<< Click to Display Table of Contents >> Backtracking on a hill 

The backtracking strategy is based on the relationship between pairs of neighbour trackers. See the description of the backtracking strategy.
It requires that the tracker's array is perfectly regular, with the same width/pitch ratio, as well as altitude. The altitude differences will necessarily be the same for all trackers (i.e. if not horizontal, the tracker array will be on a same flat EW inclined plane).
With different and irregular altitudes like on a hill, the Backtracking strategy is geometrically impossible, neither in PVsyst nor in the reality.
When the backtracking angle is calculated for the trackers A and B (at higher altitude), a shade on the tracker C (at lower altitude) is unavoidable. Here half the tracker is shaded.
Inversely when the tracker B is lower than the tracker A, the tracker C has not an optimal tilt for collecting the whole available sun's light. A better tilt would intercept more light, this corresponds to a loss due to the cosine effect.
Workaround / only way on a hill: give up the backtracking strategy
In PVsyst, we don't advise using the backtracking on a hill.
However we will allow it, with a warning to the user that this is not a "true" backtracking.
  There may be some mutual shadings. Therefore the electrical shading calculation (according to strings or Modulelayout) has to be activated. 
  There may be some irradiance loss due to useless low tilt on higher trackers. 
These two effects are taken into account in the simulation. Namely the mutual shades of irregular trackers are correctly evaluated.
With these restrictions, it is not certain that the backtracking gives a better yield than a normal tracking with shades. Both solutions should be compared with a detailed simulation.
Commercial propositions
Some people propose a situation where all trackers take a different position: this leads to extremely complex calculations (the optimization of the Phi angle of each tracker should be performed simultaneously on all trackers for each sun's position) without ensuring a perfect solution. This optimization involves machinelearning techniques for a given installation. And the real gain may be questionable.
As far as we don't have a model for the implementation of such a strategy, we cannot envisage this development in PVsyst in a foreseeable future.
Moreover on the field, you should also wonder how you will physically implement such Backtracking control in your installation.