Catastrophic mudflows and landslides, triggered by rainfall can occur suddenly and move with high speed, damaging infrastructure and threatening humans, are increasing due to the climate change. The risk of mudflows increases in post-wildfire areas mainly due to water repellent soil from burned soil organic matter gases. This research investigates mudflow initiation mechanisms by integrating micro, macro and machine learning approaches. The study shows modifications to drop dynamics on post-wildfire slopes, which is a precursor of erosion processes on granular post-wildfire slopes. High-speed video recordings of impact of a single drop on both hydrophobic and regular grains revealed the significant effect of the sand grain size and hydrophobicity on the drop shape transformation, spreading, and splash. Specifically, the drop spread factor, fitted to a power law of We/Oh ratio, is larger on the hydrophobic than hydrophilic sand, and the maximum drop spread increases approximately 40% in fine compared to coarse sand. Moreover, hydrophobizing the sand grain drops the splash initiation threshold approximately 75% in fine sand. The effect of grain size on drop mobility is attributed to Cassie-Baxter and Wenzel wetting regimes, where we observed higher drop velocities and partial bounces in case of Cassie-Baxter model on fine sands, while the velocities dropped about 6.5 times in case of Wenzel model on coarse sand. A macro-investigation on the effects of spatial variability of the post-wildfire hydrophobic layers, with different grain size and under varied rain intensities and slope gradients, identifies failure mechanisms in these layouts. The results quantify a seepage-induced infinite failure mechanism with respect to time to failure, sediment discharge and water infiltration and overflow dynamics. Furthermore, surficial erosion patterns are linked to sand grain sizes, rain intensities, and sediment discharge in the surficial hydrophobic layer. The fine sand has been identified as the most vulnerable soil type in both layouts. A logistic regression model is built to identify the probability of infinite failure based on the experimental results. Despite the small amount of data, the model predictions are promising and indicates the advantage of using machine learning technics in wiser selection of critical parameters.