HyperAI超神经

Bird migration is a fascinating natural phenomenon.It is understood that nearly one-fifth of the world's bird species migrate regularly for breeding and wintering. In ecology, studying ecological laws such as bird migration routes is of great significance for protecting endangered bird species, maintaining ecological balance, and preventing the spread of epidemics.

In recent years, due to factors such as global climate change and human activities, predicting bird migration has become more difficult. Recently, Miguel Fuentes, a graduate student at the University of Massachusetts Amherst, and Benjamin M. Van Doren of Cornell University published a new probabilistic model, BirdFlow, in the journal Methods in Ecology and Evolution.The model uses computer modeling and eBird datasets to accurately predict the flight paths of migratory birds.

This research result was published in Methods in Ecology and Evolution

Paper address:

https://besjournals.onlinelibrary.wiley.com/doi/full/10.1111/2041-210X.14052

The researchers used relative abundance estimates from the eBird Status & Trends project to simulate bird movements, but there was a problem: past relative abundance information only showed the location range of birds each week, and could not track individuals. So in this study, the researchers focused on solving this problem.The key process is shown in the figure below:

Figure 1: Data preparation and modeling process

* Data Preprocessing: Preprocessing relative abundance estimates to produce weekly population distributions;

* loss function: specifies a loss function that scores potential models using a weekly distribution and a proxy for energy cost;

* Model Structure: Select a model structure;

* Trained Model: Optimize the loss function through a numerical process to select the best model parameters;

* Validation: Calculate the average log-likelihood and PIT value of real birds to validate the trained model.

BirdFlow Modeling Overview

Researchers use ebird R downloaded relative abundance estimates for 11 bird species from eBird Status & Trends, and GPS or satellite tracking data were available for these 11 bird species.

eBird Status and Trends:

https://science.ebird.org/zh-CN/status-and-trends

Table 1: GPS tracking data of the 11 bird species used

Next, the researchers defined a loss function,The loss function is based on weekly population distribution derived from eBird Status & Trends, the energetic cost of bird movement between locations, and an entropy regularization term.

Before optimizing the loss function, a model structure needs to be specified. Here, the researchers proved that it is reasonable to limit the optimization process to searching on Markov chains. Therefore, they modeled the movement of birds as a Markov model and optimized it.Includes the use of Markov chain parameterization and optimization algorithms.

After the above steps, the researchers obtained a trained model.And relevant verification was carried out.

BirdFlow Validation Process

The verification process is divided into three parts:Hyperparameter grid search, Entropy calibration, k-week forecasting,The specific process and test results are as follows.

Hyperparameter Grid Search

To validate the model, the researchers performed a hyperparameter grid search and used the search results to study two questions.

First, the researchers conducted an ablation study.The effects of entropy regularization term and distance index on model quality were explored.The results of the ablation study are shown in the figure below. It can be seen that all BirdFlow models perform better than the baseline model that only includes the relative abundance of birds.

second,The researchers explored the model's sensitivity to hyperparameter selection using two hyperparameter selection methods.The results are shown in the figure below. For most bird species, the model using the LOO parameter (selected using validation tracking data from other bird species) performs just as well as the model using the tuned parameter (using validation tracking data from that bird species). Performance is measured as the average log-likelihood of the 1-week transition.

Entropy Correction

The figure below shows the effect of entropy regularization on model calibration.The random probability integral transform (PIT) histogram of the five versions of the American Woodcock model at different entropy weights shows the trained model's prediction of the woodcock's east-west position over the week.

As you can see, the histograms are almost identical, indicating that the model is well calibrated.

Figure 4: Effect of entropy regularization on model calibration

k-week forecast

Figures 5 and 6 show the model performance at different forecast times (in weeks). The researchers identified the best performing model from a hyperparameter grid search and evaluated it for forecast times ranging from 1 to 17 weeks.How this best model performs relative to the baseline model.

Figure 5(a) shows the results for each bird type.It can be seen that as time goes by, the performance of the best model for each bird is getting closer and closer to the baseline model.Figure 5(b) shows the gap comparison between the Woodcock-tuned model, the LOO model and the baseline model.It can be seen that during the prediction time, the performance of the tuned model and the LOO model is better than that of the baseline model.

Figure 6: Inference results of the small woodcock model

Through the above experiments, the researchers found that BirdFlow can use eBird's weekly relative abundance estimates to accurately infer the migration paths of individual birds, and the results showed thatBirdFlow predicts results much better than the baseline model.

Based on this result, the researchers believe that in addition to studying the natural phenomenon of bird migration,The BirdFlow model may also be used to study other phenomena.For example, the stopover behavior of birds and their response to global change.

However, despite the BirdFlow model's success, some researchers in North America and Europe have questioned its use of the eBird database, arguing that bird watching is not a rigorous method for collecting data. In response, BirdFlow researchers said:The team is considering integrating more data, such as satellite or GPS data tracking the location of birds.

AI may become the protector of nature

The emergence of the BirdFlow model means that humans have opened up a shortcut to machine learning in the study of bird migration. Although it is still in its early stages and there is still a long way to go before it can be put into practical application such as nature conservation, this research undoubtedly reveals an important trend.AI is being widely used in the field of nature conservation.

PAWS, developed by researchers at Carnegie Mellon University, can generate a patrol route for police to target poachers; Merlin, developed by Cornell University, can identify species based on bird songs and images; and TrailGuard AI, developed by Resolve, can protect wildlife by identifying images of suspected poachers and issuing alarms.