I am trying to smooth 8 days LAI data from MODIS and managed to work around it with several algorithms such as loess, Savitzky–Golay, smooth.spline etc. This is pretty much going well if the level of noise is not really high (please see below plot) for a target feature.

enter image description here

However, for some pixels (i.e. features), the level of noise is so high that I am not sure if the fitted smooth line is realistic or not (please see below plot)?!

enter image description here

I have already filtered low-quality MODIS pixels using QA dataset but still no improvement.

Is there any suggestion for me how to smooth values of such pixels with high levels of noise?