So, just a few additional points: First, as Michi pointed out, LPFM stations are forced to use the 3-15km terrain model, even though they don't generally reach anywhere close to 15 km and terrain within 1-3 km has much more effect than anything past 6km. Second, LPFM stations notwithstanding, the contours are reasonably accurate for most stations within the model limitations. After all, the laws of physics haven't changed in the 75 years since the FRC/FCC developed those curves. However, all models have their limits and the FCC model is designed to be simple so that it can be calculated by hand using slide rules and adding machines. Although the FCC method allows the contours to bend in and out to account for varying terrain, a huge limitation is that this method doesn't allow for pockets of higher or lower signal levels to exist within or beyond the various contours. For example, if you are near the edge of your FCC 50dBu contour, you might move to the top of a hill and suddenly find yourself within an area that has a 70dBu+ signal level.
Far better models that more closely approximate the real world exist. In broadcasting, one such model is Longley-Rice. This model allows for higher and lower signal levels at specific places rather than just the general contours of the FCC model. It also allows for propagation adjustments for receiving antenna height, the type of terrain and vegetation to be encountered. There are other models as well, most of which do a better job of predicting real world coverage than the FCC model. That said, most of the time, when Longley Rice and the FCC model are compared using the same parameters, the core signal coverage is usually surprisingly close. The Longley-Rice model is computationally intensive. For a high resolution plot with a 50km radius it can take more than an hour on a fast PC. It would probably have taken a team of engineers a month to render such a model by hand before modern computers, so that's why the FCC model was made so simple.
All this aside, another possibility for your LPFM is that they are operating at a higher power level than authorized. Within the last 24 hours, I have seen two LPFM stations operating at power levels in excess of 800% and 1000% of their authorized level. While I wouldn't classify this as a routine event, it is actually more common than it should be, so this can explain unusually large coverage areas.