Rhino population in the world require a good estimation method

http://www.savetherhino.org/rhino_info/rhino_population_figures

I imagine that rhinos clump together somewhat, so what about adaptive cluster sampling? Use some probability sampling procedure to select the initial sampling units, and then sample neighboring units if some criterion or criteria are met. Adaptive sampling has a long history in ecological research. 

You could try looking at these and contacting Richard Emslie of the IUCN rhino specialist group. The Bayesian method is based on identification of known individuals in a mark recapture framework.

Emslie R.H. (2003) Block count spreadsheet data analysis and count stratification

planning tool (available free from AfRSG)

Emslie R.H., R. Amin & K. Davey. Rhino 2.1. A Bayesian mark-recapture based

population estimation analysis software package, designed specifically for rhinoceros in

particular, but applicable to other species. Software (Emslie RH, R Amin and K Davey),

User manual (Emslie RH), Reference Guide (Emslie RH) with context sensitive

help/training videos (Emslie RH)

Huberus, I'm all in favor of Bayesian mark-recapture (ahem, DOI: 10.1111/2041-210X.12140) but I believe that Babu is looking for a way to reduce their efforts. Artificially marking the rhinos would be incredibly labor-intensive, and so would the surveys. Using natural marks (e.g., horn shape + body markings) would also require an intensive amount of effort to photograph the mark and to identify the individual back in the office. 

The advantage of an adaptive sampling approach is,

1) It's relatively cost-efficient and doesn't require marking animals,

2) you can a priori calculate your costs via simple equations,

3) you get a minimum population size from your count as well as a statistically robust population estimate along with a well-established uncertainty estimate,

and

4) it doesn't require hiring trained experts like you would need to do for a mark-recapture study.

Why distance sampling doesn't fit for Rhino? Won't that be a simple solution to count Rhinos?

If you have some geographic point data (doesn't take much) for known rhino locations, and several layers of ecological/climatological data for the same time period and region, you can use MaxEnt (https://www.cs.princeton.edu/~schapire/maxent/) to model a surface that will assign suitability scores for rhino presence. This will basically be a percent likelihood of presence for every cell in your study area. 

Many thnx Isa, Jason, Hubertu, Yadav and William for your suggestions!

@ Jason, yes I am looking for less labor intensive method which doesn't compromises the reliability and validity of the estimates! Adaptive cluster sampling seems useful.

@Hubertus, We tested block count, mark-recapture and ID based monitoring methods. Estimates from all three methods were very close but all of these are labor intensive and requires well-trained/experienced observers.

https://www.researchgate.net/publication/259434884_Population_status_structure_and_distribution_of_the_greater_one-horned_rhinoceros_Rhinoceros_unicornis_in_Nepal

@ Yadav, I was also thinking about distance sampling, never tried but it could be an option.

Article Population status, structure and distribution of the greater...

With any kind of sampling the precision of the final estimate is a function of the sampling intensity. The better the answer the more it will cost.

I do not know enough about the situation of rangers in the parks in Nepal but I believe that a combination of the current total counts (very expensive but providing good answers for a minimum number alive estimate) could be combined with a Bayesian mark recapture estimate based on individual recognition of all animals seen by the rangers on their patrols. It is possible to build up a database of known individuals from distinguishing features.

The total count can be used as an initial prior for the Bayesian approach to mark- recapture which will very significantly improve the precision of a Bayesian estimate.

Whatever method you use you will need a lot of data and getting this data is expensive unless it can be gained while doing other routine patrolling in the park.

Distance sampling is no different from other methods in respect to the data requirement.

Hi...a few thoughts on your question:

A) block counts are not based on any sort of estimation theory and, in my opinion, will come up with numbers that have little relationship with true abundance. In particular, they do not account for partial detectability. They may account for spatial sampling, depending on how blocks are selected, but this is usually ignored in practice.

B) adaptive sampling is simply one of many ways of accounting for spatial sampling ( examples of other schemes being simple random sampling, systematic sampling with a random start, stratified random sampling, etc.). I don't know much about rhinos but feel that they wouldn't be so clumped as to require adaptive sampling.

C) the other key issue (apart from spatial sampling) to address is how to account for and estimate detectability. Both distance sampling and capture recapture allow you yo do this, as long as model assumptions are reasonably satisfied.

D) I agree that spatial capture recapture approaches are far better than conventional non spatial ones, but you can use either maximum likelihood-based or Bayesian models (each has its own advantages, but produce very similar results). IF rhinos can be unambiguously identified from individual marks, folds, etc., and you are able to more or less evenly cover a large enough area, I don't think the effort required would be particularly large.

E) As Yadav pointed out, distance sampling also works well, as long as one can cover a large number of RANDOMLY located transects. Use of roads/trails will not work. My organisation successfully used distance sampling to estimate rhinos in Kaziranga, although very few transects were used.

F)SDMs such as Maxent address a very different issue, and presence only methods are easily biased by non-random sampling and/or imperfect detectability. Occupancy modelling can account for these problems

Just one final point: in the Bayesian SCR approach, the count of animals is ALREADY used as the lower bound on the uniform prior (the upper bound being the data augmentation value) for number of activity centres (= animals) in the state space S.

Less labour intensive method would be: i) Collecting fresh dungs during the dry season intensively and carry out the Genetic Analysis. It is superb, I think. 

ii) Block count has its own benefits in case of Nepal as this has been a regular practice. It produce reliable results too. Therefore, I do not see any problem using this in case of Nepal. 

iii) distance sampling does not work in case of rhino as the animals are not randomly distributed. Rhinos are habitat specific animals and have very clumped distribution. 

Iv) the use of methods and selection of methods is area specific. The method that works for Africa does not work in Chitwan because of different habitat type. 

Thank you, 

Remote cameras have proven the best method to count Javan rhinos and, in my opinion, it is also the case with Sumatran rhinos. 

Hi Babu,

People say that when you only have a hammer all problems tend to look like a nail, but I'd say that distance sampling mightbe a really good way to do these.

So instead of trying to count all the rhinos in blocks, select a random sample of blocks and do line transects from the elephants. This will allow you to model detectability and hence account for missed animals. If the transects are laid at random then you can get density estimates for the entire area by design, but probably even better you could use a density surface model to predict density as a function of available covariates (e.g. habitat, diostance to water, altitude, distance to human settlement, etc). A couple of good tutorial like paper on these are

Katsanevakis, S. Density surface modelling with line transect sampling as a tool for abundance estimation of marine benthic species: the Pinna nobilis example in a marine lake Marine Biology, 2007, 152, 77-85

Miller, D. L.; Burt, M. L.; Rexstad, E. A. & Thomas, L. Spatial models for distance sampling data: recent developments and future directions Methods in Ecology and Evolution, 2013, 4, 1001-1010

Curiosly, there are no references on distance sampling with rhino's here (a webpage I maintain on distance sampling references):

http://distancesampling.org/dbib.html

(if someone knows one, please send me an e-mail with the reference so I can add it to the list)

hope this helps

Tiago

Just to add a brief point to Tiago's response: as Tiago pointed out, the transects have to random with respect to animal distribution (especially if your density estimate for the larger area of interest is design-based), the animals themselves don't have to be random at all.