TRANSPORTATION RESEARCH RECORD 1541 133

Seismic Analysis of Large Buried

Culvert Structures
PETER M. BYRNE, D. L. ANDERSON, AND HENDRA JITNO

Field experience indicates that large buried culverts have suffered essen- The properties of the soil used in the analyses are presented in
tially no damage during past earthquakes when no significant permanent Table 1, and were obtained from published data on similar soil (2,3).
ground movements have occurred. These soil structures, which gener- These values are appropriate for first-time loading and were used to
ally comprise steel or concrete arch members and engineered soil, may
simulate construction of the backfill of the culvert. Under repeated
have spans of 15 m. Static, pseudodynamic, and dynamic finite-element
analyses have been carried out on these structures and indicate that for loading that occurs during an earthquake, values higher by a factor
horizontal seismic loading, the surrounding soil is much stiffer than the of 3 were used.
arch and results in the seismic load being taken by the soil rather than The properties used for the beam members are representative of
by the arch. Under vertical seismic loading, the arch is stiffer than the a lightly reinforced 300-mm-thick concrete section and are as fol-
surrounding soil and attracts significant load, which can essentially be lows: E 5 20,000 MPa; I 5 0.00225 m4/m; and A 5 0.30 m2/m.
accounted for by increasing the unit weight of the soil in proportion to
Gross concrete properties have been used because the arch is mainly
the vertical acceleration. Thrusts and moments in a 10-m concrete arch
are examined under combined static and seismic loading (both horizon- under high axial compression and very little cracking would be
tal and vertical). The results indicate that significant increases in thrust expected. The program used in the static and pseudostatic analysis
and moment in the arch are predicted for peak ground accelerations in has the ability to consider nonlinear flexure of the beam members;
excess of 0.3 g. The good behavior of these structures under such accel- however, the moments never reached the yield moment capacity of
eration levels in California, where they are not specifically designed for a lightly reinforced 300-mm-thick section, and thus this feature was
earthquake forces, indicates that their static design includes sufficient not used.
reserve to prevent failure under accelerations of these levels.

In general, both concrete and metal buried culvert structures have METHOD OF ANALYSIS
suffered little damage from pressures arising from ground shaking
where no significant permanent ground movements have occurred Analyses of the earthquake loading were carried out at two levels: (a)
(1). In many cases, these structures have been subjected to vertical simple pseudodynamic finite-element analyses in which inertia
and horizontal earthquake ground shaking in excess of 0.5 g with- forces corresponding to the peak ground acceleration considered are
out damage. Some damage from shaking to unreinforced concrete applied to soil elements in addition to the static gravity forces, and
structures has occurred, but it is the exception rather than the rule. (b) dynamic finite-element analyses using acceleration time histories.
On the other hand, severe damage to both metal and concrete cul-
verts has occurred when significant permanent ground movements
Static Analyses
have occurred, generally caused by soil liquefaction.
The purpose here is to examine the forces and moments in culvert
The pre-earthquake static stress conditions are an important aspect
structures arising from ground shaking and to explain, if possible,
in the evaluation of the forces on the arch. The analyses were car-
their generally good behavior. A single large concrete arch structure
ried out using Byrne and Duncan’s computer code, Non-Linear Soil
with both high and low cover is considered in these analyses.
Structure Interaction Program (NLSSIP) (4). It is a nonlinear finite-
element computer code designed to analyze soil structure interaction
STRUCTURE ANALYZED of flexible culvert structures. The soil is modeled by two-
dimensional isoparametric finite elements, and the structure is mod-
The 3-pinned culvert structure analyzed is shown in Figure 1. It has eled by beam-column members. The program uses a hyperbolic
a span of 10.5 m, a rise of 5.2 m, and a thickness of 300 mm. Cover stress-strain model for soil, uses a linear plus hyperbolic moment-
depths of 3 and 15 m were considered. The zones of compaction are curvature relationship for beam members, and considers stability of
shown on the left side of the figure and the material types shown on the beam-soil system. Backfill construction can be simulated with
the right. The finite-element mesh used for the 15-m cover is shown the program to determine the initial static stresses and deformations
in Figure 2. The backfill was modeled by 522 elements and the cul- of the soil structure before earthquake excitation. During backfill-
vert by 14 beam members with hinges at the crown and footings. ing, it was assumed that the base was fixed and the left and right
The 3-m cover mesh is the same as shown in Figure 2, with the top boundaries were fixed in the horizontal direction only.
six layers of elements removed.

Pseudodynamic Approach
P. M. Byrne and D. L. Anderson, Department of Civil Engineering, Univer-
sity of British Columbia, Vancouver, B.C. V6T IZ4 Canada. H. Jitno,
Department of Civil Engineering, Bandung Institute of Technology, Ban- In the pseudodynamic approach, which used the NLSSIP code, the
dung, Indonesia 40132. earthquake forces are simulated by additional static forces, specified
FIGURE 1 Geometry and dimension of culvert and backfill (15-m depth of backfill).

FIGURE 2 Finite element mesh used for all analyses, 15 m cover.

Byrne et al. 135

TABLE 1 Soil Properties Used in Analyses The target spectrum was based on spectral attenuation relations
(mean 1 1 standard deviation) proposed by Idriss (8) for a Richter
magnitude 6.5 event at a distance of 30 km, which gives a peak
ground acceleration of 0.2 g and a peak spectral acceleration of 0.5
g at a period of 0.2 seconds, decreasing to about 0.2 g at 1.0 second.
Dynamic analysis for vertical motion was carried out using the
vertical motion record from the San Fernando earthquake recorded
at San Vicente Boulevard, Station 252, Los Angeles. The motion
was scaled to a peak ground acceleration of 0.13 g.
The results of both pseudodynamic and full dynamic analyses are
presented in the following sections.

by seismic coefficients kh and kv, applied to the elements. The forces

on any element of the structures are increased by kW, where W is the RESULTS
weight of the element and k 5 A/g, where A is the peak ground
acceleration and g is the acceleration of gravity. Analyses were run Culvert With 15-m Depth of Cover
for peak horizontal ground accelerations of 0.2 g, 0.3 g, and 0.4 g,
and for corresponding peak vertical ground accelerations of about 2⁄ 3 Pseudodynamic Analyses—NLSSIP
the horizontal accelerations, or 0.13 g, 0.2 g, and 0.26 g. During the
application of the horizontal seismic loading, which is treated as a Axial Forces Figure 3 shows the total axial forces developed in
separate load case, the horizontal constraints on the left and right the beam elements for the static loads and three different magnitudes
boundaries were removed. Modulus values used in this part of the of horizontal seismic coefficients (k h ). The distribution of axial
analyses were based on the strain levels determined from the force in the beam at the end of construction (CR-1) is symmetrical,
dynamic analyses, discussed in the next section. with maximum values occurring in the Beam Elements 1 and 14 at
the base, which exceed the weight of the overburden, indicating that
the arch is stiffer than the soil and negative arching has taken place.
Full Dynamic Analyses The effects of the application of horizontal seismic loads (applied
to the right) increase the axial loads in the beam members to the right
Dynamic analyses were carried out using the computer code side of the crown, but decrease them on the left side. The maximum
FLUSH (5). This is a total stress finite-element program to analyze thrust occurs at Beam 11 for kh 5 0.4 and is only a 7 percent increase
dynamic soil-structure interaction problems. It solves the equation over the maximum static thrust, but note that the location of maxi-
of motion in the frequency domain using stress-strain properties and mum thrust has changed from the base to the 1⁄ 4-span point.
damping values appropriate to the magnitude of the shear strain. Figure 4 shows the effects of the application of vertical seismic
Peak ground displacements, accelerations, and strains can be deter- loads on the axial forces in the culvert. The force distributions are
mined with the program. similar to those at the end of construction but are all increased. The
The full dynamic analyses carried out here are intended to verify larger axial forces are expected since the application of the vertical
the reasonableness of the results from the pseudodynamic analyses. seismic coefficient essentially increases the weight of the fill. The
By using this approach, the amplification and phasing effects on the increase in thrust in the beam members is approximately proportional
displacements and forces can be assessed. Only one level of peak to the magnitude of kv, where kv is the vertical seismic coefficient.
ground acceleration was analyzed for each of the horizontal and ver-
tical motions.
The finite-element mesh used is the same as that used in the pseu-
dodynamic analyses, with the ground acceleration input along the
base. Although the lateral boundary was reasonably far from the struc-
ture, energy transmitting boundary elements or boundaries were used
to ensure that there was no wave reflection from the boundary. The
nodes on the lateral boundary were free during horizontal loading, and
fixed in the horizontal direction during vertical loading.
The soil properties are similar to those used during backfill con-
struction and listed in Table 1, except that the maximum shear mod-
ulus (Gmax) of the soil was taken to be three times as stiff to account
for the very low strains involved and unload-reload effects as dis-
cussed by Byrne et al. (3). The modulus and damping reduction
curves of the soils as a function of the shear strain were obtained
from Seed and Idriss (6) as described in the FLUSH manual (5).
Both vertical and horizontal motions were considered; however,
the analyses were carried out separately for each direction. For the
horizontal input motion, the San Fernando earthquake (Lake
Hughes, Station 4, Comp. S69E) record, modified using the SYNTH FIGURE 3 Axial load developed in beams (culvert) for
computer program (7 ) to fit a target response spectrum, was used. horizontal seismic loads, Chase River, 15-m cover.
136 TRANSPORTATION RESEARCH RECORD 1541

FIGURE 4 Axial load developed in beams (culvert) for vertical FIGURE 6 Moment developed in beams (culvert) for vertical
seismic loads, Chase River, 15-m cover. seismic loads, Chase River, 15-m cover.

Moments Figure 5 shows the total moments (static 1 dynamic) depth were less than the base input acceleration, except at the sur-
in the arch for horizontal seismic loadings. As indicated, a positive face, where there was a slight amplification. Above the crown, the
moment produces compressive stresses on the inside of the culvert. accelerations were about 60 percent of the free field values.
In the region of high moments, the increase in moment is approxi- Vertical accelerations for the 0.13-g peak ground acceleration
mately linear with increasing kh. A kh 5 0.4 increases the maximum record generally showed amplification in the free field except above
moment by about 170 kN ? m/m, or about 140 percent of the maxi- the crown, where the average vertical acceleration was about equal
mum static moment of 120 kN ? m/m. to the base acceleration.
Figure 6 gives the moment response to vertical seismic loads and
shows an overall increase in moment, but the increase is not as large
as that caused by horizontal seismic loading. The maximum moment
increase (corresponding to kv 5 0.26) is 57 kN ? m/m, or about 50 Axial Forces The change in thrust from horizontal seismic
percent of the maximum moment at the end of construction. loading is shown in Figure 7, where it is seen that the thrust values
from FLUSH are much lower than from NLSSIP. This likely arises
from two factors:
Dynamic Analyses—FLUSH
• The horizontal accelerations in the soil above the crown com-
The horizontal peak accelerations in the fill for locations in both the puted from FLUSH are lower than the base values used in the
free field (away from the culvert) and above the crown were evalu- NLSSIP pseudodynamic analysis; and
ated using the 0.2-g peak ground acceleration modified record. In
general, it was found that free field accelerations throughout the

FIGURE 7 Dynamic axial loads developed in culvert:

FIGURE 5 Moment developed in beams (culvert) for horizontal pseudodynamic versus dynamic analysis. Horizontal loading,
seismic loads, Chase River, 15-m cover. Chase River, 15-m cover.
Byrne et al. 137

• The accelerations in the soil above and adjacent to the crown,

which induce load in the culvert, may not be in phase.

The thrust from vertical seismic loading is shown in Figure 8.

Again the thrust computed from FLUSH is lower than from NLSSIP
and is likely caused by phasing, because the peak vertical accelera-
tions are similar to that used in the pseudodynamic analysis.
Although the increased forces from the pseudodynamic and full
dynamic analyses do not agree well, the values are small compared
with the static axial thrust in the culvert.

Moments The changes in moments are shown in Figures 9 and

10 for horizontal and vertical input motions, respectively. For hori-
zontal loading, the pattern of agreement is quite good, with peak val-
ues of 87 and 65 kN ? m/m, respectively, from NLSSIP and FLUSH.
These represent increases of 75 and 55 percent over the static values.
For vertical seismic loading, the pattern of agreement is not so FIGURE 9 Dynamic moment developed in beam:
good, but the moment values of 27 and 9 kN ? m/m are small com- pseudodynamic versus dynamic analysis. Horizontal loading,
pared with the seismic moments from horizontal motion. Chase River, 15-m cover.

Culvert With 3-m Depth of Cover For vertical accelerations FLUSH gave much higher values than
NLSSIP, which is the opposite of the findings with 15-m cover.
Both pseudodynamic and full dynamic analyses were also carried However, the values are small compared with the change in moment
out for the culvert with 3-m fill over the crown. Only a horizontal from horizontal excitation.
acceleration of 0.2 g and a vertical acceleration of 0.13 g were con-
sidered. The method of analysis and the soil properties used were
the same as those used for the culvert with 15-m depth of backfill. DESIGN IMPLICATIONS
As with the 15-m cover case, the largest increase in axial forces
arises from the vertical loading case. NLSSIP predicts increased The seismic design of arch culverts is not specifically covered by
forces about three times the values given by FLUSH. A reasonably most building or bridge codes. In many codes, earthquake loading
conservative estimate of axial forces could be obtained by dividing is not required in the design of culverts, but it is not clear whether
the pseudodynamic values by a factor of two. this applies only to small circular culverts or to all buried culvert-
For horizontal accelerations, it was found that both NLSSIP and like structures. Nevertheless, for large arch culverts it is appropriate
FLUSH gave very similar changes in moments and that they were to carry out checks for seismic resistance.
nearly identical to the 15-m cover case (Figure 9), indicating that the In structural design, the loads are customarily classified as dead
depth of cover has little effect on dynamic moments induced by hor- or live loads, with their separate load factors, and then combined
izontal accelerations. with load combination factors to produce the factored design loads.

FIGURE 8 Dynamic axial loads developed in culvert: FIGURE 10 Dynamic moment developed in beam:
pseudodynamic versus dynamic analysis. Vertical loading, pseudodynamic versus dynamic analysis. Vertical loading,
Chase River, 15-m cover. Chase River, 15-m cover.
138 TRANSPORTATION RESEARCH RECORD 1541

In culvert design, it is not clear how the earth loads should be clas-
sified: are they dead loads to be treated the same way as the self
weight of the arch, or are they live loads to be applied to the arch?
Because the soil is part of the structure in these applications, the
answer is not apparent. The AASHTO bridge code treats earth loads
as dead loads, whereas the ACI code treats earth loads as live loads
on all structures. If the earth is considered as a dead load, most codes
still require that the load factor (typically 1.25 or 1.3) be applied to
the dead loads when considering the seismic effects. This would
practically guarantee that the seismic loading case would govern the
design, especially in culverts with high fills, where the live load
effects from, say, vehicles are very small. If the earth load is con-
sidered as live load, then the load factor is higher (typically 1.5 to
1.7), and the static load case would most likely govern the design.
However, the load factor of 1.5 to 1.7 is meant to be applied to the
more normal live loads, such as congregations of people, or to vehi-
cle loads, which are much more variable, and seems too high for
earth loads, especially vertical earth loads. On the other hand, the
load factor for dead loads of 1.25 or 1.3 may be considered to be too
small for earth loads when compared to concrete or steel loads.
It is suggested that a reasonable approach would be to consider
the earth as dead load with a load factor of about 1.4 for the static
design load case (or 0.9 if this produces a more critical load situa-
tion). For the seismic load case, a lower load factor for dead loads FIGURE 11 Axial force-moment results, 15-m cover, Beam 11,
Ah 5 0.2 g 1 Av 5 0.13 g.
of about 1.1 (or 0.9) may be appropriate because the probability of
getting a dead load overload and the design earthquake would be
low. In any case, as will be shown, this has little influence on the it has the highest axial load and moment from from the dynamic
design of this culvert. A load factor of 1.0 is applied to the dynamic analysis. It is seen that an increase in the moment and a reduction
loads as the ground motion is taken as the factored seismic motion. in the axial load bring the load case closer to the interaction curve,
Table 2 presents the results of applying the load factors just dis- and the most critical condition for this example arises when the
cussed to the forces for the 15-m depth of cover case. The first row dead load factor is taken as 0.9 and the dynamic axial load is taken
gives the design loads for the static case only (it is assumed that the as tension.
live loads from vehicles and such during the static load case are neg- If the axial load and moment reach the capacity of the section, the
ligible for such a deep cover). The second row gives the factored arch will yield in flexure and form a plastic hinge, but it should con-
dead loads for use in combination with the seismic loads, and the tinue to carry the axial load if the plastic hinge rotation is small.
third row gives the seismic loads for the Ah 5 0.2 g and Av 5 0.13 g Flexural yielding allows the arch to deform and shed load to the soil;
cases. The final row presents the design loads for the seismic case, and, provided the resulting deformations are not too large, the per-
in which the seismic forces are combined using the root sum square formance can be satisfactory. Yielding of the arch is not expected to
approach and then added to the factored dead loads. For the 0.9 dead lead to increased displacements because most of the load is carried
load factor case, the seismic thrust is taken as a reduction in the total by axial forces in the arch, and flexural yielding will not apprecia-
thrust because this gives a more critical load case. bly change the axial stiffness. Thus the plastic rotations at hinge
The sectional capacity of the arch is given by an axial load- points are not expected to be large if the calculated demand on the
moment interaction diagram, Figure 11. The two lines on the right section is not much greater than the section capacity. The amount of
are representative of the factored capacity of a 300-mm-thick sec- plastic curvature or rotation that the arch can withstand depends, in
tion with 1 percent and 2 percent of longitudinal reinforcement. Also this case, mostly on the level of axial load because the level of con-
plotted are the factored axial loads and moments for the static case finement provided by the reinforcement is expected to be low in
and for the dynamic and pseudodynamic analyses of the seismic such a structure. As Figure 11 illustrates, yielding is most likely to
case. Beam 11, the 1⁄ 4-point location on the arch, is chosen because occur when the axial load is decreased, and this improves the rota-
tional capacity of the arch; however, because of the substantial axial
loads, the curvature capacity is not large.
TABLE 2 Thrust and Moment at 1⁄ 4 Point with Load Factors1
(15-m cover)

FIELD EXPERIENCE

Field experience in California indicates that culvert structures have

generally behaved well during past earthquakes for acceleration lev-
els in excess of those considered in this study. These structures were
not designed specifically to resist earthquake loading, but their static
design had sufficient reserve to accommodate earthquake forces. The
one known exception to this was a culvert 8 ft in diameter through the
Lower San Fernando dam that collapsed during the 1993 Northridge,
Byrne et al. 139

California, earthquake. However, the soil surrounding this culvert –The dynamic moment increases approximately linearly with
likely liquefied. It is known that liquefaction did occur during the 1971 horizontal acceleration level, but the depth of cover has little
San Fernando earthquake before the installation of the culvert (1). effect.
Experience in California, together with the results of the analysis • Design Implications
herein, indicates that if the structure is properly designed and con- –The analysis shows that large increases in moment are pre-
structed to resist static loads, and the foundation and surrounding dicted to occur for high acceleration levels, and these are more
materials are competent, the structure is likely to resist earthquake important than the changes in axial force.
loads corresponding to peak ground accelerations of at least 0.3 g –Field experience in California indicates that if the structure is
horizontal and 0.2 g vertical. designed and constructed to resist static loads, and the foundation
and surrounding materials are competent, the structure is likely to
resist earthquake loads corresponding to peak ground accelera-
CONCLUSIONS tions of at least 0.3 g horizontal and 0.2 g vertical. Limited flex-
ural yielding may occur, causing some cracking, but the culvert
The results of pseudodynamic analysis using NLSSIP and dynamic should remain serviceable.
analyses using FLUSH on the effects of horizontal and vertical seis-
mic loadings can be summarized as follows:
REFERENCES
• Pseudodynamic analyses generally indicate that the loads on
the culvert increase approximately linearly with depth of cover, 1. Youd, T. L., and C. J. Beckman. Highway Culvert Performance During
while the full dynamic analyses indicate that the loads are only Earthquakes. Technical Report, NCEER Highway Project 106-B(1),
State University of New York at Buffalo, 1995.
mildly affected by cover depth. The main reason for this appears to 2. Duncan, J. M., P. M. Byrne, K. S. Wong, and P. Marby. Strength, Stress-
be the phasing that takes place between the motion of the side fill Strain and Bulk Modulus Parameters for Finite Element Analyses of
and the top fill, or cover. We believe that the full dynamic analyses Stresses and Movements in Soil Masses. Report UCB/GT/80-01, College
would more closely represent the likely field condition. of Engineering, University of California, Berkeley, 1980.
• Axial forces in the culvert: The dynamic thrust in the culvert is 3. Byrne, P. M., H. Cheung, and L. Yan. Soil Parameters for Deformation
Analyses of Sand Masses. Canadian Geotechnical Journal. No. 24, 1987,
controlled more by vertical than horizontal accelerations. pp. 366–376.
–Horizontal seismic motion: The maximum axial force or thrust 4. Byrne, P. M., and J. Duncan. NLSSIP: A Computer Program For Non-
in the culvert is not significantly increased by horizontal seismic Linear Soil Structure Interaction Problems. Soil Mechanics Series, No.
forces for accelerations up to 0.4 g. However, the location of the 41. Department of Civil Engineering, University of British Columbia,
Vancouver, Canada, 1979.
maximum moves from the base to the 1⁄ 4 point, and this must be 5. Lysmer, J., T. Udaka, C.-F. Tsai, and H. B. Seed. FLUSH: A Computer
considered when combining axial forces and moments. Program For Approximate 3-D Analysis of Soil-Structure Interaction
–Vertical seismic motion: Based on pseudodynamic analyses, Problems. Report EERC 75-30. College of Engineering, University of
the maximum thrust in the culvert increases linearly with the ver- California, Berkeley, 1975.
tical seismic coefficient, and is essentially equivalent to increas- 6. Seed, H. B., and I. M. Idriss. Soil Moduli and Damping Factors for
Dynamic Response Analysis. Report EERC 70-10. Earthquake Engineer-
ing the unit weight of the soil (g) by the seismic coefficient kv. The ing Research Center, University of California, Berkeley, 1970.
maximum value occurs at the base of the culvert. Values com- 7. Naumoski, N. SYNTH Program: Generation of Artificial Acceleration
puted from full dynamic analyses using FLUSH are about one- Time History Compatible with a Target Spectrum. McMaster Earthquake
half the values computed from the pseudodynamic NLSSIP Engineering Software Library, Department of Civil Engineering and
Engineering Mechanics, McMaster University, Hamilton, Canada, 1985.
analysis, and thus g* 5 g(1 1 1⁄ 2 kv) would be appropriate to use 8. Idriss, I. M. Procedures for Selecting Earthquake Ground Motions at
in the pseudodynamic analysis. Rock Sites. U.S. Department of Commerce, 1991.
• Moment
–The maximum moment is controlled by horizontal rather than
vertical seismic motion. The pseudodynamic and full dynamic
analysis results are close, with the pseudodynamic method gen- Publication of this paper sponsored by Committee on Subsurface Soil-
erally predicting slightly higher values. Structure Interaction.