mirror of
https://github.com/zoo-apps/papers.git
synced 2026-10-30 14:56:54 +00:00
zeekayandHanzo Dev
6658236499
Leadership and team cards, board and founding prose, an AI persona modelled on a named individual, contributor and partner entries, and personal correspondence addresses all named one person. The organisation carries them now. Other people keep their names. Where a card, a board list or a founding line named someone else alongside, they stay exactly as they were; only the one entry was removed, and prose that listed two names now reads with one. Sample and fixture data uses the placeholder name these tests already use elsewhere, so the IAM username defect the notes describe still reads as written. The zoo/yapp MIT notice names the foundation that publishes it. Co-authored-by: Hanzo Dev <[email protected]>
87 lines
4.4 KiB
TeX
87 lines
4.4 KiB
TeX
\section{Introduction}
|
|
|
|
\label{sec:introduction}
|
|
% ══════════════════════════════════════════════════════════════════════════════
|
|
|
|
Climate change is reshaping the geographic ranges of species at
|
|
unprecedented rates \citep{chen2011rapid, pecl2017biodiversity}. As
|
|
temperature and precipitation regimes shift, species must either adapt in
|
|
situ, migrate to newly suitable habitats, or face local extinction
|
|
\citep{urban2015accelerating}. Understanding and predicting these
|
|
distributional changes is essential for effective conservation planning,
|
|
protected area design, and biodiversity policy.
|
|
|
|
Species distribution models (SDMs) have become the primary computational
|
|
tool for projecting range shifts \citep{elith2009species,
|
|
guisan2005predicting}. Traditional approaches, particularly MaxEnt
|
|
\citep{phillips2006maximum} and generalized additive models
|
|
\citep{hastie1990generalized}, establish correlative relationships between
|
|
species occurrence records and environmental covariates. While widely
|
|
adopted, these models suffer from several well-documented limitations:
|
|
|
|
\begin{enumerate}[label=(\roman*)]
|
|
\item \textbf{Linearity assumptions}: Most correlative SDMs struggle to
|
|
capture the complex, non-linear interactions between environmental
|
|
variables that determine habitat suitability
|
|
\citep{elith2006novel}.
|
|
\item \textbf{Spatial independence}: Standard SDMs treat occurrence
|
|
records as independent samples, ignoring spatial autocorrelation and
|
|
dispersal constraints \citep{dormann2007methods}.
|
|
\item \textbf{Static landscapes}: Projections typically assume
|
|
instantaneous equilibrium with climate, ignoring transient dynamics,
|
|
dispersal limitation, and habitat fragmentation
|
|
\citep{zurell2016benchmarking}.
|
|
\item \textbf{Connectivity neglect}: The physical connectivity of
|
|
landscapes---whether corridors exist for species to track shifting
|
|
climate envelopes---is rarely incorporated into distributional
|
|
predictions \citep{hodgson2009climate}.
|
|
\end{enumerate}
|
|
|
|
Recent advances in deep learning offer pathways to address these
|
|
shortcomings. Neural network-based SDMs can model arbitrary non-linear
|
|
relationships \citep{botella2018deep, chen2020deep}, and graph neural
|
|
networks (GNNs) provide a natural framework for spatially structured
|
|
ecological data \citep{jetz2019essential}.
|
|
|
|
\subsection{Contributions}
|
|
|
|
This paper introduces \textbf{HabitatNet}, a graph neural network
|
|
framework for species distribution modeling that addresses the above
|
|
limitations. Our contributions are:
|
|
|
|
\begin{enumerate}
|
|
\item A \textbf{spatial graph representation} of landscapes where
|
|
nodes encode local environmental features at 1\,km resolution and
|
|
edges represent dispersal potential weighted by terrain resistance.
|
|
|
|
\item A \textbf{multi-species GNN architecture} that shares learned
|
|
environmental representations across taxa while learning
|
|
species-specific habitat preferences through attention-weighted
|
|
readout layers.
|
|
|
|
\item Integration with \textbf{CMIP6 climate projections} under
|
|
multiple SSP scenarios (SSP1-2.6, SSP2-4.5, SSP3-7.0, SSP5-8.5)
|
|
for temporal forecasting through 2100.
|
|
|
|
\item A \textbf{corridor identification algorithm} built on learned
|
|
graph embeddings that identifies critical connectivity bottlenecks
|
|
and proposes optimal corridor placements.
|
|
|
|
\item Comprehensive evaluation on \textbf{847 terrestrial vertebrate
|
|
species} across six biogeographic realms, with temporal validation
|
|
against independent resurvey data.
|
|
\end{enumerate}
|
|
|
|
\subsection{Relation to Zoo Labs Foundation}
|
|
|
|
Zoo Labs Foundation operates as an open AI research network dedicated to
|
|
decentralized science (DeSci) and decentralized AI (DeAI). This work
|
|
advances the Foundation's mission by: (a) developing open-source AI
|
|
tools for biodiversity conservation, (b) establishing reproducible
|
|
benchmarks through the Zoo Open Science Network, and (c) enabling
|
|
community-driven conservation planning through transparent, verifiable
|
|
models. All artifacts are governed through Zoo Improvement Proposals
|
|
(ZIPs) at \texttt{zips.zoo.ngo}.
|
|
|
|
% ══════════════════════════════════════════════════════════════════════════════
|