Methodology

Transparent by Default

PelagicLabs publishes model assumptions, validation boundaries, and uncertainty handling so users can evaluate the system on merit.

Validation Boundaries

Regional and model scope

Uncertainty Handling

Confidence and freshness logic

Biology Models with Evidence Trails

Species behavior assumptions are tied to documented sources and regional validation work, not anecdotal lore.

Species	Status	Accuracy	Data Points	Temp Range	Depth
European Seabass Dicentrarchus labrax	Validated	84%	42,000+	12-18°C	0-100m
Atlantic Cod Gadus morhua	Validated	79%	38,000+	4-10°C	0-600m
European Pollack Pollachius pollachius	Validated	76%	28,000+	8-14°C	0-200m
Northern Pike Esox lucius	Beta	72%	15,000+	10-20°C	0-30m
Atlantic Mackerel Scomber scombrus	Validated	81%	35,000+	8-14°C	0-200m
Brown Trout Salmo trutta	Beta	68%	12,000+	6-18°C	0-20m

Validation Methodology: Accuracy is tracked against observed outcomes within predicted activity windows, with thresholds that tighten as sample size and regional coverage mature.

Method Note

Nearshore Physics, Operationalized

Wave guidance incorporates propagation and transformation behavior to better represent break-level reality.

Data Sources

NOAA WaveWatch III: Global spectral wave model, 0.5° resolution
Copernicus Marine: European regional wave forecasts
NDBC Buoy Network: Real-time wave height, period, direction
Sentinel-1 SAR: Ocean surface wind fields

Physics Model

We use spectral wave energy balance equations to track swell evolution across the ocean basin:

∂E/∂t + ∇·(C_gE) = S_in - S_ds + S_nl

Where E = spectral energy density, C_g = group velocity, and S terms represent wind input, dissipation, and nonlinear interactions.

Local Calibration

Each surf spot requires local calibration factors:

Bathymetry: Shoaling and refraction coefficients
Exposure: Swell window angles (which swells reach the spot)
Historical fit: Adjustment from observed vs predicted

Swell Journey: Basin to Break

Computation

What We Compute and Why

Formula-level visibility exists to make outputs interpretable, contestable, and continuously improvable.

Bite Probability Score

P(bite) = Σ(wᵢ × fᵢ) × C_data × C_time

Weighted sum of environmental factors (wᵢ = weight, fᵢ = normalized factor score) multiplied by data quality and temporal confidence coefficients.

Variables

fᵢ = {SST, tide_phase, solunar, wind, chlorophyll}
C_data = min(1, data_freshness / 24h)
C_time = exp(-λ × hours_until_prediction)

Solunar Major Period

T_major = T_transit ± (duration / 2)

Major feeding periods centered on lunar transit (overhead) and opposition (underfoot), with duration scaled by lunar phase intensity.

Variables

T_transit = local time of moon overhead
duration = 2h × (1 + 0.3 × |cos(phase)|)

Wave Group Velocity

Cg = gT / (4π) ≈ 1.56T (m/s)

Deep water group velocity determines swell arrival time. Waves travel in groups at half the speed of individual wave crests.

Variables

g = 9.81 m/s² (gravitational acceleration)
T = wave period in seconds

Confidence Decay Function

C(t) = C₀ × exp(-t/τ)

Confidence in predictions decays exponentially with time. τ varies by data source and environmental volatility.

Variables

C₀ = initial confidence (based on data quality)
τ = decay constant (6-24h depending on source)
t = time since last data update

Thermal Preference Index

TPI = exp(-(T - T_opt)² / (2σ²))

Gaussian distribution modeling species thermal preference. Score decreases as water temperature deviates from species optimum.

Variables

T = current sea surface temperature
T_opt = species optimal temperature
σ = tolerance width (species-specific)

These equations are simplified representations; production models include additional terms for edge cases, seasonal effects, and regional corrections.