Fitting of Soil-Water Characteristic Curves (SWCC) of Bukit Mewah, Malaysia soil using field monitoring dataset

Rainfall-induced landslides are a frequent geohazard for tropical regions with prevalent residual soils and year-round rainy seasons. The water infiltration into unsaturated soil can be analyzed using the soil-water characteristic curve (SWCC) and permeability function which can be used to monitor and predict incoming landslides, showing the necessity of selecting the appropriate model parameter while fitting the SWCC model. This paper presents a set of data from six different sections of the studied slope at varying depths that are used to test the performance of three SWCC models, the van Genuchten-Mualem (vG-M), Fredlund-Xing (F-X) and Gardner (G). The dataset is obtained from field monitoring of the studied slope, over a duration of 6 months. The study discovered that the van Genuchten-Mualem model provided the best estimation based on RMSE and evaluation metric, R2 followed by Fredlund and Xing, and Gardner, however, the difference between them is minor. The R2 obtained varies as the value at the crest with 1.0 m depth has a mean of 0.44, the lowest among the overall data fitted but it also has the best RMSE value with a mean of 0.00473. Whereas the location mid-section at a depth of 1.0 m has the highest R2 with a mean of 0.97, and an average value of RMSE of 0.0145 which is the middle of the group that was fitted. This indicates that R2 measurement for model performance relies highly on the dispersion of the variables collected. The dispersion of the data set is mainly due to the sensors' inability to detect effectively at exceedingly high matric suction and zero matric suction. The investment in improving the equipment's precision will boost reliability and reduce the number of assumptions as the data is collected from the site rather than laboratory testing.