Australian bird prediction project

Bird predictions for Australia, built from data and ecology.

LikelyBird is an Australian bird prediction project in development. It uses bird observations, environmental data and ecological modelling to estimate which species are likely in different places and seasons.

In development — not yet a public prediction service

Purpose

What LikelyBird is doing

Bird observations are not evenly collected. People bird more often near roads, towns, reserves, tracks and familiar places. Weather, time of day, observer behaviour and checklist effort all affect what gets reported.

LikelyBird is designed to account for those issues rather than simply mapping where birds have already been seen.

The modelling target is checklist-level encounter probability: the probability that a species would be reported on a suitable checklist under given conditions.

Methods

A modelling pipeline, not just a sighting map.

LikelyBird treats bird observations as structured survey evidence. Complete checklists can provide detections and inferred non-detections, while incomplete checklists provide positive evidence only. The modelling pipeline keeps survey quality, detectability, environment and range plausibility separate and auditable.

Checklist-level evidence

The modelling target is encounter probability at the checklist level, not a simple map of individual sightings.

Weighted non-detections

Complete checklist non-detections are useful evidence, but they need to be weighted by survey quality, detectability and range plausibility.

Public-safe outputs

Public products will primarily expose derived predictions and summaries. Observation detail or checklist-level locations would only be included where appropriate approvals and safeguards allow.

SIL

Survey Information Layer

The umbrella checklist-level layer used for audit, modelling weights and downstream joins.

CQL

Checklist Quality Layer

Species-agnostic observer, effort and checklist quality information.

WDL

Weather Detectability Layer

Weather conditions that affect bird detection, such as wind, rain, temperature and recent rainfall.

TDL

Temporal Detectability Layer

Time-of-day, solar timing, seasonality and other temporal observation effects.

ECL

Environmental Covariate Layer

Completeness and reliability checks for joined environmental and access covariates.

RPL

Range Prior Layer

Species-specific ecological plausibility, used as a soft range prior rather than a hard exclusion.

Data safety

Public outputs are derived, not raw records

LikelyBird is designed to respect eBird data governance and sensitive-species concerns. Public products will default to derived outputs; any observation records, linked records or precise locations would require appropriate approvals and safeguards.

Public outputs are intended to show derived products only: prediction maps, uncertainty summaries, model validation results, explanatory reports and public-safe species summaries.

Sensitive species will be handled cautiously. Public predictions may be suppressed, coarsened, delayed or downgraded where needed.

What may be public

Derived predictions, model summaries, uncertainty classes, validation results, public-safe maps and explanatory methodology notes.

What requires approval

Observation records, checklist-level coordinates, observer-level detail and downloadable source data require appropriate approvals and data safeguards before publication.

Sensitive species

Some species may require suppression, coarsening, delay or descriptive-only treatment before any public release.

Progress

Current status

LikelyBird is in active development.

  • Building a national reference grid for Australia.
  • Joining environmental and access covariates.
  • Constructing checklist-level modelling layers.
  • Developing species-level prediction models.
  • Testing validation methods.
  • Designing safe public outputs.

Contact

Project enquiries

LikelyBird is in development. For project enquiries, collaboration or feedback, contact:

Coming soon

Future versions of this site may include project updates, methodology notes, public-safe species summaries, validation reports, static prediction examples and eventually an interactive bird prediction tool.

Project notebook

Background pages will grow as the work develops.

The prediction service is still in development, but the supporting data work can be explained as it is audited, decoded and recoded into modelling layers.

Data sources

Data & layer overview

A plain-language guide to the main source datasets, what they contribute and how they become public-safe modelling evidence.

Read about eBird checklist evidence

Architecture

Modelling layers

SIL, CQL, WDL, TDL, ECL, RPL and the species model explained as separate, auditable components.

Explore the layer map

Next notes

Dataset explainers

Future pages can cover eBird checklists, weather, hydrology, vegetation, fire history and land cover as each layer is documented.

For more information

Project updates.

Follow occasional LikelyBird progress notes, release updates and public-facing project material.

Email updates

Infrequent project updates — roughly fortnightly while active.