Predator-Prey Model Challenges and Robust Consumption Formulations
The predator-prey model, specifically a deer-wolf model with vegetation, is a classic in system dynamics. This discussion focused on building a simple yet robust model and analyzing its behavior, with a particular emphasis on the challenges encountered during the modeling process and specific formulations for different components.
Common Modeling Challenges
Participants highlighted several difficulties in conceptualizing and operationalizing the model:
- Predation and Consumption: Accurately modeling predation rates was challenging. Rates needed to be lower when prey density was low and higher when prey was abundant, while also accounting for maximum and minimum consumption levels for predators. Similar difficulties arose in modeling vegetation consumption by prey.
- Interference from Classical Models: Preconceived notions from classical mathematical models often hindered the development of more realistic conceptualizations.
- Operationalization of Extreme Conditions: Ensuring the model behaved realistically under extreme conditions, such as the absence of prey leading to no predator growth, was a common hurdle.
- Agent-Based vs. Aggregate Thinking: Shifting from an agent-based model perspective (individual deer and wolves) to an aggregate, system dynamics approach proved difficult for some, as microscopic interactions often complicated aggregate rate calculations.
- Time Step Sensitivity: The importance of time step size was noted. Significant changes in model behavior with different time steps indicate the time step is too large. The heuristic is to halve the time step until behavior stabilizes, reflecting the continuous nature of the real world.
- Single vs. Separate Flows: A recurring question was whether to use a single net flow for population changes or separate birth and death rates, and further, whether to distinguish between natural death and death by predation.
Vegetation Growth Formulations
The discussion then moved to specific model formulations, starting with vegetation growth:
- Fractional Regeneration Rate: Many models used a simple fractional regeneration rate (stock multiplied by a growth rate).
- Carrying Capacity Integration: A more robust approach incorporated a maximum vegetation level using the formulation
(Max Vegetation - Vegetation) / Time Step. This prevents indefinite exponential growth and accounts for competition. However, a downside is that if vegetation is extremely low, the regeneration rate also collapses, which may not reflect all ecosystems. - Two-Flow Model for Vegetation: An alternative is a two-flow model: an exponential regeneration rate (inflow) and an outflow that increases significantly with density, representing natural death or competition. This separates the physics of growth from density-dependent mortality.
- Logistic Function for Inflow: Using a logistic function in the inflow can capture both the reinforcing loop at low levels and saturation at higher levels, preventing indefinite growth.
Prey Population Dynamics: The Challenge of Consumption
Modeling the consumption of prey by predators (and vegetation by prey) proved to be one of the most complex aspects:
- Combined Flows and Complexity: Combining many factors into a single flow for birth or death rates makes the model logic difficult to follow and debug. Breaking down the logic into separate pathways is generally recommended.
- Unit Correction Issues: Using "unit correction" parameters to force units to match often indicates a fundamental flaw in the model's logic. Mechanisms should naturally align units.
- Infinite Consumption Problem: A common issue in early formulations was the "infinite consumption" problem: if prey (or vegetation) is abundant, a single predator (or prey) could consume an infinite amount, violating physical constraints.
Robust Consumption Formulations
Several approaches were discussed to address the infinite consumption problem and ensure realistic behavior:
- Minimum Function (Gerardo's Model): An effective starting point uses a
MINfunction:MIN(Desired Prey Consumption, Prey Available / Minimum Time to Eat Prey).- Pros: Prevents infinite consumption and ensures consumption doesn't exceed available prey.
- Cons: Creates a sharp, discontinuous behavior (a "kink") where the model abruptly switches between consuming the desired amount and being limited by availability. This can make analysis and optimization more challenging.
MAX(0, ...): The use ofMAX(0, ...)to prevent negative flows was discussed. While it ensures non-negativity, it can hide underlying errors and add unnecessary complexity if the equation is inherently robust against negative values.
- Softmax Functions (Analytical Alternatives): To address the discontinuity of the
MINfunction, "softmax" analytical functions were proposed:x / (1 + x): This function starts at 0, approaches 1 asxgoes to infinity, and never exceeds they = xline (meaning consumption never exceeds availability). However, it approaches saturation slowly and lacks flexibility.x / (1 + x)^alphaor(x / (1 + x))^alpha: Adding an exponentalphacan increase flexibility. Ifalpha < 1, it can lead to a sharper increase towards saturation. Care must be taken to ensure the function doesn't exceed they = xline.1 - e^(-alpha * x): This exponential approach also starts at 0 and approaches 1, offering a different curve shape.- Combining with
MIN: Often, aMINfunction is still needed to ensure the consumption rate never exceeds they = xline, even with softmax functions, which can reintroduce a kink.
- Table Functions: System dynamicists often use table functions to define these non-linear relationships, allowing for precise shaping of the curve without complex analytical equations. However, they can be less transparent and require manual adjustment.
- Jason's Formulation (
x^2 / (1 + alpha * x^2)): This formulation, wherexrepresents deer population, also addresses the infinite consumption problem.- Pros: Starts at 0, saturates at
1/alphafor largex, and incorporates a non-linear dependence on density (x^2) to reflect that hunting success might increase disproportionately with density. - Cons: The micro-mechanisms behind the
x^2term are less intuitive, and it's not immediately clear if it violates they = xconstraint without detailed analysis. It also lacks a parameter to easily adjust the speed of convergence.
- Pros: Starts at 0, saturates at
Starvation and Death Rates
The discussion also covered how starvation impacts death rates:
- Separating Death Causes: Separating natural death (old age) from starvation-induced death is a valid approach, allowing for distinct mechanisms.
- Satiation as a Driver: Starvation can be linked to "satiation" – the ratio of actual consumption to desired consumption. A lower satiation level indicates higher starvation risk.
- Non-Linear Impact of Starvation: A direct linear relationship between lack of satiation and death rate might be too strong. Biologically, animals can endure periods of reduced food intake without immediate death. A non-linear relationship (e.g.,
(1 - satiation)^alphawherealpha < 1) can better reflect this. - Impact on Birth Rates: Starvation can also reduce birth rates, adding another feedback loop.
- Hunter Lifespan Adjustment: Instead of a direct death rate, starvation can reduce the "hunter lifespan," effectively increasing the death rate over time.
Hunting Efficiency and Desired Consumption
The concepts of "hunting efficiency" and "desired consumption" were explored:
- Conceptual Overlap: These two concepts are often closely related. "Desired consumption" represents how much a predator wants to eat, while "hunting efficiency" (or actual consumption) reflects how much it can eat given prey availability and its skill.
- Single Parameter Approach: It's often more parsimonious to use a single parameter for "desired prey per predator" and then use a non-linear function (like the softmax functions discussed) to model how actual consumption deviates from this desired level based on prey availability and hunting effectiveness.
- Non-Linear Search Process: The effectiveness of hunting can scale non-linearly with prey density. At very low densities, prey might be harder to find than a simple linear relationship would suggest, leading to a sharper drop in consumption.
The discussion highlighted the importance of balancing mathematical robustness with conceptual clarity and biological realism in building effective system dynamics models.
Predator-Prey Model Analysis
The discussion then focused on refining a predator-prey model, specifically addressing issues related to consumption rates and death rates.
Consumption Rate and Reference Density
A key point of contention was the "reference density of prey." Initially, it was considered the maximum carrying capacity. However, a student clarified that their anchoring was based on the prey death rate, implying a normal or desired consumption rate that then anchored the reference density.
The instructor, Hazhir Rahmandad, highlighted a potential issue: if the reference density is not the maximum for the prey population, the ratio of actual prey density (y) to reference prey density (reference y) could exceed 1. When this ratio is raised to a sensitivity power, it could lead to consumption per predator going above the normal consumption per predator, which is undesirable. While some argued that higher density might lead to higher consumption, it should still saturate and not go to infinity, as the current formulation allowed. This was identified as a primary issue, suggesting that the effect should saturate at 1, multiplied by normal consumption to define maximum consumption.
Food Scarcity Effect on Predator Death Rate
The model includes a "food scarcity effect" on the predator death rate, formulated as e to the power of some parameter multiplied by food scarcity. Food scarcity is defined as MAX(0, (normal consumption - actual consumption) / normal consumption).
Key interpretations of food scarcity:
- Food scarcity of 1: Means famine, where predators get nothing they want to eat.
- Robustness: The
MAX(0, ...)ensures that even if actual consumption exceeds desired consumption, food scarcity doesn't go below 0.
Fractional Death Rate Unit Problem
A significant unit and conceptual problem was identified in the formulation of the predator death rate. The equation involved 1 - fractional death rate of r with ample food.
- Unit Issue: Fractional death rate can exceed 1 (e.g., 50 if the unit is 1/year and death occurs in two weeks). It's not a probability but a rate per unit of time.
- Conceptual Issue: The
1in the equation should represent the maximum fractional death rate, not just a generic1. This is crucial to prevent the death rate from going beyond a realistic maximum. - Clunky Formulation: The use of an exponential leading to a ceiling imposed by another term creates a "clunky" and unrealistic non-linearity. Despite this, it provides meaningful feedback from deer consumption to predator death.
Model Analysis and Oscillations
The discussion then moved to analyzing a student's model, which included prey, vegetation, and other factors. The model was initialized at an analytical equilibrium, and a deforestation shock was introduced.
The model exhibited an oscillatory mode, even with small pulses. To understand these oscillations, a causal tracing process was initiated, focusing on a shorter time horizon (e.g., 9 to 15 time units) to clearly observe the dynamics.
Causal Tracing Steps:
- Start with Vegetation: Vegetation is influenced by restoration and destruction. Restoration, depending only on vegetation, forms a reinforcing loop and is unlikely to cause oscillations.
- Focus on Destruction Rate: The initial jump in destruction rate triggers the decline in vegetation.
- Identify Main Driver: The destruction rate is partly driven by vegetation, but the primary driver of oscillation is likely the prey population.
- Analyze Prey Population: Prey population dynamics depend on birth and death rates. When the prey population goes down, it's because the birth rate is going down, not because the death rate is going up.
- Trace Prey Birth Rate: Prey birth rate is influenced by vegetation and prey population. By observing the timing of minimums (valleys) in the graphs, it was determined that the valley in prey birth rate occurs after the minimization of vegetation but before the minimization of the prey population. This suggests vegetation is driving the prey birth rate.
- Hypothesis Formation: This leads to a candidate hypothesis for the oscillatory feedback loop: Less vegetation leads to less prey population, which leads to less vegetation consumption, which leads to more vegetation. This is a higher-order balancing feedback loop (involving two stocks: vegetation and prey population), capable of generating oscillations.
Testing the Hypothesis:
To test this hypothesis, the prey birth rate was "disrupted" by overriding its default behavior and setting it to a constant value (its equilibrium value of 15) after a certain time.
- Intervention: An
if/then/elsestatement was used to switch the prey birth rate from its calculated value to a constant 15 at a specific time (e.g., time 14). - Result: The oscillations started as expected, but as soon as the switch was activated, the oscillations ceased. This strongly supported the hypothesis that the identified feedback loop was responsible for the oscillatory behavior.
Best Practices for Intervention:
- It's beneficial to see oscillations start and then be disrupted instantaneously by the intervention, providing a strong signal.
- Interventions should be as minimally disruptive as possible to isolate the effect of the targeted feedback loop. Setting a parameter to zero might disrupt too many other dynamics.
Next Week's Assignment: Drug Development Pipeline
The next assignment involves modeling a drug development pipeline, focusing on deterministic behavior despite its real-world stochasticity.
Model Stages and Failure Rates
The pipeline consists of multiple stages:
- Compound discovery
- Animal trials/Laboratory tests
- Phase I trials
- Phase II trials
- Phase III trials
- Market launch
Each stage has a failure rate, which can be quite high (e.g., 90% or more across the three trial phases).
Assignment Goal
The primary goal is to formulate the start rate for Phase I (assuming earlier stages are easily managed) to achieve a desired revenue trajectory in the long term.
- Revenue Projection: The desired revenue should follow an exponential curve, such as growing at a 5% rate per year. The model should be robust to different growth rates (e.g., 0% for steady state, or even declining rates).
- Stock Management: This is essentially a stock management problem. The company needs to manage the "work in progress" stocks (Phase I, II, III trials) to meet future revenue goals.
- Robustness: The model should quickly move towards the desired trend without overshooting or oscillating when exposed to a step increase in desired revenue.
Key Considerations and Simplifications
- Stochasticity: Ignore stochasticity for this assignment.
- Resource Constraints: Do not include resource constraints on project starts unless specifically aiming for an advanced model.
- Revenue Generation: Drugs that make it to the market generate revenue per year.
- Patent Expiration: Assume revenue from a drug stops once its patent expires.
- Market Dynamics: Do not worry about market dynamics like the availability of other drugs impacting acceptance rates.
- Failure Rates: Assume constant failure rates for each stage.
- Delays: Use first-order delays for the time taken in clinical trials, even if they are not strictly first-order in reality.
- Drug Life Cycle: Drugs on the market also have a life cycle (e.g., patent expiration).
The assignment builds upon previous stock management concepts but incorporates failure rates and growing demand. It requires adjusting desired levels for each "work in progress" stock (Phase I, II, III).
Takeaways
- Modeling predation rates requires density-dependent functions that lower consumption at low prey densities and cap consumption at realistic maximums, avoiding infinite consumption.
- Vegetation growth can be represented with carrying-capacity-based formulations or two-flow models that separate exponential regeneration from density-dependent mortality.
- Using MIN or softmax functions for predator consumption prevents unrealistic flows, but MIN introduces discontinuities while softmax offers smoother saturation.
- Starvation effects on predator death and birth rates are best modeled with non-linear satiation functions, separating natural death from food-scarcity-induced mortality.
- Oscillations in the deer-wolf-vegetation system arise from a higher-order feedback loop where reduced vegetation lowers prey birth, which then allows vegetation to recover, and disrupting the prey birth rate eliminates the cycles.
Frequently Asked Questions
Why can a reference prey density ratio greater than 1 lead to consumption per predator exceeding normal consumption?
Because the model multiplies the normalized prey density (actual/reference) by a sensitivity exponent, a ratio above 1 raises the term above the baseline, allowing the calculated consumption per predator to surpass the intended normal consumption value. This violates the intended saturation behavior, so the formulation should cap the effect at 1 before applying the normal consumption rate.
How does a softmax function like x/(1+x) improve consumption modeling compared to a hard MIN function?
A softmax function such as x/(1+x) provides a smooth, continuously increasing consumption curve that asymptotically approaches the availability limit without a sudden kink. Unlike the hard MIN, it maintains differentiability for analysis and optimization while still ensuring consumption never exceeds the prey stock, though additional parameters may be needed to adjust the saturation speed.
Who is MIT OpenCourseWare on YouTube?
MIT OpenCourseWare is a YouTube channel that publishes videos on a range of topics. Browse more summaries from this channel below.
Does this page include the full transcript of the video?
Yes, the full transcript for this video is available on this page. Click 'Show transcript' in the sidebar to read it.
was whether to use
single net flow for population changes or separate birth and death rates, and further, whether to distinguish between natural death and death by predation.
Helpful resources related to this video
If you want to practice or explore the concepts discussed in the video, these commonly used tools may help.
Links may be affiliate links. We only include resources that are genuinely relevant to the topic.