Aggregation times of tuna schools to FADs estimated by echosounder data

We perform a systematic study of aggregation and disaggregation times of tuna schools to drifting
Fish Aggregating Devices (dFADs), using the signal provided by the echo-sounder buoys attached
to dFADs deployed across all major oceans in the period 2018-2020. The tuna biomass estimation
for each day in the time series has been obtained by applying the TUN-AI Machine Learning model
(Precioso et al., 2021), which incorporates oceanographic information and hourly echo-sounder data in
10 depth layers on a time window of 72 hours prior to the prediction. We preprocess the data collected
from the buoys to select around 10 000 series with daily estimations where no human intervention
has occurred. A statistical analysis of these time series with different smoothing techniques shows
that tuna schools remain aggregated to dFADs for a median time of 3-9 days, and that the aggregation
and disaggregation processes are symmetrical.