Back to the roots: The reliability of a pullout test for assessing the erosion resistance of grassed dike covers

Grass cover erosion by wave overtopping is potentially a major failure mechanism for earthen dikes. Grass
cover erosion resistance (quantified by the critical velocity, 𝑈𝑐 ) can be assessed using full-scale, destructive
tests with the wave overtopping simulator (WOS) in combination with the erosion model ‘cumulative
overload method’ (COM). Although these tests provide valuable information, they are relatively expensive
and time consuming. Therefore, a small-scale grass pullout test (GPT) may be an attractive alternative.
With this method, multiple (typically 30) pullout tests within a test plot of roughly 10 m x 15 m are done,
after which the maximum pullout force for each test is translated to a representative tensile strength
parameter. This is currently the critical normal stress at the ground level, 𝜎𝑔𝑟𝑎𝑠𝑠,𝑐 (0). From all tests
performed, a probabilistic design value for the weakest spot within the test plot is translated to the 𝑈𝑐 via
a dedicated formula.
Given its easy execution and the relatively straightforward calculation of 𝑈𝑐 , the GPT, developed in 2015,
is considered a quick and easy way to gain insight into the erosion resistance of a grass cover. However,
apart from five initial test sites with grass on clay, the 𝑈𝑐 obtained with the GPT did not correspond well
with the 𝑈𝑐 obtained with the WOS for grass on sand. Moreover, since pinpointing the cause for observed
discrepancies has been difficult and the relation between 𝜎𝑔𝑟𝑎𝑠𝑠,𝑐 (0) and determinant vegetation
properties (e.g. root mass) has not been strong, the consistency of the method may be questioned.
This EngD project comprises a reliability analysis and subsequent improvement of the GPT as an estimator
for the 𝑈𝑐 of grass covers. The study concerns the consistency of the tensile strength, the accuracy of the
𝑈𝑐 , as well as a critical reflection on the methodological steps. If necessary, these methodological steps
may be adapted to improve the reliability of the GPT. The ultimate aim of the EngD project is to generalize
the applicability of the GPT for all existing types of grass covers, comprising grass on clay, grass on sand,
and species-rich covers.
First, the consistency was assessed via the explainability of the tensile strength using a logical (i.e. in line
with literature) set of soil and vegetation parameters. It was found that the pullout force of an intact grass
sod, 𝐹𝑖𝑛𝑡𝑎𝑐𝑡, is reasonably consistent with the natural variation of root mass. This was to a lesser extent
the case for 𝜎𝑔𝑟𝑎𝑠𝑠,𝑐 (0), which is derived through dividing 𝐹𝑖𝑛𝑡𝑎𝑐𝑡 by the surface area of the pulled grass
sod.
Next, the accuracy of the 𝑈𝑐 was assessed via the correspondence between the 𝑈𝑐 obtained with the GPT
and the WOS. This was done by comparing the 𝑈𝑐 𝑠 of the specific sites where both methods were applied
and by comparing the 𝑈𝑐 distributions (i.e. mean and standard deviation) constructed based on all sites
for a specific grass cover type. There appears to be a significant discrepancy between the current GPT and
the WOS, whereby the GPT typically underestimates the 𝑈𝑐 obtained with the WOS for all considered grass
cover types.
Subsequently, via a reflection on the steps in the current GPT method, several design adaptations were
proposed. Three design variants, comprising different combinations of the proposed design adaptations,
were compared by the correspondence between the estimated 𝑈𝑐 and the 𝑈𝑐 obtained with the WOS.
The preferred variant gave the highest accuracy for the site-specific 𝑈𝑐 as well as for the 𝑈𝑐 distributions.
Compared to the current GPT method, the adaptations of the selected variant consist of: the inclusion of
the self-weight of the pulled grass sod in the derivation of 𝐹𝑖𝑛𝑡𝑎𝑐𝑡, the use of (depending on the sample
skewness) either the 1st, 2.5th or 5th percentile of the sample distribution of 𝐹𝑖𝑛𝑡𝑎𝑐𝑡 as a design value for
the weakest spot, and the use of an adapted formula which directly translates 𝐹𝑖𝑛𝑡𝑎𝑐𝑡 to a 𝑈𝑐 .
Finally, the methodology of the selected variant was further simplified by determining the minimum
number of tests required for a sufficiently accurate estimate of the 𝑈𝑐 per test plot. It was found that at
least a sample size of n=20 is required to obtain a comparable consistency of the tensile strength (i.e.
𝐹𝑖𝑛𝑡𝑎𝑐𝑡) and accuracy of the 𝑈𝑐 as for n=30.
Ultimately, the adapted GPT method proposed in this EngD project provides a more accurate 𝑈𝑐 for all
considered grass cover types (i.e. grass on clay, grass on sand, and species-rich covers) compared to the
current GPT method. Furthermore, due to its direct translation of 𝐹𝑖𝑛𝑡𝑎𝑐𝑡 to 𝑈𝑐 , the results can be better
explained using the environmental (root) characteristics of a site. Moreover, due to the simplified
procedure of the adapted GPT method, the duration of the physical field tests is shorter than for the
current GPT method.