The reading group will meet this Friday (23rd October) at 1pm. We're going to read a paper on analysing data along stream networks by Jay ver Hoef and colleagues. They have been building spatial models for stream networks which cleverly take into account the linear dependence along streams, as well as the direction of flow.
We will be reading a paper from Environmental and Ecological Statistics titled Spatial statistical models that use flow and stream distance.
Hope you can join us.
We thought this was an interesting paper but a hard one to get into if you don't already know about "geostatistical models" - as it essentially extends ideas from standard spatial stats to stream networks, where distance between two sites is measured along the stream network (accounting for flow) rather than using straight lines (Euclidean).
ReplyDeleteWe talked a fair bit about the weighting idea (sqrt of flow), which was a key step, and thought a bit about some of the challenges in terms of extending the method: non-normal response (like counts), flow upstream not down (like fish). And what about situations where some of the spatial correlation is flow-related but some of it is straight line? So this paper is the start of something interesting with lots of opportunities for extension.