Let us break up the problem into four parts: GPS, Barometer, the Naza, and terrain.
GPS:
The altitude error of GPS is known to be around 1.5x horizontal error, due to the simple fact of geometry: In order to obtain higher accuracy, the angular spread in altitude of the GPS satellites from point of observation (the sensor) needs to be high. The constraint on this desired spread is due to signal attenuation and signal path distortion when a satellite is close to the horizon.
For civilian GPS (such as the Naza), the best case scenario for altitude accuracy turns out to be around +/- 25 meters, for 95% confidence. This holds true only when the GPS antenna has unobstructed view of the sky at all times, and a number of GPS satellites (at least 4, preferably 6 or more) are accessible throughout the flight duration. Also, this height is relative to the center of rotation of the respective satellites - for a reasonable approximation, take it to be relative to the center of the earth. There is no correspondence to height above the actual ground wherever you might be.
This 50 meter spread is pretty much unusable for low-level flight. Even if this variance were acceptable, for up to 5% of sensor samples the variance would exceed this, ending up with the hexacopter trying to embed itself into the ground at some point.
On the bright side, this will happen just once, since the copter would need to be replaced or rebuilt after that.
Barometer:
A typical low-cost MEMS barometer promises +/- 0.8 to 3.5 meter accuracy, with the Naza claiming +/- 0.8 meters. This sounds great in theory, but in practical use, air pressure variations due to weather activity, and even more so on mountainous terrain due to rapidly moving pressure vortexes, reduces the actual accuracy to perhaps +/- 10 meters. The barometric height reading is based on air pressure, and thus relative to an arbitrary "mean sea level", varying widely with air pressure conditions. Again, no direct correlation to the actual height of the ground wherever you might be.
This is still significantly better than what GPS alone can deliver. For programmed flight, this would suffice if the copter were to be consistently flown 20 to 30 meters above the height of any ground-based obstructions, in level flight. By introducing ground contour variations, all bets are off.
The Naza:
This is where things get better. The combination of barometer and GPS in the Naza controller provides checks and measures to reject outlier samples and jitter from one or the other of those sensors. In addition, the integrated accelerometer provides a further set of inputs for the damping algorithm of the controller. In tandem, and with a fairly conservative damping setting in place, the hexacopter should behave tolerably well, with effective vertical accuracy of a couple of meters in stable weather.
Gusty wind conditions, and especially low-pressure pockets during impending storms, would reduce the accuracy to perhaps +/- 10 meters in the real world. Further, calibration of the sensors would be needed each flying day, perhaps even once every few hours, to keep the altitude readings meaningful.
Terrain:
Unlike flying over flat terrain, flying over ski slopes can make things rather interesting. For one thing, the terrain contours will necessarily be sloping, so the flight path design needs to take that into account, that too with an accuracy of a few meters. Further, the tops of trees and jutting rocks need to be taken into account.
In addition, signal distorting factors such as water bodies (reflection of GPS signal causing false readings) and metal cables of cable cars or ski lifts, will adversely impact the accuracy of the GPS.
Finally, the obstruction of line-of-sight from GPS antenna to individual satellites, due to mountain tops, trees, or rocks, will cause wide variation in GPS accuracy over the duration of a flight.
Fly by eye. Control the hexacopter manually, do not depend on GPS or barometric settings. Also keep in mind that if some form of sensor dump takes place during a flight, and the hexacopter drops out of the sky, one must ensure that it does not endanger any skiers.
Ideally, one would fit the hexacopter with a camera with a medium telephoto lens - not the ultra-wide of the GoPro cameras, for example - and fly it at well over 20 meters above local terrain contour at all times. Also, one would ensure that the flight path, and especially a fall trajectory in case of failure, would be far removed from the route taken by skiers.
Best Answer
The accuracy when following lines of longitude via GPS is easy enough to calculate. It's just a question of how much cross-track error you have relative to how long your baseline is (the distance you fly along the line of longitude).
Specifically, suppose you have a GPS receiver with an RMS error of 5m. This is in any direction, so it affects both the cross-track error and the length of your baseline. If you fly a nominal 100m north along a line of longitude as reported by the GPS, you'll have a cross-track error of up to 5m × √2 = 7m. (The √2 is applied because you have error in your start position, and error in your end position, too. These are assumed to be independent.) There will be a length error of up to 7m, too, but we can ignore that since its effect on the final answer is much less.
So, the overall error in your direction is up to 7m for the 100m you traveled, so the error in your concept of due north is \$\sin^{-1}\frac{7}{100}\$, or approximately 0.07 radians = 4.2° RMS.
Since the cross-track error remains constant regardless of the distance traveled, if you double the length of the baseline, the error in the angle will be halved, and so on. If you fly 500m north, then your error will be less than ±1°.
If you use an RTK-capable GPS receiver, the measurement error will be reduced by about two orders of magnitude — centimeters rather than meters. This would allow you to achieve much higher accuracy for a given baseline distance, or to use a shorter baseline for a given level of accuracy.
If your drone/UAV is large enough, you can mount two receivers on it and use a technique known as "short-baseline interferometry" to get an accurate heading measurement directly without any motion at all. The receivers need to be capable of giving you carrier phase measurements, and their antennas need to be on the order of 24" apart. One of my consulting clients has done pioneering work in this area.