Avian Species Population Forecaster Using Machine Learning

Likhitha Morampudi

, Ramya Rajanala

and Pothuri Surendra Varma

Department of Artificial Intelligence and Data Science, Velagapudi Ramakrishna Siddhartha Engineering College,

Andhra Pradesh, India

Department of Computer Science and Engineering, Velagapudi Ramakrishna Siddhartha Engineering College,

Andhra Pradesh, India

Keywords: Ecosystem Dynamics, SARIMA Model, Temporal Patterns, Historical Data Analysis

Abstract: Birds help to link various ecosystems. Ecosystems like farmland, woodland, water and wetlands, wildfowl.

Migration patterns link biodiversity by facilitating gene flow, spreading seeds, transferring nutrients, and

maintaining ecological balance across different ecosystems. This research analyzed historical data of birds

from 1960 to 2015, and forecasted the future bird's population. and focused on predicting the bird's

population in various ecosystems. We employed the Seasonal Autoregressive Integrated Moving Average

(SARIMA) model is used to achieve accurate forecasting. bird population trends by integrating seasonal

and temporal patterns, thereby enhancing predictive precision for ecological monitoring and conservation

planning.

1 INTRODUCTION

Birds play a crucial role in maintaining ecological

balance by helping with pollination, spreading seeds,

controlling pests, and cycling nutrients through eco

systems. As ecological indicators, bird populations

reflect environmental health and alert us to

challenges such as climate change, habitat

destruction, and ecosystem disruptions. Accurate

forecasting of bird populations is critical for

conservation, as failure to predict declines could

lead to species extinction and ecological imbalances.

This research focuses on developing an avian

species population forecaster using the Seasonal

Autoregressive Integrated Moving Average

(SARIMA) model. With its ability to account for

both seasonality and trends in time-series data,

SARIMA is employed to predict bird populations

and provide insights for conservation strategies. This

approach aims to safeguard both biodiversity and the

ecosystem services essential to agriculture and

ecological stability.

1.1 Contribution

Development of a Bird Population Forecaster:

Introduced a SARIMA-based forecasting model

designed to capture both seasonal and long-term

trends in bird populations.

Time-Series Analysis of Bird Data: Applied

historical data to accurately predict bird population

dynamics across different habitats.

Identification of Key Environmental Stressors:

Provided insights into how climate change and

habitat loss are influencing bird populations,

enabling proactive conservation measures.

Support for Agricultural Practices: Highlighted

the importance of avian species in agriculture by

predicting the consequences of population declines

on pest control and seed dispersal. Conservation

Policy Implications: Offered actionable insights for

policymakers and conservationists to maintain

ecological balance and promote sustainable

ecosystems.

1.2 Motivation

Birds are important to ecological balance through

pollination, pest control, seed dispersal and nutrient

cycling. As ecological indicators, changes in bird

populations serve as warning signals for

environmental disruptions, including climate

change, habitat destruction, and pollution. If

conservationists fail to predict these population

changes, critical species could face extinction,

leading to cascading effects across ecosystems.

Morampudi, L., Rajanala, R. and Varma, P. S.

Avian Species Population Forecaster Using Machine Learning.

DOI: 10.5220/0013607700004664

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 2, pages 929-935

ISBN: 978-989-758-763-4

929

Furthermore, bird population declines could

negatively affect agriculture by increasing pests and

reducing crop yields, impacting both biodiversity

and economic stability. Therefore, timely and

accurate forecasting of bird populations is

essential

for devising proactive conservation strategies

and ensuring sustainable ecosystems.

1.3 Objectives

• Forecast Bird Populations: Develop a

model to predict future bird populations

for effective conservation planning.

• Incorporate Seasonality and Trends:

Leverage seasonal and non-seasonal

patterns in historical data to improve

prediction accuracy. • Identify

Environmental Impacts: Use forecasting to

assess how factors like climate change and

habitat loss impact different bird species.

• Support Agricultural Sustainability:

Highlight the importance of bird species in

pest control and seed dispersal to inform

agricultural strategies.

• Inform Conservation Policy: Provide

insights to environmentalists and

policymakers to develop targeted actions

for preserving bird populations and

maintaining ecosystem health.

2 RELATED WORK

2.1 Review

Christiaan Both et al. (Both, Bouwhuis, et al. , 2006)

embedded about |Population trends of Dutch pied

flycatcher populations. They proposed climate-

change-induced badly timed leads to population

declines in a migratory songbird, used linear

regression and correlation tests. In result Spearman

rank correlation was used to relate the trends in

population decline with the timing of the caterpillar

food peak, calculated the annual median laying date

from 1980–2002.

Birgit Erni et al. (Erni, Liechti, et al. , 2003)

embedded about Simulations of individual bird

migration paths across a grid-based environment,

considering fuel loads, stopovers, flight costs, and

directions. Vector Summation. The combination of

spatial modeling, vector summation (Navigation

Algorithm) constant endogenous direction,

evolutionary algorithms, and energy cost functions.

Spatially Explicit Individual-Based Model, which is

a simulation algorithm and a Directional Adaptation

Algorithm.

James A. Smith et al. (Smith, Deppe, et al. ,

2007) embedded about individual-based modeling to

predict how environmental changes might impact

migratory birds and maximum entropy model. The

focus is on predicting migration patterns by

integrating environmental data, bird physiology, and

energetics and also modeled the spring migration of

the Pectoral Sandpiper (Calidris melanotos) in North

America and observed how environmental

conditions and stopover habitat quality affect the

success of migration.

Hiromi Kobori et al. (Kobori, Kamamoto, et al. ,

2012) Using 23 years of citizen-scientist

observations, they analyzed the first arrival and final

departure dates of birds at a wintering site in

Yokohama and correlated these dates with

temperature changes. In their observations on

average, birds are arriving 9 days later and departing

21 days earlier than in the past, shortening their stay

by about one month. These changes are linked to

rising temperatures. Their study is limited to one

location in Japan, making it difficult to generalize

findings to other regions without additional data. But

our data covers whole Europe continent. During data

collection process they have used manual power

rather than using Weather Radars and Dedicated

Avian Radars. Speed Conscious Recurrent Neural

Network (Varma, Anand, et al. , 2022) and federated

KNN (Varma, Anand, et al. , 2024) also propose

efficient machine learning algorithms.

Anders P. Tøttrup1 et al. (Tøttrup, Thorup, et al.

, 2008) Utilized NDVI as a proxy for ecological

conditions, such as food availability. NDVI was

calculated for both African wintering areas and

European stopover sites to assess how vegetation

growth impacted the timing of migration. linear

mixed models to analyze the relationship between

NDVI, timing of migration, and migration duration

across different latitudes and migratory phases (first

5%, 50%, and 95% of the migrating

population).Their study divides migration into three

phases based on when portions of the population

migrate (early, mid, late) and examines the impact of

environmental conditions on these phases .It also

asserted the complexity of migration systems and

the need for further studies to distinguish between

phenotypic plasticity and evolutionary changes.

Bidirectional LSTM was proposed in (Sowmya;,

Pothuri, et al. , 2024).

Advaith S Pillai et al. (Pillai, Sathvik, et al. ,

2024) demonstrated the potential of integrating IoT

INCOFT 2025 - International Conference on Futuristic Technology

930

and ML techniques in wildlife conservation, helping

address challenges like habitat loss and urbanization

that make traditional bird monitoring difficult.

Researchers mainly considered bird species like

Eurasian Curlew, Blue-tailed Bee-eater, and others

found in Kerala, India. Mel-Frequency Cepstral

Coefficients (MFCCs) for extracting features from

bird calls. Additional feature extraction techniques

include Δ and ΔΔ MFCCs, which are derived from

MFCCs to capture changes in frequency over time.

ANN architecture consists of 4 dense sequential

layers and is used for multi-class classification of

bird sounds. The CNN architecture includes three

2D convolutional layers with different filter sizes to

detect bird images for classification. They have

utilized some IOT devices like Raspberry Pi, to

collect and process data from sensors (cameras and

microphones), transmitting it to the cloud for further

research. Microphones capture bird calls, and

cameras capture images of birds for classification.

Standard Scaling to normalize feature data for the

ML model. Applied Principal Component Analysis

(PCA) to reduce the dimensionality of the feature

space before feeding it into the ML model. Aspired

to involve, integration of Long Short-Term Memory

(LSTM) with the Recursive Neural Network (RNN)

to treat bird calls as time-series data, which can

improve classification accuracy and combining both

image (CNN) and audio (ANN) classification results

is suggested to further enhance the performance of

the system.

A semi centralized architecture is proposed in

(Varma, Anand, et al. , 2023) for efficient prediction

using machine learning. Multivariate regression is

used in (Varma, Anand, et al. , 2023). Some of the

interrupting issues are Noise Interference, Multiple

Bird Calls, Lighting Conditions which affects the

quality of the images captured. Movement of Birds

(high frame rates and image resolution are needed,

which adds complexity to the system design). Power

consumption of microphones, cameras, and

processing units can be a challenge, especially for

long-term deployments.

Jianxi Zhang et al. (Zhang, Shao, et al. , 2018) in

their research Density-Based Spatial Clustering of

Applications with Noise (DBSCAN) algorithm,

focuses on identifying spatial patterns of bird

habitats and uncovering clusters of bird presence

from geospatial data, improving the understanding

of bird distribution. It detects clusters based on

proximity and density. Parameter Tuning, to

optimize the DBSCAN algorithm, careful selection

of parameters like ε (radius) and MinPts (minimum

points). One of the key strengths of DBSCAN is its

ability to identify noise or outliers and works well

with large datasets, which is an advantage when

dealing with extensive geospatial data. But it may

not accurate on Handling High-Dimensional Data,

Parameter Sensitivity, Less Quality Data.

3 METHODOLOGY

In this study, we employed the SARIMA (Seasonal

Auto Regressive Integrated Moving Average) model

for predicting bird population trends based on

historical data. The dataset, stored in a CSV file,

consisted of a Date column representing the timeline

and a Value column indicating bird population

counts. During data preprocessing, missing values

were handled using mean imputation to ensure

consistency, and the dataset was cleaned to align

with the model’s requirements. The SARIMA model

was chosen for its capability to capture both

seasonal variations and long-term trends, making it

suitable for time series forecasting. We used the

Akaike Information Criterion (AIC) and Bayesian

Information Criterion (BIC) to optimize the non-

seasonal (p,d,q) and seasonal parameters (P,D,Q,m)

with minimal error of the forecast. Model training

was performed using the statsmodels library in

Python, leveraging data from 1970 to 2015 to build

the model. Monthly forecasts were generated for 12

months of 2016, allowing the model to capture

short-term seasonal fluctuations.These forecasts

were aggregated to obtain the annual predicted

population value by averaging the individual

monthly forecasts. Visualizations comparing

forecasted values with actual data were created using

matplotlib to identify trends, patterns, and

discrepancies. Additionally, a flowchart depicting

the entire process from data collection to forecast

generation was included to enhance clarity. The

methodology was designed to be replicable, with all

steps, libraries, and tools—such as Python, pandas,

statsmodels, and matplotlib—documented. Also,

evaluation metrics like MAE — Mean Absolute

Error and RMSE — Root Mean Squared Error were

regarded to evaluate model effectiveness and verify

forecasts accurateness. A flowchart illustrating the

SARIMA model is provided in Fig 1.

Avian Species Population Forecaster Using Machine Learning

931

Figure.1: Flowchart of SARIMA Algorithm

3.1 Dataset details

The dataset used in model for predicting the birds

population using time series forecasting contains the

yearly data across various habitats like woodland,

farmland, water and wetland. The data was taken

from (European Environment Agency, 2024)

contains the data from year 1970 to 2015. The

values in this dataset are normalized such that the

initial population for each habitat is set to 1, and

subsequent values are represented as ratios relative

to this initial population. The dataset contains three

main variables year, habitat, and population count.

The dataset contains bird population counts

expressed as ratios, with the initial value for each

habitat type set as 1.

This approach allows for straightforward

comparative analysis, enabling us to track relative

changes in population over time across different

habitats. By using this normalized data, we can more

effectively assess the impacts of various ecological

factors on bird populations.

This study utilizes a comprehensive dataset that

includes bird population records from 1970 to 2015.

In total, there are 368 records encompassing various

categories of birds. These categories include all

species, woodland birds (which are further divided

into all, specialist, and generalist), farmland birds

(comprising all and generalist), wetland birds, and

wildfowl. It is important to note that the dataset

contains missing values for water and wetland

habitats, as well as wildfowl habitats. To address

this issue, missing values have been replaced with

mean values, ensuring continuity in the analysis.

This approach allows for consistent comparison of

trends across different habitats.

In this study, we leveraged both the smoothed

2.5 confidence interval (CI) and the smoothed 97.5

CI derived from our dataset to enhance the accuracy

and reliability of our predictions regarding bird

populations. bound for our predictions, indicating

the maximum expected population. This information

is valuable for understanding the best-case scenarios

and planning for optimal conservation strategies. By

incorporating both the lower and upper confidence

intervals, we aimed to capture a comprehensive

range of uncertainty in our forecasts. To train our

predictive model, The smoothed 2.5 CI served as a

critical lower bound, allowing us to identify the

potential minimum expected population, which is

essential for assessing risks related to population

decline. This interval highlights scenarios where

intervention may be necessary to prevent further

decreases in species numbers. Conversely, the

smoothed 97.5 CI provided an upper we included

both the 2.5 CI and 97.5 CI as features, alongside

the smoothed population estimates.

This approach allowed us to account for the

variability in the data and provided a more nuanced

understanding of potential outcomes. The model’s

predictions, enriched by these confidence intervals,

facilitate informed decision making in conservation

efforts, enabling us to allocate resources effectively

and develop strategies that address both risks and

opportunities in bird population management.

Figure.2:Bird population trends with Confidence Intervals

4 RESULT ANALYSIS

We present the findings from our SARIMA model

analysis, which predicted monthly bird populations

across different habitats—farmland, woodland, and

INCOFT 2025 - International Conference on Futuristic Technology

932

wetland. The model effectively captured population

dynamics, as illustrated in Figures 3-8, where we

observe notable trends and fluctuations supported by

confidence interval.

Figure 3: Forecasted Values and Confidence Intervals for

All Birds: 2.5% range

Figure 4: Forecasted Values and Confidence Intervals for

All Birds: 97.5% range

The forecasting results from the SARIMA model

for the "all birds" column in the dataset indicate a

range of values defined by the 2.5% and 97.5%

confidence intervals (CIs). The lower bound (2.5%

CI) shows a stable trend, with values ranging from

0.9135 in the first month to 0.9198 by the twelfth

month, resulting in an average of approximately

0.9195. In contrast, the upper bound (97.5% CI)

exhibits a slight decline, starting at 1.0406 and

decreasing to 1.0451, yielding an average of about

1.0475. This analysis underscores the relative

stability in the "all birds" data, with the upper and

lower bounds illustrating the inherent uncertainty in

forecasting process.

Figure 5: Forecasted Values and Confidence Intervals for

Woodland Column: 2.5% range

Figure 6: Forecasted Values and Confidence Intervals for

Woodland Column: 97.5% range

The forecasting results from the SARIMA model

for the woodland column in the dataset indicate a

range of values represented by the 2.5% and 97.5%

confidence intervals (CIs). The lower bound (2.5%

CI) shows a declining trend, with values decreasing

from 0.7017 in the first month to 0.6528 by the

twelfth month, averaging approximately 0.6784. In

contrast, the upper bound (97.5% CI) exhibits a

slight increase, ranging from 0.8603 to 0.8646, with

an average of about 0.8621. This analysis highlights

the notable variation in the woodland data, with the

confidence intervals reflecting the inherent

uncertainty in the forecasting process.

Avian Species Population Forecaster Using Machine Learning

933

Figure 7: Forecasted Values and Confidence Intervals for

Water and Wetland Column: 2.5%

Figure 8: Forecasted Values and Confidence Intervals for

Water and Wetland Column: 97.5%

The forecasting results from the SARIMA model

for the woodland column in the dataset indicate a

range of values represented by the 2.5% and 97.5%

confidence intervals (CIs). The lower bound (2.5%

CI) shows a declining trend, with values decreasing

from 0.7017 in the first month to 0.6528 by the

twelfth month, averaging approximately 0.6784. In

contrast, the upper bound (97.5% CI) exhibits a

slight increase, ranging from 0.8603 to 0.8646, with

an average of about 0.8621. This analysis highlights

the notable variation in the woodland data, with the

confidence intervals reflecting the inherent

uncertainty in the forecasting process.

5 CONCLUSIONS

This study successfully utilized the SARIMA model

to forecast bird population trends across diverse

ecosystems, analyzing historical data from 1960 to

2015. The results, supported by various validation

metrics, underscore the model's effectiveness in

capturing seasonal and temporal patterns, thereby

enhancing predictive accuracy for avian populations.

Despite the insights gained, challenges remain,

including the need for more comprehensive datasets

and the integration of real-time environmental

factors. Future research should address these gaps to

further refine forecasting methods and improve

conservation strategies. The inclusion of a metrics

table in this study highlights the model's

performance and provides a foundation for

comparing future methodologies in bird population

forecasting.

Table 1

HABITA

CONFIDENC

INTERVAL

MAE

MAPE

All

ecies

2.5 CI

0.029

0.03

3.3112

All

Species

97.5 CI 0.041

0.04

4.0423

Wood

Lan

2.5 CI

0.008

0.00

1.0994

Wood

Lan

97.5 CI 0.004

0.00

0.5509

Water

Wetlan

2.5 CI 0.011

0.01

1.2662

Water

Wetlan

97.5 CI 0.019

0.02

1.6577

REFERENCES

C. Both, S. Bouwhuis, C. Lessells, and others, "Climate

change and population declines in a long-distance

migratory bird," Nature, vol. 441, pp. 81–83, 2006.

[Online]. Available:

https://doi.org/10.1038/nature04539.

B. Erni, F. Liechti, and B. Bruderer, "How does a first

year passerine migrant find its way? Simulating

migration mechanisms and behavioural adaptations,"

OIKOS, vol. 103, pp. 333–340, 2003.

https://doi.org/10.1034/j.1600-0706.2003.12176.x.

J. A. Smith and J. L. Deppe, "Simulating bird migration

using satellites and biophysics," in Proc. IASTED

Symposium on Environmental Modeling and

Simulation, vol. 509, pp. 6–11, Aug. 2007.

H. Kobori, T. Kamamoto, H. Nomura, and others, "The

effects of climate change on the phenology of winter

birds in Yokohama, Japan," Ecological Research, vol.

27, pp. 173–180, 2012.

https://doi.org/10.1007/s11284-011-0891-7.

Varma, P.S., Anand, V., ”Fault-Tolerant indoor

localization based on speed conscious recurrent neural

network using Kullback– Leibler divergence”, Peer-

INCOFT 2025 - International Conference on Futuristic Technology

934

to-Peer Netw. Appl. 15, 1370–1384. 2022.

https://doi.org/10.1007/s12083-022-01301-y.

P. Surendra Varma, V. Anand and P. K. Donta, ”Federated

KNNBased Privacy-Preserving Position

Recommendation for Indoor Consumer Applications,”

in IEEE Transactions on Consumer Electronics, vol.

70, no. 1, pp. 2738-2745, 2024.

doi:10.1109/TCE.2023.3329391.

A. P. Tøttrup, K. Thorup, K. Rainio, R. Yosef, E.

Lehikoinen, and C. Rahbek, "Avian migrants adjust

migration in response to environmental conditions en

route," Biology Letters, vol. 4, no. 6, pp. 685–688,

Dec. 2008. https://doi.org/10.1098/rsbl.2008.0290.

Markapudi Sowmya; Pothuri, Surendra Varma; Katarapu

Deepika. “Emotion recognition from EEG signal data

of the brain using bidirectional long short-term

memory”, International Journal of Advanced

Technology and Engineering Exploration, Vol. 11, Iss.

115, pp. 930-942, 2024.

DOI:10.19101/IJATEE.2023.10102441.

A. S. Pillai, S. U. Ajay Sathvik, N. B. Sai Shibu, and A.

Remanidevi Devidas, "Monitoring urban wetland bird

migrations using IoT and ML techniques," in 2024

IEEE 9th International Conference for Convergence in

Technology (I2CT), Pune, India, 2024, pp. 1–7. doi:

10.1109/I2CT61223.2024.10543277

Varma, P.S., Anand, V. ”Intelligent scanning period

dilation based Wi-Fi fingerprinting for energy efficient

indoor positioning in IoT applications”. J

Supercomput 79, 7736–7761. 2023.

https://doi.org/10.1007/s11227-022-04980-9

Varma, P.S., Anand, V. ”ReMAPP: reverse multilateration

based access point positioning using multivariate

regression for indoor localization in smart buildings”.

Telecommun Syst 83, 303–322. 2023.

https://doi.org/10.1007/s11235-023-01021-5

J. Zhang and Z. Shao, "Research on bird habitat discovery

based on DBSCAN algorithm," in 2018 Eighth

International Conference on Instrumentation &

Measurement, Computer, Communication and Control

(IMCCC), Harbin, China, 2018, pp. 746–750. doi:

10.1109/IMCCC.2018.00161.

S. R. Archer, "Regulation of ecosystem structure and

function: climatic versus non-climatic factors," in

*Meteorology*, J. F. Griffiths Ed., Oxford, U.K.:

Oxford University Press, 1994, pp. 245-255.

K. Balevičius, *Red Data Book of Lithuania*. Vilnius,

Lithuania: Environmental Protection Department of

the Republic of Lithuania, 1992, p. 368.

R. G. B. Brown, "Marine birds and climatic warming in

the northwest Atlantic," in Studies of High-Latitude

Seabirds. 1. Behavioural, Energetic, and

Oceanographic Aspects of Seabird Feeding Ecology,

W. A. Montevecchi and A. J. Gaston, Eds., Occasional

Paper no. 68. Ottawa, Canada: Canadian Wildlife

Service, 1991, pp. 49–54.

G. Rheinwald, "Population counts in the White Stork

Ciconia ciconia," in Bird Numbers 1992. Distribution,

Monitoring and Ecological Aspects, E. J. M.

Hagemeijer and T. J. Verstrael, Eds. Beek-Ubbergen,

Netherlands: SOVON, 1994, pp. 163–168.

L. V. Sokolov and V. A. Payevsky, "Spring temperatures

influence year-to-year variations in the breeding

phenology of passerines on the Courish Spit, eastern

Baltic," Avian Ecology and Behaviour, vol. 1, pp. 22–

36, 1998.

Žalakevičius, M. (1999). Global Climate Change Impact

on Bird Numbers, Population State and Distribution

Areas. Acta Zoologica Lituanica, 9(1), 78–89.

https://doi.org/10.1080/13921657.1999.10512266

M. Zalakevicius and R. Zalakeviciute, "Global climate

change impact on birds: A review of research in

Lithuania," Folia Zoologica, vol. 50, no. 1, pp. 1–17,

2001.

S. K. Skagen, "Migration stopovers and the conservation

of Arctic-breeding Calidridine sandpipers," The Auk,

vol. 123, no. 2, pp. 313–322, Apr. 2006. [Online].

Available: https://doi.org/10.1093/auk/123.2.313.

European Environment Agency, Wild Bird Populations in

England [Data set], European Union Open Data Portal,

2024. Available:

https://data.europa.eu/data/datasets/wild_bird_populati

ons_in_england/embed?locale=en

Avian Species Population Forecaster Using Machine Learning

935