the online meeting place for all who love our amphibians and reptiles |
|
Adder Dispersal Distances |
Post Reply | Page <1234 8> |
Author | |||
calumma
Senior Member Joined: 27 Jun 2003 Location: United Kingdom Status: Offline Points: 375 |
Post Options
Thanks(0)
|
||
Steve - just got it!
At least we are thinking along the same lines It may also be worth considering a Level 4 that incorporates more refined land classifications that correspond to data available at local records centres. Take a look at the Kent Landscape Information Service website as an example of what I mean. |
|||
calumma
Senior Member Joined: 27 Jun 2003 Location: United Kingdom Status: Offline Points: 375 |
Post Options
Thanks(0)
|
||
Actually the KLIS website doesn't look like it shows all of the different habitat
breakdowns. I believe more detailed information is available on their GIS - and I aim to get a copy of this. The following screenshot illustrates what we should be able to do once we overlay estimated occupancy areas on top of the habitat data. This example is based on individual adder records, but I aim to take a more landscape level approach to the analysis. |
|||
calumma
Senior Member Joined: 27 Jun 2003 Location: United Kingdom Status: Offline Points: 375 |
Post Options
Thanks(0)
|
||
Just for fun, here is a map illustrating occupied gcn ponds for Kent. Larger
circles this time represent 1 km radii and indicate waterbodies that have another gcn record (any record, not just occupied pond) within 1 km. Small circles represent more isolated records and are drawn with a radius of 500 m. Bias in survey effort is a problem for any recording project and the above map has a classic example. Spot the gcn ponds along the Tonbridge pipeline survey! Looking at the large number of gaps in the low weald indicates that we still have a long way to go with this species. That Tonbridge pipeline demonstrates what the whole of the low weald should look like |
|||
calumma
Senior Member Joined: 27 Jun 2003 Location: United Kingdom Status: Offline Points: 375 |
Post Options
Thanks(0)
|
||
I should also point out that there are nearly 2000 gcn records on the KRAG
db. However, several hundred of these are either terrestrial observations or do not specify whether the observation was on land or in water. For the map above, I was interested in plotting records that were confirmed as originating in a waterbody. There would be quite a few more red dots if I was less demanding ! |
|||
calumma
Senior Member Joined: 27 Jun 2003 Location: United Kingdom Status: Offline Points: 375 |
Post Options
Thanks(0)
|
||
Happy to suck eggs on things when it saves helps make the db more efficient Using the squared distance is relatively straightforward to incorporate into my nearest neighbour script so I will updating things after work. One issue that I have had is that I developed the routines for calculating distances for other purposes - listing records in order of distance from a point of interest, whilst also providing a bearing. However, it is fairly trivial to use a different field for the large scale nearest neighbour calculation and for that I can rely on the squared distance. Just in case anybody else is trying to keep up with some of these more geeky discussions on calculating distance between two points, there is a really excellent summary available here. BTW apologies for the poor formating of the post. However, I seem to have problems retaining line break formatting and links when posting to the forum from Safari. Not sure if this is a browser specific issue (I use Safari). Edited by calumma |
|||
calumma
Senior Member Joined: 27 Jun 2003 Location: United Kingdom Status: Offline Points: 375 |
Post Options
Thanks(0)
|
||
Not yet addressed some of Vicar's points.
What I am suggesting is that we can use maximum likely dispersal as a proxy for predicting distribution from confirmed locations. I completely agree that the prediction is woolly. However, I am less concerned about this at the moment, because it is a prediction that will become less woolly as more records are collected. In areas where animals cannot disperse we may expect less records. I am therefore minded to use the count of nearest neighbour(s) as a proxy for estimating the prevalence of dispersal barriers ('dispersal potential'). Of course, this all relies on a good data set. But I would hope that as more records are collected the prediction would become more refined. The estimate is also likely to be less woolly for species with larger potential dispersal distances. The smaller the dispersal distance, the more records that are needed to adequately identify distribution boundaries. Grass snake may have been an easier species to start with, but since adders have such a well defined range in Kent they are perhaps a more interesting target.
We wouldn't. However, a site with only summer sightings can be flagged within the db for early spring survey work. If I chose to use nearest neighbour as a proxy for dispersal potential, it may not be necessary to control for season. Of course the beauty of a relational database is that this information can still be stored and turned on at some point when the dataset is considered sufficiently large.
Yep. My aim is to correlate the predicted presence of a species in a specified habitat type against (1) the area that that habitat covers across the county (potential range) and (2) habitat within the polygons defined by available records and maximum likely dispersal (predicted range). Maximum likely dispersal is a variable that can be defined by species, lifestage, observation date and dispersal potential. Sorry, this all probably sounds horribly complicated to other folks reading the thread! It isn't really - honest. Both Vicar and I are aiming to achieve the same ends, the difference at the moment is in the methods that we may use to estimate overall distribution. Vicar, if I remember correctly you use a statistical routine to predict presence based on nearest neighbour. As I understand it, your method would involve dividing the area of interest into a number of blocks (defined by 6 figure grid references?) and then predicting the likely presence of each species in that block using statistical analysis of the nearest neighbour dataset? Presumably you would then compare predicted presence against HSI to further refine the occupancy estimate? Or perhaps only run the calculation on those squares already flagged by the HSI? I guess what I am suggesting is to incorporate maximum likely dispersal into the calculation as a way of helping to further refine the estimate. A last comment - honest! We should also remember that presence of a species on a site may well be a consequence of historical dispersal. Just because a site is now isolated doesn't mean that is wasn't colonised at an earlier time. |
|||
Vicar
Senior Member Joined: 02 Sep 2004 Location: United Kingdom Status: Offline Points: 1184 |
Post Options
Thanks(0)
|
||
Lee, yes.... Its a three-stage process:
|
|||
calumma
Senior Member Joined: 27 Jun 2003 Location: United Kingdom Status: Offline Points: 375 |
Post Options
Thanks(0)
|
||
Have modified the relevant script and decreased processing time by 63% If I cleaned up some other elements within the table I would probably see further gains as well |
|||
calumma
Senior Member Joined: 27 Jun 2003 Location: United Kingdom Status: Offline Points: 375 |
Post Options
Thanks(0)
|
||
I suspect that since it may be difficult to obtain dispersal distance values
that have been quantitatively derived for all species, what I may do is use interquartile nearest neighbour values for minimum and maximum likely distribution (from each confirmed record). To me, this smells better that just picking a number out of the air. For interest, 75% (upper quartile) of Kent adder records are situated within 0.86 km of each other. This seems a reasonable figure to use for the upper limit. Steve, do you have any similar values for Surrey? Edit: To clarify, the upper quartile is a reasonable figure to use for the distance in which the majority of observations are likely to be encountered (what I meant by upper limit). Nearest neighbour values in the upper quartile would therefore represent the minimum distribution range. Hope this makes sense? Edited by calumma |
|||
calumma
Senior Member Joined: 27 Jun 2003 Location: United Kingdom Status: Offline Points: 375 |
Post Options
Thanks(0)
|
||
And when I control for possibly spurious outliers by excluding unconfirmed records, the value drops to 0.64 km. I'll post a graphic showing what this looks like. |
|||
Post Reply | Page <1234 8> |
Tweet
|
Forum Jump | Forum Permissions You cannot post new topics in this forum You cannot reply to topics in this forum You cannot delete your posts in this forum You cannot edit your posts in this forum You cannot create polls in this forum You cannot vote in polls in this forum |