing of Environment (2008), doi:10.1016/j.rse.2008.03.018
ARTICLE IN PRESS
6
C.D. Rennó et al. / Remote Sensing of Environment xxx (2008) xxx-xxx
Fig. 4. Processing steps for the height above the nearest drainage descriptor.
below their associated drainage points (impoundment). Considering the
1 m elevation accuracy of the SRTM and the disturbances caused by the
forest canopy, these small negative differences fell within the noise in the
data, and were zeroed. Therefore, in the HAND algorithm we combined
the more meaningful need to reach a topologically coherent LDD
(requiring the topological correction of the DEM, leading to an accurate
drainage network), with the use of the original non-corrected DEM
(which has dispersed sinks) for the computation of the final HAND grid,
thus avoiding the canyon aberrations associated with the corrected DEM.
Fig. 3 shows an example of the HAND procedure. Based on a LDD
grid, overlaid with the computed drainage network, all flow paths are
coded according to the nearest associated drainage point. The marked
point with height 72 in the DEM is connected to the drainage point with
height 53 (both coded 2), resulting in a HAND of 19. This means that the
marked grid point is 19 m above its corresponding drainage point. Fig. 4
summarizes all operations involved in obtaining the HAND grid.
3. Application
3.1. Study area
Data from a 37 km square study area located in the Cuieiras Biological
Reservation (central Amazonia NW of Manaus, Fig. 6), was used to test
the new HAND terrain descriptor. The area, representative of large areas
in Amazonia and including the K34 LBA fluxtower site (Araujo et al.,
2002) and the Asu hydrological catchment (Waterloo et al., 2006;
Tomasella et al., 2007; Cuartas et al., 2007), is covered by pristine, terrafirme (terrain not subject to flooding by the annual flood cycle of the
Amazon and its major tributaries) rain forest vegetation, with the
canopy height varying from 20 m to 35 m. Forest biomass in the
vicinities is highly heterogeneous and has been reported ranging from
215 to 492 ton/ha (Laurance et al., 1999; Castilho, 2004). Details about
the wet equatorial seasonal climate (annual rainfall larger than
2000 mm) can be found in Araujo et al. (2002) and Cuartas et al.
(2007). The landscape in and around the Asu catchment (Fig. 5), is
composed mostly of flat plateaus (90–105 m ASL) incised by a dense
drainage network within broad swampy valleys (45–55 m ASL).
Ground truth data were collected in various field campaigns. A total
of 120 points were visited in the Asu catchment for testing the HAND
application. These points fell along a hydrological transect (site C1),
across two 1st order sub catchments, and along two 2.5 km-long N–S
and E–W orthogonal transects, crossing the catchment edge-to-edge.
Field positions of streams and stream heads were also logged for
verification of the calculated drainage network. Points beneath the
forest canopy were located in the field using accurate geo positioning
(obtained with a 30 m horizontal accuracy) in conjunction with
Please cite this article as: Rennó, C. D., et al., HAND, a new terrain descriptor using SRTM-DEM: Mapping terra-firme rainforest environments
in Amazonia, Remote Sensing of Environment (2008), doi:10.1016/j.rse.2008.03.018
ARTICLE IN PRESS
C.D. Rennó et al. / Remote Sensing of Environment xxx (2008) xxx-xxx
7
Fig. 5. Study area NW of Manaus. (a) Landsat TM 5,4,3 (RGB) 2001, (b) SRTM-DEM, showing rich details in the forest topography. Flood lands in the SRTM-DEM, identified using JERS-1
data (Melack and Hess, 2004), were removed from the analysis of this study (masked in black).
vegetation and soil pattern evaluation, aided by subsurface water table
depth data (multiyear time series, logged by an irregular sampling
network of 27 piezometers installed in valleys, major stream heads, and
along a hillslope hydrological transect). Hydrological details about the
Asu instrumented catchment can be found in Tomasella et al. (2007).
Non-floodable local environments were clearly identified in the field
through topography, vegetation, soils and hydrological cues.
3.2. Results
The SRTM data for the study area were corrected in order do
produce the hydrologically coherent LDD. Using this corrected LDD,
the contributing area grid was computed and with it the drainage
network was derived, using a selected contributing area threshold.
The HAND grid was then obtained using this drainage as reference.
The results of these steps are presented below.
3.2.1. Drainage network sensitivity
Drainage density is an important parameter for the accuracy of the
HAND descriptor. Fig. 6 shows networks extracted from the SRTM data
using three different contributing area thresholds. For the HAND
descriptor, the contributing area is the tunable parameter. Note that
the lower the threshold, the higher the density of the resulting drainage
network. Automatic extraction of the drainage network from a DEM still
lacks competence to represent channel heads realistically and with
robustness, which requires field verification for an appropriated
representation of drainage density. The ground truth for channel heads
carried out in this study indicated that the contributing area threshold of
50 grid points produces the most accurate drainage network density.
From an exploratory analysis of the relationship between drainage
density and contributing area threshold, we found that varying thresholds within the range of 47 through 92 did not change channel hierarchy
at a verification point in a 3rd order stream. Smaller or larger thresholds
produced higher or lower orders respectively. Fig. 7 shows the impact of
these three contributing area thresholds (47, 50 and 92), plus two chosen
extreme points (5 and 500) on the HAND grid distribution of heights.
The skewness in the HAND distribution of heights is directly proportional to the smoothness of the HAND grid. Higher frequencies of the
small HAND values, for example, result in a smoother topography of the
HAND grid, which implies a lower ability to distinguish and resolve
contrasting local environments. If the calculated drainage network
remains within the range that realistically captures the order hierarchy
Fig 6. Contributing area (left) and drainage networks plotted on SRTM image, with varying contributing area thresholds (in number of grid points).
Please cite this article as: Rennó, C. D., et al., HAND, a new terrain descriptor using SRTM-DEM: Mapping terra-firme rainforest environments
in Amazonia, Remote Sensing of Environment (2008), doi:10.1016/j.rse.2008.03.018
ARTICLE IN PRESS
8
C.D. Rennó et al. / Remote Sensing of Environment xxx (2008) xxx-xxx
Fig. 7. Sensitivity of the HAND grids based on different drainages with varying contributing areas.
of the drainage network, then the effect of slightly varying channel
heads on the HAND grid (and therefore on other estimations based on
it), will not be significantly great. In a forthcoming paper we report this
to be a robust result, verified for a wider area around the Asu catchment.
But for other areas with distinct geomorphologies this finding still needs
verification.
3.2.2. The HAND grid DEM
Fig. 8 shows a HAND grid shaded relief of the test area. Altimetric
differences along the drainage channels are no longer visible as they
were in the original DEM (Fig. 5b). The capacity of the HAND
descriptor to make evident locally significant terrain-controlling
factors is apparent.
Fig. 8. HAND grid DEM (using 50 grid points threshold for drainage contributing area).
Please cite this article as: Rennó, C. D., et al., HAND, a new terrain descriptor using SRTM-DEM: Mapping terra-firme rainforest environments
in Amazonia, Remote Sensing of Environment (2008), doi:10.1016/j.rse.2008.03.018
ARTICLE IN PRESS
C.D. Rennó et al. / Remote Sensing of Environment xxx (2008) xxx-xxx
9
Fig. 11. Box-plot of HAND for ground truth points: 36 points from the hydrological
transect, 84 spread out within the Igarapé Asu catchment, with water table inferred
from data of 27 piezometers.
Fig. 9. Profile comparing original SRTM data and the HAND grid normalized for the
drainage network.
The main difference between the original SRTM-DEM and the
HAND grid DEM (Fig. 9) lies in the effect of the topographic
normalization for the drainage network. In fact, the HAND grid
converts into absolute the relative nature of the distributed frame of
topographic reference in the drainage network. Thus, although the
HAND grid loses the height reference to sea level, it enhances
meaningful local relative variations in height. This local relevance is
especially useful because height differences now found in the HAND
grid have hydrological significance, and can potentially reveal
previously hidden local environments.
3.2.3. Mapping local environments
The HAND algorithm can be applied to the SRTM-DEM of any terrain,
producing HAND grid DEMs with implicit geomorphologic and hydrological meaning. However, the significance in practical applications is
provided by finding HAND classes (ranges of heights) that match with
relevant soil water and land cover characteristics. For this application,
the most detailed field data was produced along the hydrological
transect (Fig. 10), running orthogonally from the top of the K34
fluxtower plateau to the 2nd order Asu stream (Hodnett et al., in
preparation). This transect encompassed and represented all topographic features for the area, containing measurement and sampling
points for soil water, vegetation, soil and topographic ground truth
verification.
The vertical cross-section of the hydrological transect (Fig. 10a),
reveals the relation of the HAND grid profile with the surveyed ground
topography and water table data. Note the convergence of the water
table with the topography in the lowland, towards the stream. Note also
that the HAND grid profile hovers above the ground topography by the
distance of the forest vegetation height (the C-band radar data from the
SRTM interacted strongly with the forest canopy). Local environments
and their relative extents were defined by matching ground truth of
vegetation and topography with groundwater data. These environments
were easily recognized in the field by pattern observation, and were
classified as waterlogged (considering only the zone where soil is
perennially saturated to the surface), ecotone (shallow water table,
usually covered by Campinarana – sensu, Anderson et al., 1975) or upland
Fig. 10. a) Cross-section of the Igarapé Asu hydrological hillslope transect (from A to A') with HAND grid profile superimposed on ground topography and water table data. b) Overlay
of ground truth points onto SRTM-based HAND grid DEM (grayscale, with top contour lines in red) along the hillslope hydrological transect (from A to A').
Please cite this article as: Rennó, C. D., et al., HAND, a new terrain descriptor using SRTM-DEM: Mapping terra-firme rainforest environments
in Amazonia, Remote Sensing of Environment (2008), doi:10.1016/j.rse.2008.03.018
ARTICLE IN PRESS
10
C.D. Rennó et al. / Remote Sensing of Environment xxx (2008) xxx-xxx
Fig. 12. Four-class HAND map overlaid with ground truth points to site C1.
(deep water table). The last category was arbitrarily split into slope
(upland with slope N3°) and plateau (flat upland).
The overlaying of the ground truth points onto the HAND grid
(Fig. 10b) suggests a coherent matching between local environments
identified in the field and corroborated by groundwater data, with
drainage-normalized canopy-topography represented by the HAND
grid. This coherence suggests that local environments can be
associated with height classes in the HAND grid. An exploratory
quantitative analysis matching the distribution of ground truth points
in the Asu catchment with respective heights in the HAND grid (Fig. 11)
suggests a compelling separation of HAND classes, especially between
waterlogged, ecotone and upland environments.
Taking these into account, HAND grid values of 5 m and 15 m were
selected as preliminary best-guess thresholds between the three
classes. To optimize this separation (lessen errors on class inclusion)
we applied the simplex algorithm (Cormen et al., 2001), finding 5.3 m
and 15.0 m as the best thresholds between classes for the set of ground
truth points available.
The upland class, in comparison to the other two lowland classes,
represented well and in a relatively homogeneous way a single soil
Fig. 13. Four-class HAND map for entire study area.
Please cite this article as: Rennó, C. D., et al., HAND, a new terrain descriptor using SRTM-DEM: Mapping terra-firme rainforest environments
in Amazonia, Remote Sensing of Environment (2008), doi:10.1016/j.rse.2008.03.018
ARTICLE IN PRESS
C.D. Rennó et al. / Remote Sensing of Environment xxx (2008) xxx-xxx
water condition (well drained soil, deep water table). However there are
many distinct substrate effects in the definition of an environment on a
slope compared to one on a plateau. The waterlogged and plateau classes
are quite well defined in that they share low slope angles, and are well
separated from ecotone and slope. This analysis reveals that the HAND
height is a good separator between the three soil water relevant classes.
It also shows that, in the case of the Igarapé Asu catchment — with its
fairly uniform plateau heights, it can also separate the upland class into
slope and plateau. However, for other catchments with plateaus at
variable HAND heights, it might not score so well. Slope angle, on the
other hand, can be used to identify flat surfaces independently of HAND
height. However the HAND height can still be useful to set aside plateaus
from flat alluvial terrain in the valley bottoms. As both slope and aspect
are terrain descriptors that do not require field verification, slope will be
a better separator when applied exclusively for the upland class. The
upland class (HANDN 15.0 m) was then split on the basis of slope, with
the initial threshold value arbitrarily selected at 6.5% slope, and then
optimized with the simplex algorithm to 7.6% (Fig. 12).
The ground truth survey was accurate in identifying the local
environment for each chosen point, and as a result, for most points, the
matching between field environments with HAND predicted environments was exceptionally good. Nevertheless, unavoidable localization
errors were responsible for a few mismatches, which happened only to
the extreme values. Area estimation indicates that the distinctive
waterlogged plus ecotone environments (valley bottoms) occupied 43%
of the surface, whereas slope and plateau occupied only 26% and 31%
respectively (Fig. 13). The two valley bottom classes have been widely
considered as a minor proportion of the landscape and as a result have
been largely disregarded by most integrative studies of terra-firme. The
HAND descriptor clearly and quantitatively reveals that the valley
bottom classes are very important in this landscape. The HAND class test
carried out in this study might suffer from some site specific bias, but as
it uses site-independent gravitational potential as the main driver for its
description of terrain, we expect that it will show robustness when
applied in the Amazon outside the test area, and even anywhere else
with a similar terrain covered by thick forests with a relatively uniform
canopy height across the landscape.
4. Discussion
Topography in the SRTM shaded relief image is plainly discernible to
the trained eye. When the computed drainage network is plotted on the
11
SRTM image of a rainforest area, hidden local environments, such as
riparian zones, appear to pop up out of the continuous canopy carpet.
However this perception reveals an imaginary presence, which at best
represents only a qualitative indication. An objective and quantitative
descriptor of these hidden environments was lacking. Pure hypsometry,
as that in the shaded relief, does not offer much. A given valley area at
200 m ASL, for example, might have a similar environment to another
valley area at 650 m ASL, the latter lying kilometers upstream.
Conversely, at 200 m ASL one can find a range of different local
environments. Geographic Information System (GIS) buffer zoning is a
common tool used for delineating local environments, most often in
association with streams and other superficial waters or distinctive
landscape features. Buffer analysis to define areas of riparian influence
creates a zone of predefined width around the drainage, usually based
on simple Euclidean distance (Burrough and McDonnell, 1998).
However, without functional variables driving the buffer zoning process,
geometric distances present little topographic, pedological and hydrological meaning or consistency. The uncertainty in this local environment delineation is especially troublesome for tropical rain forests
(Bren, 2000; McGlynn and Seibert, 2003). The widely variable zone of
floodable forests in Amazonia (Hess et al., 2003), for example, suggests
that it would be grossly misrepresented if a simple geometric GIS buffer
were applied. Therefore, geometric buffering zones also fail to provide a
local environment descriptor, as the spatial extents of these perceived
environments are highly irregular, and do not have extractable
geometries that would correlate with the drainage network or other
emergent feature in any easy way. Topographic patterns associated with
contrasting environments, which are very evident from within the rain
forest, needed to be rendered in quantitative manner.
A porous medium (soil) presents a fine and interconnected network of
void spaces and channels through which water can flow, following
gradients of potential energy, always seeking equilibrium in the condition
of least energy. This is a fundamental hydraulic principle. Flow routes
from any given point on the terrain, or hill slope flow paths (overland or
groundwater in the saturated zone), will seek trajectories of least
resistance and will be propelled by gradients of gravitational energy
(neglecting both matric and osmotic potentials for a saturated soil or for
overland flow). These hill slope flow paths will invariably end somewhere
on the drainage network, with the principle of least energy requiring a
discharge into the nearest stream. Therefore, for each grid point in the
SRTM-DEM, there must be a unique and topologically consistent flow
path connecting it with its stream outlet. These connecting flow paths
Fig. 14. Comparison between the HAND (vertical distance do the nearest drainage) and horizontal distance to the nearest drainage (along water flow path) values, according to
algorithm proposed by Tucker et al. (2001).
Please cite this article as: Rennó, C. D., et al., HAND, a new terrain descriptor using SRTM-DEM: Mapping terra-firme rainforest environments
in Amazonia, Remote Sensing of Environment (2008), doi:10.1016/j.rse.2008.03.018
ARTICLE IN PRESS
12
C.D. Rennó et al. / Remote Sensing of Environment xxx (2008) xxx-xxx
bear all topological components, extractable from the SRTM-DEM, which
allowed for the development of the new HAND terrain descriptor. The
results of the test of the HAND descriptor reported here demonstrate its
capacity to resolve meaningfully the delineation of local environments.
There are a host of generic terrain descriptors, e.g. topographic
indexes, which could be compared to the HAND, but an exhaustive
comparison is beyond the aim of the present work. The closest
descriptor to the vertical distance along a flow path is the nonEuclidian horizontal distance along the same flow path (Tucker et al.,
2001). To test the similarity between the two descriptors, 1000
independent points (pixels) were randomly drawn from the test area
SRTM-DEM. According to the linear regression, the horizontal distance
can only explain 55% of the total variance that can be explained by the
HAND (Fig. 14).
There is an obvious relation between horizontal and vertical
distances because it is expected that as one moves away from the
drainage, the terrain will get higher. However, the other 45% of the
variance is new information that only the HAND descriptor can give.
Points with large horizontal distances but low HAND are indicative of
great flat areas connected to the drainage (swampy areas). The innate
swampy characteristic of these terrains, for example, would never be
seen with horizontal distances alone. With a large HAND value
(vertical flow path distance) and a small horizontal flow path distance,
the terrain is quickly rising away from the stream, i.e. a well etched
drainage valley. But, although the HAND has this evident advantage,
the horizontal flow path distance is a great advance over plain
Euclidian distances, because the latter may connect and relate points
on the terrain that do not belong to the same catchment. Furthermore,
horizontal flow path distances to the nearest drainage might have a
bearing in terms of span of time spent in draining flows, determined
by hydraulic conductivity and slope along the flow path. Height above
the nearest drainage correlates directly with gravitational potential,
and this fact alone sets this descriptor aside as unique.
5. Conclusions
The height above the nearest drainage algorithm was developed on
top of the local drain directions and drainage networks, two well
established and basic topographic descriptors. The HAND has added
the height difference along flow paths, or draining potential, as a
significant and unique terrain descriptor. The HAND terrain descriptor
produces a normalized digital elevation model (HAND grid) that can
be applied to classify terrain in a manner that is related to local soil
water conditions. The HAND grid, a DEM where all pixels have been
mapped according to their draining potential, has a wide range of
potential applications. The application of the HAND descriptor in
classifying the terrain within a monitored hydrological catchment in
Amazonia revealed strong correlations between soil water conditions,
like classes of water table depth, and topography. This correlation
obeys the physical principle of soil draining potential, or relative
vertical distance to drainage, which can be detected remotely through
the topography of the vegetation canopy found in the SRTM-DEM
data. To our knowledge no previous study has noticed or reported the
height above the nearest drainage as likely being a good terrain
descriptor. It increases usability of the SRTM-DEM data and provides a
new quantitative view on the steady state landscape, one that was
missing in the repertoire of terrain descriptors.
Acknowledgements
This work was developed within the Environmental Physics Group
of GEOMA modeling network (Brazil's Ministry of Science and
Technology funding), with invaluable collaboration and co-funding
from the LBA project (Igarapé Asu instrumented catchment study).
This work would not have been possible without full support of INPE
and INPA. The Igarapé Asu catchment study was also funded by the
PPG7/FINEP Ecocarbon project and by CTEnerg and CThidro (Energy
and Water Resources Sectorial Funds). Brazilian Science Funding
Agency, CNPq and Fapeam, also contributed with grants and scholarships of personnel involved in the project. Thanks to Gerard Banon
(DPI/INPE), two anonymous reviewers and the editor for their valuable
comments and questions. We thank dearly Antonio Huxley for support
in the field work. We thank also Evlyn Novo, Adriana Affonso and John
Melack for the cession of the JERS-1 based flood land numerical mask.
References
Anderson, A. B., Prance, G. T., & Albuquerque, B. W. P. de (1975). Estudos sobre a vegetação
das Campinas Amazônicas — III. A vegetação lenhosa da Campina da Reserva
Biológica INPA-SUFRAMA (Manaus Caracaraí, Km 62). Acta Amazônica, 5(3),
225−246.
Araujo, A. C., Nobre, A. D., Kruijt, B., Elbers, J. A., Dallarosa, R., Stefani, P., Randow, C.,
Manzi, A. O., Culf, A. D., Gash, J. H. C., Valentini, R., & Kabat, P. (2002). Comparative
measurements of carbon dioxide fluxes from two nearby towers in a central
Amazonian rainforest: The Manaus LBA site. Journal of Geophysical Research, 107
(D20), 8090. doi:10.1029/2001JD000676.
Band, L. E. (1986). Topography partition of watersheds with digital elevation models.
Water Resources Research, 22(1), 15−24.
Barbosa, C., Hess, L., Melack, J., & Novo, E. (2000). Mapping Amazon basin wetlands
through region growing segmentation and segmented-based classification of JERS1 Data. IX Simposio Latinoamericano de Percepción Remota, Puerto Iguazú, Argentina.
Bren, L. J. (2000). A case study in the use of threshold measures of hydrologic loading in
the design of stream buffer strips. Forest Ecology and Management, 132, 243−257.
Burrough, P. A., & McDonnell, R. A. (1998). Principles of Geographical Information
Systems. Spatial Information Systems.New York: Oxford University Press 333p.
Castilho, C. (2004). Variação espacial e temporal da biomassa arbórea viva em 64 km2 de
floresta de terra-firme na Amazônia Central. 87p. Doctoral Thesis, ecology, Instituto
Nacional de Pesquisa da Amazônia, Manaus-AM.
Cormen, T. H., Leiserson, C. E., Rivest, R. L., & Stein, C. (2001). Introduction to algorithms,
Section 29.3: The simplex algorithm (pp. 790−804)., Second Edition Cambridge: MIT
Press and McGraw-Hill 0-262-03293-7.
Cuartas, L. A., Tomasella, Nobre, A. D., Hodnett, M., Waterloo, M. J., & Munera, J. C. (2007).
Interception water-partitioning dynamics for a pristine rainforest in Central
Amazonia: Marked differences between normal and dry years. Agricultural and
Forest Meteorology, 145(1–2), 69−83.
Curkendall, D. W., Fielding, E. J., Cheng, T. -H., & Pohl, J. M. (2003). A computational-grid
based system for continental drainage network extraction using SRTM digital
elevation models. Proceedings of the HPSECA/IPCC conference, Taiwan.
Goudie, A. (2004). Encyclopedia of Geomorphology.New York: Routledge publisher
1250p.
Guyot, G., Guyon, D., & Riom, J. (1989). Factors affecting the spectral response of forest
canopies — A review. Geocarto International, 4, 3−18.
Hess, L. L., Melack, J. M., Novo, E. M. L. M., Barbosa, C. C. F., & Gastil, M. (2003). Dualseason mapping of wetland inundation and vegetation for the central Amazon
basin. Remote Sensing of Environment, 87, 404−428.
Hodnett, M.G., Tomasella, J., Cuartas, L.A., Waterloo, & Nobre, A.D. (in preparation). Sub
Surface Hydrological Flow Paths in a Ferralsol (Oxisol) landscape in Central
Amazonia.
Hirano, A., Welch, R., & Lang, H. (2003). Mapping from ASTER stereo image data: DEM
validation and accuracy assessment. Journal of Photogrammetry & Remote Sensing,
57, 356−370.
Jacobsen, K. (2003). DEM Generation from Satellite Data. EARSeL Symposium Ghent:
Remote Sensing in Transition (pp. 273−276). 90-77017-71-2.
Jenson, S. K., & Domingue, J. O. (1988). Extracting topographic structure from digital
elevation data for geographic information system analysis. Photogrammetric
engineering and remote sensing, 54, 1593−1600.
Kellndorfer, J., Walker, W., Pierce, L., Dobson, C., Fites, J. A., Hunsaker, C., Vona, J., & Clutter,
M. (2004). Vegetation height estimation from shuttle radar topography mission and
national elevation datasets. Remote Sensing of Environment, 93, 339−358.
Laurance, W. F., Fearnside, P. M., Laurance, S. G., Delamonica, P., Lovejoy, T. E., Rankin-deMerona, J. M., Chambers, J. Q., & Gascon, C. (1999). Relationship between soils and
Amazon forest biomass: A landscape-scale study. Forest Ecology Management, 118
(1–3), 127−138.
Lindsay, J. B., & Creed, I. F. (2005). Removal of artifact depressions from digital elevation
models: towards a minimum impact approach. Hydrological processes, 19,
3113−3126.
McGlynn, B. L., & Seibert, J. (2003). Distributed assessment of contributing area and
riparian buffering along stream networks. Water resources research, 39(4),
TNN2.1−TNN2.7.
Melack, J. M., & Hess, L. L. (2004). Remote sensing of wetlands on a global scale. SIL
News, 42, 1−5.
Montgomery, D. R., & Dietrich, W. E. (1988). Where do channels begin? Nature, 336,
232−234.
Montgomery, D. R., & Foufoula-Georgiou, E. (1993). Channel network source
representation using digital elevation models. Water Resources Research, 29(12),
3925−3934.
Nobre, A. D., Hirata, Y., Honzak, M., & Campos, A. C. (1998). Vegetation classification and
mapping of Manaus region using passive remote sensing. In N. Higuchi, M. A. A.
Campos, P. T. B. Sampaio, & J. Santos (Eds.), Pesquisas florestais para a conservação da
Please cite this article as: Rennó, C. D., et al., HAND, a new terrain descriptor using SRTM-DEM: Mapping terra-firme rainforest environments
in Amazonia, Remote Sensing of Environment (2008), doi:10.1016/j.rse.2008.03.018
ARTICLE IN PRESS
C.D. Rennó et al. / Remote Sensing of Environment xxx (2008) xxx-xxx
floresta e reabilitação de áreas degradadas da Amazonia, Projeto Jacaranda (pp. 28−49).
Manaus: MCT-INPA & Jica pub.
Novo, E. M. L. M., Leite, F. A., Avila, J., Ballester, M. V., & Melack, J. M. (1997). Assessment
of Amazon floodplain habitats using TM/Landsat data. Ciência e Cultura, 49(4),
280−284.
O'Callaghan, J. F., & Mark, D. M. (1984). The extraction of drainage networks from digital
elevation data. Computer Vision, Graphics and Image Processing, 28, 323−344.
Oldeman, R. A. A. (1990). Forests: Elements of Silvology.Berlin: Springer Verlag 624p.
Rodriguez, E., Morris, C. S., & Belz, J. E. (2006). A global assessment of the SRTM
performance. Photogrammetric Engineering and Remote Sensing, 72(3), 249−260.
Siqueira, P., Chapman, B., & McGarragh, G. (2003). The coregistration, calibration, and
interpretation of multiseason JERS-1 SAR data over South America. Remote Sensing
of the Environment, 87, 389−403.
Siqueira, P., Hensley, S., Shaffer, S., Hess, L., McGarragh, G., Chapman, B., & Freeman, A.
(2000). A continental-scale mosaic of the Amazon basin using JERS-1 SAR. IEEE
Transactions on Geoscience and Remote Sensing, 38(6), 2638−2644.
Soille, P., Vogt, J., & Colombo, R. (2003). Carving and adaptive drainage enforcement of
grid digital elevation models. Water Resources Research, 39(12), 1366−1375.
Tarboton, D. G. (2003). Terrain analysis using digital elevation models in hydrology.
23rd ESRI International Users Conference, San Diego, California.
Tarboton, D. G., & Ames, D. P. (2001). Advances in the mapping of flow networks from
digital elevation data. World Water and Environmental Resources Congress, Orlando,
Florida.
Tarboton, D. G., Bras, R. L., & Rodriguez-Iturbe, I. (1991). On the extraction of channel
networks from digital elevation data.. Hydrological processes, 5, 81−100.
13
Tarboton, D. G., Bras, R. L., & Rodriguez-Iturbe, I. (1992). A physical basis for drainage
density. Geomorphology, 5, 59−76.
Tomasella, J., Hodnett, M. G., Cuartas, L. A., Nobre, A. D., Waterloo, M. J., & Oliveira, S. M.
(2007). The water balance of an Amazonian micro-catchment: The effect of interannual
variability of rainfall on hydrological behavior. Published online, Hydrological
Processes.
Tucker, G. E., Catani, F., Rinaldo, A., & Brás, R. L. (2001). Statistical analysis of drainage
density from digital terrain data. Geomorphology, 36, 187−202.
Valeriano, M. M., Kuplich, T. M., Storino, M., Amaral, B. D., Mendes, J. N., & Lima, D. J.
(2006). Modeling small watersheds in Brazilian Amazonia with shuttle radar
topographic mission-90 m data. Computer Geosciences, 32, 1169−1181.
van Zyl, J. J. (2001). The shuttle radar topography mission breakthrough in remote
sensing of topography. Acta Astronautica, 48(5–12), 559−565.
Waterloo, M. J., Oliveira, S. M., Drucker, D. P., Nobre, A. D., Cuartas, L. A., Hodnett, M. G.,
Langedijk, I., Jans, W. W. P., Tomasella, J., Araújo, A. C., Pimentel, T. P., & Estrada, J. C.
M. (2006). Export of organic carbon in run-off from an Amazonian rainforest
blackwater catchment. Hydrological Processes, 20(12), 2581−2597.
Wulder, M. (1998). Optical remote-sensing techniques for the assessment of forest
inventory and biophysical parameters. Progress in Physical Geography, 22(4),
449−476.
Zhou, Q., & Liu, X. (2002). Error assessment of grid-based flow routing algorithms used in
hydrological models. International journal of geographical information science, 16(8),
819−842.
Please cite this article as: Rennó, C. D., et al., HAND, a new terrain descriptor using SRTM-DEM: Mapping terra-firme rainforest environments
in Amazonia, Remote Sensing of Environment (2008), doi:10.1016/j.rse.2008.03.018