Directable Weathering of Concave Rock
Using Curvature Estimation
Michael D. Jones, Member, IEEE, McKay Farley, Joseph Butler, and Matthew Beardall
Abstract— We address the problem of directable weathering of exposed concave rock for use in computer- generated animation or
games. Previous weathering models that admit concave surfaces are computationally inefficient and difficult to control. In nature, the
spheroidal and cavernous weathering rates depend on the surface curvature. Spheroidal weathering is fastest in areas with large
positive mean curvature and cavernous weathering is fastest in areas with large negative mean curvature. We simulate both processes
using an approximation of mean curvature on a voxel grid. Both weathering rates are also influenced by rock durability. The user
controls rock durability by editing a durability graph before and during weathering simulation. Simulations of rockfall and colluvium
deposition further improve realism. The profile of the final weathered rock matches the shape of the durability graph up to the effects of
weathering and colluvium deposition. We demonstrate the top- down directability and visual plausibility of the resulting model through a
series of screenshots and rendered images. The results include the weathering of a cube into a sphere and of a sheltered inside corner
into a cavern as predicted by the underlying geomorphological models.
Index Terms— Physically based modeling, modeling packages.
1 INTRODUCTION
T HIS work simplifies the creation of weathered rock for
use in computer- generated animation or games. Weath-ered
rock is difficult to create algorithmically because it
almost always includes overhangs or concavities, which do
not have a simple heightmap representation. Tools for
creating weathered rock should also support the creative
artistic process and fit into the existing animation produc-tion
pipeline.
The prevalence of weathered rock as visual cue in film
indicates that directors find weathered rock useful in telling
a story. Sandstone rock, in particular, has been widely used
to create a feeling of being somewhere remote, exotic, or
unique. Sandstone terrain as a visual cue can be found in
live- action film [ 1], numerous automobile commercials,
documentary works [ 2], and in both hand- drawn [ 3] and
computer- generated animation [ 4]. In one live- action film
[ 5], sandstone rock formations were used as the setting for
scenes on an alien planet.
Many algorithms have been published for terrain
generation in computer graphics but none have adequately
addressed the artistic creation of weathered rock. Most
algorithms are difficult to control and do not admit concave
surfaces. With a few exceptions, terrain generation algo-rithms
operate on heightmaps in which a 2D array of
elevations represents the height of the terrain at grid points.
The heightmap representation is compact and easily
processed but is poorly suited for concave surfaces in
which each grid point may be associated with more than
one elevation.
Three algorithms for concave terrain generation have
been presented. Gamito and Musgrave present a purely
procedural approach to concave terrain, which is limited to
terrain described by analytically differentiable functions [ 6].
Both Ito et al. and Bene s et al. give algorithms on voxel
grids. Ito et al. simulated rockfall using graph component
analysis [ 7] and Bene s et al. simulated hydraulic erosion [ 8]
using computational fluid dynamics. Both Ito et al. and
Bene s et al. create convincing concave terrain, but neither
address the problems of weathering and top- down control.
Algorithms for weathering exposed rock have also been
presented but these algorithms are either too slow or do not
admit concave surfaces. Musgrave et al. reduce steep slopes
into more gradual slopes until the slope reaches the angle of
repose [ 9] but do not include concave surfaces. Dorsey et al.
convincingly weather exposed concave stone [ 10] but the
process is slow and not designed for directability.
Commercial landscape modeling tools, such as [ 11], [ 12],
do not support the manipulation of 3D shapes using
physically based algorithms such as weathering or erosion.
These tools do support physically based manipulation of
height fields but do not apply to concave rock. Weathering
and erosion could be performed manually using 3D
modeling packages, such as [ 13], [ 14], but this would
involve a tedious process of moving vertices and edges.
Wang and Kaufman created a volumetric sculpting tool that
allows a user to remove varying sizes of sections from a
voxel representation of clay [ 15]. Such a tool could also be
used to manually simulation weathering and erosion but at
the cost of moving voxels rather than vertices and edges.
This paper builds on our preliminary work on spheroidal
weathering of sandstone goblins ( which are a kind of
hoodoo), as reported in [ 16]. In this paper, we extend the
original algorithm by including the sign of the surface
curvature rather than just the magnitude. This allows us to
simulate two kinds of weathering using curvature estimation.
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, VOL. 16, NO. 1, JANUARY/ FEBRUARY 2010 81
. M. D. Jones, M. Farley, and J. Butler are with the Department of Computer
Science, Brigham Young University, 3328 TMCB, Provo, UT 84602.
E- mail: jones@ cs. byu. edu, macfarley5@ gmail. com, joebutler@ byu. net.
. M. Beardall is at Caselle, Inc.— Engineering, 1570 N Main St, Spanish
Fork, UT 84660. E- mail: beardalls@ gmail. com.
Manuscript received 23 May 2008; revised 5 Dec. 2008; accepted 4 Mar. 2009;
published online 24 Mar. 2009.
Recommended for acceptance by B. Guo.
For information on obtaining reprints of this article, please send e- mail to:
tvcg@ computer. org, and reference IEEECS Log Number TVCG- 2008- 05- 0068.
Digital Object Identifier no. 10.1109/ TVCG. 2009.39.
1077- 2626/ 10/$ 26.00 2010 IEEE Published by the IEEE Computer Society
We also integrate the user interface with the simulation to
improve directability, and add rockfall and colluvium
deposition to improve realism. In our previous work, the
user interface allowed durability to be set to constant values at
a fixed number of fixed locations in the rock. In this paper, we
present a user interface, which allows the user to vary rock
durability during simulation at any altitude and for any
number of vertical locations.
This work uses an estimation of the signed mean surface
curvature and top- down interactive control to address the
problem of directable weathering of concave rock. Efficient
estimation of the signed curvature by counting voxels
containing air or stone in bubbles centered on the rock
surface allows us to simulate both spheroidal and caver-nous
weathering. The geological foundation [ 17], [ 18], [ 19]
for this approach is that exposed rock with high positive
curvature has more exposure to air and wind than less
exposed rock. Protected rock, which has high negative
curvature, retains moisture that accelerates decomposition
of the cementation between particles in rock. Rockfall and
colluvium deposition are added to enhance the look of
weathered rock shapes.
Top- down interactive control is provided by a durability
graph in which the user edits a collection of line segments,
which represent the durability of each rock layer as a
function of the altitude. The profile of the final rock shape
matches the shape of the durability graph and the durability
graph can be edited during the weathering simulation. This
creates a feedback loop in which the user can control the
weathering simulation by changing simulation inputs based
on simulation outputs.
The user interface mitigates problems with the run-tweak-
run cycle typically found in simulation for graphics
as the artist tweaks the parameters and repeats the process.
Our approach to top- down directability was inspired by
Prusinkiewicz et al.’ s use of curves for artistic control of
plant generation [ 20] and is similar to Dobashi et al.’ s use of
profiles to suggest cloud shapes [ 21].
When the user is satisfied with the weathered rock
shape, a triangular mesh is extracted, which can be
imported into the production pipeline. Any of a number
of third party modeling and rendering tools can then be
used to tweak, texture, and render the object.
Examples of weathered rock shapes are given in
Section 4. These results illustrate the control provided by
the interface and the variety of shapes generated by the
weathering simulation. In the next section, we review
related work in terrain generation and weathering. We then
discuss spheroidal and cavernous weathering, which are
the geological foundations for our algorithm. The algorithm
appears in Section 4 and we discuss the user interface in
Section 5. We conclude in Section 7 and discuss ideas for
future work in concave terrain generation.
2 RELATED WORK
This work is most closely related to other work in terrain
generation and weathering for computer graphics. Our
algorithm is a geomorphological based approach to concave
rock formation, which relies on a phenomenological
approach to weathering simulation in that we do not
simulate the chemistry of rock decomposition through
exposure to the elements. The geomorphological theories of
weathering on which our simulation is based are discussed
separately in the next section.
2.1 Foundational Work
Our work is built on the foundation created by several key
papers, which we highlight here.
Colluvium is deposited using an algorithm similar to
Musgrave et al.’ s thermal weathering simulation [ 9] and has
been adapted to work on a voxel grid rather than a
heightfield. Rockfall is simulated using a simplified version
of Ito et al.’ s rockfall simulation [ 7] in which we perform the
component analysis to identify rocks that are not connected to
the ground but we do not compute friction angles and
perform a slippage analysis to move rocks that are connected
to the ground but lie on an unstable slope.
Editing durability graphs in the user interface is a kind of
top- down approach to artistic modeling, which was in-spired
by Prusinkiewicz et al.’ s use of curves to describe
position information for L- systems [ 20] and is similar to
Dobashi et al.’ s more recent work on controllable cloud
generation [ 21]. As with Dobashi et al.’ s work, the final
shape may not match the desired profile due to the nature
of the underlying natural phenomena. However, this is
desirable because it creates a plausible shape, which
matches the artist’s intent.
We estimate curvature using a new application of a
technique first used to model diffusion- limited aggregation
[ 22]. Later, Pimienta et al. verified that the actual curvature
of a circle is linearly related to the estimated curvature and
applied the technique to sintering simulation [ 23]. Vicsek
and Pimienta both estimated curvature in two dimensions.
We independently reinvented the idea but in three
dimensions.
2.2 Terrain Generation
Terrain generation has a long and productive history in
computer graphics. Here, we discuss only the most relevant
work. Terrain generation algorithms can be divided into
phenomenological approaches and geomorphological ap-proaches.
Phenomenological approaches begin with the
appearance of terrain and design algorithms that create
shapes, which mimic that appearance. Geomorphological
approaches begin with a geological process, which might
have created the terrain, and design algorithms which
simulate that process with the goal of generating the right
appearance. Phenomenological approaches mimic the ap-pearance
but ignore the process while geomorphological
approaches mimic the process to get the appearance.
It should be noted that all geomorphological approaches
become phenomenological at some level. Otherwise, geo-morphological
approaches would require the simulation of
physics down to the atomic or even subatomic level. As
computational power has increased, geomorphological ap-proaches
have become more complex and realistic.
We briefly summarize key work in both approaches.
2.2.1 Phenomenological Approaches
Phenomenological approaches can be traced back to
Mandelbrot’s observation that certain mountain ranges
resemble the trace of a particle moving in Brownian motion
in 1D. Generalizing to 2D gives a striking approximation of
82 IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, VOL. 16, NO. 1, JANUARY/ FEBRUARY 2010
a mountain range [ 24]. Further generalization of fractional
Brownian motion ( fBm) gives terrain with varying degrees
of ruggedness. Later, Fournier et al. simultaneously
invented a more efficient approximation of fBm using
midpoint displacement [ 25].
The primary drawback of fBm terrain is that the
frequency distribution of the resulting terrain does not vary
with altitude as it does in nature 1 and that it does not include
rivers and river networks, which are an important part of
mountain range appearance. Geomorphological approaches
to rivers and erosion were then added to fundamentally fBm
models ( i. e., [ 26], [ 9], [ 27], [ 28]) and will be discussed below
with other geomorphological approaches.
Other purely phenomenological approaches to improv-ing
fBm terrain were undertaken by Mandelbrot, Voss, and
Prusinkiewicz. Mandelbrot used scaling in postprocessing
as well as other displacement methods to differentiate
valleys and mountains in fBm terrain [ 24]. Voss also scaled
fBm terrain as a postprocessing step for a similar purpose
[ 29]. Prusinkiewicz et al. used rewriting systems on fBm
terrain to model rivers as squig- curve fractals [ 30]. The
resulting terrain includes rivers, but the displacement of the
terrain around river channels looks unrealistic in places.
Later, Gamito and Musgrave presented a new model of
phenomenological terrain based on the displacement of
vector fields. Vector field displacement describes a new
kind of terrain, which allows overhangs [ 6]. The terrain
images in [ 6] are not particularly realistic and require the
surface to be described by a differentiable function, but the
work is significant because it both presents a different
approach to phenomenological terrain and allows over-hangs.
The method does generate realistic models of
breaking ocean waves.
The compelling advantage of phenomenological terrain
is that it can efficiently be generated and is viewable at
many scales using a small amount of data. This can be done
because the surface can procedurally be generated at
arbitrary resolutions on- the- fly. This is useful for long
tracking shots, which include viewing terrain from hun-dreds
of kilometers above and from millimeters away from
the terrain surface. These kinds of shots are unusual in film
and are not addressed by geomorphological approaches
including our work on weathering presented here.
The main problem with phenomenological terrain is that
it is not always clear how the input to a phenomenological
algorithm relates to the final terrain shape. In phenomen-ological
approaches, designing terrain requires under-standing
seemingly unrelated mathematical objects. Our
work is fundamentally geomorphological and avoids this
problem by simulating a process. This, however, creates a
new set of problems.
2.2.2 Geomorphological Approaches
Geomorphological approaches began in an effort to make
phenomenological mountain ranges look more like moun-tain
ranges by mimicking the effects of river systems. Kelley
et al. gave the first model of river flow in a mountain range
[ 26]. The results are visually plausible and even more
impressive given the computational resources of the time.
Musgrave et al. then presented an erosion and weathering
algorithm for fractal terrain, which begins with a heightmap
generated through fBm [ 9].
As computing power increased, it became possible to
write increasingly accurate models of hydraulic erosion.
Nagashima simulated erosion on geologically distinct rock
layers in a flat plain [ 31]. The resulting terrain resembles
entrenched meandering rivers in sandstone canyons. Later,
Chiba et al. calculated velocity fields from particle flow
simulations and used these fields to erode terrain [ 27].
Bene s et al. perform hydraulic erosion by simulating 3D
fluid flow on voxel grids using the Navier- Stokes equations
[ 8]. This model includes overhangs and other types of
surfaces created through fluid flow. The results are
extremely realistic and the process is fully general, but the
simulation is tractable on terrains consisting of on the order
of only 120 32 120 voxels. St’ava et al. integrated several
hydraulic erosion processes into a single algorithm that is
implemented on the GPU [ 32]. This method can erode
scenes consisting of 2,048 1,024 pixels at 20 frames per
second, but is based on a heightmap rather than a voxel
grid, and thus, cannot represent concave terrain.
Other work in geomorphological terrain generation has
centered on weathering. Weathering was beyond the scope
of Kelley’s early work on geomorphological terrain genera-tion
but is included in Musgrave’s work on fractal terrain.
Musgrave et al. give an algorithm for simulating a process
they call “ thermal weathering.” Thermal weathering mod-els
the process by which steep cliffs become sloping hills
through the accumulation of talus at the base of the cliff.
Musgrave’s weathering model is effective but does not
admit concave surfaces. We use Musgrave’s model as the
foundation of our colluvium deposition algorithm. Ap-proaches
to weathering for objects other than terrain
features will be discussed separately in the next section.
Ito et al. modeled the geomorphological process of
rockfall in jointed cliff faces [ 7]. The algorithm uses a
search through an undirected graph representing connec-tions
between blocks of rock to find blocks, which form
connected components. A simulation of gravity and slope is
used to determine which component is most likely to fall
next. The algorithm results in convincing images of cliff
faces particularly those found by the sea and in the presence
of prevailing winds and tides.
The collective advantage of geomorphological ap-proaches
is that they generate a wider variety of terrain,
which is more visually plausible than most terrain generated
using phenomenological approaches. Phenomenological
approaches have had success generating mountainous
terrain, which is often viewed from above at a distance.
Geomorphological approaches can be used to create terrain
features, such as rock columns, which look realistic from
smaller viewing distances. This is particularly useful in film
for scenes, which are set in or near terrain features.
A key limitation of geomorphological approaches is that
they are highly specialized and can be difficult to control.
Even Bene s’ work, which simulates generic fluid flow,
mimics only a single geomorphological process. Our work
is specialized as well but we are able to generalize curvature
estimation to mimic two weathering processes. We address
JONES ET AL.: DIRECTABLE WEATHERING OF CONCAVE ROCK USING CURVATURE ESTIMATION 83
1. The accumulated effects of weathering and erosion reduce high-frequency
variations in older rock which tend to lie at lower altitudes.
problems with controllability using a top- down interactive
control during the simulation.
2.3 Weathering
Previous work in weathering simulation has focused on
either convex terrain, as mentioned above, or man- made
objects as will be discussed here.
Dorsey et al. designed a weathering algorithm based on a
simulation of fluid flow beneath the surface of an object
[ 10]. Slabs of voxels were aligned perpendicular to surface
normals and used in the weathering simulation. Voxel slabs
were used to create a detailed simulation grid near the
surface of an object while ignoring the center. The resulting
algorithm includes what Dorsey calls “ corestone weath-ering,”
which is similar to spheroidal weathering. We use a
single axis- aligned voxel grid rather than slabs of voxels
aligned to surface normals.
Chen et al. simulate weathering by tracing the flow of
- ton particles through a scene [ 33]. The main advantage
of - ton tracing is that it can propagate weathering effects
globally throughout a scene. In the simulation, - ton
particles cause weathering or staining when they strike a
surface. If a - ton ray causes a particle to separate from
the surface, then subsequent interaction points between
the - ton particle and the surface can deposit accumu-lated
material. Emitters of - ton particles can be defined
to simulate, for example, stains created by dripping
water. Computing global weathering effects in a scene
is much like computing global illumination in a scene
and is computationally difficult. Chen’s model required
90 minutes of computation time to weather a scene with
50,000 vertices.
Hsu and Wong also use a ray casting technique but do so
to simulate dust accumulation on 3D objects [ 34]. They
derive an equation, which is based on the angle between the
surface normal and direction of the dust source and an
approximation of the surface friction. We use a simpler
method to calculate where eroded voxels fall and accumu-late
colluvium at the base of the rocks.
Our algorithm is less physically based than Dorsey’s or
even Chen’s but can only simulate a single process, while
Dorsey’s and Chen’s can mimic several processes including
staining. Like Dorsey’s and Chen’s models, our algorithm
allows the deposition of weathered material at the base of
the object being weathered, which is similar to how Hsu
and Wong simulate dust accumulation.
2.4 Estimating Curvature
We estimate curvature by computing the ratio of rock to air
in a fixed volume. This approach to estimating curvature
avoids the differentiation of functions, which describe the
surface. Sarracino et al. give a series of differential equations,
which describes weathering due to fluid flow through rock
in three dimensions, but the resulting equations have been
solved only for the two- dimensional case [ 35].
Vicsek first proposed estimating curvature at point x by
counting discrete units of volume within a fixed distance
from x [ 22]. Later, Pimienta et al. verified that the actual
curvature of a sphere is linearly related to the estimated
curvature and applied the technique to sintering simulation
[ 23]. Like Sarracino, Vicsek and Pimienta estimated curva-ture
in two dimensions.
Hsu and Wong use ray casting to determine exposure of
surface points as part of their dust accumulation algorithm
[ 34]. They cast random rays at each point on the surface to
determine the surface exposure. The surface exposure is
used to scale dust accumulation at each point. This is
similar to our approach to concave weathering using
negative curvature. This approach may produce better
surface curvature estimates than our bubble- based method
for points that lie inside thin- walled caverns.
Miller proposed a voxel- based method for determining
accessibility of 3D objects for shading purposes [ 36]. His
method involves calculating an offset surface in voxel space
by placing a voxel sphere at each voxel. The accessibility at
each point is radius of the smallest sphere that touches
empty space in the offset representation. Wang and
Kaufman use a similar approach for rendering [ 15]. They
center spheres around each point in voxel space and use ray
tracing to render a smooth surface representation of the
voxel grid.
Our curvature estimates are based on the same concept
as of Sarracino, Vicsek, and Pimienta, but are done in three
dimensions. Miller also worked in three dimensions but did
not use spheres to estimate curvature. We also modify
curvature estimates by counting discrete units of volume in
skewed volumes, rather than spheres, in order to exagge-rate
curvature in a given dimension and change the
resulting shape of the weathered rock.
3 WEATHERING
In this section, we describe spheroidal and cavernous
weathering from a geological perspective and discuss three
other factors, which influence the weathering rate in natural
rock. These factors are: durability variations between rock
layers, durability variations within rock layers, and the ratio
of exposed surface area to volume per unit volume of rock.
Variations in the weathering rate are responsible for the
visually interesting shapes of natural rock, and therefore,
important elements in our weathering algorithm.
In weathered rock, a pile of weathered material called
colluvium can be found at the base of the rock formation.
Colluvium is soft and can quickly be transported by air or
water and is important to the appearance of a rock formation.
The discussion of spheroidal weathering is based on
Milligan’s geological description of the rock formations in
Goblin Valley State Park, Utah [ 19], and the discussion of
cavernous weathering is based on Huinink et al.’ s and
Turkington’s description of small cavern formation in a
variety of sandstone formations [ 17], [ 18].
3.1 Spheroidal Weathering
Spheroidal weathering is a rock changing, or geomorpho-logical,
process by which sharp corners are reduced to
rounded corners in exposed rock. When rock cracks, sharp
angular edges may be formed. Spheroidal weathering
gradually reduces those sharp edges to rounded corners.
Milligan’s theory is that spheroidal weathering occurs in
certain sandstones because sharp angular edges have more
exposed surface area per unit of volume than flat surfaces.
84 IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, VOL. 16, NO. 1, JANUARY/ FEBRUARY 2010
It is called “ spheroidal” weathering because it tends to
produce sphere- like shapes.
The spheroidal weathering rate is proportional to the
ratio of surface area to volume in a unit volume of rock. A
unit volume of rock on a corner has more exposed area than
a unit volume on an edge, which has more exposed area
than a unit volume on a face. Because the unit volume on a
corner has more exposure to air and other weathering
elements, a unit volume on a corner will weather more
quickly than a unit volume on an edge or face.
The ratio of surface area to volume at a given point is
proportional to the positive surface curvature at that point.
A sharp corner or edge has high positive curvature while a
flat surface has zero curvature. In this framework, spher-oidal
weathering is a process that equalizes positive
curvature across the surface of a rock.
The photograph in Fig. 1 highlights the effects of
spheroidal weathering in soft sandstone near Cohab
Canyon in Capitol Reef National Park, Utah. The sharp
edges on the left are the result of recent rockfall and have
not yet been rounded by spheroidal weathering. The
rounded edges are the result of spheroidal weathering
following past rockfall events.
3.2 Cavernous Weathering
Cavernous weathering is a geomorphological process by
which caverns, or pits, form in the rock surface. Fig. 2
contains example of caverns formed through cavernous
weathering in Forgotten Canyon, Utah. The rate of
cavernous weathering depends on both material durability
and surface curvature, as with spheroidal weathering.
However, the cavernous weathering rate is proportional
to negative curvature rather than positive curvature.
Cavernous weathering equalizes negative curvature on
the inside of a curve up to a limit.
Cavernous weathering increases with negative curvature
because surface points with negative curvature tend to be
protected from exposure by adjacent rock. These protected
regions hold moisture longer than neighboring, exposed
rock, and this leads to increased deposition of salt ( when
compared to exposed neighboring rock). Increased salt
concentrations accelerate the breakdown of chemical
cementation between adjacent rock particles and cause the
protected rock to disintegrate more quickly.
The foregoing description of cavernous weathering
follows the salt weathering theories advanced by Huinink
and Turkington. Huinink explains that as water collects at
the surface of the rock, this water is absorbed by the rock
at a rate related to the porosity of the rock. When this wet
layer reaches a certain saturation level, the transportation
of salts dissolved by the water becomes possible. The
folds and curvature of the rock are important because
they allow for uneven drying across the rock surface. If
parts of the surface dry at slower rates, they will remain
above the saturation threshold where salt transportation is
possible, thus allowing these areas to collect more salt
from the surrounding rock.
Turkington also suggests that caverns are likely to begin
on areas of the surface that are more shaded from the sun
than neighboring areas. These shaded areas are more likely
to retain the water for more time than areas that are not
shaded, thus adding to the amount of salt deposited.
Huinink states that depressions that exist in the rock create
a difference in the airflow above the surface of the rock,
which further slows the rate of evaporation. With higher
levels of salt being deposited in these areas, the rate of salt
weathering will be higher than surrounding areas, thus
weakening and eventually creating and enlarging these
depressions into caverns.
4 ALGORITHM
The weathering algorithm uses surface curvature estima-tion
to simulate two weathering processes. The weathering
rates depend on curvature as well as durability variations
between and within rock layers. Weathering is simulated
for surface voxels in a voxel grid in a time- based simulation.
We first discuss the data structures used in the algorithm,
then present the algorithm, and finally, give a detailed
discussion of curvature estimation. Important implementa-tion
details are given at the end of the section.
The simulation also includes colluvium accumulation
and rockfall. Rockfall is a reimplementation of Ito et al.’ s
rockfall algorithm [ 7] based on component analysis, but
without friction, and is not discussed further.
JONES ET AL.: DIRECTABLE WEATHERING OF CONCAVE ROCK USING CURVATURE ESTIMATION 85
Fig. 1. Spheroidal weathering in soft sandstone. Spheroidal weathering
will eventually reduce the sharp edges on the left to the rounded edges
seen on the right.
Fig. 2. Cavernous weathering on a canyon wall.
4.1 Data Structures
The weathering algorithm is defined on a uniform voxel
grid in three dimensions. Each voxel in the grid represents a
cubic unit of either air, rock, or colluvium.
Voxels containing air have no associated properties
because air is modeled as a static object. The rock voxels
are organized into layers by altitude as defined by the user.
Layer boundaries are limited to horizontal planes. Each
layer has a durability, which is defined using the durability
graph. Perlin noise [ 37] can also be added to vary layer
durability. Variations in layer durabilities cause variations
in weathering rates.
Each rock and colluvium voxel is associated with a
durability and a decimation. The durability is the sum of the
durability of the enclosing layer and the value of a noise
function evaluated at the voxel location.
Voxel decimation is a measure of the degree to which a
voxel has weathered. Voxels are initialized with a decima-tion
of 100. When decimation reaches 0, the voxel is fully
decimated, removed from the surface, and added to a
colluvium pile. Colluvium voxels fall and settle to an angle
of repose using a process similar to Musgrave et al.’ s
thermal weathering algorithm [ 9].
4.2 Simulation Algorithm
The weathering simulation is based on computing the ratio
of air to rock in a bubble centered on each voxel. The
spheroidal weathering rate is proportional to this ratio and
the cavernous weathering rate is inversely proportional.
The geological justification for this approach to spheroidal
weathering is that the weathering rate increases as the
amount of exposed surface area per unit of volume
increases. Fig. 3 illustrates the approach to spheroidal
weathering in the two- dimensional case. Fig. 3a shows two
bubbles centered on unweathered rock and Fig. 3b shows
two bubbles centered on weathered rock. Initially, the
bubble centered on the corner contains a greater percentage
of air than the bubble centered on an edge. After weath-ering,
the corner is rounded and the bubbles contain similar
percentages of air. The final effect of spheroidal weathering
is to equalize the rate of change of positive curvature across
the rock surface.
Fig. 4 illustrates the process for cavernous weathering.
Fig. 4a contains rock before cavernous weathering and
Fig. 4b shows the same rock after weathering. The
geological justification for this approach to cavernous
weathering is that moisture accumulates in caverns as the
surrounding rock, which is more exposed, dries. Cavernous
weathering tends to equalize the rate of change of negative
curvature on the surface by increasing the cavern size.
The simulation algorithm is a synchronized time- based
simulation in which each simulation step represents a single
atomic unit of time. This approach is simple but looses
accuracy because all particles which erode in a given unit of
time erode at the same instant regardless of when they
actually eroded during that time unit.
The simulation can be designed as an asynchronous
event- based process in which particles erode at any time.
This can be done using a time- sorted event queue rather
than a for- loop over time. We have implemented both
methods and found that the time- based simulation, as
described below, is more efficient and stable than the event-based
method.
The pseudocode for the weathering and voxel erosion
algorithms is shown in Fig. 5. The core of the algorithm is a
loop ( line 3), which traverses the surface voxels in a voxel
grid V to compute the decimation experienced by each
voxel during a single time step. The weathered surface is
redisplayed after each simulation step and the user can
terminate weathering and export a surface at any time.
For each voxel v at or near the surface, we estimate a
weathering rate by computing the percentage of voxels,
which contain air in a bubble b centered on v by counting
the number of voxels containing air va inside b and dividing
by the number of voxels jbj in b ( line 4):
vair
va
jbj
:
We then use this percentage to determine the decimation
experienced by the voxel. Decimation due to spheroidal
weathering is computed ( line 5) using the durability at v,
vdurability, and a function f ( lines 6 and 8), which can scale or
clamp the curvature estimate to produce different effects.
The relationship between curvature and decimation rate is
not established in the geology literature. The function f is a
tunable parameter, which is set to produce plausible and
controllable results. Decimation due to cavernous weath-ering
is computed ( line 7) using a similar process, but vair is
inverted to make the weathering inversely proportional to
the ratio of air to stone in the bubble.
86 IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, VOL. 16, NO. 1, JANUARY/ FEBRUARY 2010
Fig. 3. Weathering rates are computed using curvature estimates based
on the percentage of air contained in a bubble centered on each voxel
on the rock surface.
Fig. 4. Cavernous weathering equalizes the rate of change of negative
curvature across the surface. The surface curvature is estimated to be
negative if more than half of the bubble contains rock.
Bubbles are stored as a three- dimensional cube of Boolean
values, which indicate whether or not the corresponding
voxel is in or out of the bubble. In order to determine which
voxels are in the bubble, we treat the bubble space as a
coordinate space with the center of the bubble at the origin.
We then calculate the distance to each point in bubble space
from the origin. For sphere- shaped bubbles, the array value
for a voxel is set to true if the distance from the center to the
voxel is less than the radius. Bubbles with different shapes can
be defined by modifying the criteria used to determine
whether or not a voxel lies in the bubble.
The bubbles used to estimate curvature can also be defined
as a mixture of primitive bubble types with the mixture
controlled by the value of a Perlin noise function. As with
durability variations within a layer, varying bubble types also
produces irregular rock shapes. The bubble distribution is set
by partitioning the range of a noise function into b equal
intervals, where b is the number of bubble types to be mixed,
then assigning the bubble type based on the range in which
the noise value falls.
If the new decimation value vdecimation is less than or equal
to the minimum decimation value, then v is fully decimated
at which point it erodes ( line 9) and becomes colluvium. The
colluvium then settles to a stable location and begins to
weather. The colluvium voxels take on the characteristics of
the voxels within the layer in which they settle. However, if
v is already colluvium ( line 15), then v is simply removed
from the terrain and that voxel becomes air.
Musgrave’s thermal weathering algorithm is the inspira-tion
for our colluvium distribution algorithm. In Musgrave’s
algorithm, material on a steep slope is distributed to
neighboring locations, which results in talus 2 slopes at the
base of mountains. Our weathering simulation also mimics
the accumulation of material but does so in three dimensions
on a voxel grid rather than on a heightmap.
Colluvium distribution is a two- step process, which is
repeated until the colluvium comes to rest in a stable
location ( lines 22- 32). First, we allow the colluvium to slide
( lines 25- 29) down steep slopes until reaching the angle of
repose. If the downward slope in the direction s of the
steepest downward gradient g is steeper than the angle of
repose, then the colluvium particle moves one unit in the
direction s. Then, we allow the colluvium to fall by iterating
toward the ground in the y direction until reaching a
nonair voxel ( lines 30 and 31). We repeat both steps until
the colluvium particle lies on either rock or colluvium on a
gradual slope at or below the angle of repose ( line 32).
Gradient is estimated by counting the number of voxels
containing air beginning at vs elevation continuing down-ward
to the first nonair voxel. This is done in each of the
four connected columns surrounding v. The direction of the
steepest gradient s points toward the column adjacent to v,
which contains the most air voxels below v. The steepest
gradient g is compared with the angle of repose by
estimating the slope angle using the number of voxels
containing air between vs elevation and the first nonair
voxel below. For example, if the angle of repose is
45 degrees and the steepest gradient lies in a column with
one voxel containing air between vs elevation on the first
nonair voxel, then colluvuium movement halts because
arctan1= 1Þ 45.
When the user decides that the weathering is complete
( line 11), the marching cubes algorithm [ 38] is implemented
to generate a surface, which can be imported into third
party rendering tools. The colluvium and rock are exported
as different surfaces in the same file so that different
textures can easily be applied to each. We duplicate
geometry wherever the rock and colluvium are adjacent
to prevent gaps between the colluvium and rock surface.
Before the surfaces are split into rock and colluvium, the
vertex positions are averaged with their neighbors’ to
smooth the surface boundaries.
4.3 Curvature Estimation
The algorithm estimates the mean curvature at surface
voxel v by counting voxels in a fixed volume around v.
Given a three- dimensional curved surface, the mean
curvature H at some point p on the surface is
H
1 2
2
;
JONES ET AL.: DIRECTABLE WEATHERING OF CONCAVE ROCK USING CURVATURE ESTIMATION 87
Fig. 5. Weathering and erosion algorithms.
2. Talus is a kind of colluvium which is composed of rock.
where 1 and 2 are the principle curvatures at p. The
principle curvatures at p are the maximal and minimal
normal curvatures at p. The sign of 1 or 2 is positive if the
surface curves away from the surface normal and negative
if the surface curves toward the surface normal. In general,
points which lie in a concavity have negative curvature and
points which lie on corners have positive curvature.
The algorithm estimates mean curvature rather than
Gaussian curvature ( which is the product 1 2) because
the sign of the mean curvature is a better predictor of the
ratio of exposed surface area to volume than the sign of
the Gaussian curvature, and this ratio is a key factor in the
underlying geomorphological process. For example, the
Gaussian curvature is positive for points which lie in a
concavity because both principle curvatures are negative.
The ratio of rock to air in a fixed volume around surface
voxel v is a reasonable approximation of the mean
curvature at v if the surface is smooth and thick in the
neighborhood of v. The neighborhood of v is the area
contained in the volume used to approximate curvature at
v. A surface is smooth at v if the rate of change of the
direction of the surface normal in the neighborhood of v is
constant or small. A surface is thick at v if the neighborhood
of v contains no surface points with surface normal equal to
the negation of the surface normal at v.
Let r represent the voxels containing rock and a
represent the voxels containing air within the neighborhood
of v. If the surface is smooth and thick at v, then the mean
curvature at v is proportional to a r. If the surface is flat at
v, then a r 0 and the mean curvature is 0. If the surface
is concave at v, then a r < 0, because more voxels contain
rock than air and the mean curvature is also less than 0.
Similarly, if the surface is convex at v, then a r > 0,
because more voxels contain air than rock and the mean
curvature is also greater than 0.
There are several significant cases in which the estimated
curvature is incorrect. A voxel, which lies on a saw- tooth
surface, will have estimated curvature at or near 0 if the
neighborhood is sufficiently large. A voxel which lies on a
flat thin wall on the inside of a concavity will have
estimated large positive curvature because air outside the
wall will be included in the neighborhood.
We can skew the estimated curvature to exaggerate the
contribution of the normal curvature to the mean curvature
in a given direction by changing the shape of the bubble.
For example, an ellipsoid bubble elongated in the vertical
direction includes more voxels containing rock if the
normal curvature in the vertical direction is negative. This,
in turn, overestimates the contribution of vertical curvature
in the estimate of the mean curvature.
4.4 Implementation Details
The implementation is designed to save space and time. We
save space by instantiating only the voxels on or near the
rock surface. A three- dimensional array of pointers to voxel
objects represents the simulation space. This reduces space
overhead by allowing us to store a pointer per voxel cell
rather than an entire voxel object per voxel cell.
Fig. 6a identifies three groups in the voxel cell space
which can be handled differently to save space. Air and
unexposed rock can be represented by copies of the null
pointer and copies of a pointer to a default rock object. Only
surface voxels require representation by unique objects.
Array entries for voxels on the surface point to the voxel
object for that surface location. When default rock voxels
become exposed to weathering, a new voxel object is
created and the pointer in the array is updated. Colluvium
voxels are handled like exposed rock voxels.
The weathering algorithm, as written, has a time bound
that contains a cubic term in the bubble radius because the
entire volume of each bubble must be visited in order to
compute the percentage of the bubble, which contains air. If
the bubble radius is R and the number of voxels within d
units of the surface is V , then each round of simulation will
require time on the order of
OðR3V Þ:
Although the bubble radius R is often small ( on the order of
four or five) compared to the number of voxels V ( on the
order of 10,000s to 100,000s), the operation is costly because
it must be performed once per voxel per weathering step.
We reduce the time bound on each weathering step by
computing the percentage of air in a bubble once and
storing information related to the weathering rate. How-ever,
that information must also be updated as the rock
shape changes.
After the initial computation of the percentage of air in a
bubble, the ratio is used along with the durability to
compute a weathering rate and determine when the voxel
will become fully decimated. If voxel v weathers at rate rv,
then rv is stored with v and reused during subsequent
simulation steps. The bubble is then ignored until v
becomes fully decimated or until a voxel in the bubble for
v becomes fully decimated.
If the bubble for voxel v contains a voxel w and w
becomes fully decimated before v, then v must update its
weathering rate based on the loss of w, as illustrated in
Fig. 6b. When voxel w becomes fully decimated, voxel w
notifies all neighbors within a distance of r from w on the
rock surface. The update radius r is chosen so that all voxels
which may contain w in their bubbles are notified. 3 When
voxel v is notified that a voxel in its bubble has become fully
decimated, voxel v increases its weathering rate rv.
88 IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, VOL. 16, NO. 1, JANUARY/ FEBRUARY 2010
3. The update radius r is not just the radius of the bubble centered on w
because neighboring voxels may have differently sized bubbles.
Fig. 6. ( a) Voxels are instantiated only for voxels, which lie on the rock
surface. Voxels containing air and unexposed rock are collectively
represented by either the null pointer or a pointer to a default rock object.
( b) Time is saved by storing rather than recomputing the weathering
rate. However, this requires voxels to notify each other when the
weathering rate may change. When voxel w becomes fully decimated,
voxel v must update its weathering rate.
If D voxels become fully decimated and N voxels are
newly exposed behind the decimated voxels, then new time
bound for a single simulation cycle is, on average,
OðR3N þ R2D þ RV Þ; ð1Þ
because the weathering rate for the newly exposed voxels
must be initialized, the neighbors of the fully decimated
voxels must be notified, and the rest of the voxels checked
to see if they are fully weathered.
5 USER INTERFACE
The user interface is designed to give the user top- down
artistic control over the weathering process while allowing
the simulation to manage the problem of creating the look
of weathered rock based on that input.
User input is required to set the initial rock shape, the
durability of rock layers, and to determine when the
simulation is complete. User input is given by editing
durability and bubble graphs. The durability graph allows
the user to specify the target profile of the weathered rock. The
final rock profile may not exactly match the durability graph
profile, but results in a weathered rock with a similar shape.
Fig. 7 illustrates each step in the creation of a weathered
rock column. First, the initial shape of the rock column is
specified by a black and white image as shown in Fig. 7a. That
image is used to create a rock column as shown in Fig. 7b.
The durability for each rock layer is set using the
durability graph shown on the right side of Fig. 7b. In the
durability graph, the vertical axis corresponds to altitude
and the horizontal axis corresponds to durability with
greater durability to the right. The user clicks on the line
segment to create or select a vertex, which can then be
dragged to alter the durability. Clicking the right mouse
button on a durability graph line segment allows the user to
increase or decrease the period or amplitude of the Perlin
noise for that line segment. The color of the line segment
and the surface of the rock are keyed to the durability with
red hues representing durable rock and blue hues repre-senting
less durable rock. Later, in the simulation, purple
will be used to indicate accumulated colluvium.
The bubble sizes and shapes are set using the bubble graph
shown on the left side of Fig. 7b. As with the durability graph,
clicking on the graph creates or selects the boundary between
two bubble types. The bubble type includes a bubble
collection type in which a set of bubbles is distributed across
the surface and used to estimate the curvature.
Fig. 7c shows the final bubble and durability graphs
before weathering. The bubble graph specifies three bubble
shapes and sizes. In the durability graph, the segments
between two layer endpoints are curved based on the
amplitude and frequency of the noise added to the
durability of each segment. The effects of the noise can
also be seen in the coloration of the rock column.
Specifying durability by editing curves gives the user
visual feedback on the approximate profile of the future
appearance of the weathered rock. Durability can also be
edited during the weathering process. This allows the user
to interactively change the weathering effect during the
weathering simulation. The process is reminiscent of wheel-thrown
pottery in which the artist adjusts the shape of the
piece during the creation of the piece.
Fig. 7d shows the rock column after 12 cycles of
weathering simulation. The durability graph was edited
during weathering to modify the Perlin noise and reduce
the durability of the small layer in the middle of the base.
The profile of the final rock is similar to the profile of the
final durability graph. Parts of the rock that are covered by
colluvium are shown in purple.
The user can then implement the marching cubes algo-rithm
to export the rock and colluvium surfaces. The resulting
surfaces can be imported into third party tools for rendering.
The weathered shape in Fig. 7d was used to create the image
shown in Fig. 8 using Vue- 6 from e- On software.
JONES ET AL.: DIRECTABLE WEATHERING OF CONCAVE ROCK USING CURVATURE ESTIMATION 89
Fig. 7. Steps in making a weathered rock column.
Fig. 8. A scene containing the final weathered rock shape from Fig. 7d
along with several other weathered rock shapes which were also
created using the weathering simulation.
6 RESULTS
The results demonstrate the use of positive and negative
curvature in weathering simulation, user control over the
weathering process, and the variety of rock shapes which
can be created using the algorithm.
6.1 Curvature in Weathering Simulation
Estimates of mean curvature are used to control the weath-ering
rate. Surface points with large positive curvature
experience spheroidal weathering while surface points with
large negative curvature experience cavernous weathering.
Fig. 9 shows the estimated curvature on the surface of a
cube. Red and orange colors correspond to large positive
curvature while green corresponds to low or no curvature.
After 13 rounds of spheroidal weathering, the cube weathers
into a sphere as predicted by the underlying geomorpholo-gical
theory. Curvature on the final sphere is more uniform
than curvature on the original cube. The final image shows
the surface of the sphere modeled at a higher resolution.
Resolution in the user interface ( from which the first three
images were captured) is kept low to reduce the time spent
extracting a surface from the internal voxel representation.
Fig. 10 contains a rock shape with a fold, which creates a
region of negative curvature. Surfaces with large negative
curvature are colored blue. Cavernous weathering equalizes
negative curvature by creating spherical caverns inside the
rock shape. The image on the right side of Fig. 10 contains a
cavern in which the negative curvature is more uniform than
in the original rock shape. This shape was generated by
applying 20 rounds of cavernous weathering with a round of
spheroidal weathering between the 17th and 18th rounds.
Spheroidal weathering was applied to make the cavern
visible by removing the outer cavern walls ( which have high
positive curvature). Rockfall simulation removed material
left suspended by the weathering process. Colluvium
deposition was also disabled. The rock durability was set to
restrict the cavern to the middle of the rock by placing a less
durable layer between two durable layers.
The curvature estimation overestimates positive curva-ture
for voxels on flat faces near corners, edges, or caverns.
Curvature is estimated by checking voxel contents in a
bubble surrounding a voxel. The curvature estimate is
incorrect near edges, corners, and caverns because these
voxels are surrounded by more voxels, which contain air
rather than rock. Similarly, curvature estimation under-estimates
negative curvature for voxels inside thin- walled
caverns. Curvature estimation through ray casting, as in
[ 34], may reduce this error.
Errors in the curvature estimation can be seen in Figs. 9 and
10. The cube in Fig. 9 includes regions of high estimated
positive curvature near the corners. These regions are shown
in red. Similarly, the weathered rock on the right side of
Fig. 10 includes high estimated positive curvature on the rock
face to the right of the cavern. In both cases, the actual surface
has zero curvature on a flat rock face. In the case of caverns,
the curvature estimate is more skewed because the volume
behind the flat cavern face contains air rather than rock.
Skewed curvature estimates using asymmetric bubbles
can be exploited to obtain different weathering effects.
Fig. 11 shows the shapes created by weathering a cube
using curvature estimated by, respectively, a sphere, a
gumdrop, and a disc. Colluvium deposition was disabled
during weathering simulation for these shapes. Fig. 12
shows the effects of colluvium deposition on weathering.
The shape in Fig. 12 was created using the same process as
90 IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, VOL. 16, NO. 1, JANUARY/ FEBRUARY 2010
Fig. 9. Spheroidal weathering equalizes positive curvature across the rock surface. This process transforms a cube of uniform material into a sphere.
The first three images depict that process and the final image shows the final sphere at a higher resolution. Surface color for the first three images is
a function of the estimated curvature with red corresponding to estimated positive curvature.
Fig. 10. Cavernous weathering equalizes negative curvature by creating
spherical caverns inside the rock. Color is based on estimated curvature
with blue corresponding to estimated negative curvature.
Fig. 11. A cube weathered into different shapes using different bubbles
shapes to estimate curvature. The weathering shape resembles the
bubble shape.
the spheres in Figs. 9 and 11 but with colluvium deposition
enabled. Colluvium protects rock at the base, which causes
the base to weathering more slowly.
Fig. 13 shows the effect of varying the function, which
relates the curvature estimate to the decimation experienced
by a voxel ( this is the function f in lines 6 and 8 in Fig. 5). The
rock column on the left was created with the decimation rate
equal to the estimated curvature c and the rock column on the
right was created with the decimation rate equal to 3c4.
Setting the decimation rate per unit time equal to the fourth
power of the curvature smooths the surface because protrud-ing
regions with large positive curvature experience more
decimation per unit time than regions with low or no
curvature. All other figures in this paper were created with
decimation equal to 3c3.
6.2 User Control
The user controls the weathering process by setting
durability within a layer, varying durability within a layer,
and changing the way curvature is estimated. All three can
be done during weathering simulation. The first two tools
correspond to geomorphological processes while the third
is a useful artifact of the weathering algorithm.
Fig. 14 contains a screenshot of an unweathered rock
column as seen in our weathering tool. This rock column
will be used throughout the section to illustrate the user
control over different weathering effects.
The rock column contains 23,452 voxels either on the
surface or one voxel below the surface. The time required
to complete one cycle of the weathering simulation for
this column varies with the number of voxels being
removed, instantiated, or ignored, and the bubble radius,
as described by ( 1), but ranges from around 0.2 second
( all times exclude extraction of the surface used in the
display) for bubbles with radius 3 to around 3.0 seconds
for bubbles with radius 7. The weathered shapes based
on this column, except Fig. 18b, were each generated
using between 10 and 15 seconds of weathering time.
Fig. 18b requires more weathering simulation time
because using a collection of bubbles in a layer adds an
additional layer of indirection inside the critical loop of
the weathering simulation.
The first tool for artistic control of the weathering process
is the durability graph. The durability graph is used to set
the durability within each rock layer. The spheroidal
weathering simulation creates rock shapes with silhouettes
that match the profile of the durability graph while still
maintaining the appearance of weathered rock.
Fig. 15 contains two weathered rock columns with
different layer durabilities. The bubble graph is omitted as
the bubble graph, shown in Fig. 14, is used in both cases.
The silhouette of the final weathered rock column resem-bles
the shape of the durability graph with two exceptions
that are related to weathering. First, the top layer of rock
weathers more quickly than similarly durable rock lower in
the column. This can be seen on the right side of Fig. 15 and
JONES ET AL.: DIRECTABLE WEATHERING OF CONCAVE ROCK USING CURVATURE ESTIMATION 91
Fig. 12. A cube weathered using spheroidal weathering but with
colluvium deposition enabled.
Fig. 13. Effect of varying the relationship between the decimation rate
and the estimated curvature.
Fig. 14. A column of rock which will be used to illustrate different
weathering effects created by varying the layer durability. In this image
as well as in Figs. 15, 16, 17, 18, surface color corresponds to durability
with green representing less durable and red representing more durable
rock.
Fig. 15. The weathering simulation creates visually plausible models of
weathered rock columns, which match the shape of the durability graph.
happens because the top layer of rock is more exposed to
weathering elements, such as air, than the lower layers.
The second exception is due to the colluvium accumula-tion.
Both rock columns in Fig. 15 contain accumulated
colluvium at the base of the column. The column on the right
side of the figure also contains accumulated colluvium on
intermediate layers.
The relationship between the durability graph and the
result of cavernous weathering is more difficult to see because
cavernous weathering does not influence the final shape
profile. Tohelp the user see the effect of cavernous weathering
before applying cavernous weathering, we set the surface
color to a color between bright red and purple depending on
the decimation to be generatedbycavernous weathering. This
is computed by taking the product of the negative curvature
and the durability for each voxel on the rock surface. Fig. 16
highlights the locations of cavernous weathering for an
intermediate step in the creation of the shape in Fig. 10.
Perlin noise is the second user tool for controlling the final
rock shape. The user can add different kinds of Perlin noise
with varying frequency and amplitude in order to create
durability variations within layers. Adding durability varia-tions
within layers creates organic shapes, which more
closely resemble natural rock. Durability variations within
layers are created and displayed in the durability graph.
Fig. 17 demonstrates the effect of varying durability
within rock layers. The same durability graph, bubble, and
simulation duration were used in both cases with two kinds
of Perlin noise added to the rock column on the right. The
top layer contains low frequency, low amplitude noise, as
might be seen in more durable rock, and the bottom layer
contains high frequency, high- amplitude noise, as might be
seen in less durable rock.
The final user control of weathering is the size and shape of
the bubble used to estimate curvature. Altering the bubble
size and shape allows the user to exaggerate or suppress
horizontal or vertical curvature. The effect of bubble size and
shape is more subtle than the variations in durability.
Fig. 18 illustrates the use of bubbles to change weathered
shape of the rock column from Fig. 14. In Fig. 18a, a sphere is
used to estimate curvature in the top layer and a small disc is
used to estimate curvature in the bottom layer. The disc-shaped
bubble results in faster weathering because a disc-shaped
bubble centered on a surface voxel contains a larger
percentage of air than a sphere- shaped bubble at the same
location. In this case, the influence of the vertical curvature
reduces the weathering rate because the curvature is constant
when traveling vertically on the local surface.
Fig. 18b shows the effect of using Perlin noise to
distribute different bubble types and sizes within a layer.
The resulting effect is similar to using Perlin noise to vary
durability within a layer but is harder to control because it
is more difficult to visualize and understand distributions
of bubble types than it is to visualize and understand
variations in durability.
6.3 Variety of Rock Shapes
Figs. 19 and 20 contain a rendering of weathered rock
shapes generated by our algorithm. The shapes in these
92 IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, VOL. 16, NO. 1, JANUARY/ FEBRUARY 2010
Fig. 16. The surface is colored bright red and purple to highlight the
areas where cavernous weathering will produce decimation.
Fig. 17. Adding Perlin noise to layer durability creates variety within rock
layers as found in natural rock.
Fig. 18. ( a) The disc- shaped bubble suppresses vertical curvature
estimates in the weathering algorithm, which results in faster weathering
compared to the sphere- shaped bubble. ( b) Perlin noise can also be
used to distribute bubble sizes and types.
Fig. 19. Scenes with rock shapes created using simulation of spheroidal
and cavernous weathering.
images demonstrate the range of effects that can be
achieved by varying the simulation settings using the
durability and bubble graphs.
7 CONCLUSION AND FUTURE WORK
Directable weathering of exposed rock with concave surfaces
can be achieved through mean curvature estimation on voxel
grids. Curvature estimation supports the simulation of
spheroidal and cavernous weathering. The user interface
allows the specification of a profile shape and the weathering
simulation fills in the details to create the look of weathered
rock with the given profile. The weathering simulation is not
intended to be scientifically accurate, but the underlying
geomorphological process provided a useful metaphor for
both the algorithm and the user interface.
The weathering algorithm presented here may be more
efficient if defined on a geometric representation other than a
voxel grid. This would avoid the extraction of a surface mesh
from the voxel grid and may simplify curvature estimation.
For example, local surface curvature on a mesh representa-tion
can be computed directly from the mesh representation.
We have discussed weathering primarily in the context
of sedimentary sandstone rock. More specifically, the
layered rock model used here targets sedimentary rock
with horizontal layers. However, both cavernous and
spheroidal weathering occur in other rock types, such as
granite [ 39], [ 40]. Future work might include extending the
rock model to include characteristics of other rock types.
The weathering algorithm based on curvature estimation
could then be applied directly to the resulting rock models.
The algorithm may be parallelizable as the weathering
behavior depends on localized interactions between surface
points, which can be distributed across processing engines.
The primary difficulty in parallelizing the algorithm lies in
synchronizing colluvium deposition. Colluvium deposition
allows particles to interact in ways that are not easily limited
to one part of the rock surface, which makes it difficult to
partition the work. Modifying the weathering algorithm for
implementation on graphics processing units ( GPUs) may
allow the algorithm to be applied on larger scales.
This work, like other work in geomorphological terrain
generation, focuses on a small set of specific processes in
isolation. Further work to unify several geomorphological
approaches, such as [ 27], [ 8] with the algorithm presented
here, on a single data structure may result in a tool for
creating a wider range of terrains.
REFERENCES
[ 1] Indiana Jones and the Temple of Doom, Lucasfilm, Ltd., 1984.
[ 2] The Mormons Part 2: Church and State, Public Broadcasting Service,
Inc., 2007.
[ 3] Wile E. Coyote and Roadrunner: Zoom and Bored. Warner Brothers,
1957.
[ 4] Boundin. Pixar, 2003.
[ 5] Galaxy Quest. Dreamworks SKG, 1999.
[ 6] M. Gamito and F. K. Musgrave, “ Procedural Landscapes with
Overhangs,” Proc. 10th Portuguese Computer Graphics Meeting,
pp. 33- 42, 2001.
[ 7] T. Ito, T. Fujimoto, K. Moraoka, and N. Chiba, “ Modeling Rocky
Scenery Taking into Account Joints,” Proc. Computer Graphics Int’l
Conf., pp. 244- 247, 2003.
[ 8] B. Bene s, V. T e si ´ nsk y, J. Horny s, and S. Bhatia, “ Hydraulic
Erosion,” Computer Animation and Virtual Worlds, vol. 17, no. 2,
pp. 99- 108, 2006.
[ 9] F. K. Musgrave, C. E. Kolb, and R. S. Mace, “ The Synthesis and
Rendering of Eroded Fractal Terrains,” Proc. ACM SIGGRAPH,
pp. 41- 50, 1989.
[ 10] J. Dorsey, A. Edelman, H. Jensen, J. Legakis, and H. Pedersen,
“ Modeling and Rendering of Weathered Stone,” Proc. ACM
SIGGRAPH, pp. 225- 234, 1999.
[ 11] Bryce 6.0. DAZ Productions, 2006.
[ 12] Vue- 6 infinite. e- on Software, 2006.
[ 13] Maya 2008. Autodesk, 2008.
[ 14] Blender 2.48a. Stichting Blender Foundation, 2008.
[ 15] S. W. Wang and A. E. Kaufman, “ Volume Sculpting,” Proc. Symp.
Interactive 3D Graphics ( SI3D ’ 95), pp. 151- 156, 1995.
JONES ET AL.: DIRECTABLE WEATHERING OF CONCAVE ROCK USING CURVATURE ESTIMATION 93
Fig. 20. A variety of weathered rock shapes created through simulation of spheroidal and cavernous weathering.
[ 16] M. Beardall, M. Farley, D. Ouderkirk, J. Smith, C. Rheimschussel,
M. Jones, and P. Egbert, “ Goblins by Spheroidal Weathering,”
Proc. Eurographics Workshop Natural Phenomena, pp. 1- 8, 2007.
[ 17] H. Huinink, L. Pel, and K. Kopinga, “ Simulating the Growth of
Tafoni,” Earth Surface Processes and Landforms, vol. 29, pp. 1225-
1233, 2004.
[ 18] A. V. Turkington, “ Cavernous Weathering in Sandtone: Lessons to
be Learned from Natural Exposure,” Quarterly J. Eng. Geology,
vol. 31, pp. 375- 383, 1998.
[ 19] M. Milligan, “ Geology of Goblin Valley State Park, Utah,” Geology
of Utah’s Parks and Monuments, Utah Geological Assoc. and Bryce
Canyon Natural History Assoc., pp. 421- 432, 2003.
[ 20] P. Prusinkiewicz, L. Mundermann, R. Karwowski, and B. Lane,
“ The Use of Positional Information in the Modeling of Plants,”
Proc. ACM SIGGRAPH, pp. 289- 300, 2001.
[ 21] Y. Dobashi, K. Kusumoto, T. Nishita, and T. Yamamoto, “ Feed-back
Control of Cumuliform Cloud Formation Based on Compu-tational
Fluid Dynamics,” Proc. ACM SIGGRAPH, vol. 27, no. 3,
2008.
[ 22] T. Vicsek, “ Pattern Formation in Diffusion- Limited Aggregation,”
Physical Rev. Letters, vol. 53, no. 24, pp. 2281- 2284, Dec. 1984.
[ 23] P. Pimienta, W. Carter, and E. Garboczi, “ Cellular Automaton
Algorithm for Surface Mass Transport Due to Curvature
Gradients: Simulation of Sintering,” Computational Materials
Science, vol. 1, pp. 63- 77, 1992.
[ 24] B. Mandelbrot, The Fractal Geometry of Nature. W. H. Freeman and
Co., 1982.
[ 25] A. Fournier, D. Fussell, and L. Carpenter, “ Computer Rendering
of Stochastic Models,” Comm. ACM, vol. 25, no. 6, pp. 371- 384,
June 1982.
[ 26] A. Kelley, M. Malin, and G. Nielson, “ Terrain Simulation Using a
Model of Stream Erosion,” Proc. ACM SIGGRAPH, pp. 263- 268,
1988.
[ 27] N. Chiba, K. Muraoka, and K. Fujita, “ An Erosion Model Based on
Velocity Fields for the Visual Simulation of Mountain Scenery,”
J. Visualization and Computer Animation, vol. 9, no. 4, pp. 185- 194,
1998.
[ 28] F. Belhadj and P. Audibert, “ Modeling Landscapes with Ridges
and Rivers: Bottom up Approach,” Proc. Third Int’l Conf. Computer
Graphics and Interactive Techniques in Australasia and South East Asia
( GRAPHITE ’ 05), pp. 447- 450, 2005.
[ 29] Fractals in Nature: From Characterization to Simulation, H.- O. Peitgen
and D. Saupe, eds., ch. 1, pp. 21- 70. Springer- Verlag, 1988.
[ 30] P. Prusinkiewicz and M. Hammel, “ A Fractal Model of Mountains
with Rivers,” Proc. Graphics Interface Conf., pp. 174- 180, 1993.
[ 31] K. Nagashima, “ Computer Generation of Eroded Valley and
Mountain Terrains,” The Visual Computer, vol. 13, pp. 456- 464,
1997.
[ 32] O. St’ava, B. Bene s, M. Brisbin, and J. K riva ´ nek, “ Interactive
Terrain Modeling Using Hydraulic Erosion,” Proc. ACM Euro-graphics/
Symp. Computer Animation, 2008.
[ 33] Y. Chen, L. Xia, T.- T. Wong, X. Tong, H. Bao, B. Guo, and H.- Y.
Shum, “ Visual Simulation of Weathering by - Ton Tracing,” Proc.
ACM SIGGRAPH, pp. 1127- 1133, 2005.
[ 34] S. Hsu and T. Wong, “ Simulating Dust Accumulation,” IEEE
Computer Graphics and Applications, vol. 15, no. 1, pp. 18- 22, Jan.
1995.
[ 35] R. Sarracino, G. Prasad, and M. Hoohlo, “ A Mathematical Model
of Spheroidal Weathering,” Math. Geology, vol. 19, pp. 269- 289,
1987.
[ 36] G. Miller, “ Efficient Algorithms for Local and Global Accessibility
Shading,” Proc. ACM SIGGRAPH, pp. 319- 326, 1994.
[ 37] K. Perlin, “ An Image Synthesizer,” Computer Graphics, vol. 19,
no. 3, pp. 287- 296, 1985.
[ 38] W. Lorensen and H. Cline, “ Marching Cubes: A High Resolution
3D Surface Construction Algorithm,” Proc. ACM SIGGRAPH,
pp. 163- 169, 1987.
[ 39] C. R. Twidale and J. R. V. Romani, Landforms and Geology of Granite
Terrains. CRC Press, 2005.
[ 40] Y. Matsukura and Y. Tanaka, “ Effect of Rock Hardness and
Moisture Content on Tafoni Weathering in the Granite of Mount
Doeg- Sung, Korea,” Geografiska Annaler, vol. 82, pp. 59- 67, 2000.
Michael D. Jones is an associate professor of
computer science at Brigham Young University,
where he directs the Laboratory for computer-generated
natural phenomena. His research
interests include animation and modeling of
natural phenomena using algorithms based on
physical models. He is a member of the IEEE.
McKay Farley received the BS degree in
computer science from Brigham Young Univer-sity,
where he is currently working toward the
MS degree in computer science. His research
interests include graphics, modeling, rendering,
and special effects.
Joseph Butler received the BS degree in
computer science from Brigham Young Univer-sity,
Idaho, where he is currently working toward
the MS degree in computer science. His
research interests include graphics, computer
vision, and biometrics.
Matthew Beardall received the BS degree in
computer science from Brigham Young Univer-sity.
He is currently working as a software
engineer at Caselle, Inc. His research interests
include graphics and persistent languages.
. For more information on this or any other computing topic,
please visit our Digital Library at www. computer. org/ publications/ dlib.
94 IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, VOL. 16, NO. 1, JANUARY/ FEBRUARY 2010