Others may chime in with more specific answers, but having read a lot literature on this topic or others related, my sense is: not much, unless the land use change is huge (that is, a major change in land use, and over most or all of the watershed, and all at the same time). The water quality response to land use change is hard to track for several reasons, even through monitoring the water quality directly.
One big reason: hydrology is often the major source of variation in water quality from year to year. So in order to have good data for assessing WQ responses to land use change, you have to have long enough records for both the pre- and the post- that you can "average out" the hydrology variations. I have seen great studies in medium/small watersheds where the WQ response to pretty significant land use change was only barely detectable after ~5 years.
On the other hand, it's defensible to qualitatively describe WQ impacts from land use change (mapped remotely or not), because the wealth of the science can inform what kind of changes are likely even if it's difficult to determine the extent of those changes quantitatively. For example, conversion from perrennial prairie to row crop agriculture generally increases runoff and thus soil erosion and thus sediment-bound WQ pollutants like phosphorus. Conversion from rural land use to urban changes the hydrology and can also change the nature of pollutants likely to end up in receiving water bodies.
For very intensive land cover transformation, let say toward heavy agriculture, there should be changes in the patterns of nutrients concentration and their fate. Sediment's regime also should change on dependance of geology. Nevertheless the autodepuration factor of streams can disolve the effects. This factor depends mainly on basin's morphometry and water availability, being the latter a result of climate regime. So I think basin's morphometry and the hydrologic regime have to be well established and understood in order to clarufy the effects of land cover perturbation over water quality.
That is what usually done in Total Maximum Daily Loads (TMDLs) programs. In TMDLs a water quality model is calibrated and validated using measured water quality data. The load reductions of TMDLs are calculated using the calibrated model and further translated into control measures that may include landuse changes.
if you talk about prediction, it means that you need to evaluate time series data to identify the pattern and create a model based on the past
during this time series analysis, you can try to match the rate and the type of land conversion with the change in the water quality. Different type of land conversion will have different effect on the water quality. This will give an idea weather there is a significant relationship between land conversion change and water quality degradation. this time series analysis should also consider the variation of season during the selection of multi temporal data. this is important because some extreme event may only be identified at specific time of the year i.e. first rain after dry season.
If the pattern can be identified properly and the model can be formulated, the prediction can be made. It can be performed via spatial prediction analysis i.e. cellular automata. in this case, remote sensing data can provide valuable information i.e. multitemporal land cover/use map and water quality modeling