The following briefly describes some of the more common problems encountered in backcalculation.
More detailed discussions are available elsewhere ⁸⁾.

In many cases pavement deflection measurements include "irregularities" which are generally related to differences between measured pavement response and the theoretical models used to predict that response.
These may result from a number of reasons, including pavement distress, variations in layer thickness, non-linear material response, presence of bedrock or other stiff layers, moisture and temperature effects etc.
Anomalies within the pavement structure, such as culverts and utility ducts are not discussed here since they can be observed and are considered atypical.

It should be pointed out that as backcalculation techniques mature some of the problems are being addressed by software modification.

(I) Input Data Effects
These include seed moduli, modulus limits and layer thicknesses as well as program controls such as number of iterations and convergence criteria.

A typical result may show, as an example, subgrade modulus that is significantly higher than expected for the material type, while the base layer modulus is far too low and the surfacing modulus is too high.
This probably occurs most commonly for a significantly stress softening subgrade, where the subgrade stress level for the outer sensors in a FWD test is very much lower than the subgrade stress level directly beneath the load plate.
The apparent subgrade modulus for the outer sensor location is therefore higher than the apparent subgrade modulus directly beneath the load plate.
If the subgrade is modeled as a linear elastic material, then, since most backcalculation routines first calculate subgrade modulus from the outer sensors, the higher modulus value is calculated and assumed to be constant throughout.
At the next iteration, when the base modulus is being calculated, the too high subgrade modulus is compensated for by calculating a modulus that is too low for the base, in order to match the deflections measured in this region.
In other words, alternating layers exhibit a high or low compensating effect.

Ideally, correctly modeling non-linear material response will remove this type of error, and this is becoming more and more common (e.g., ELMOD^®, MODCOMP3, EVERCALC, BOUSDEF can all use non-linear material models).
If an elastic subgrade is used, then the inclusion of a stiff layer, or the use of a layered subgrade, can help alleviate the problem.
This is at least partially the reason why some backcalculation routines include a stiff layer by default at some depth (usually approximately 6 meters or 20 ft.).

It is also worth noting here that the effect of too rapidly decreasing deflections with distance is often due to the dynamic nature of the impulse load.

(III) Subgrade "stiff" layer
For the purposes of a general definition, a "stiff" layer is one below which there is little or no apparent contribution to the measured surface deflections.

"Stiff" layers can be real or "apparent" and are possibly the most common problem encountered during the evaluation of deflection basins.

The stiff layer may in fact consist of rock or other stiff materials.
However, the effect has also been observed where a water table is encountered near the surface.
Possibly the most common phenomenon is due to the subgrade non-linearity effects described above resulting in an apparent stiff layer effect with backcalculated moduli exhibiting the compensating effect.
For the case where an actual rigid layer exists, computer backcalculation programs such as MODULUS, BISDEF, and WESDEF have a rigid layer subroutine built in.

Bedrock information can be obtained from geologic maps, by coring or by penetration resistance measurement.
Depth to the stiff layer can also be estimated from the deflection data as done in ELMOD^® and MODULUS. The best approach is to model the actual situation as closely as possible.

One approach used for the "apparent" stiff layer problem, if a layered-elastic backcalculation program is used, is to divide the subgrade into two or more layers, allowing the backcalculation program to assign modular ratios which achieve the best fit.

(IV) Pavement layer thickness effects
Due to limitations in the backcalculation software, and the limited time available to perform backcalculation activities in a production environment, pavement layer thicknesses are generally assumed to be constant over the pavement section under test.
This is seldom the case. Pavement layer thickness variations result from various construction and maintenance details, even under specially controlled conditions.

On Texas SHRP sections, it has been found that asphalt concrete thicknesses may vary up to 2 in. within 500 ft. Pavement layer thickness variations will produce variations in the deflections from point to point which are indistinguishable from layer moduli variations.
The net result is that this variation manifests itself in the backcalculated moduli for the various layers.
In some cases, these moduli variations are not significant.
However it is desirable to use correct layer thicknesses and various techniques, such as GPR, are improving the ability to obtain thickness data.

It should be noted that surface layer thicknesses of less than 75 mm (3 in.) cannot be reliably characterized with Falling Weight Deflectometer (FWD) data, primarily due to the geometry of the loading and measuring system.
Moduli of thin layers are generally difficult to determine from FWD data.

(V) Relative layer stiffness effects
Backcalculation can describe a pavement layer's stiffness only to the degree to which that layer affects the deflections.
Thin layers contribute only a small portion to the overall deflection and as a result, the accuracy of their backcalculated values is reduced.

To some extent, the layer thickness discussion covers relative layer stiffness effects.
However, the intent of this section is to emphasize that the layer stiffness (i.e., combination of thickness and modulus) needs to be relatively significant (compared with other pavement components) for it to influence the surface deflections.

If this is not the case, then backcalculation approaches will not be successful in identifying the effect of the layer.
As an example, consider a 200 mm (8 in.) thick natural gravel base course.
If this layer is placed on an average subgrade and surfaced with a chip seal, it is relatively stiff and backcalculation will easily evaluate the difference in modulus between the base and subgrade.

On the other hand, if this base material occurs beneath a 400 mm (16 in.) PCC slab, it is not relatively stiff and it is unlikely that the backcalculation process will be able to reliably separate the contribution of this layer from the subgrade effect.

Similar problems occur for many unbound base/subbase combinations.
These materials often differ only in terms of gradation and indicator specifications and their moduli are relatively similar, so that their contributions to the deflection response are difficult to separate.

Similarly, if the surfacing is made up of more than one AC layer, these should be considered as a single layer.
There is generally not enough difference between the response of an asphalt concrete surfacing layer and an asphalt treated base to evaluate these layers separately.