Observer performance is one of the biggest issues in aerial survey, and probably the hardest to control; this is something we’ll be discussing again and again in aerial survey methods, at least until the time when we switch to using entirely remote sensing techniques. We accept that aerial sample counts usually miss groups of animals – some are completely hidden from view, observers miss some as they search other parts of the sampling strip, animals move out of the strip before you arrive, and so on.
Just how many groups are we missing? How important is experience, and how much training might be needed? Seeing animal groups from the air is not a trivial job – it is a demanding one, requiring you to pay close attention to a task for long periods of time, and in a rather horrible environment with vibration, noise and (often) heat to contend with.
In training, with a brand-new observer, I find that the new observer is usually spotting the great majority of animals by the end of the second training session, which is the second of two 2.5 hour sessions (this is after eliminating others in the first session or before flying due to species recognition and eyesight issues). They often have great difficulty on the first flight, especially the first hour – spotting animals from the air is a different skill to spotting from the ground, and the moving field of view is difficult to get used to.
In his 2001 book, Jachmann1 discusses an experiment using double-counts to examine the percentage of animals seen based on a number of factors:
- The total hours of the observer’s previous experience in aerial survey
- The observer’s current experience on the current survey;
- The duration of each flight.
Twelve observers with varying levels of experience flew survey flights using double-counts (which allow one to determine the percentage of observations missing per observer). A model was developed to show percentage of animals seen based on the three factors2:
%=0.28⋅Previous+4.01⋅Current — 7.68⋅Duration+77.49
What this means is that, in their experiment, you start with a base chance of seeing something of 77.49%, and this is modified by the other factors: for every hour you fly, the chance of seeing something goes down by 7.68%; every hour of previous survey experience gave 0.28% higher chance; every hour of current experience added 4.01%.
What I find particularly interesting about this is that it lets us look at the relative value of current and previous experience. How many hours of experience are needed to see 100% of animals at the beginning of a flight?
- Previous experience needed = (100-77.49) / 0.28 = 80 hours
- Current experience needed = (100-77.49) / 4.01 = 5.6 hours
It seems that current experience far outweighs previous experience by a factor of 14. Less than 6 hours of current experience would be needed to bring an observer up to the same level as an observer with previous experience on several surveys! Obviously there are a lot of other factors, and we don’t know how observers were pre-selected for Jachmann’s study, but this is a really important finding.
Just as importantly, the effect of flight duration is large – the model implies that after 3 hours of flying you’d have 23% less chance of seeing animals in the strip. This could be offset partly by observer experience, but the effects of fatigue are non-linear (something not examined in the study), increasing dramatically after 30 minutes on a single task and probably after 3 hours in a single session.
The observer-experience effects Jachmann saw accords with what I’ve experienced in training – observers typically learn very rapidly, becoming accustomed to the difficult conditions so that by the third flying session I begin to be confident that they can do the job accurately. However, this is something we really need to look into – particularly because experience and fatigue are almost certainly non-linear.
1 Jachmann, Hugo, 2001. Estimating abundance of African wildlife: an aid to adaptive management, Kluwer Academic Publishers.
2 The model showed a significant fit3, but only explained 50% of the variation. Other factors that were probably having an effect would be group size, habitat, sunlight, observer eyesight, variable cloud cover, other weather conditions, and many others.
3 A “multiple correlation coefficient of 0.704 and a significant F-statistic (P<0.05)”.
