Economic consequences of inland waterway disruptions in the Upper Mississippi River region in a changing climate

To understand the impact of the interdependency of different systems, Folga et al. (2009) introduced three types of interdependencies. These types include facility interdependencies, geographic interdependencies, and physical interdependencies. Facility interdependencies describe connections among infrastructures, such as locks and dams, levees, highway and rail bridges, electric power plants, and petroleum refineries co-located with or close to waterways. On the other hand, geographic interdependencies refer to the links between infrastructures located in a common corridor or close proximity, where the failure of one infrastructure in one region can cause a joint loss in other regions. Physical interdependencies highlight that the output of one infrastructure is utilized as input for another infrastructure, making the latter system vulnerable to disruptions in the output from the former.

Several studies have attempted to quantify the impact of interdependent infrastructure system failures. For instance, Nachtmann et al. (2015) evaluated the economic benefits of the interdependencies between different industries to the national economy, using a multi-regional social accounting matrix model (MRSAM), which is a multi-regional variable input–output modeling framework. Similarly, Pant et al. (2015) assessed the economic loss generated by the closure of the Port of Catoosa in Oklahoma using a Dynamic Inoperability Input–Output Model, which considered the interdependencies among regions and various industries. Their study showed that a two-week closure of the port facility could result in a direct output loss of $111.8 million and an indirect output loss of $72.9 million among ten states.

In addition, network analysis can measure geographic interdependencies of disruptions from one location to another or even the entire transportation network. For example, Blandford and Souleyrette (2015) measured the impact of lock closures on the freight network using network models that focused on the change in volumes of coal shipped by rail after disruptions in the barge. They found that the closure of the Greenup locks or the Newburgh locks could increase coal shipment by rail by 15–19 million tons. Similarly, Güler et al. (2012) measured the economic impact of disruption in inland waterways using network models, focusing on waterside coal-fired power plants along the Ohio river.

Madson and Lather (2022) proposed a conceptual framework that depicts the relationships between affected communities, transportation systems, and the economy after disruptions. The study indicated that transportation system expansion generates societal impacts on communities locally and spatially. Although several studies have attempted to analyze the interdependencies among industries or sectors, and between different locations and facilities, none of these studies have empirically examined the spillover effects of inland waterway system disruptions on the regional economy, and the effects of the interdependencies among different infrastructure systems have also been overlooked.

To address the research gaps and provide decision-makers with new insights into the economic consequences of inland waterway system failures and the role of resilience, we focused on the seven major locks and dams in the Upper Mississippi River-Illinois River (UMR-IR) region. Figure 1 shows the major locks and dams considered, which were selected based on data availability: MISS 15, MISS 25, MISS 26, MISS 27, ILL 8, OHIO 52, and ARK 1.¹

In terms of factors affecting barge rate and movement, we considered the following elements based on data availability and the literature, including Scheepers et al. (2018) and Yu et al. (2016): weather-related factors (temperature and precipitation), water levels (gage height and discharge), and various lock disruptive events (scheduled and non-scheduled disruptions). The daily weather data were collected from the Daily Temperature and Precipitation Reports of the National Centers for Environmental Information in the National Oceanic and Atmospheric Administration (NOAA). The daily water level data, including gage height and discharge, were collected from the US Geological Survey (USGS) National Water Dashboard. Gage height is defined as the height of the stream above a reference point, while discharge reflects the volume of water flowing past a given point in the stream in each period. Both variables were introduced to test the extent to which the change in waterway conditions may influence the barge’s performance. Eventually, both variables were aggregated into weekly data based on the weekly mean values to match other key investigation variables, such as the barge rate.

The disruption data of each lock and dam were collected from the Navigation System Status website of the Army Corps of Engineers (USACE).² This dataset includes information on the stoppage start and end time, reasons, and types of disruption (scheduled or non-scheduled). The duration of each interruption is calculated based on the start and end time of the stoppage. 97% of disruption events lasted no more than one day, only 1.52% lasted more than seven days, 0.72% lasted more than 30 days, and 0.58% disrupted more than 60 days. There were no disruption events lasting more than 180 days. Hence, we consider an event with more than seven days a severe disruptive event. In addition, the disruption events were classified into eight types: drought, flood, fog, snow and ice, other weather, system failure, system maintenance, and water level, which are converted into different dummy variables. Furthermore, each type of disruption during each week and the duration of scheduled and unscheduled disruptions were calculated. Supplemenatary 1 (in the supplementary document) summarizes the descriptive statistics of all the variables used to analyze barge performance and environmental conditions.

Figure 2 illustrates the three major steps involved in the analytical process of the assessment. In the first step, we estimated the impact of different disruptive conditions, including dam and lock system failures caused by water levels, temperature, extreme weather events, and other disruptive events, on changes in barge rate using the spatial econometric regression analysis. The model was adopted for two reasons. Firstly, spatial autocorrelation tests (as shown in S-2) indicated that the key variables, such as the barge rate, water conditions, and lock performance, exhibit strong spatial dependence. Failure to account for spatial dependence can lead to biased estimates. Hence, spatial econometric models are needed. Secondly, to capture the spatial spillover effects of lock disruptions on the variations of barge rate, the spatial econometric model is an essential tool.

Since our data are in a panel structure covering seven locks and 468 weeks for 2013–2021, we implemented a data balancing procedure to ensure strong data balance. Next, following Chen and Haynes (2015a) and Chen and Haynes (2015b), we adopted spatial fixed and time fixed effects that are necessary for the modeling structure’s robustness. Considering both spatial and time-fixed effects is necessary because it allows for the simultaneous control of spatial heterogeneity and time-specific factors in empirical analysis. Specifically, spatial fixed effects capture unobserved heterogeneity across spatial units (e.g., locks along the upper Mississippi River). Each spatial unit may have unique characteristics or conditions influencing the outcome variable (e.g., barge rates). By including spatial fixed effects, we control for these unobservable factors, ensuring that our estimates are not biased due to spatial variation. Conversely, time-fixed effects control for time-specific factors that affect all spatial units uniformly. These factors could include changes in market conditions, policy interventions, or technological advancements over time. By including time-fixed effects, we ensure that our estimates capture changes in the outcome variable over time while holding constant spatial differences. Finally, by controlling for spatial and temporal heterogeneity, fixed effects models help address endogeneity concerns and provide more credible estimates of causal relationships (Elhorst and Fréret 2009).

Separate analyses based on different key variables, representing the overall disruption, scheduled disruption only, non-scheduled disruption only, and disruption by specific reasons, were conducted. Following LeSage (2014)’s recommendation, we used the spatial Durbin model (SDM) as the generalized global spatial model. The specific modeling structure is outlined as follows:where Yi,t is a vector representing changes of barge rate at lock i and in week t. \(W{Y}_{i,t}\) represents the spatially weighted lag of Yi,t, with W representing an N x N weight matrix derived from different nearest-neighbors (often referred to as KNN). Xi,t represents a matrix of independent variables at lock i and in week t, including variables representing various durations of disruptive events and environmental factors, such as temperature, water discharge volume, and water level. \(W{X}_{i,t}\) refers to a matrix of spatially lagged independent variables.

The results are compared regarding R-squared values, maximum log-likelihood value, and overall estimation outcomes. In the end, we summarized the regression results of five models, which represent the influences of the overall lock disruption, a scheduled disruption, a non-scheduled disruption, and two specific disruption cases (including flooding-related and system maintenance-related).

We also conducted sensitivity analyses of the three key scenarios (overall disruption, scheduled disruption, and non-scheduled disruption), using different spatial weight matrices (shown in S-3). The purpose of conducting sensitivity analyses with different spatial weight matrices is essential for robustness checking, exploring spatial connectivity, assessing spatial scale effects, examining spatial dependence, and enhancing the interpretation of results in spatial econometric studies of various environmental impacts on barge rate changes across different disruption scenarios.

In the second step, we estimated the elasticity of substitution between barge transportation and railroad transportation using panel regression analysis, with the objective to evaluate the impact of resilience, as captured by the elasticity of substitution between barge and railroad. The estimates were used in the CGE model to 1) improve the accuracy of regional economic impact analysis; and 2) analyze the resilience effect in terms of modal substitution after lock disruptions. We conducted the analysis in three steps. First, we used the US Surface Transportation Board (STB) waybill and the GIS data of rail stations to calculate monthly aggregated rail shipping demand and average shipping rate per ton from the seven locks to New Orleans. Secondly, we calculated the corresponding data based on three types of buffer zones using two distance thresholds at 100 miles and 175 miles, respectively, based on the suggestion obtained from Sytsma and Wilson (2021). Thirdly, we estimated the cross-elasticity of demand as the percentage change in barge transportation volume influenced by a marginal change in the price of rail services shipping from rail stations nearby a lock to New Orleans. The specific estimation procedures were included in S-4.

It is important to note that for the transportation of bulk commodities over long distances, the cost of modes such as trucking is prohibitively high. Therefore, trucking is not considered an economically viable option for resilience in this case. The analysis allows us to measure the varying levels of modal substitution under normal and disruptive conditions while controlling for various environmental factors. By measuring the substitution level through historical data reflecting the performance of barge and rail transportation services, the analysis provides a relatively realistic estimation of the effect of transportation resilience in muting the economic losses of disruptive events.

It is important to highlight that the initial two steps in our approach serve as preparatory phases aimed at determining inputs for the CGE model in the subsequent step. As Zhou and Chen (2020) pointed out, while many studies have utilized CGE models for evaluating disaster impacts, those relying on hypothetical scenarios tend to overestimate the impacts compared to those based on real-world scenarios. Therefore, it is advisable that future disaster impact assessments proceed with caution, ensuring the adoption of appropriate data, models, and shock scenarios.

To enhance the validity of CGE modeling outcomes, we opted to use barge rate changes as a response to environmental shifts, estimated exogenously through spatial econometric analysis, as inputs for the CGE model. Note that in a similar integration of spatial econometric models with CGE, Chen and Haynes (2015c) demonstrated that this approach helps control for the issue of spatial dependence under equilibrium. This integration is crucial, as failing to account for spatial dependence between environmental conditions and barge performance can lead to inaccurate estimates of their relationship and, consequently, impair the validity of inputs for the CGE assessments. This sequential approach ensures that our economic impact evaluation is anchored in more realistic estimations, moving beyond arbitrary specification methods.

In the third step, we estimated the regional economic impact of inland waterway failures in step three using a multi-regional computable general equilibrium (CGE) model. CGE models are widely considered state-of-the-art models for the economic consequences of a disaster. The model has several advantages over conventional models, such as benefit–cost analysis and econometric models. It utilizes real-world economic data to simulate the economic impacts caused by changes in technology, policy, or other external factors.

One should note that the CGE formulation incorporates many advanced features of other models. For example, CGE models inherited the strengths of input–output models, such as capturing multi-sectoral interdependencies. In addition, it also overcomes the limitations of IO, such as linearity, lack of behavioral responses, the effect of price, and substitution possibilities (Rose 1995; Zhou & Chen 2020). The CGE model has been widely used for disaster impact assessments, including measuring the negative economic impacts of Hurricane Katrina and the earthquake on various transportation systems (Chen & Rose 2018; Rose 2017; Wei et al. 2020, 2022).

The CGE model used for this study is The Enormous Regional Model (TERM)-USA, which was developed by The Centre of Policy Studies (CoPS) at Victoria University in Australia. TERM-USA follows the standard CGE modeling structure (Horridge 2012). For instance, producers are assumed to minimize production costs subject to the constraints of intermediate and primary factor inputs. The substitution is specified in a Constant Elasticity of Substitution (CES) nesting structure form. A representative household is assumed to maximize utility by purchasing optimal bundles of goods per its preferences, subject to its budget constraint. In addition, the model also incorporates activities for government and trade. The Model is considered bottom-up, meaning that each region is designed as an independent economic system, while their linkages are captured through inter-regional trade flow.

In addition to the above standard CGE features, the substitution between water and rail transportation services is also explicitly reflected in the model. Specifically, as denoted in Eq. 2, the demand of transportation margin m on good c,s going from region r to region d (xtradmar) is determined by the direct demand xtrad delivered by the margin. However, suppose water-rail substitution is considered (illustrated as the binary dummy ISWATERRAILm when it equals one). In that case, the price substitution is added (SIGWATERRAIL), which reflects the price of composite margin m on goods from r to d (\(psuppma{r\_p}_{m,r,d}\)), and the tech change of margin m on good c,s going from r to d (\(atradma{r}_{c,s,m,r,d}\)), minus the average effective water/rail price on good c,s going r to d (pWaterRail). Hence, the substitution level between water and rail can be modeled by its influence on the final demand for water transportation services. Further detailed specifications of the modal substitution function of the model were included in S-5.

The TERM-USA database used in this assessment is custom-built by Glyn Wittwer at the CoPS, which consists of six regions and 39 economic sectors, with a base value measured in 2019 US dollars. The regions contain five UMR-IR states (Missouri, Illinois, Iowa, Wisconsin, and Minnesota), and the rest of the USA.

Simulations of the TERM-USA model were conducted based on a short-run closure rule (also known as Keynesian’s closure rule). This involved assuming aggregate investment as exogenous to allow a slack variable to adjust, while regional consumption was assumed to follow wage income. In this specification, the wage is exogenous, while the labor supply (or employment) is endogenous. Such a specification is suitable for measuring the more immediate economic impact following an inland waterway system disruption, as most disruptive events are short-run in nature. Given that TERM-USA is a comparative static multi-regional CGE model, the default simulation outcome represents the impact of a shock on the economy from one equilibrium to another equilibrium.

The performance of barge transportation is evaluated based on two metrics: total grain movement and barge rate. Figure 3(a) illustrates the trends of total grain movement at seven locks from 2013 to 2021. Overall, an increasing trend in total grain movement was observed during the period of investigation, although a clear seasonal variation was also noticeable. In 2016 and 2020, the total volume of grain movement by barge reached relatively high levels. However, a sharp decline occurred in 2019, which could be attributed to two primary reasons. The decline in the first part of the year could be due to flooding, while the decrease in the second could be attributed to drought, causing a dramatic drop in water levels, particularly along the locking portion of the Mississippi River System. A regional comparison indicates that locks on the Mississippi river have higher barge volume, with Lock MISS 27 and MISS 26 showing very similar patterns as they are in close proximity. Lock ARK 1 has the lowest levels of grain movement volume. Figure 3(b) demonstrates the moving average barge rate at the seven locks during 2013–2021. The barge rate varies, with relatively high rates in 2014 and 2018. Since 2020, the rates have also increased substantially, likely due to the COVID-19 Pandemic and the US-China trade war. Regarding regional comparison, Lock ILL 8 had the highest rate during 2013 – 2019, but it was then surpassed by MISS 15. Similar to the pattern of volume change, Lock ARK 1 presented the lowest barge rate.

Figure 4 displays the variations in temperature, precipitation, and water levels at the seven locks based on the weekly data adopted in this study. Figure 4a shows that Lock ARK 1 has a relatively higher temperature than other locks. There is no significant difference in temperatures among the other six locks. Figure 4b reveals a relatively high volume of precipitation in 2015 and 2019 when compared with other years. In terms of regional differences, Lock AKR 1, MISS 25, and OHIO 52 experienced a relatively higher volume of precipitation than other locks. Figure 4c shows the variations of water levels at the seven locks, represented by discharge volume and gage height. The figure shows that MISS 25, 26, MISS27, and OHIO 52 have higher discharge volumes, while Lock ILL 8 and MISS 15 present relatively lower gage heights than other locks. The water level was found to be the highest in 2019.

Figure 4(d) presents the variation in the duration of lock disruption (measured in the annual aggregated number of disruptive days) for different reasons at the seven locks during 2013–2021. Eight types of disruption are compared: drought, fog, flood, snow and ice, other weather conditions, system failure, system maintenance, and water level. Disruptions caused by floods, system failure, and system maintenance present longer durations than other types of disruption. The disruptions at Lock ARK 1 appear to have been influenced mainly by system maintenance. Lock ILL8 has been largely disrupted by fog, snow and ice, and other severe weather conditions. MISS 15 is primarily influenced by flood, system maintenance, and other severe weather conditions, while Lock MISS 25 and MISS 26 are primarily disrupted by fog, snow, and ice. Lock MISS 27 was primarily disrupted by a lock system failure in 2014. Lock OH 52 has primarily been disrupted by factors such as fog, system maintenance, other weather types, and water level changes.

Table 1 presents the impact of lock disruption on barge rate variation, revealing that the duration of the disruption has a significant but varying influence on the barge rate. Specifically, an increase in the duration of lock disruption leads to a strong positive impact on the barge rate, indicating that disruption of lock operation can have severe negative consequences for barge operation, resulting in an increase in the shipping rate.

Furthermore, the results show that lock disruptions have direct and indirect (spillover) effects on the barge rate. The former pertains to the influence of environmental factors on the barge rate within the local locks, whereas the latter effect relates to the influences on barge rate variations in neighboring locks. Interestingly, the spillover effects are generally larger than the direct effects, implying that a lock disruption can not only affect the barge rate and operation locally, but can also generate considerable effects on the barge rate in neighboring states.³

The study also compares disruptive causes and reveals that the influences vary substantially by different causes and effects. For instance, non-scheduled disruptions have a larger effect on barge rate than scheduled disruption. However, a slight difference is observed in the comparison of the results of flood-related and maintenance-related disruptions. While the former represents non-scheduled disruption, the latter represents scheduled disruption. The direct effect of flood-related disruption on the barge rate is relatively larger than that caused by maintenance-related disruption, while the indirect effect of these two types of disruption is the opposite.

Note that the results presented were derived from the specification of various spatial weighting matrices with the best fitted outcomes according to model explanatory power. Overall, a high R-squared value indicates that the model is effectively capturing the spatial autocorrelation present in the data. This suggests that the spatial dependencies captured by the model are contributing to a better fit of the model to the observed data.

The sensitivity analyses of the spatial econometric modeling using different spatial weight matrices (as shown in S-3) indicate that the estimates’ signs are generally consistent across the various scenarios for the different comparison categories, except for the estimate of the indirect effect of non-scheduled disruption with a specification of the KNN-2 spatial weighting matrix. In addition, the magnitude of the estimates generally presents a monotonically increasing trend as the spatial weighting matrix adopts a higher number of neighbors. To account for the uncertainty that may result from the spatial weighting matrix’s specifications, the mean estimates of various effects for each scenario were calculated (shown in Table 2), which also serve as input data for the CGE simulation analysis.

The table presents a concise overview of key estimates reflecting the influences of various types of lock disruptions on the variations of barge rate. Each coefficient within the table represents a mean estimate derived from coefficients that exhibit statistical significance at the 5% level across models, with a consistent structure across the board and utilizing various spatial weighting matrices. It is important to note that while the magnitude of the direct effect of the overall case falls between that of the scheduled and non-scheduled disruption scenarios, the indirect effect is much more significant in the latter two scenarios. This difference is mainly due to the variation in mean estimate results in the spatial econometric modeling procedure. One possible explanation for this variation is the indirect (spillover) effect of lock disruption on changes in barge rates, which appears to be contingent on how neighboring states are defined, as indicated by the spatial weight matrix. This further confirms the necessity of relying on mean estimated coefficients rather than specific estimates derived from a particular spatial weight matrix for CGE simulation purposes. By utilizing mean estimates, a more generalized relationship between disruptions and barge rate changes is elucidated, transcending the limitation of any specific models.

The results of the estimation of the elasticity of substitution between barge and railroad confirm that the rail prices do affect the volume of barge transportation services, with the effect varying by commodity types, lock disruptive conditions, and other buffers used to compare the barge-rail services. For instance, Table 3 shows the estimated results of eight models, varying by commodity types and normal versus disruptive conditions. The cross-elasticities of barge and rail for corn, wheat, and soybean were statistically significant, with values of 0.281, − 0.205, and 0.509, respectively. A positive value suggests a substitutional relationship, while a negative value suggests a complementary relationship. However, the estimation of the aggregate commodity type was not significant, possibly due to the complexity of the mixed relationship.⁴ In addition, the results show that cross-elasticities under lock disruptions were generally higher than under normal conditions, suggesting an enlargement of the substitution/complementary relationship between barge and rail service during disruptive events.

We tested the robustness of the results by conducting the same analysis using a different rail shipping rate, which was calculated based on the 100-mile buffer zone near the corresponding seven locks. Table 4 shows that the results demonstrate two salient characteristics compared with the previous findings. First, the magnitude of elasticities for soybeans was relatively smaller than the previous results, suggesting that smaller numbers of rail stations included in the comparison are associated with a more negligible observed impact. Second, the cross-elasticities of demand for corn and the aggregate case were negative and statistically significant, suggesting a complementary relationship between barge and rail shipment of corn from the Mississippi regions to New Orleans, with the magnitude of effects becoming substantial when the lock systems are disrupted.

Finally, we calculated the mean estimates based on the two groups of analyses using different buffers and treated them as the exogenously estimated parameters of barge-rail elasticity of substitution for the CGE model. Specifically, the substitution rates were specified as 0.336 in the normal condition and 0.588 under lock disruptive conditions. The former was calculated as the mean estimate of the cross-price elasticity estimates of rail-barge based on the soybean model estimated in both Model 3 and 11, while the latter reflects the mean estimate of the cross-price elasticities of Model 7 and 15. The selection was based on the significant estimates with consistent signs identified in both sets of models. A higher parameter value indicates that agricultural commodities are more likely to be substituted from barge to rail during a lock disruption period.

It is important to acknowledge a caveat regarding the specified parameter: The substitutional relationship was derived from patterns identified within a specific type of agricultural commodity. Nevertheless, it is crucial to recognize that real-world relationships can be significantly more intricate than theoretical frameworks suggest, given the diverse array of commodities and factors under examination. We acknowledge this complexity as a limitation of our analysis.

However, by integrating empirical estimations of substitution elasticities, we aim to improve the model’s fidelity to real-world dynamics, thereby facilitating more accurate simulations and policy assessments. Therefore, the estimates utilized in this study should be regarded as illustrative rather than definitive.

Our analysis assumes that the shutdown of Lock 25 in St Louis during a disruptive period will result in a shipping rate increase due to the shortage of service supply. This will indirectly reduce gross output and employment in affected sectors, ultimately reducing the value-added regional GDP. Furthermore, due to the interdependency of the inland waterway systems, such disruptions may cause a cascading effect on other states in the Upper Mississippi River and Illinois River streams, leading to a barge rate increase in the affected states. Consequently, gross output and employment losses become inevitable in the barge and related sectors. Hence, the regional economic consequences of barge disruption can be estimated by adjusting the change in barge rate in the CGE model.

Table 5 summarizes the three scenarios for evaluating the regional economic impact of inland waterway system disruptions. These scenarios are designed to check and compare the various effects of disruptions with different durations based on the distribution of historical lock disruptive events. Specifically, the direct effect pertains to the change in local barge shipping rates in Missouri, where lock disruption was assumed to have occurred. The indirect effect pertains to the changes in barge shipping rates in neighboring states, including Iowa, Minnesota, Wisconsin, and Illinois. In general, the table indicates that indirect (spillover) effects are much greater than the direct (local) effects.

In addition, we evaluated the effects of transportation resilience, which is defined as the ability to maintain transporting services when facing an external shock. In our case, resilience is achieved by substituting barge shipment with rail for transporting agricultural commodities from locks to the destination, New Orleans. The effect of resilience was calculated following Rose (2017)’s Direct Static Economic Resilience (DSER) metric. An operational metric of DSER is denoted as the degree to which the estimated decrease in direct output diverges from the anticipated maximum potential decrease following an external shock, such as the reduction or cessation of one or more critical inputs (Chen and Rose 2018). The impact with and without the resilience tactic is simulated and compared to calculate the avoided losses in terms of GDP and employment, measured in both a level change and a percentage change.

Although the lower bound scenario is considered more realistic, the Middle bound (30-day) and Upper bound (180-day) scenarios are included to illustrate the spectrum of uncertainty surrounding the outcomes of these scenarios. While these projections extend beyond observed data, they are predicated on the premise that the magnitude of impact escalates proportionally with the duration of the disruption, as delineated by the spatial econometric estimates.

While these extreme scenarios may be improbable, understanding the corresponding shifts in impact scale offers valuable insights for policymaking and decision-making, particularly in formulating strategies to mitigate economic losses to the greatest extent feasible.

Our analysis assumes that the direct impact of the Lock 25 shutdown is limited in Missouri, while the indirect impact is assumed to occur in the four neighboring states in the UMR-IR region. We calculated the percent changes in barge rate for the direct and indirect impacts in the different scenarios using the corresponding base estimated parameters, which reflect the mean estimated effects of various types of disruptions on barge rate from the spatial econometric model analyses. For instance, in the overall case, the percent changes in barge rate were calculated using 0.012 times the number of disruptive days and 0.038 times the number of disruptive days, respectively, in different scenarios. Three cases were introduced, representing the overall disruption case, scheduled disruption, and non-scheduled disruption cases. Each case contains three scenarios, and various disruptions, ranging from a 7-day disruption to a 180-day disruption. In addition, we included a 30-day disruption as the mid-bound to compare the distributional impacts by the duration of the disruption. We also simulated the impacts with and without considering resilience to examine the avoided economic losses given the substitution between barge and rail transportation services. In total, 18 CGE simulations were conducted based on the various data. We applied the CGE simulation shock to the variable atradmar_csm,r,d, representing the technology change of margin m for commodities transported from region r to region d. It is an generic form of atradmarc,s,m,r,d as shown in Eq. 2 since we assumed that the technological change parameter is uniform across various commodities. In this case, we assume that r denotes different states within the UMR-IR regions, while d represents Louisiana, which is part of the rest of the USA.

Table 6 summarizes the regional economic impact results measured in GDP change from the CGE simulation for the three cases: overall, scheduled, and non-scheduled disruptions. Each case contains two sets of results (with and without considering the resilience effect). The results show that inland waterway system failures can negatively impact real GDP, with the impact being more significant as the duration of the disruptive events increases. For example, in the overall case (which includes both scheduled and non-scheduled disruptive events), a 7-day disruption is likely to cause a $0.752 billion or 0.003% decrease in real national GDP. A 30-day disruption is likely to cause a $3.1 billion or 0.013% decrease in the real national GDP, and a 180-day disruption can cause a $15.2 billion loss of GDP. In terms of the regional impacts, real GDP in Iowa and Minnesota was found to experience a relatively substantial decrease (in terms of percent change measure) due to a complete closure of Lock 25 compared to other states. For instance, if Lock 25 were to close for 180 days, our research suggests that it could result in a 0.467% and 0.44% decrease in GDP for Iowa and Minnesota, respectively. Wisconsin, Illinois, and Missouri could also experience a decline in real GDP of 0.334%, 0.199%, and 0.131%, respectively.

Two possible reasons could explain these outcomes. Firstly, since lock disruption has a much higher indirect effect than direct effect, the results suggest that a lock disruption in Missouri could have a relatively greater economic impact on the two states upstream of the Mississippi River due to the physical interdependence of the inland waterway navigation system. When locks located downstream are disrupted, such a negative effect may quickly affect barge rate and operation upstream. Secondly, the relatively high impact on Iowa and Minnesota may also reflect the important role that barge plays in agricultural commodity transportation in both states. In addition, our findings demonstrate that the impacts increase proportionally as the duration of the disruption extends, indicating that a longer period of disruption can have even more severe negative consequences on the regional economies.

Overall, the results suggest that a complete closure of Lock 25 will lead not only to a negative consequence on the local economy where it is located but also to relatively substantial negative consequences on its neighboring states. In the case of UMR-IR as a whole, it is evident that a major closure of lock 25 near the St. Louis region can unavoidably severely affect the economic growth of the entire UMR-IR region, with a particularly strong dampening effect on the agricultural states in the upper stream of Mississippi. The total losses of regional GDP account for 38.4 – 42% of the total national GDP losses.

Table 6 also summarizes regional economic impacts by different types of disruptive events, regions, and the duration of the disruption. For instance, in the case of a 7-day disruption without resilience scenario, the total GDP losses were found to be -$1.045 billion and -1.249 billion in a scheduled disruption and non-scheduled disruption, respectively. The results suggest that, on average, the GDP losses caused by a scheduled disruption (such as maintenance-related disruption) were found to be relatively smaller than those caused by non-scheduled disruptions (e.g., caused by extreme weather conditions). In terms of the regional impacts, the share of impacts is found consistent in the overall case, which suggests that economic structuring remains the same across the scenarios, given the short-run nature of the analysis.⁵

In terms of the impact of resilience, the results confirm a significant dampening effect in reducing economic losses. By assuming a higher level of modal substitution between barge and rail services for agricultural commodity shipment, the GDP losses can be substantially avoided. For instance, were the level of barge-rail substitution rate increased by 75% (in other words, by adjusting the substitution elasticity from 0.336 to 0.588), 5.2%, 8.5%, and 17.6% of the losses in GDP can be avoided in the disruptive scenarios of a 7-day, a 30-day, and a 180-day, respectively.

In terms of the resilience comparison at the regional level, the percentage of avoided losses is found to differ in various scenarios. For instance, in a 7-day disruptive case, the avoided GDP losses were found to be the highest at 9.3% in Missouri, whereas the highest avoided losses were found to be 12.8% in Iowa in the case of a 30-day disruption. The most significant resilience effect was found during a 180-day disruption, because the avoided GDP losses range from 18.3 to 23.3% among the UMR-IR states. The variation of the resilience impacts is due to several reasons, including the volume of shipment, the magnitude of substitution, and the duration of the disruption. States with a relatively higher regional freight flow by barge and rail are likely to experience larger avoided losses if the level of modal substitution is high. In addition, the positive effect from modal substitution is also likely to be more substantial in a longer disruptive duration than in a shorter one.

In addition, the regional economic impacts of inland waterways system disruption are also measured in employment changes. The number of employment changes was calculated using the percent change of real wage times the corresponding regional employment in the base year. As shown in Table 7, the overall distributional impacts from the same scenarios are consistent with the GDP results. However, there are two observable differences. First, the positive effect of resilience was found to be relatively more substantial on employment than GDP, suggesting that substituting modes during disruptive events contributes to stabilizing the factor market. In addition, the results also show that the resilience effect varies by the duration of a disruption. For instance, the study shows that in a 7-day disruption event, a higher number of job losses can be avoided in a scheduled disruption than in a non-scheduled disruption. However, in both the 30 and 180 days of disruptive events, a stronger level of modal substitution contributes to a higher degree of job-saving in a non-scheduled disruption than in a scheduled disruptive event. Such a result suggests that transportation resilience (achieved through modal substitution) may generate various positive outcomes, subject to various conditions and factors, such as duration of disruptions, types of disruption, and regions being affected.