Seasonal Variations in The Moduli of Unbound Pavement Layers

Chapter 4: Evaluation of Volumetric Moisture Predictions From The Integrated Climatic Model

INTRODUCTION

As discussed in Chapter 2, the ICM was developed in the late 1980s to simulate temporal variations in the temperature, moisture, and freeze-thaw conditions internal to the pavement, and their impact on key pavement material properties.^[78] It is based on a one-dimensional model of the pavement, but does consider both vertical and lateral drainage of the base. More recently, the ICM was updated and enhanced to improve the user interface, predictive capabilities, and operational aspects of the program.^[79] The updated model is the EICM. Further details of the theory on which the program is based may be found in the referenced reports. As discussed in Chapter 2, the EICM will be used in the 2002 Guide for Design of New and Rehabilitated Pavement Structures.

This chapter discusses the application of data collected at the 10 LTPP Seasonal Monitoring Program test sections, identified in Table 30, to evaluate the volumetric moisture prediction capabilities of the EICM. This work was performed in collaboration with efforts to enhance the moisture-predictive capabilities of the model, in preparation for its use in the 2002 guide. Several versions of the EICM were considered in this work: Version 2.0, Version 2.1, and Version 2.6. The evaluation of Version 2.0 was limited in scope, but is important because it provided the initial impetus for the modifications reflected in Version 2.1. Similarly, the evaluation of Version 2.1 provided the impetus for the more extensive model revisions reflected in Version 2.6 of the model.

The moisture prediction capabilities of the EICM were evaluated by applying the model to predict subsurface moisture contents for the selected LTPP test sections, and then comparing the results obtained to the data collected at those test sections. An overview of the data used in this evaluation is provided in Chapter 3. The actual data are presented in Appendix B.

EVALUATION OF EICM VERSION 2.0

Two test sections were considered in the evaluation of EICM Version 2.0: section 271018, which is located in Minnesota; and section 091803, located in Connecticut. The input data used in modeling these sections are summarized in Table 60, found in Appendix B. Two simulation periods were considered in the trial runs for these sections: a 1-day simulation period and a 365-day simulation period. This was done to test whether the model was prone to initialization effects–i.e., whether the model output varied with the length of the simulation period.

*HMAC – hot–mix asphalt concrete

**Manitoba province, Canada

The trial moisture predictions obtained for the two sections yielded poor agreement between the predicted moisture contents and the monitored data, as illustrated in Figure 10 and Figure 11. The data points for the two different simulation periods are essentially coincident, showing that varying the time period used in the simulation from one day to one year did not significantly alter the predicted moisture content for these sections. Other findings from initial trials with EICM Version 2.0 are as follows.

Figure 10. Predicted and monitored moisture contents for section 091803 (Connecticut), 6/30/94

Figure 11. Predicted and monitored moisture contents for section 271018 (Minnesota), 8/8/94 metric units.

• Varying the time increment used in the calculations from 1.0 to 0.01 hours had no effect on computed moisture contents.
• The EICM "generate" function, which is used to "fill in" (via interpolation) missing data for climatic data elements did not function properly when used with input data in metric units.
• Varying the time period used in the simulation from 1 day to 1 year may significantly alter the predicted moisture content when EICM-generated pore-pressure profile data are used in the simulation. (The results shown in Figure 10 and Figure 11 used porepressure profiles estimated from section-specific water table depth data.)

As a result of the poor agreement observed in these initial trials, the EICM was modified (by the model developer) to provide the capability to use section-specific moisture profile data to initialize the computations and correct the program "bugs" identified through the evaluation, with the result being EICM Version 2.1. The author’s evaluation of EICM Version 2.1 is discussed in the next section of this chapter.

EVALUATION OF EICM VERSION 2.1

Sections Considered

Six of the test sections identified in Table 30 were used in the evaluation of Version 2.1 of the EICM–sections 091803 (Connecticut), 231026 (Maine), 271018 (Minnesota), 331001(New Hampshire), 501002 (Vermont), and 831801 (Manitoba). The input data used in modeling these sections with EICM Version 2.1 are summarized in Table 60 and Table 61 of Appendix B. Pertinent details unique to the application of Version 2.1 are as follows.

Layer Structure

In applying EICM Version 2.1, a single model layer was used for each of the nominal pavement layers, with the number of elements in each layer selected to yield nodes close to the depths at which the moisture monitoring instrumentation was installed in each test section.

Layer Porosity

In most cases, the maximum observed volumetric moisture content for each pavement layer was used as the basis for estimating the porosity of that layer, as only some of the data required to determine material-specific porosity values were available to the author at the time of this analysis. For section 271018, the porosity values used were derived from in situ density and moisture data, using an assumed soil specific gravity of 2.65 g/cm³ (0.1 pound/inch³).

Soil-Water Characteristic Curve Parameters (Gardner Coefficients)

EICM Version 2.1 and earlier versions use the Gardner model for the soil-water characteristic curve to model moisture movement through the soil. With Version 2.1, entry of the Gardner coefficients was made optional, and the capability to provide an initial moisture content profile to "calibrate" the model was added. Gardner coefficients are not among the data available for the LTPP test sections. Thus, user-supplied Gardner coefficients were not used.

Results

The moisture contents predicted by EICM Version 2.1 are compared with the moisture content data collected at the LTPP test sections in Figure 12 and Figure 13. In comparing the EICM Version 2.1 output to the field data, linear interpolation was used to estimate the EICM output for the specific depths at which the TDR probes are placed. Inspection of moisturedepth profile plots for the EICM output, such as that shown in Figure 14, suggests that this is a reasonable approximation for the small differences in depth under consideration. In addition, data for monitoring dates associated with actual or predicted frozen conditions (as reflected in one or more of the following indicators) have been omitted.

• Data in LTPP database table SMP_FROST_PENETRATION indicating the presence of one or more frozen layers in the pavement structure.
• Temperature data in table SMP_MRCTEMP_AUTO_DAY_STATS indicating that the minimum observed temperature at one or more depths in the pavement was less than 0 °C.
• Predicted moisture content of less than 1 percent occurring in winter months.

These data points were excluded to provide a "fair" comparison in light of the uncertain meaning of monitored moisture contents obtained in frozen or partially frozen soils. The EICM assumes that the unfrozen moisture content of frozen soil is zero. In contrast, the LTPP data show a reduced, but nonzero, moisture contents in frozen soils.

The error bands shown on the lines of equality are the maximum and minimum 95 percent confidence intervals applicable to the TDR moisture contents (see Table 28). (Only one error band is evident for the base materials, as the minimum and maximum applicable error limits are the same.) These error bands are plotted in the figure to provide a means of judging the extent to which the poor agreement between the monitored and predicted moisture contents may be attributable to error in the TDR-based moisture contents.

Figure 12. Comparison of monitored and predicted base-layer moisture for EICM Version 2.1

Figure 13. Comparison of monitored and predicted subgrade moisture for EICM Version 2.1

While good or excellent agreement is observed for some data points, overall agreement between the model and the field data is poor, as evidenced by both the low fraction of variation explained by the trend lines (R² = 0.01 and R² = 0.30) and the many data points that fall outside the estimated 95 percent confidence intervals for the monitored moisture data. Based on discussions with the model developer and others, it is believed that the discrepancies are attributable to several deficiencies, some in the EICM itself, and others in its application. These deficiencies point to the need for improvements in both the EICM and user documentation to support its use. Recommended changes arising from these discussions included the following.

EICM Changes

• Improvements to the soil-water characteristic curve model used in the simulation.
• Expanded capabilities for generating parameters for the soil-water characteristic curve, which reduce the need to input parameters not commonly available for pavement materials.

These changes were eventually implemented by others as a result of this study. This work was conducted as a part of the development of the 2002 Guide for Design of New and Rehabilitated Pavement Structures under NCHRP project 1-37A.^[19]

Figure 14. Sample EICM Version 2.1 moisture-profile plot, section 091803 (Connecticut)

Application Changes

• The user of the EICM is required to choose between two internal boundary conditions, one known as the "flux" condition, and the other referred to as "suction." The flux boundary condition assumes that water may enter the subgrade through a saturated subbase, whereas the suction condition assumes that subgrade moisture is controlled primarily by suction induced by the water table. EICM trials with Versions 2.0 and 2.1 used the suction boundary condition. The model developers recommended use of the flux condition in most cases.
• Subdivision of pavement subbase and subgrade layers to allow better characterization of the initial moisture profile. (Subdivision of the base layer is not recommended.)

These changes were implemented in the author’s evaluation of subsequent versions of the EICM.

APPLICATION OF EICM VERSION 2.6

All sections identified in Table 30 were used in the evaluation of EICM Version 2.6. Four sections not used in the evaluation of Version 2.1 were considered, to provide more complete coverage of the range of climatic conditions found in the United States and Canada. The data used as input to this version of the model are presented in Table 62 to Table 64 of Appendix B. As noted previously, these data differ from those used with Version 2.1 due to: (1) differences between the two versions of the model; (2) implementation of the application changes noted previously; and (3) additional data that became available in the intervening time period. Pertinent details unique to the application of Version 2.6 are as follows.

Layer Structure

In applying EICM Version 2.6, the pavement layers were subdivided, such that each TDR probe depth corresponded to a mid-depth node for a layer in the model. In cases where a single pavement layer corresponds to multiple model layers (e.g., pavement layer 3 for the Arizona section (041024) corresponds to model layers 3–13), a single entry in Tables 62–64 for any material property indicates that the same value was used for all model layers. Multiple entries indicate that different values were used for each model layer. The values used are listed in order from top to bottom. For example, model layer 5 at the Arizona section was 1.2 cm thick modeled with 2 elements, and had an initial volumetric water content of 19.95.

Layer Porosity

In the time frame between the application of EICM Version 2.1 and the release of EICM Version 2.6, the author obtained specific gravity data for some of the materials considered in this work, which enabled the computation of porosity for those materials. Where available, these values were used; otherwise, this field was left blank.

Specific Gravity

Specific gravity was not among the input parameters required for Version 2.1, and is an optional (but highly recommended) input parameter for Version 2.6. Specific gravities were entered where material-specific data were available. Otherwise, this field was left blank.

Saturated Permeability

Saturated permeability data are not available for the LTPP test sections. This field was left blank in all cases.

Initial Volumetric Water Content

Whereas EICM Version 2.1 required entry of an initial moisture profile (i.e., moisture on a node-by-node basis), Version 2.6 requires input of an initial moisture content for each model (as opposed to pavement) layer. The data used as input are identical. The manner in which they are used in the model differs. Details of the difference are discussed elsewhere.^[19]

Soil-Water Characteristic Curve Model and Parameters (Gardner Coefficients)

Among the changes reflected in EICM Version 2.6 is the addition of the capability to use the soil-water characteristic curve model developed by Fredlund and Xing (discussed in reference 19) as an alternative to the Gardner model. The soil parameters required as input for this model are automatically selected, based on the soil classification and other routine soils data entered by the user. The Fredlund and Xing model was used in all applications of Version 2.6.

Results

The EICM V2.6 predicted moisture contents are compared with the LTPP moisture data in Figure 15 and Figure 16. As before, data points associated with actual or predicted frozen conditions were omitted, and the error bands on the lines of equality are the maximum and minimum 95 percent confidence intervals applicable to the TDR moisture contents. (Only one error band is evident for the base material because the minimum and maximum applicable error limits are coincident.)

Figure 15. Comparison of monitored and predicted base moisture for EICM Version 2.6

Figure 16. Comparison of monitored and predicted subgrade moisture for EICM Version 2.6

While not perfect, the agreement between the measured and predicted base moisture contents is reasonable, with the majority of the data points clustered about the line of equality within the 95 percent confidence limits for the monitored moisture data. These results are substantially better than those obtained with Version 2.1 of the model (see Figure 12). The results for the subgrade soils are not as good as those for the base, but are better than those obtained with Version 2.1 (Figure 13).

The data were examined to identify what conditions are associated with the data falling outside the maximum applicable confidence limits. Sections for which more than one data point fell outside the applicable confidence limits are discussed below.

As illustrated in Figure 17, the predicted base moisture content for section 041024 is relatively constant with time, whereas the monitoring data show a marked (approximately 10 percent on a volumetric basis) increase in moisture content between winter and summer months. The magnitude of this increase in monitored moisture content decreases with depth, and the data for the deeper monitoring depths show a decrease in moisture in this time frame, suggesting an upward migration of soil moisture in the hotter months of the year. It is likely that the discrepancy between the monitored and predicted moisture contents is attributable to incompatibility between the theory on which the EICM is based and the true mechanisms of soil moisture movement in the arid climate in which this section is located.

Figure 17. Variation in monitored and predicted base moisture, section 041024.

Figure 18 shows the variation in monitored and predicted subgrade moisture content with time for section 091803 (Connecticut). The monitored data indicate that the base moisture content at this section is relatively constant with time, with the moisture content at the shallower depth being slightly lower than that at the greater depth. The predicted base moisture content for both depths is quite similar to the monitored moisture for the shallower depth through the summer and fall, but increases markedly in the spring. For the subgrade, the model tends to underpredict the actual moisture content for the first half of the simulation period (summer - early fall), and overpredict the moisture content for the later portion of the simulation period (late fall - spring). It is believed that the observed discrepancies can be attributed, at least in part, to use of model-generated (rather than material-specific) material parameters that (apparently) do not accurately reflect the drainage characteristics of the soils at this section.

Figure 18. Monitored and predicted subgrade moisture content, section 091803 (Connecticut)

Figure 19. Monitored and predicted subgrade moisture, section 231026 (Maine)

The upper subgrade data for section 231026 (Maine) are plotted in Figure 19. For this section, the model consistently underpredicts the actual moisture content for all but the deepest (1.969 m) monitoring depth (not shown). The seasonal trends in moisture predicted by the model are reasonably consistent with those observed in the data, though the model tends to overpredict the magnitude of the changes. This discrepancy may be attributable, at least in part, to inconsistencies in the data for this section. The porosity for the subgrade material, computed from available materials data, and used as input to the EICM was 0.30, while the maximum observed volumetric moisture content for the layer was 0.37–greater than the computed porosity. This discrepancy may be due to spacial variations in porosity, disturbance of the soil that occurred when the TDR probes were installed, or errors in the TDR moisture data. The result is that the model will always underpredict (relative to the monitored moisture data) moisture for monitoring dates where the observed moisture content was greater than 0.30.

The predicted subgrade moisture content for section 501002 (Vermont), illustrated in Figure 20, remains at or near the saturation value throughout the year. The monitored moisture contents are both lower and more variable. As with section 091803, this discrepancy is attributed to the use of model-generated values for many of the key material parameters.

The time trends for the monitored and predicted subgrade moisture contents for section 831801 are illustrated in Figure 21 and Figure 22. For this section, the model consistently underpredicts subgrade moisture, though the monitored and predicted moisture contents for depths of 0.65 and 0.96 m are very similar. For the other depths, the model predicts increases in moisture where the data show decreasing moisture content, and vice versa. In contrast, there is good agreement between the monitored and predicted base moisture contents for this section. As with the sections discussed previously, it is probable that some of the observed discrepancy is attributable to the use of model-generated material parameters. However, the author is at a loss to fully explain the very poor agreement for the deeper monitoring depths in the subgrade soil.

Figure 20. Monitored and predicted subgrade moisture, section 501002 (Vermont)

Figure 21. Monitored and predicted subgrade moisture (shallow depths), section 831801 (Manitoba)

Figure 22. Monitored and predicted subgrade moisture (greater depths), section 831801 (Manitoba)

EVALUATION OF EICM VERSION 2.6 USER SENSITIVITY

Data assembled by the author to evaluate EICM Version 2.6 were also applied by members of the NCHRP 1-37A research team in their efforts to assess the outcome of the model changes. Although the basic data used in the 1-37A evaluation of the model were identical to those used by the author, there were subtle differences in the application of those data by virtue of the judgments that must be made in applying the model. Thus, two parallel and independent sets of applications of the EICM for the same nominal input data set were developed, creating an opportunity for a limited evaluation of the user-sensitivity of the model.

Between-User Differences in Model Application

The differences between the author’s application of the EICM Version 2.6 and that of the NCHRP 1-37A research team are summarized in Appendix C, Table 69. The most prevalent differences were related to the entry of specific gravity, percent passing 200, and plasticity indices for base and subgrade materials. Whereas the author entered specific gravity values only when material-specific data were available, the NCHRP 1-37A research team entered estimated values when soil-specific data were not available. It is likely that this difference in the application of the model had some impact on the results obtained.

In several instances, the number of sublayers used differed between the two applications. These differences are, by definition, associated with differences in the number of elements in the sublayers and in equilibrium moisture content. Small differences in the number of sublayers and equilibrium moisture associated with differences in the number of sublayers are not noted.

The NCHRP 1-37A team reported entry of values for percent passing 200 and plasticity index (PI) in all cases, whereas the author entered values only when soil-specific values were both available, and required, based on the following understanding of their use: (1) for granular soils or nonplastic fine-grained soils, only the D60 is used; and (2) for fine-grained soils with PI greater than 0, only the percent passing 200 and the PI are used. Entry or failure to enter the parameters that are not used in the computations should have no impact on the predicted moisture contents.

Impact of Between-User Differences in Model Application

The overall impact of the application differences on the predicted moisture contents is explored in Table 31, Figure 23, and Figure 24. The information presented in Table 31 includes the two mean predicted moisture contents for each pavement layer at each section, the mean, standard deviation, minimum, and maximum values for the difference between the two predicted moisture contents, the number of paired moisture observation for each layer (n), and the Student’s t statistic for the difference, for a paired t-test on the null hypothesis that the difference is equal to zero. The null hypothesis is accepted for only four layers–all of them base layers–at a 5 percent level of significance: Those layers, denoted by bold type in table 31, are the base layers for section 041024, 081053, 091803, and 501002.

The overall results for the two sets of EICM applications are compared to each other in Figure 23, and to the monitored moisture data in Figure 24. As in previous figures in this chapter, the dashed lines in the figures denote the maximum applicable 95 percent confidence intervals for the related field moisture data. Overall, the results presented in Table 31, Figure 23, and Figure 24 show that between-user differences in the application of the EICM can affect the simulation results. The observed variation in the magnitude and significance of the betweenuser differences in the predicted moisture contents is to be expected, as the nature and extent of the between-user differences in the model application varied from one section (and layer) to another. A more detailed examination of the results obtained follows.

Figure 23. Comparison of EICM Version 2.6 moisture predictions for all sections

A graphic comparison of the two sets of predicted moisture contents for section 081053 is provided in Figure 25. This figure and the numerical information presented in Table 31 indicate that the differences in the application of the model for this section do not result in any meaningful difference in the predicted moisture contents. Similarly good agreement is observed for sections 041024 and 331001.

Figure 24. Monitored and predicted moisture contents for all sections

Figure 25. Comparison of moisture predictions, section 081053

Larger differences in the predicted moisture contents are observed for the remaining sections: 091803, 131005, 231026, 271018, 481077, 501002, and 831801. The results for section 481077 are illustrated in Figure 26 and Figure 27. Differences between the two applications for this section included differences in both the starting date (and thus, equilibrium moisture conditions) for the simulation, and data characterizing the base material, so the fact that significant differences are observed is not surprising. In Figure 26, one can see that both sets of predicted base moisture content fall within the 95 percent confidence limits for the monitored moisture data. The NCHRP 1-37 application tends to overpredict moisture, while most of the data points for the author’s application fall closer to the line of equality. In Figure 27, one may observe that both applications tend to overpredict subgrade moisture, though the NCHRP 1-37A application does so to a greater degree than the author’s application.

Figure 26. Monitored and predicted base moisture content, section 481077 (Texas)

The two sets of predicted moisture contents for section 091803 are compared to each other in Figure 28 and to the monitored moisture data in Figure 29. Agreement between the two sets of predictions is relatively poor. Thus, the observed relationships between the monitored and predicted values illustrated in Figure 29 are quite different, though the correlations are similar. The fact that the majority of the data points for the NCHRP 1-37A predictions fall within the 95 percent confidence limits for the monitored data, while those for the author’s predictions do not, suggests (but does not by itself definitively prove) that use of one or more of the parameters estimated and entered by the NCHRP 1-37A research team, but not entered by the author (absent soil-specific data) improves the quality of the EICM results.

Figure 27. Monitored and predicted subgrade moisture content, section 481077 (Texas)

Figure 28. Comparison of predicted volumetric moisture contents for section 091803 (Connecticut)

Figure 29. Monitored and predicted moisture content, section 091803 (Connecticut)

Predictions for section 131005 are compared with the monitored data in Figure 30 and Figure 31. The predicted values for the base and upper subgrade layers are very similar, but the differences for the subgrade are quite substantial. For this layer, the author did not enter the specific gravity, because soil-specific data were not available, while the 1-37A research team used estimated values. Soil-specific values for percent passing 200 and PI were entered by the 1-37A team, but not by the author (who understood them to be unnecessary). Differences in the start dates (and thus, the equilibrium water contents) used in the predictions may have also contributed to the observed differences. The NCHRP 1-37A model yields better overall agreement with the monitored moisture data.

Figure 30. Monitored and predicted base moisture, section 131005 (Georgia)

Figure 31.Monitored and predicted subgrade moisture, section 131005 (Georgia)

DISCUSSION AND CONCLUSIONS REGARDING EICM MOISTURE PREDICTIONS

The evaluation of the moisture-prediction capabilities of EICM Versions 2.0 and 2.1 presented here is in many respects imperfect because several key material parameters required by the model are not among the data collected for the test sections used in the evaluation. Thus, no judgments can be made as to whether the model itself does or does not yield accurate moisture predictions when applied with complete, material- and section-specific data for the pavements being modeled. However, the results of this evaluation do point out the practical limitations of EICM Versions 2.0 and 2.1, as the data used here are more complete than is often the case in a typical pavement design application. It can be concluded (and should come as no surprise) that the model may not yield accurate moisture predictions when assumed values are used for several key input parameters.

Furthermore, the program documentation guidance on the selection of appropriate values for key material parameters in the absence of complete section-specific data is insufficient to support use of the model in routine practice, where incomplete input data may be expected to be the norm, rather than the exception.

In contrast to Version 2.1, it can be concluded, based on the findings here, that EICM Version 2.6 can provide reasonable estimates of the variation in the in situ moisture content of unbound pavement materials. This is sometimes, but not always, true even when model-generated values are used for several key material parameters. The findings for section 041024 (Arizona) suggest that the model may not work well for sections in arid climates; however, more extensive evaluation will be needed to draw definitive conclusions in this regard.

It is apparent that the revisions made in the transition from EICM Version 2.1 to Version 2.6 have greatly enhanced the practical applicability of the model. While it may be very appropriate for use in a research setting, EICM Version 2.1 is difficult to use in practice because it requires input parameters, most notably the Gardner coefficients, that are not commonly obtained in the laboratory characterization of pavement materials. While soilspecific values of these parameters can be used with Version 2.6 if they are available, they are no longer essential to obtain reasonable results. Thus, in the author’s judgment, Version 2.1 is most appropriately thought of as a research tool, while Version 2.6 can be considered a practical engineering tool. There remains room for improvement, however, in the user interface for the EICM, as a great deal of manual data manipulation is required to generate input data sets compatible with the EICM.

Lastly, both t-test results and the detailed results illustrated in Figure 23 through Figure 31 indicate that between-user differences in EICM-predicted moisture contents can be significant in both statistical and practical terms. These differences point to the need for: (1) very specific guidance on the use of data that are recommended, but not required; (2) care in entering correct information relative to material gradation; and (3) careful consideration of the selection of appropriate initial conditions for the moisture prediction. It would be appropriate to conduct further sensitivity analyses to evaluate the relative importance of these factors in causing the observed differences in predicted moisture.