papers/zoo-satellite-ecology/sections/05-conservation-alert-system.tex

\section{Conservation Alert System}

\label{sec:alerts}

\subsection{Alert Generation Pipeline}

The CAS processes change detections through a multi-stage pipeline:

\begin{figure}[H]
\centering
\begin{tikzpicture}[
    node distance=1.2cm,
    box/.style={rectangle, draw, rounded corners, minimum width=2.5cm, minimum height=0.7cm, text centered, font=\footnotesize},
    edge/.style={->, thick}
]
    \node[box, fill=blue!10] (detect) {Change Detection};
    \node[box, fill=green!10, right=1.5cm of detect] (filter) {Noise Filtering};
    \node[box, fill=orange!10, right=1.5cm of filter] (classify) {Impact Assessment};
    \node[box, fill=red!10, below=1cm of detect] (priority) {Priority Ranking};
    \node[box, fill=purple!10, below=1cm of filter] (alert) {Alert Generation};
    \node[box, fill=yellow!10, below=1cm of classify] (delivery) {Stakeholder Delivery};

    \draw[edge] (detect) -- (filter);
    \draw[edge] (filter) -- (classify);
    \draw[edge] (classify) -- (priority);
    \draw[edge] (priority) -- (alert);
    \draw[edge] (alert) -- (delivery);
\end{tikzpicture}
\caption{Conservation Alert System pipeline.}
\label{fig:cas_pipeline}
\end{figure}

\subsection{Priority Ranking Model}

Detected changes are ranked by conservation priority using a composite score:

\begin{equation}
    P(e) = w_1 \cdot S_{\text{eco}}(e) + w_2 \cdot S_{\text{prot}}(e) + w_3 \cdot S_{\text{bio}}(e) + w_4 \cdot S_{\text{mag}}(e) + w_5 \cdot S_{\text{trend}}(e)
\end{equation}

\noindent where:
\begin{itemize}[leftmargin=*]
    \item $S_{\text{eco}}$: Ecosystem sensitivity score based on biome rarity, intactness, and irreplaceability.
    \item $S_{\text{prot}}$: Protected area status (WDPA category, buffer zone proximity).
    \item $S_{\text{bio}}$: Species richness estimate from the Zoo Data Commons and species range maps.
    \item $S_{\text{mag}}$: Change magnitude (area, intensity, rate of expansion).
    \item $S_{\text{trend}}$: Trend indicator from HFI trajectory ($\dot{\text{HFI}}$ and $\ddot{\text{HFI}}$).
\end{itemize}

The weights $\mathbf{w}$ are learned from conservation expert preferences using a pairwise ranking model trained on 2,400 expert-ranked change event pairs.

\subsection{Alert Delivery}

Alerts are delivered through multiple channels depending on priority level:

\begin{table}[H]
\centering
\caption{Alert delivery tiers.}
\label{tab:alert_tiers}
\begin{tabularx}{\textwidth}{@{}llcX@{}}
\toprule
\textbf{Tier} & \textbf{Priority} & \textbf{Latency} & \textbf{Delivery Channel} \\
\midrule
Critical & $P > 0.9$ & < 2 hours & SMS, push notification, email, dashboard \\
High & $0.7 < P \leq 0.9$ & < 24 hours & Email, dashboard, weekly digest \\
Medium & $0.4 < P \leq 0.7$ & < 1 week & Dashboard, monthly report \\
Low & $P \leq 0.4$ & Archive & Searchable archive, annual summary \\
\bottomrule
\end{tabularx}
\end{table}

Each alert includes a map visualization, satellite image comparison (before/after), change classification, HFI impact assessment, and recommended response actions based on the change type and local context.

% ============================================================
% 6. Evaluation
% ============================================================