link: https://www.nature.com/articles/s41467-024-48239-8

long paper with loads of domain knowledge. This summary is more focussed on the statistical techniques they have used and the models they have built.

Motivation and context

Emperor pengs live in harsh env with tough to monitor and observe conditions. But climate change will affect their habitat realllyy badly. Out of all colonies only 2 have been properly monitored with on ground data points and other metrics being closely recorded.

incubating and caring for their young ones takes a couple of months. During this period, different habits tend to show up: one parent might stay, other might go for foraging trip, to regulate body temp they will form huddles, etc. So at any given time, abundance and density will depend on their behaviour.

Satellite imaging allows us to make use remote sensing easily but has problems: Polar nights, off nadir images, sun angle, etc.). Using satellites it is easier to estimate area covered by the huddled colony or groups rather than count individuals. So, we an potentially get the area from this imagery and use formulas based on their behaviours to estimate the type and number of individuals in that area.

Useable satellite images due to polar nights are present for sept-jan, when chicks and only fraction of adults present in colony.

Goal

The aim of this study is to develop a method to compensate for the uncertainties of satellite-based surveys and to provide an estimate of the annual number of breeding pairs as well as the annual breeding success of a colony, based on the colony area measured during the austral spring and summer (September to December).

Broadly

The method can be broken down into three separate steps. First, we convert colony covered areas from ground based or satellite imagery to individual counts by modeling the colony density as a function of temperature, wind speed, solar radiation, and humidity at the colony site (“windchill model”). Second, we present a phenological model that describes how the number of individuals present at the colony on each day depends on the number of breeders and the breeding success. We benchmark the model with ground based individual counts. Third, we invert the phenological model to infer the number of breeding pairs and the breeding success from sparse counts of adult animals at the site of the colony, obtained near the end of the breeding season. We benchmark this method with data from ground-based and satellite-based images.

Satellite images

Manual annotations are drawn as polygons to cover the entire colony area. These are then projected into a top-down view or Nadir view using cameratransform. Not sure how reliable this is.