density surface modeling (dsm) for replicate surveys

[Samantha Ball]

Dear List, 

  Hoping someone here can help an extremely frustrated and confused list member. I have been playing around with distance data on and off for a while now but have been stuck on something for a few weeks and feel like it is potentially something small that I am missing. I am using the distance and dsm packages in r. 

I have 9 replicate surveys of a site across the space of a year, with a single observer to estimate the size of a lagomorph population in an enclosed site. Each season has 2 surveys except for summer with 3 surveys. My survey area is ~212ha.  There are three line transects used for each survey, resulting in a total of n=338 observations (including individuals and groups). 

I have successfully modelled the population size with the dht2 function and the results are quite satisfactory with appropriate CV/ SE values. However, as the site that I am conducting the surveys in is quite complicated, I am trying to use DSM to estimate the size and plot the abundance over the site. This is where I am running into trouble. Whereas the dht2 estimates are giving an average estimate of ~172 animals due to the ability to post stratify, I am getting wild estimates from the dsm model output of ~1,600 animals. This just isn't possible for an area 212 ha in size. 

Is there a way for me to tell the dsm function that they are indeed replicate surveys as I believe that estimates are currently being based on replicates being a single survey. I get crazy estimates with or without adding covariates to the detection function and with or without pooling together the surveys into the Region.Label for the detection function. 

Is dsm appropriate for multiple, replicate surveys? 

Example code below: 

dataframe=Tran.com

summer<- ifelse(Tran.com$Survey_Number=="3" | Tran.com$Survey_Number=="4"| Tran.com$Survey_Number=="11", TRUE, FALSE)

autumn <- ifelse(Tran.com$Survey_Number=="5"| Tran.com$Survey_Number=="6", TRUE, FALSE)

winter <- ifelse(Tran.com$Survey_Number=="7"| Tran.com$Survey_Number=="8", TRUE, FALSE)

spring <- ifelse(Tran.com$Survey_Number=="9"| Tran.com$Survey_Number=="10",TRUE, FALSE)

Tran.com$Region.Label[summer] <- "summer"
Tran.com$Region.Label[autumn] <- "autumn"
Tran.com$Region.Label[winter] <- "winter"
Tran.com$Region.Label[spring] <- "spring"

#Detection Function

tran.com.hr.80 <- ds(Tran.com, transect = "line", key="hr",adjustment = NULL,
                  formula=~as.factor(Region.Label), truncation=80,
                  convert.units = unit_conversion_line) #AIC= 2754

#segment data set to 160m as truncated to 80m

segment data= segs.2_df

Joined observation data=obs_df.2

#DSM- abundance.est being used as covariates in detection function and because covariates differ between segments

tran.com.hr.80.dsm <- dsm(abundance.est~s(X, Y), ddf.obj=tran.com.hr.80, 
                        segment.data=segs.2_df,
                        observation.data=obs_df.2, Transect.Label = "Transect.Label",
                        family=tw(),transect="line")

## prediction 

Trans.pred <- predict(tran.com.hr.80.dsm, Trans.preddata, Trans.preddata$Area) #
Trans.preddata $Tran_xy <- Trans.pred.2

#estimate sum of animals 

sum( Trans.preddata $Tran_xy ) = 1590

I have tried the above code with covariates (season) added into the segment data and the dsm model but estimates for this were even higher (~2,600) and individually modelling each season has also overestimated the population. 

I would greatly appreciate any insight or point in the right direction as I have spent weeks at this stage trying to work this out! 

Best Wishes, 

Sam

[Eric Rexstad]

Sam

It sounds as if you are producing reasonable estimates using the design-based approach.  When using the model-based approach, it is possible to use replicate surveys; the Gulf of Mexico spotted dolphin data set used in Miller et al. (2013) was data accumulated over replicate surveys.

Unusual estimates from density surface models is often the result of extrapolation.  You note you have three transect in your survey and your spatial model is using (x,y) coordinates as predictors.  Do the transects cover the full range of x- and y-coordinates in your study area?  If not, there might be some extreme extrapolation. 

Lots of visual model criticism is needed when interpreting density surface models.  Plot the gams; where is there support (from the rug plot) and is extrapolation outside the range of the data occurring.  Plot the predicted density surface; is there evidence of hot spots where extremely high densities are predicted?  It may be that (x,y) are poor predictors of the distribution of your study animals.

---

From: distance...@googlegroups.com <distance...@googlegroups.com> on behalf of Samantha Ball <samant...@ucc.ie>
Sent: 04 March 2022 17:41
To: distance-sampling <distance...@googlegroups.com>
Subject: {Suspected Spam} [distance-sampling] density surface modeling (dsm) for replicate surveys

 

--
You received this message because you are subscribed to the Google Groups "distance-sampling" group.
To unsubscribe from this group and stop receiving emails from it, send an email to distance-sampl...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/distance-sampling/495409f4-fdd5-422b-bdc5-2a5177b1ad20n%40googlegroups.com.

[Samantha Ball]

Dear Eric, 

 Thank you for getting back to me. There are some locations at my study site where transects did not reach, as they were inaccessible. Even when adding in a 4th transect I had data for, the estimates are not improved much, but probably because this 4th transect still excludes the areas which were inaccessible, as demonstrated on the plot with the transects and detections visible below. The resulting predicted density plot does show a higher density where you would expect to at the site. However, this prediction is still providing a rather large estimate of ~750, which is in line with what the detection function is estimating as the total number across all surveys. Does predicting the abundance when 'abundance.est' has been used to fit the dsm average across surveys or provide an estimate of the total abundance across all surveys? 

Best Wishes, 

Sam

[Eric Rexstad]

Sam

It is not my belief that density surface model abundance estimates using temporal replicates produce summed estimates of abundance across replicates.

Density surface modelling is estimating density in each prediction grid cell based upon the fitted gam model.  That density estimate is the average density across replicate surveys.  The abundance per prediction grid cell is the density estimate multiplied by the grid cell area.

---

From: distance...@googlegroups.com <distance...@googlegroups.com> on behalf of Samantha Ball <samant...@ucc.ie>

Sent: 08 March 2022 21:25
To: distance-sampling <distance...@googlegroups.com>
Subject: {Suspected Spam} Re: {Suspected Spam} [distance-sampling] density surface modeling (dsm) for replicate surveys

 

To view this discussion on the web visit https://groups.google.com/d/msgid/distance-sampling/f5b51b27-4ef3-48da-af67-e94536703798n%40googlegroups.com.