Implement temporal filter to remove non-coastline noise

[robbibt]

Throughout the time series, some individual annual composites are affected by noise in ocean areas, possibly caused by unmasked clouds or sunglint. As these areas of noise appear semi-randomly with no relationship to the preceding or subsequent year, we should be able to apply a temporal filter to remove them from the dataset.

Proposed method:

1. Take annual composites and threshold to identify land/water
2. Dilate land area by X m to account for moving coastlines. Threshold TBA, but based on sensible max threshold of year-to-year change
3. Compute a rolling sum of this dilated layer
4. Create mask to remove any pixels that do not have a neighbor in either the preceding or subsequent annual composite (e.g. n > 1 neighbor).