Skip to content

Latest commit

 

History

History
60 lines (48 loc) · 4 KB

README.md

File metadata and controls

60 lines (48 loc) · 4 KB

Probabilistic PISCES

Transforms PISCES HUC12 level species occurrence data to NHD Stream Segment level by assigning probabilities of occurrence based on stream order.

Animation of transition from ranges to probabilistic segments

The problem

While California contains about 4,600 HUC12 subwatersheds, it contains ~160,000 NHD Stream Segments. The HUC12s are the coarsest valid observation unit in PISCES, so this scale adjustment adds additional uncertainty to the species records. In order to attach this additional uncertainty, the code here uses an approach based on Strahler stream order and the species' entire range to probabilistically assign species to segments within their broader range. This approach is more valid for wide ranging species than those with narrow ranges, or which are resident in lakes.

While PISCES data contain range information instead of direct locational observations, it is granular enough to often be used to represent actual presence in streams within a watershed. Based on the scale mismatch between HUC12s and stream reaches, it is risky to make this assumption. This code is an effort to better represent uncertainty at the stream segment level if the data are being used that way. It also represents a first pass estimate, not a rigorous evaluation of species presence at the stream segment level. Use it only very carefully and if you understand both the transformation described below and what it means for the data.

Downscaling approach by stream order

The approach

We start by calculating each taxon’s primary stream order, the stream order value above which we assume the taxon is present and below which we decrease probability they are present. The calculation of primary stream order is based on an assumption that most fish taxa in California will be present in the stream segment in a subwatershed that connects it to the downstream subwatershed, since that segment ensures connectivity of the range. Depending on the taxon, they may or may not be present in other segments, but a conservative estimate that works for most taxa suggests they are at least in this connecting segment. This assumption will not hold for all species. Some are highly endemic, and others like tributaries. But it works as a first pass for downscaling a large dataset and can be refined with more information later.

The segment that connects each subwatershed to a downstream subwatershed will have the largest stream order in the upstream subwatershed. Once we have largest stream order value for each subwatershed, we examine each taxon individually and gather all of the connecting stream order values for subwatersheds within the taxon’s range. For each taxon, we select the lowest value as their primary stream order (see image).

Once we have a primary stream order value for each taxon, we apply presence probabilistically to all segments within their subwatershed-based range. Any stream segment in their range that exceeds the primary stream order is assigned a presence probability of 1, while segments with the primary stream order are assigned a probability of 0.9 to reflect some uncertainty. Stream segments with smaller stream orders receive smaller presence values according to the function 0.9 * 0.5^(S−S) where S is the primary stream order for the taxon and S is the stream order of the segment being evaluated.

S* S=5 S=4 S=3 S=2 S=1
1 1 1 1 1 0.9
2 1 1 1 0.9 0.45
3 1 1 0.9 0.45 0.225
4 1 0.9 0.45 0.225 0.1125
5 0.9 0.45 0.225 0.1125 0.0563

The table shows the probabilities used for each stream order based on the primary stream order for a taxon

Results

See Releases for the latest data and information on using it.