See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/281618238

International Freight and Related Carbon Dioxide Emissions by 2050

Article  in  Transportation Research Record Journal of the Transportation Research Board · August 2015
DOI: 10.3141/2477-07

CITATIONS READS
4 202

3 authors, including:

L.M. Martinez Jari Kauppila

Organisation for Economic Co-operation and Development (OECD) Organisation for Economic Co-operation and Development (OECD)
57 PUBLICATIONS   1,016 CITATIONS    15 PUBLICATIONS   118 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

InnoVshare: Viability analysis of different carsharing system configurations through an innovative agent-based simulation model View project

SCUSSE: Smart Combination of passenger transport modes and services in Urban areas for maximum System Sustainability and Efficiency (MIT-Portugal) View project

All content following this page was uploaded by Jari Kauppila on 09 May 2016.

The user has requested enhancement of the downloaded file.

International Freight and Related
Carbon Dioxide Emissions by 2050
New Modeling Tool
L. Miguel Martínez, Jari Kauppila, and Marie Castaing
A new method to project the impacts of global trade scenarios on inter-
national freight flows is presented. The model uses international trade
scenarios developed by the Economics Department of the Organisation
for Economic Co-operation and Development for the period 2004–2060
with alternative liberalization scenarios. A global freight network model
is developed, and current exports data from Eurostat and the United
Nations Economic Commission for Latin America and the Caribbean are
used to calibrate a model converting trade in value into freight volumes
(in tonne-kilometers) assigned to a routable transport network for the
period 2010–2050. The results highlight the impact of changes in global
trade flows and carbon dioxide emissions related to international trade
and raise several policy implications. The outputs suggest that increases
in transport flow will be significant, especially within Asia and on routes
used for export–import activities from and to this region.
International trade has grown rapidly in the postwar era with trade
volume growing twenty-seven-fold between 1950 and 2007, three
times faster than the growth of the world gross domestic product
(GDP) (1). Growth in trade is expected to outpace the GDP growth
also over the next 50 years, according to recent projections by the
Organisation for Economic Co-operation and Development (OECD).
The value of international trade is estimated to grow by a factor of 4
by 2050 in real terms (2). Trade patterns will, however, change because
of the fragmentation of production processes and the integration of
emerging economies into global markets. Trade liberalization, either
at the global or regional level, will also have an impact on global
patterns.
Global value chains, dependent on relatively inexpensive and
reliable transport links, are key to economic development and
will be potentially affected by changes in trade and manufacturing.
The transport sector is interrelated with changes in international
pro­duction and consumption patterns as well as location choices
of multinational companies. Freight transport as derived demand
depends on the volume and type of goods produced and consumed
at different locations. Location of global production and consumption,
structure of trade in terms of the nature of goods trade, and transport
costs all influence the volume of freight as well as related mode and
route choice.
International Transport Forum, 2/4 Rue Louis David, Paris 75016, France.
Corresponding author: L. M. Martínez, Luis.Martinez@oecd.org.
Transportation Research Record: Journal of the Transportation Research Board,
No. 2477, Transportation Research Board, Washington, D.C., 2015, pp. 58–67.
DOI: 10.3141/2477-07
58
Yet the long-term impact of the evolution of trade flows on global
transport flows has been overlooked in research, with the exception
of some long-term aggregate projections over the RUBMRIO model
(3). There is also limited evidence on environmental effects caused
by the implied transport movements. This shortcoming is probably
partly due to lack of data and related projections and partly due to
the inability to estimate the transport component of international
trade. The literature has therefore focused on examining local or
regional transport flows in production centers and evaluating them
under traditional location theories to understand companies’ strat-
egies (4). There is also extensive literature on how transport costs
affect international trade (1, 5–8).
The most comprehensive effort to estimate international transport
flows and related greenhouse gas emissions associated with changes
in trade flows was carried out by Cristea et al. (9). They use data on
trade value, weight, and modal share for different origin–destination
(O-D) pairs for a base year; distance between each bilateral pair; and
the greenhouse gas intensity of each transport mode to arrive at related
tonne-kilometers and carbon dioxide (CO2) emissions. The results for
this initial year are then projected up to 2020 by using a computable
general equilibrium model for changes in value and composition of
trade resulting from tariff liberalization and GDP growth.
This method presents some shortcomings. The authors underline
the lack of heterogeneity in terms of products and geographical
location but also the linearity of the approach, not accounting for
the reduction in fixed costs with distance, especially in the maritime
option. Furthermore, the outputs of the model are based on great-
circle distances, not accounting for the geographical and network
availability specificities, which can produce significant biases.
This earlier work is built on by development of a new international
freight model that takes into account changes in location, changes in
the value–weight ratio for products over time, and reduction in fixed
costs with distance. This work is part of the Transport Outlook pre-
pared by the International Transport Forum at OECD. This model
projects international freight flows (and related CO2 emissions) for
26 regions and 25 product groups depending on alternative trade
liberalization scenarios.
A four-step model is developed based on a global freight network
model (with actual routes and related real distances) that converts
trade in value into freight volumes (in tonne-kilometers) and into
related CO2 emissions by transport mode and route for the period
2010–2050.
Since cargo does not stop at ports but continues to economic
centers, this domestic link is also estimated by introducing centroids
for production and consumption, allowing estimation of the freight
transported by road or rail from or to ports from consumption and
production centers.
Martínez, Kauppila, and Castaing 59

Methodology
General Framework
The model projects international freight transport activity until 2050
based on alternative world trade scenarios. The following five main
components are presented in Figure 1:
• A general equilibrium model for international trade developed
by OECD and covering 26 world regions and 25 commodities (2),
• A global freight transport network model based on 2010 data,
• An international freight mode choice model (in value) calibrated
with data from Eurostat and the United Nations Economic Commission
for Latin America and the Caribbean (ECLAC),
• A weight and value model using the same data to convert trade
value into weight and calibrated for each commodity and transport
mode, and
• CO2 intensities and technological pathways by mode.
Each model component output is used to feed the next step. The final
model output is international trade–related freight tonne-kilometers
by transport corridor by mode and related CO2 emissions.
Underlying Trade Projections
The Economics Department of OECD has published, in coopera-
tion with the Centre d’Etudes Prospectives et d’Informations Inter-

Inputs Method Output

Trade O-Ds by
World trade projecons by Producon/
commodity type
region and commodity type Consumpon
(dollars)
(2004–2060) centroids
(2010–2050)

Travel mes and distances

Network model
Trade O-Ds by
Trade value mode commodity type and
share model mode (dollars)
Geographic, trade, and (2010–2050)
economic profiles
(scenario based)

Travel mes and distances

Network model
Trade weight O-Ds by
Internaonal trade commodity type and
weight-value model mode (tonnes)
Geographic, trade, and (2010–2050)
economic profiles
(scenario based)

Shortest path Trade volume O-Ds by

between O-D pairs by commodity type and
mode mode (tonne-km)
Network model (2010–2050)

Internaonal trade
CO2 intensies and
CO2 emissions by
technological
pathway by mode mode (CO2 tons)
(2010–2050)

FIGURE 1   Schematic description of model.

60
nationales (CEPII), trade scenarios up to 2060 using a framework
integrating long-term macroprojections for the world economy with
a sectorial trade model reproducing the key trade driving forces and
specialization past trends (10).
The long-run growth model in the OECD Economic Outlook (11)
provides long-term projections for GDP, saving, investment, and cur-
rent accounts for OECD and non-OECD G20 countries augmented
with projections by Fouré et al. (12) for other countries.
The trade model is a version of MIRAGE, a multicountry,
sectorial dynamic micro-founded model developed by CEPII (2,13).
This computable general equilibrium model analyses the global evolu-
tion of bilateral trade and sectorial specialization and covers the world
economy aggregated into 26 regions and 25 sectors (19 of them requir-
ing transport). These 26 regions can represent either large countries
(e.g., United States, China) or economic regions with special trade
agreements (e.g., the Euro area, European Free Trade Association).
Although the model aggregates results into regions, it considers all
international trade within regions as well. The model excludes only
domestic trade.
Combining aggregate projections and individual behaviors
(consumers and firms) underlines the impact of both global trends and
country-specific policies on future trade and specialization patterns,
acknowledging international spillovers. Trade projections are given
in value terms in fixed 2004 U.S. dollars.
The model results indicate that over the coming half-century, world
trade is expected to continue to outpace GDP growth but at a slower
rate than previously. An important conclusion of the forecast is that
the geographical center of trade is likely to continue shifting toward
emerging economies:
• China, India, and other Asian economies will continue to
strengthen their role in manufacturing trade with exports climbing
up the global value-added ladder and
• In parallel with shifts in trade patterns the industrial structure
in emerging economies should gradually become similar to that of
the current OECD, whereas it remains fairly unchanged in OECD
economies.
The OECD model was computed for three trade policy scenarios
(10):
• Baseline trade scenario (BS), which considers that the current
world trade agreements are maintained until 2060.
• Bilateral trade agreement scenario, which considers the following:
– A free-trade agreement (FTA) is established in 2012 between
the North American Free Trade Association, the European
Union (EU), Switzerland, European Free Trade Association,
Australia, New Zealand, Japan, and South Korea. Trade barriers
are progressively phased out.
– In the FTA, tariffs on goods are abolished by 2060 and trans-
action costs for goods are reduced by 25%.
– Regulatory barriers in services as measured by ad valorem
tariff equivalents converge to half the average intra-EU level
for non-EU countries within the FTA, and barriers between EU
members are reduced by a further 10%.
– In 2030, bilateral trade agreements are negotiated with key
partners of the FTA including South Africa, the Russian Federation,
Brazil, China, India, Indonesia, other Association of Southeast
Asian countries, and Chile. With these countries, the FTA bilaterally
reduces tariffs by 50% progressively.
Transportation Research Record 2477
• Multilateral trade agreement scenario, which considers the
following:
– From 2013 tariffs on goods are reduced on a multilateral
global basis by 50% by 2060 and transaction costs are reduced
by 25%.
– In the FTA, tariffs on goods are abolished by 2060 and trans-
action costs for goods are reduced by 25%.
– Regulatory barriers in services converge to half the average
intra-EU level for non-EU countries within the FTA, and barriers
between EU members are reduced by a further 10%.
– From 2013 agricultural support is reduced 50% by 2060 in
the EU, the United States, Japan, Korea, Canada, and in European
Free Trade Association countries.
Centroids
The regional aggregation of the model, although very convenient
for economic modeling of trade activity, introduces significant
uncertainties from a transport perspective since it does not allow
proper discretization of the travel path used for different types of
product.
Therefore, the regional O-D trade flows were discretized into a
larger number of production and consumption centroids. To iden-
tify centroids, an adapted p-median procedure was used over all the
cities around the world classified by the United Nations in 2010 rel-
ative to their population (2,539 cities). The objective function for
this aggregation is based on the minimization of a distance func-
tion that includes two components: GDP density and geographical
distance.
Instead of imposing a predefined number of p centroids, no more
than one centroid or region closer than 500 km within the same
country was considered. With these premises the algorithm obtained
a solution with 294 centroids, and a rather stable and spatially balanced
result was obtained for all the continents (America, 69; Europe, 74;
Africa, 34; Asia, 110; and Oceania, 7); each centroid is designated by
the name of the respective city.
The trade flow between economic regions provided by the OECD
study is then converted into trade flows between centroids, con-
sidering the regional GDP around each centroid as weights (2).
The regional GDP for the base year (2010) was obtained from the
Brookings Institution (14) and Pricewaterhouse Coopers (15). The
growth projections of centroids are based on the growth rates at
the country level obtained from the OECD 2013 economic projec-
tions (16). The share of trade flow between O-D pairs within the same
country will be constant over time since the growth rates of centroids
are at the country level.
The resulting equation for the estimation of trade flow between
the centroid O-D pairs for each type of commodity is
GDPit GDPjt
trade flow ijCt = trade flow[Region i][Region j]Ct i i
∑ GDPkt ∑ GDPit
k ∈Region i l ∈Region j
(1)
where
i, j = origin and destination centroids, respectively;
C = commodity;
t = year of analysis;
k = all centroids that belong to region containing i; and
l = all centroids that belong to region containing j.
Martínez, Kauppila, and Castaing 61

Network Model TABLE 1   Speed in Kilometers per Hour Used in Network Model,
by World Region
One of the key components and innovations of this model is the
development of a new geographic information system–based global Speed (km/h)
freight transport model based on open geographic information sys-
Australia,
tem data for different transport modes. Its main contribution is the Europe Canada, and Rest of
consolidation and integration of all different modal networks into a Infrastructure Type and Japan United States the World
single freight network.
For the road network, two main sources were integrated: Global Roada
Major highways 80 80 70
Roads Open Access Data Set (gROADS) (http://sedac.ciesin.columbia Secondary highways 65 65 55
.edu/data/set/groads-global-roads-open-access-v1) and OpenStreetMap Beltways and national roads 40 40 30
(www.openstreetmap.org). Only the first and secondary road networks Tracks or unpaved roads 10 10  5
were considered (i.e., motorways, main roads, and truck roads). For Railb
the rail network, the model integrates data from the Princeton project Electrified 40 35 25
Digital Chart of the World (http://www.princeton.edu/∼geolib/gis Nonelectrified 30 30 15
/dcw.html) updated with the OpenStreetMap data on rail lines and rail Seac
Bulk vessels NA 22.2 NA
stations as intermodal points of connection between road and rail.
Container vessels NA 41 NA
The actual maritime routes are taken from the Global Shipping Lane
Aird 600
Network data of the Center for Transportation Analysis at Oak Ridge
Connectorse
National Laboratory’s Transportation Network Group (http://www
Road connector NA 30 NA
-cta.ornl.gov/transnet/Intermodal_Network.html), which generates Sea connectors to ports NA 15 NA
a routable network with actual travel times for different sea segments.
This network was connected to ports on the basis of data from the latest Note: NA = not available.
a
World Port Index Database of the National Geospatial-Intelligence Authors’ estimates based on speed limits and road infrastructure quality.
b
International Railway Statistics, Union Internationale des Chemins de Fer, 2010.
Agency (http://msi.nga.mil/NGAPortal/MSI.portal). The commercial c
Review of Maritime Transport, U.N. Conference on Trade and Development, 2010.
d
air links between international airports were integrated by using data Authors’ estimates based on average travel times for commercial flights in
Europe to other continents (www.travelmath.com), accessed April 2014.
from the OpenFlights.org database on airports, commercial airlinks, e
Authors’ estimate for setting of road and sea connectors.
and airline companies (www.OpenFlights.org). Finally, all these net-
works were interconnected by transport links between the centroids
using road network and rail stations together with data on intermodal
dwell times. share was estimated in value by product group. Data on travel times
In order to estimate travel times for the different types of infra- and distances for each mode were taken from the global network
structure as well as dwell times between transport modes, average model at the centroid level. Two geographical and economic con-
speeds based on available information by region were used (Tables 1 text binary variables were added: one describing whether the pair
and 2). The network model is then used to compute the shortest paths of countries has a trade agreement and the other for the existence of
between centroids for each transport mode (if the transport is available a land border.
as an option), generating two main inputs: The data set contained 17,427 observations, with an average
weighted mode share in value of 26% for road, 22% for air, 50% for
• The average travel time and distance by mode to link each coun- sea, and just 2% for rail. The calibrated model has ρ2 = .22, show-
try based on a weighted average for all pairs of centroids between the ing a satisfactory explanatory power of the mode choice, and all the
country O-D pairs and variables are statistically significant except for one panel term (coal),
• The shortest paths between the centroids for each transport mode. which presented a similar value to the base commodity considered
(chemicals) (Table 3).
These inputs are used either in the calibration of the other model
components or as an element of the model itself.
TABLE 2   Dwell Time in Hours Used in Network Model,
by World Region
Mode Share Model for International Freight Dwelling Times (h)
The mode share model (in value) for international freight flows Europe Australia, Canada, Rest of
assigns the transport mode used for trade between any O-D pair of Intermodal Connection and Japan and United States the World
centroids. The mode attributed to each trade connection represents the
longest transport section. All freight will require intermodal transport Air to air 4 4 72
in both the origin and the destination. This domestic component of Air to road—road to air 4 4 72
international freight is usually not accounted for in the literature, but Air to rail—rail to air 4 4 72
the current model integrates this component. Air to sea—sea to air 22 22 72
A standard multinomial logit formulation was used including Rail to sea—sea to rail 49 49 96
a commodity-type panel term. It was calibrated with exports data Rail to road—road to rail 31.5 31.5 31.5
from Eurostat and ECLAC, which contain information on the value, Road to sea—sea to road 49 49 96
weight, and mode of transport of export activities from the EU and
Latin America to the rest of the world. For each O-D pair, the modal Source: Fluidity Indicator Project (Canada’s Gateways), Transport Canada 2012.
62 Transportation Research Record 2477

TABLE 3   Trade Value of International Freight Mode Share time given by travel speed: sea routes, requiring a big diversion from
Calibration Results the direct link, such as Europe to Asia with links not using the Suez
Canal, may not be very attractive.
Robust Robust Robust
Variable Coeff. SE t-Test p-Value
An opposite relation is observed for the air mode, for which long
distances are favorable. However, if no direct flights are found
Mode Specific Constant (increasing the travel time) the utility is reduced significantly. Road
Air −0.75 0.02 −41.50 .00 and rail present similar behavior to air. However,
Rail −1.30 0.03 −40.22 .00
• Rail distance presents a huge penalty; this factor suggests that
Road 0.15 0.02 12.42 .00
the alternative may be more attractive for shorter connections; this
Sea 0.00 na na na
penalty may stem from border control difficulties as well as inter­
Commodity Panel Term in Sea Utility Function
operability of systems, which introduce large delays and loss of lia-
Chemicals 0.00 na na na bility of the cargo with the large number of rail operators involved
Coal −0.02 0.08 −0.25 .82 to cross several countries;
Crude oil 2.45 1.15 2.16 .03 • Trade agreements between countries appear to favor road trans-
Electronics −2.37 0.02 −95.67 .00 port over rail transport; this aspect again indicates border crossing
Fishing −0.80 0.09 −8.00 .00 issues with freight rail services; and
Food 0.50 0.02 21.17 .00 • A land border between countries favors exports through road
Forestry −1.03 0.04 −23.89 .00 and rail.
Iron and steel 0.17 0.02 7.14 .00
Livestock −1.01 0.04 −22.77 .00
Weight–Value Model for International Trade
Metal products −0.20 0.02 −10.43 .00
Nonferrous metals −0.15 0.03 −5.74 .00 Several Poisson regression models were formulated to estimate the
Nonmetallic minerals 0.18 0.03 5.51 .00 rate of conversion of value units (dollars) into weight units of cargo
Other manufacturing −0.37 0.02 −20.41 .00 (tonnes) by mode, using the natural logarithm of the trade value in
Other mining 0.73 0.03 23.63 .00 millions of dollars as an offset variable and panel terms by commodity.
Paper, pulp, and print −0.06 0.03 −2.83 .00 The selection of this regression method was based on observation of
Petroleum and coke 0.07 0.03 2.51 .01 the statistical distribution of the sample, which showed more suit-
Rice and crops 0.30 0.02 13.52 .00 able behavior in a discrete statistical distribution than with continu-
Textiles −0.09 0.02 −4.72 .00 ous distributions (i.e., normal, lognormal), especially for low trade
Transport equipment −0.22 0.02 −9.44 .00 connections.
Transport-Related Attributes
The model was calibrated by using Eurostat and ECLAC exports
data. Travel time and distance information were added as discussed
Distance air (10,000 km) 0.03 0.00 19.58 .00
earlier, and geographical and cultural variables, the binary variables
Distance rail (10,000 km) 0.06 0.00 11.70 .00
for trade agreements and land borders used earlier, and a binary
Distance road (10,000 km) 0.02 0.00 5.52 .00 variable identifying whether two countries present the same official
Distance sea (10,000 km) −0.07 0.00 −21.66 .00 language. Moreover, economic profile variables were included to
Time air (1,000 h) −0.02 0.00 −1.97 .05 describe the natural trade relationship between countries with different
Time rail (1,000 h) −1.77 0.00 −12.19 .00 types of production sophistication and scale of trade intensity:
Time road (1,000 h) −0.40 0.00 −35.42 .00
Time sea (1,000 h) 0.37 0.00 40.96 .00 • GDP percentile of the origin country ( p% GDPi),
Context Variables • GDP percentile of the destination country ( p% GDPj),
TA rail −0.07 0.04 −1.65 .10 • GDP per capita percentile of the origin country (p% GDP capitai),
TA road 0.38 0.01 26.21 .00 • GDP per capita percentile of the destination country list ( p%
LB rail 0.98 0.07 14.14 .00 GDP capitaj), and
• Natural logarithm of the GDP per capita ratio between O-D
LB road 1.10 0.02 45.84 .00
countries,
Note: Coeff. = coefficient; SE = standard error; TA = trade agreement effect;
LB = land border; na = not available.
 ln GDP capita i 
 GDP capita 
j

The results show greater relation with the sea alternative, mainly for
low raw material values and the nonperishable nature of the products
but also because of security issues (e.g., crude oil). Transport-related
variables present interesting behavior that clearly distinguishes sea
transportation from the other available options. Although an increase
in distance may reduce the utility of transporting value by sea, longer
journeys make this mode more attractive. The utility of a sea trade
connection depends on the balance between distance and travel
All the economic variables were constructed by using a relative
order of countries instead of their absolute values to avoid their dis-
proportional effect in the future relation of the value–weight ratio.
This effect is not expected to happen since the products will not
become lighter for the majority of the commodities, especially from
raw materials, nor will a disruption in the market change the valuation
of some commodities over others. This will ensure the stability of the
effect over time and improve the model’s forecast ability.
Martínez, Kauppila, and Castaing 63

The result was a data set for each transport mode with individual
model calibrations. All models include an overdispersion Poisson
term since they all showed significant overdispersion.
Table 4 presents the calibration results for the four models and
shows significant differences by mode. The sea model performs well
in reproducing market patterns. Time is the only relevant trans-
port factor; it shows that more expensive goods are transported
farther away (less weight per value). Of the economic variables only
the p% GDPi and the ln GDPcapitai /GDPcapitaj were found significant;
this finding presented a positive impact on the weight that is being
transported by value unit for both. This result shows that larger
economies export greater quantities to less developed countries
that are more weight intensive than those for more developed
countries. The cultural relationship between countries (set as a
binary value that takes a value of 1 if the O-D countries have a
common official language or had a colonial relationship in the
past) was also found significant, leading to exports of greater value
products.

TABLE 4   Weight Value Model Calibration Results for Each Transport Mode

Sea Road Rail Air

Parameter Coeff. Sig. Coeff. Sig. Coeff. Sig. Coeff. Sig.

Shortest Path Characteristics

Timeij (100 h) −0.11 .00 −1.09 .00 0.38 .00 0.50 .00
Distanceij (1,000 km) 0.00 .00 0.269 .00 −0.24 .00 −0.03 .04
Economic Profile of Countries
p% GDPi 0.24 .00 0.77 .00 −0.25 .03 1.56 .00
p% GDPj 0.00 .00 −0.19 .00 0.59 .00 −0.58 .00
p% GDP capitai 0.00 .00 −1.77 .00 −1.55 .00 −2.52 .00
p% GDP capitaj 0.00 .00 −0.81 .00 0.79 .08 −1.23 .00
GDP capita i
ln 0.16 .00 −0.30 .02 0.75 .01 −0.43 .00
GDP capita j
Economic, Geographical, and Cultural Relations Between Countries
LBij 0.00 .00 −0.39 .00 0.00 .00 −1.34 .00
Same languageij −0.11 .00 0.48 .00 0.00 .00 −0.40 .00
TAij 0.00 .00 0.14 .00 0.74 .00 0.66 .00
Commodity Panel Term
Chemicals −7.13 .00 −3.41 .00 −5.78 .00 −2.99 .00
Coal −4.27 .00 −0.91 .00 −3.56 .00 2.12 .45
Crude oil −6.17 .00 −3.66 .00 −33.72 .00 0.19 .98
Electronics −9.45 .00 −2.30 .00 −9.26 .00 −3.47 .00
Fishing −7.28 .00 −3.17 .04 0.00 .00 −2.92 .00
Food products −6.30 .00 −3.29 .00 −5.71 .00 −1.82 .00
Forestry −4.66 .00 −1.94 .00 −3.42 .11 −1.53 .01
Gas −5.49 .00 −2.68 .00 0.00 .00 −23.12 .00
Iron and steel −6.63 .00 −3.00 .00 −5.49 .00 −1.03 .01
Livestock −7.50 .00 −3.04 .00 −4.06 .07 −3.26 .00
Metal products −8.18 .00 −3.73 .00 −6.67 .00 −1.87 .00
Nonferrous metals −8.37 .00 −4.65 .00 −7.38 .00 −5.94 .00
Nonmetallic minerals −5.85 .00 −1.85 .00 −4.62 .00 −1.82 .00
Other manufacturing −8.51 .00 −4.43 .00 −6.81 .00 −2.77 .00
Other mining −6.50 .00 −2.66 .00 −5.17 .00 −0.70 .03
Paper, pulp, and print −7.48 .00 −3.95 .00 −7.82 .00 −2.73 .00
Petroleum and coke −6.33 .00 −2.44 .00 −5.27 .00 −0.39 .31
Rice and crops −5.79 .00 −2.50 .00 −5.89 .00 −1.61 .00
Textiles −7.36 .00 −3.05 .00 −5.58 .00 −3.18 .00
Transport equipment −8.46 .00 −4.54 .00 −7.71 .00 −2.09 .00

Note: Sig. = significance; GDP = gross domestic product; correl = correlation.

Sea Road Rail Air
(Scale) 19,379.55 8,819.49 3,266.30 722.91
Pseudo-ρ2 .65 .60 .68 .38
Correl (y, y– )2 .90 .93 .35 .21
64 Transportation Research Record 2477

The road model presents good fit indicators. Distance should
increase the weight intensity of trade flows, whereas time produces
the inverse effect. This situation highlights the importance of speed
in road freight to determine the typology and value of goods that
are being exported by this mode. Larger economies tend to export
larger quantities, whereas exports to larger economies are simulta-
neously less weight intensive. This finding may be explained by a
larger internal market that requires importing less differentiated and
valuable products. As expected, more developed countries (higher
GDP per capita) trade more value-intensive goods. Land borders
also lead to more valuable exports, whereas the presence of the same
language and trade agreements may potentiate the quantities being
exported.
The rail model presents acceptable quality fit indicators; the ρ2
of the model is high, but the estimates obtained from the model are
less able to reproduce the calibration data. This finding suggests
that other factors not included in the model may affect the value–
weight relationship. Transport variables suggest an inverse logic to
that obtained for the road sector. This result may reflect problems
related to border crossing of the rail mode (i.e., interoperability).
Trade agreements may significantly facilitate the flow of products
between countries using this transport mode. The economic-related
variables present a different configuration. Although larger economies
tend to export more valuable goods by rail, the opposite occurs from an
importing perspective, reinforced by the variables for GDP per capita.
This finding may indicate problems in infrastructure availability and
quality in less developed countries, reducing the rail efficiency for
freight transport.
The air model presents the worst quality-of-fit indicators, which
underlie the great variability of the sector. Nevertheless, the model
shows significant effects of the variables considered. The transport
variable coefficients present behavior similar to that of the rail
model, although the reasons behind this behavior may differ. The
positive impact of time may indicate that the absence of direct flight
connections with large dwell times at airports may reduce signifi-
cantly the value of the transported good, whereas distance potenti-
ates the transport of more valuable goods. The economic profile
variables have an impact similar to that in the road model, in which
exports from larger economies may lead to the transport of more
quantity but the most developed countries export more valuable
goods. The panel commodity terms show that most of air transport
is less weight intensive. For some commodities air weight trans-
port may be much reduced, especially because of safety reasons but
because some products do not need to reach markets quickly to be
consumed (i.e., nonperishable products).
Generation of Model Outputs
Once all the components are set, the model is computed sequentially
as presented in Figure 1. The results include the value, weight, and
distance traveled (with path specification) between 2010 and 2050
for each centroid pair, mode, type of commodity, and year that stem
from international trade.
These results (in tonne-kilometers) are then combined with infor-
mation on related CO2 intensities and technology pathways by mode
obtained from the International Energy Agency’s Momo model (17)
and the International Maritime Organization (18). In the case of
road and rail, these coefficients and pathways are geographically
dependent, whereas the maritime and air CO2 efficiency is considered
to be uniform worldwide.
Model Results
Initial Benchmark of Results for 2010
This section provides a comparison of the results with statistical data
on global international trade–related freight transport (Table 5). This
model yields results for total maritime and air tonnes and tonne-
kilometers that are very close to data provided in other sources;
these results show that the model is able to reproduce current market
behavior adequately. For maritime CO2 emissions, the model esti-
mate falls between the International Energy Agency figures and
slightly below the calculations done by the International Maritime
Organization.

TABLE 5   Model Validation for Baseline Trade Scenario in 2010

Model Available
Variable Estimates Statistics Source

Maritime international trade volume 8,372 8,408 UNCTAD review of

(million tonnes) Maritime Transporta
Air international trade volume 34 31.8 ICAOb
(million tonnes)
Maritime international trade-related freight 60,053 65,599 UNCTAD Review of
(billion tonne-km) Maritime Transport
Air international trade-related freight 191 158 ICAO
(billion tonne-km)
Maritime international trade-related CO2 779 644 IEAc
emissions (million tonnes) 870 IMOd

Note: ICAO = International Civil Aviation Organization; IEA = International Energy Agency;
IMO = International Maritime Association; GHG = greenhouse gas.
a
Review of Maritime Transport, U.N. Conference on Trade and Development (UNCTAD), 2012.
b
Annual Report of Council, International Civil Aviation Organization, 2012.
c
CO2 Emissions from Fuel Combustion Statistics, International Energy Agency (IEA).
d
GHG Study 2009, International Maritime Organization (IMO).
Martínez, Kauppila, and Castaing
Analysis of International Freight Flow Patterns
by Trade Liberalization Scenario
Some preliminary results are presented for international freight
volumes and CO2 emissions by 2050. Figure 2 shows the evolution
of transport volumes related to international trade by different trans-
port modes from 2010 to 2050. International freight is estimated to
grow by 334% by 2050 in the BS scenario. Sea remains the most
relevant transport mode (measured in tonne-kilometers), accounting
for about 85% of the total volume in 2010 and around 83% in 2050
for all trade scenarios. Road freight is estimated to increase its share
of the total during the period. These calculations include the domestic
link of international freight carried by road or rail. Excluding that
link, sea accounts for 95% of total tonne-kilometers.
Trade liberalization may have an increasing effect on international
freight volumes. In the bilateral scenario, freight volumes are estimated
to grow by 350%, whereas in the case of the multilateral scenario,
the estimated growth is 380% by 2050. This increase stems from the
increasing intensity of trade as well as growth in average distance.
A more in-depth analysis by geographical region and corridor is
also carried out. The world is divided into 12 transport regions or
corridors: North America, North Atlantic, Europe, Mediterranean
and Caspian Seas, Asia, North Pacific, South Pacific, South America,
South Atlantic, Africa, Indian Ocean, and Oceania.
Figure 3a presents the corridors’ spatial location along with the vol-
umes by corridor in 2010 and 2050 in the BS scenario: the growth in
freight volume is far from uniform around the world; it is significantly
stronger in maritime routes and inland connections in Asia.
The North Pacific (6) corridor is expected to surpass the North
Atlantic (2) as the main world freight corridor. (The numbers in
parentheses refer to the regions enumerated in Figure 3a.) This find-
ing partly reflects the shift of the economic center of gravity toward
Asia. This shift produces a large increase in volume in the Indian
Ocean and the Suez Canal resulting from the greater trade in Asia–
Africa and Asia–Europe. A marked rise in inland connections for all
400,000
International Freight Volume (billion tonne-km)
350,000
300,000
250,000
200,000
150,000
100,000
50,000
0
2010 2030 2050
Baseline Trade Scenario
FIGURE 2   Estimated international freight volume by world trade scenario.
65
continents is also observed. Significant growth is envisaged to take
place in intra-Asian volumes (5), which are projected to grow by over
380% by 2050. Intra-African freight volumes are also projected to
grow even more significantly (+480%). These results mirror the trade
increase within Asia and Africa and the increasing traffic from and to
ports from and to the consumption and production centers. Because
of the lack of an efficient rail network, these movements are mostly
carried by trucks.
Overall, the results show that bilateral trade liberalization will
not significantly affect freight volumes in any of the regions or cor-
ridors. In contrast, in the multilateral liberalization scenario, trade
is reoriented toward the non-OECD area; this change reflects com-
paratively larger reductions in tariffs than in OECD countries as well
as stronger underlying growth performance in this area. As a result,
global freight will grow by 380% in the multilateral trade scenario by
2050. For the growth by trade corridor or continent, it can be seen that
more liberalized trade produces significantly more transport volumes in
Africa (10), South America (8), and the South Atlantic (9) (Figure 3b).
CO2 Emissions
International trade–related CO2 transport emissions are estimated to
grow by 286% by 2050 in the BS scenario. The bilateral trade sce-
nario results in only a 2% increase in CO2 emissions compared with
the baseline emissions in 2050. However, multilateral trade liberal-
ization would yield 15% more CO2 emissions than in the baseline
by 2050 (Figure 4).
Relevance of Domestic Transport Linked
to International Freight
As already discussed, the domestic freight related to international trade
is often not accounted for. It is estimated that this component represents
Sea
Road
Rail
Air
2010 2030 2050 2010 2030 2050
Bilateral Trade Agreements Multilateral Trade Agreements
Scenario Scenario
 Scale
(billion tonne-km)

80,000

60,000

40,000

20,000

(a)

1,000
Tonne-km, 2050 (2010 = 100)

800

600

400

200

0
North America North Atlantic Europe Mediterrian Asia North Pacific South Pacific South America South Atlantic Africa Indian Ocean Oceania
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Freight Corridors
(b)

FIGURE 3   Model results in tonne-kilometers by corridor: (a) BS scenario at transport corridors (2010–2050) and (b) tonne-kilometers
by corridor for alternative trade liberalization scenarios, 2050 (2010 5 100).

7,000

6,000
CO2 Emissions (million tonnes)

5,000

4,000
Sea
3,000 Road
Rail
2,000
Air

1,000

0
2010 2030 2050 2010 2030 2050 2010 2030 2050
Baseline Trade Scenario Bilateral Trade Agreements Multilateral Trade Agreements
Scenario Scenario
Scenario

FIGURE 4  CO 2 emissions from international freight, by trade scenario.

 Martínez, Kauppila, and Castaing
around 10% of the total trade–related freight globally. Domestic freight
presents great variability depending on the geographic location of the
main producers and consumers in each country. In China, where most
of the economic activity is concentrated in coastal areas, the domestic
link presents 9% of the total international trade–related freight vol-
umes. In India, however, the share is 14% as a result of the inland
location of production and consumption centers.
Domestic transport linked to international trade represents a large
share of the total surface freight volume (national and international)
in some countries. In China, this share is estimated to grow from 9%
in 2010 to 11%, assuming that the coastal pattern of GDP concentra-
tion in China remains. In the United States, this share is estimated
to be 15% in 2010 (according to the U.S. Bureau of Transportation
Statistics, this share was 15% in 2010) but can grow up to 40% by
2050 depending on future trade patterns. Overall, domestic freight
related to international trade puts significant pressure on national
infrastructure capacity.
Conclusions
A new model is introduced to project transport volumes and related
CO2 emissions resulting from international trade under different trade
liberalization scenarios. The model aims to fill the gap emphasized
in earlier studies on modeling and measuring the environmental
effects of international transport (3).
Broadening international trade links have brought a greater vol-
ume of goods; these goods are moving farther and in increasingly
complex and interdependent ways. Freight is a derived demand, and
a critical factor in the future growth of freight transport is related
to the location of future production facilities and consumers. The
results discussed here show that increasing international trade will
lead to increasing freight volumes in the future. This increase in
volume puts significant pressure for infrastructure development in
critical areas, especially at ports and their hinterland connections.
These results also suggest that changes in the global production and
consumption patterns would lead to an increase in average freight
distance. Especially in developing countries, the current infrastructure
may be insufficient to match the estimated demand; this insufficiency
may hinder development of these countries if they do not expand
their transport networks in the next decades.
The model in its current form has several limitations. It does not
take into account capacity constraints or changes in the structure of
global supply chains, among others. Given the complexity of global-
ization and future supply chain configuration, it is difficult to assess
the scale of the environmental effects of future global trade. However,
these results suggest that increasing global trade would lead to a par-
allel increase in CO2 emissions, and multilateral trade liberalization
especially can lead to a greater increase in emissions.
The effects of international trade on the environment are the
result of changes in the scale and structure of trade combined with
technological and product effects. Current trends can be potentially
changed, for example, by improving the emission intensity of the
existing fleet through development of alternative transport modes,
improvement of the efficiency of supply chains, and introduction
of new technologies. This change would, however, require targeted
policies to ensure positive benefits from increasing trade while simul-
taneously improving the energy efficiency of the transport system.
Alternative technological pathways are not assessed here, however.
Already in its current form, the potential use of existing outputs
for policy analysis is broad. Apart from the traditional analysis of
transport activity and related CO2 emissions, the model may be used
View publication stats
67
to assess the capacity of current infrastructure (port terminals, airports,
or road and rail infrastructure) to deal with the expected trade flow
changes or to identify the main bottlenecks that may emerge in the
worldwide transport network, among others.
One future extension of this study may also be the introduction of
trade barriers in some corridors related to security (e.g., piracy, politi-
cal stability) and not only the assignment of freight to the shortest
path between O-D pairs but also use of an equilibrium assignment
procedure.
Acknowledgments
A modeling framework is presented as a part of the Transport Out-
look developed by the International Transport Forum at the OECD.
The authors acknowledge the willingness of OECD and the Inter-
national Energy Agency to share their models for the development
of this model.
References
 1. International Trade Statistics. World Trade Organization, Geneva, 2007.
  2. Chateau, J., L. Fontagné, J. Fouré, A. Johansson, and E. Olaberría. Trade
Patterns in the 2060 World Economy. OECD Economics Department
Working Paper 1142. OECD Publishing, Paris, 2014.
  3. Du, X., and K. M. Kockelman. Tracking Transportation and Industrial
Production Across a Nation: Applications of RUBMRIO Model for U.S.
Trade Patterns. In Transportation Research Record: Journal of the Trans­
portation Research Board, No. 2269, Transportation Research Board of
the National Academies, Washington, D.C., 2012, pp. 99–109.
  4. van Veen-Groot, D. B., and P. Nijkamp. Globalisation, Transport and the
Environment: New Perspectives for Ecological Economics. Ecological
Economics, Vol. 31, No. 3, 1999, pp. 331–346.
  5. Hummels, D. Transportation Costs and International Trade in the Sec-
ond Era of Globalization. Journal of Economic Perspectives, Vol. 21,
No. 3, 2007, pp. 131–154.
  6. David, H., and S. Alexandre. Shipping the Good Apples Out? An Empir-
ical Confirmation of the Alchian-Allen Conjecture. Journal of Political
Economy, Vol. 112, No. 6, 2004, pp. 1384–1402.
  7. Milner, C., and D. McGowan. Trade Cost and Trade Composition.
Economic Inquiry, Vol. 51, No. 3, 2013, pp. 1886–1902.
  8. Venables, A. J., and A. Behar. Transport Costs and International Trade.
Department of Economics, University of Oxford, Oxford, United
Kingdom, 2010.
  9. Cristea, A., D. Hummels, L. Puzzello, and M. Avetisyan. Trade and
the Greenhouse Gas Emissions from International Freight Transport.
Journal of Environmental Economics and Management, Vol. 65, No. 1,
2013, pp. 153–173.
10. Johansson, Å., and E. Olaberría. Global Trade and Specialisation Patterns
Over the Next 50 Years. OECD Economic Policy Papers, No. 10, OECD
Publishing, Paris, 2014, 45 pp.
11. OECD Economic Outlook, No. 93. OECD Publishing, Paris, 2013.
12. Fouré, J., A. Bénassy-Quéré, and L. Fontagné. The Great Shift: Macro­
economic Projections for the World Economy at the 2050 Horizon.
Working Paper 2012-03. CEPII Research Center, Paris, 2012.
13. Fontagné, L., and J. Fouré. Opening a Pandora’s Box: Modelling World
Trade Patterns at the 2035 Horizon. Working Paper 22. CEPII Research
Center, Paris, 2013.
14. Istrate, E., A. Berube, and C. A. Nadeau. Global MetroMonitor 2011:
Volatility, Growth and Recovery. Brookings Institution, Washington,
D.C., 2012.
15. Global City GDP Rankings, 2008–2025. PricewaterhouseCoopers UK
Economic Outlook, 2009, p. 20.
16. OECD Economic Outlook No. 95. OECD, Paris, 2014.
17. Momo ETP 2014. Energy Technology Policy Division, International
Energy Agency, Paris, 2014.
18. Second IMO GHG Study 2009. International Maritime Organization,
London, 2009.
The Standing Committee on Freight Transportation Planning and Logistics
peer-reviewed this paper.