📄 earth_science

A reproducible algorithm and R implementation for classifying ecological flood pulse phases from flood-area time series

This paper introduces the Flood Pulse Phase Discrimination Index (FPDI), a reproducible R-based algorithm that classifies continuous flood-area time series into distinct ecological phases to operationalize flood-pulse theory for small floodplain systems, as demonstrated through a robust analysis of Lake Yardi in Ethiopia.

Original authors: Kibru Zeleke

Published 2026-08-03

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Kibru Zeleke

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the world's rivers not just as flowing water, but as giant, breathing lungs. For decades, scientists have known that these "lungs" expand and contract in a rhythmic dance called the Flood Pulse. When a river swells, it spills onto its floodplain, turning dry land into a temporary, life-filled wetland. This pulse brings food, opens up new swimming lanes for fish, and triggers plants to grow. It's the heartbeat of the ecosystem. But here's the tricky part: nature doesn't keep a calendar. A river doesn't say, "Okay, it's June, time to be a 'wet season'." Instead, the water rises, peaks, and falls in a continuous, messy line.

The problem for scientists is that nature is a smooth, flowing movie, but our data often looks like a stack of frozen snapshots. If you just look at a snapshot of a flooded lake, you might see a lot of water and think, "Great, it's high water!" But is the water arriving (which means new habitats are opening up)? Is it staying (a stable party)? Or is it leaving (the party is over, and things are drying up)? These different moments feel the same in terms of water depth, but they are totally different for the animals living there. Fish might spawn when the water is rising, but hide when it's falling. Until now, there hasn't been a simple, repeatable way to turn that continuous, wiggly line of water data into clear, distinct "phases" that biologists can actually use to understand what's happening.

Enter Kibru Zeleke and a new tool called the Flood Pulse Phase Discrimination Index (FPDI). Think of this paper as the instruction manual for a clever translator that turns a confusing, continuous river story into a clear, five-act play.

The paper doesn't invent the idea that floods are important; that's already a well-established fact. Instead, it solves a methodological puzzle: how do we chop up a continuous flood timeline into meaningful chapters? The author developed a step-by-step recipe (an algorithm) that looks at two things for every single data point: how much water is there and which way is it moving?

Imagine you are watching a tide pool.

Low Water: The pool is small and shrinking or staying small.
Flood Expansion: The water is rising, but it's still relatively low. It's the "getting ready" phase.
Flood Build-up: The water is rising fast and is now high. This is the "main event" starting.
Flood Recession: The water is still high, but it's starting to go down. The party is winding down.
Flood Contraction: The water is low and still dropping. The show is over.

The magic of this paper is that it proves you can't just look at the water level; you must know the direction. A high water level while the water is rising (Flood Build-up) is ecologically different from a high water level while the water is falling (Flood Recession), even if the depth is identical.

To test this, the author applied the recipe to Lake Yardi in Ethiopia. They used satellite images (specifically Sentinel-2) to measure the lake's surface area, creating a timeline of 47 observations between 2017 and 2020. The flood area in this lake ranged from 77.67 km² to 117.28 km².

The results were clear. The algorithm successfully sorted the data into the five distinct phases. When they checked the math, they found that these phases weren't just random labels; they actually separated the water levels significantly. The statistical test showed an F value of 19.71 with a p-value of 3.37e-09, meaning the separation was extremely strong. The phases explained about 65% of the variation in the flood area (an eta-squared of 0.65).

Crucially, the paper tested how "tweakable" the recipe is. They tried changing the rules slightly—making the threshold for "rising" water smaller or larger, or changing the cutoff for what counts as "high" water. They tested 27 different combinations of these settings. The result? The system was robust. Even when they changed the rules, the labels only shifted by a tiny amount (between 0.0% and 4.65% of the monthly labels). This suggests the method is sturdy and won't fall apart if you adjust the settings a little bit for a different lake.

The author is careful to state what this tool is not. It is not a new law of nature, and it doesn't replace the big "Flood Pulse Concept." It's also not a magic crystal ball that guarantees how fish will react; it just defines the opportunity the environment provides. A fish might still ignore the rising water if it's sick or if there are too many predators. The tool simply makes the hydrological story readable.

Finally, the paper offers a reproducible R implementation (a piece of computer code) called fpdi. This means any other scientist can take their own lake data, run this exact same code, and get the same kind of clear, five-phase story. It turns a messy, continuous river into a structured, understandable narrative, allowing researchers to finally ask, "What happened to the fish during the build-up phase?" rather than just, "What happened during the wet season?" It's a small but vital step toward making river science more precise and more useful for protecting the creatures that call these pulsing waters home.

Technical Summary: A Reproducible Algorithm and R Implementation for Classifying Ecological Flood Pulse Phases

Problem Statement
Ecological studies of river-floodplain ecosystems rely on the Flood Pulse Concept, which posits that periodic hydrological expansion and contraction regulate habitat availability, nutrient exchange, and biological processes. However, a methodological gap exists in empirical research: hydrological data are typically continuous trajectories, whereas biological observations are often aggregated into discrete, arbitrary categories (e.g., "wet season," "high water," or specific months). These simple categories fail to capture the functional direction of a flood pulse. A flood of a specific magnitude may have distinct ecological implications depending on whether the system is expanding, stable, or receding. There is a lack of a reproducible, transparent method to convert continuous hydrological time series into specific "Ecological Flood Pulse Phases" (EFPP) that account for both magnitude and trajectory.

Methodology
The paper introduces the Flood Pulse Phase Discrimination Index (FPDI), a workflow designed to translate continuous flood-pulse magnitude time series into five distinct EFPP categories:

Low Water
Flood Expansion
Flood Build-up
Flood Recession
Flood Contraction

The algorithm operates through the following steps:

Input Standardization: The input is a chronologically ordered time series of flood-pulse magnitude (e.g., flood area, water level). The system performs an audit for missing values, duplicates, and date ranges.
Hydrographic Direction Calculation: The change between consecutive observations ( $\Delta_t = X_t - X_{t-1}$ ) determines the direction. A user-settable threshold (default 0.50 km²) classifies observations as Rising ( $\Delta > 0.50$ ), Receding ( $\Delta < -0.50$ ), or Stable. The first observation is classified as Initial.
Annual Normalization: To account for interannual variability in absolute flood magnitude, flood values are normalized within each calendar year: $FloodNorm_t = (X_t - min(X_y)) / (max(X_y) - min(X_y))$ .
Phase Assignment: EFPP labels are assigned by combining the hydrographic direction with the normalized flood magnitude using user-defined thresholds (default low-water threshold = 0.35; high-flood threshold = 0.60).
- Low Water: Stable and below the low-water threshold.
- Flood Expansion: Rising and below the high-flood threshold.
- Flood Build-up: Rising and at or above the high-flood threshold.
- Flood Recession: Receding and at or above the high-flood threshold.
- Flood Contraction: Receding and below the high-flood threshold.
Monthly Consolidation: For compatibility with biological sampling, the observation-level series is consolidated to a single label per calendar month, retaining the latest observation within that month.

Implementation and Validation
The workflow is implemented as a lightweight, pure R package named fpdi. It does not require compiled code and includes functions for data auditing, classification, and hydrological validation.

Case Study: The method was demonstrated using a flood-area time series (2017–2020) for Lake Yardi, Ethiopia. Flood areas were derived from Sentinel-2 imagery using the Modified Normalized Difference Water Index (MNDWI).
Hydrological Validation: One-way ANOVA confirmed that the assigned phases significantly separated the hydrological data ( $F = 19.71, p = 3.37 \times 10^{-9}$ ), with the phases explaining approximately 65% of the flood-area variation (eta-squared = 0.65).
Sensitivity Analysis: The system was tested across a grid of direction, low-water, and high-flood thresholds. Results indicated that the phase structure was robust; varying thresholds within tested ranges resulted in minimal label changes (0.0% to 4.65% for monthly labels), and the hydrological distinctness of the phases remained substantial.

Key Contributions

Operational Bridge: FPDI provides a reproducible, transparent method to link continuous hydrological dynamics with discrete ecological analytical units without proposing a new ecological theory.
Directional Sensitivity: Unlike static categories, FPPI distinguishes between states of similar magnitude but opposite trajectories (e.g., Flood Build-up vs. Flood Recession), which is critical for understanding ecological processes like spawning or migration.
Reproducibility: The R package ensures that the classification logic is auditable, allowing researchers to inspect thresholds, consolidation rules, and input data transformations.
Modularity: The package separates the core classification logic from upstream remote sensing processing, making it accessible to ecologists who already possess hydrological time series while allowing for future integration of satellite-derived data tools.

Significance and Claims
The author explicitly states that FPDI is not an alternative to the Flood Pulse Concept but rather a "nested operational method" for applying the theory to specific systems, particularly small connected floodplain lakes and oxbow systems.

The method is designed to make local flood dynamics "auditable, reproducible, and analytically usable" in ecological and fisheries research.
It addresses the limitation of calendar-based or simple high/low water categories by defining phases based on the actual functional state of the system.
The author acknowledges that while FPDI identifies hydrological "phase opportunity," it does not guarantee specific biological responses, which depend on species-specific life histories and other environmental factors.
The paper concludes that the Lake Yardi demonstration successfully produced interpretable, hydrologically distinct phases and a robust software implementation, validating the workflow's utility for transferring flood-pulse theory into practical data analysis.

Technical Summary: A Reproducible Algorithm and R Implementation for Classifying Ecological Flood Pulse Phases

More like this