---
title: 'Biscay: Marine Dynamics and Mobile Networks'
url: https://www.emergentmind.com/topics/biscay
type: topic
---

# Biscay: Marine Dynamics and Mobile Networks

Biscay denotes, in the literature considered here, a North Atlantic marine region centered on the Bay of Biscay and its southern sector, the Cantabrian Sea, and it also appears as the name of a distinct transport-systems acronym, **BISCAY**, in mobile networking. As a geographic and oceanographic domain, Biscay is used to analyze coastal upwelling, larval connectivity, internal-wave finestructure, and offshore wave climate; as a technical acronym, BISCAY refers to a radio-KPI-driven congestion control system for 4G/5G networks [2401.09513] [2203.00475] [2205.13325] [2509.02806].

## 1. Geographic and oceanographic setting

The Bay of Biscay is treated in these studies as a multi-scale physical domain. Its southern sector is the Cantabrian Sea, and one study focuses on the Asturian coast in central Cantabria, with implications for adjacent Basque Country and Galicia. Key nearshore sites include Cabo Peñas and Cudillero-Oviñana, with an ADCP mooring at \(43^\circ 34.18'N, 6^\circ 8.43'W\), while genetic sampling spans Corme, Punta de la Cruz, Ribadesella, Monpas, and Jaizkibel [2401.09513].

Several circulation features organize this setting. Summer northeasterly winds drive the **NW Spanish Upwelling system**, producing westward alongshore surface flow and upwelling, with compensating onshore return flow at depth over a narrow shelf. This regime dominates April–September, when *Pollicipes pollicipes* spawns. By contrast, the **Iberian Poleward Current** is strongest in late autumn–winter, flows east or northeast along the shelf break, is typically offshore of the nearshore strip where barnacle larvae are released and settle, and is weak during peak spawning. Shelf-break fronts and anticyclonic slope water eddies occur in the southern Bay of Biscay, but are described as less relevant to nearshore larval recruitment during summer upwelling [2401.09513].

The Biscay basin is also represented by deep-ocean moorings in the NE Atlantic Bay of Biscay. Yearlong current observations were obtained at a continental slope site at \(46^\circ 39' \, \mathrm{N}, 05^\circ 29' \, \mathrm{W}\), \(H = 2450\) m, and an abyssal plain site at \(45^\circ 48' \, \mathrm{N}, 06^\circ 50' \, \mathrm{W}\), \(H = 4810\) m. Stratification from nearby CTD profiles is summarized by

$$
N(z) = (1 \pm 0.5)(20 + 0.0034 z)\ \mathrm{cpd}
$$

for \(-4480 < z < -2740\) m, with an abrupt change above \(z = -2740\) m and \(N \approx 28\) cpd at \(z \approx -1500\) m [2203.00475].

A separate offshore engineering study places a wave-prediction target point at \((45.2^\circ \mathrm{N}, 1.6^\circ \mathrm{W})\). There, the relevant forcing domain is not only local: the Bay of Biscay sea state is described as swell-dominated, and swells generated by North Atlantic storms can take on the order of days, approximately \(3\)–\(5\) d, to reach Biscay [2205.13325].

## 2. Coastal recruitment, larval connectivity, and barnacle productivity

On the Cantabrian coast, Biscay is a managed ecological system for the gooseneck barnacle (*Pollicipes pollicipes*). The fishery is co-managed through a fine-scale, interspersed set of protected rocks where each rock receives a distinct level of protection. The underlying management question is whether this spacing is consistent with mean larval dispersal distances [2401.09513].

The dispersal analysis contrasts two larval seasons: 2009, a year of high upwelling activity, and 2011, a year of low or interrupted upwelling. Ocean data came from a Nortek Aquadopp ADCP moored 400 m from shore in 20 m water depth, with 10 depth cells of 2 m thickness from 2–20 m, extrapolated to 0–2 m, and velocities measured every 30 minutes from July to November. A decorrelation time scale of 12 hours was used to estimate stochastic spreading [2401.09513].

Larvae were represented with an advection–diffusion kernel in two horizontal dimensions. Settlement probability from release point \(i\) to settlement point \(j\) was parameterized as

$$
D_{ij} = \frac{1}{L_s\sqrt{2\pi}} \times \exp\!\left(-\frac{(d_{ij}-L_A)^2}{2L_s^2}\right),
$$

computed independently for alongshore and cross-shore directions and multiplied to form the 2D kernel. The coast was treated as a “sticky boundary”: inland probabilities with \(Y < 0\) were removed and added to the shoreline bin to conserve total probability, reflecting enhanced settlement and reduced flow at the shore. Pelagic larval duration was fixed at 60 days, with ontogenetic vertical migration such that nauplii occupied the 0–10 m layer during days 0–30 and cyprids the 10–20 m layer during days 30–60. Releases were uniform from July 1 to October 2, with 4,465 release events per year [2401.09513].

The resulting dispersal contrasts are marked. In 2009, net advective displacement was westward alongshore, \(L_{Ax} = -56.12 \pm 21.8\) km, and landward cross-shore, \(L_{Ay} = -67.79 \pm 47.3\) km. In 2011, net advective displacement became slightly eastward, \(L_{Ax} = +12.95 \pm 9.8\) km, and seaward, \(L_{Ay} = +31.61 \pm 8.4\) km. Stochastic spreading was also higher in 2009 than in 2011 [2401.09513].

Recruitment outcomes follow the circulation change. In 2009, the dispersal kernel was tightly constrained near the shoreline, peak recruitment was predicted 56 km west of the emission point, theoretical recruitment success was approximately 94%, and maximum coastal probability density reached approximately \(173.8 \times 10^{-3}\ \mathrm{km}^{-2}\). In 2011, the kernel was displaced offshore, peak recruitment was predicted 13 km east of the emission point, theoretical recruitment success dropped to approximately 15.4%, and maximum coastal \(D_{ij}\) was approximately \(49.9 \times 10^{-5}\ \mathrm{km}^{-2}\). The mechanistic explanation given is that, during strong upwelling in 2009, an onshore return flow spanned most of the water column and favored nearshore retention, whereas in 2011 onshore flow was limited to the bottom layer, enabling offshore export of nauplii at the surface [2401.09513].

Fishery productivity was linked to seasonal upwelling through the Integrated Upwelling Index. The best CPUE model used IUI with a 4-year lag and a TAC dummy, explaining 93.1% of variance, with adjusted \(R^2 = 0.9308\), \(AIC = -15.244\), and Akaike weight \(= 0.9961\). After Bonferroni correction, TAC had \(p < 0.00001\) and IUI had \(p = 0.005\). The fitted relationships were

$$
\mathrm{CPUE} = 0.24 \times \mathrm{IUI} + 6.125 \quad (\mathrm{TAC}=6\ \mathrm{kg})
$$

and

$$
\mathrm{CPUE} = 0.24 \times \mathrm{IUI} + 6.918 \quad (\mathrm{TAC}=8\ \mathrm{kg}),
$$

implying that CPUE increases by \(0.24\ \mathrm{kg\ day^{-1}}\) per unit of seasonal Ekman transport [2401.09513].

Genetic connectivity reinforces the physical picture. A mitochondrial COI dataset of 243 sequences across five populations was analyzed with MIGRATE-N. Migration rates were well above one migrant per generation, the full bidirectional nearest-neighbor model fit better than either constrained eastward-only or westward-only alternatives, and parameter estimates were consistent with a net westward bias in gene flow. This aligns with summer upwelling-driven westward flow rather than winter IPC-driven eastward transport [2401.09513].

A recurrent misconception addressed by this study is that long-distance mesoscale dispersal or slope-current transport should dominate recruitment. The reported result is the opposite: small-scale, asymmetric connectivity, typically within 10–60 km and biased westward in high-upwelling summers, is more consistent with the scale of co-management in this fishery [2401.09513].

## 3. Internal-wave spectra and deep-basin nonlinear dynamics

In deep Biscay, the internal-wave band is described as spectrally structured rather than smooth. Tidal harmonic analysis isolates a time-coherent narrowband “barotropic” semidiurnal signal from the remainder, termed “baroclinic,” and at \(M_2\) the barotropic current amplitude is about twice the baroclinic amplitude. Kinetic-energy spectra show dominant peaks at the inertial frequency \(f\), semidiurnal \(M_2\), higher harmonics \(M_4, M_6, \ldots\), and interaction bands \(M_2+f, M_4+f, \ldots\). Higher harmonics are detected up to \(M_{10}\) at the deep site and up to \(M_{16}\) at the shallower site [2203.00475].

The continuum between peaks slopes roughly as \(PKE \sim \omega^{-1}\) for \(f < \omega < 7\)–\(10\) cpd and steepens at higher \(\omega\). Heavily smoothed spectra show \(\omega^{-2}\) at the shallower site and \(\omega^{-3}\) at the deeper site, both steeper than the canonical Garrett–Munk range \(-2.5 < p < -1.5\). When inertial energy is strong, interaction bands and higher tidal harmonics show \(PKE \sim \omega^{-3}\); when inertial energy is weaker, \(M_n+f\) bands scale like \(\omega^{-2}\) [2203.00475].

To interpret this structure, a simple model is proposed for forced nondispersive motions with forward energy cascade among baroclinic harmonics. The current components are written as \(\hat{u}_j = \hat{U}_j \exp[i(k_j \xi - \omega_j t)]\), with discrete kinetic-energy spectrum

$$
P_j(\omega) = \frac{1}{2\omega}\hat{U}_j^\ast \hat{U}_j,
$$

and the nonlinear interaction is represented by

$$
\partial_t \hat{u}_j = -\sum_{i=1}^{j-1}\hat{u}_{j-i}\,\partial_\xi \hat{u}_i, \qquad j=2,3,\ldots
$$

leading to non-resonant higher harmonics

$$
\omega_j = j\omega, \qquad k_j = jk, \qquad c_j = \frac{\omega_j}{k_j} = c.
$$

The mechanism is explicitly distinguished from freely propagating internal waves: the higher harmonics are bound, non–freely propagating, and do not satisfy the free-wave dispersion relation [2203.00475].

The model fit yields a single dimensionless parameter,

$$
y = \frac{\hat{U}}{c} \equiv \epsilon,
$$

with best-fit value \(y = 0.48 \pm 0.05\). Using harmonic analysis with \(\hat{U}_0 / \hat{U} = 2.0 \pm 0.15\), the study obtains \(\hat{U}_0 = (0.96 \pm 0.11)c\), so \(c \approx \hat{U}_0\). The interpretation is that the barotropic \(M_2\) current amplitude sets the phase-speed scale underlying the nonlinear cascade among baroclinic harmonics, even though barotropic advection itself does not directly generate the observed nonlinear higher harmonics [2203.00475].

This leads to a more specific physical implication: Biscay deep-basin spectra are consistent with a wave–turbulence transition, with observations suggesting similarity of particle displacement speed and phase speed of the semidiurnal signal and an inferred \(Ri \sim 1\). The model therefore treats the fitted \(y\) as a measure of steepness limited by a balance between nonlinear forcing and turbulent mixing. A stated limitation is that direct turbulence measurements are lacking, so this interpretation remains statistically supported rather than directly validated [2203.00475].

## 4. Offshore wave climate and data-driven prediction

At an offshore Biscay site, significant wave height \(H_s\) has been modeled with a two-stage deep-learning architecture that links North Atlantic winds to local sea state. The target point is \((45.2^\circ \mathrm{N}, 1.6^\circ \mathrm{W})\). Wind predictors come from the CFSR reanalysis on a \(0.5^\circ \times 0.5^\circ\) grid, restricted to grid points whose great-circle paths to the site are not blocked by land. The resulting domain contains \(p = 5651\) masked grid points [2205.13325].

The global wind predictor is based on “projected wind” toward the target point,

$$
W_j(t) = U_j(t)\,\cos^{2}\!\left(\tfrac{1}{2}(b_j - \theta_j(t))\right),
$$

with

$$
X^{(g)}(t) = \left(W_1^2(t), \ldots, W_p^2(t)\right).
$$

The local predictor is

$$
X^{(\ell)}(t) = \{U(t),U^2(t),U^3(t),U^2(t)F(t),U(t-1),U^2(t-1),U^3(t-1),U^2(t-1)F(t-1)\},
$$

where \(F(t)\) is the fetch length along the local wind direction, defined as the minimum of the distance-to-shore along that direction and 500 km. Data cover 1994–2016 at 3-hourly resolution, with calibration on 1994–2011 and validation on 2012–2014 [2205.13325].

The first stage uses CNNs with 3×3 convolutions, ReLU activation, 2×2 max-pooling, Batch Normalization, flattening, and a dense multi-output head to estimate immediate and future wave contributions from a single wind field. The second stage uses a single LSTM layer to integrate the sequence of Stage-1 contributions over a maximum swell travel-time window \(t_{\max}\), then concatenates the LSTM output with the 8-element local predictor and maps the result through two fully connected layers to final \(H_s(t)\) [2205.13325].

A 5-fold cross-validation selected \(t_{\max} = 30\) time steps of 3 h, approximately 90 h or 3.3 days. This is consistent with the cited travel times from common North Atlantic storm tracks to Biscay. The physical interpretation built into the architecture is explicit: the first column of the Stage-2 input represents wind sea, while entries with delayed arrival represent swell generated by earlier wind events [2205.13325].

Validation at the Biscay site shows \(r = 0.98\), \(RMSE = 0.21\) m, and bias \(= -0.006\) m for the two-stage CNN–LSTM. Weather-types statistical downscaling and H-CNN each achieved \(r = 0.97\) and \(RMSE = 0.27\) m, with biases of \(-0.03\) m and \(-0.04\) m respectively. The mean \(H_s\) during validation was approximately 1.9 m, with standard deviation approximately 1.1 m. Training time was approximately 5 minutes on a system with 30 GB RAM, 2 CPU cores, and a 16 GB Nvidia K80s GPU [2205.13325].

Two limitations are stated directly. First, the model is trained and validated for one Biscay location, so transfer to nearby sites is proposed rather than demonstrated. Second, handling of rare extremes is not explicitly assessed: the time series suggests that peaks are tracked, but quantitative peak-bias or tail-error statistics are not reported [2205.13325].

## 5. Management, monitoring, and methodological implications

For coastal Biscay fisheries, the dominant recommendation is to align spatial management with the observed scale and direction of connectivity. Since typical dispersal distances fall within 10–60 km and are biased westward in high-upwelling summers, closures around overharvested target areas should be complemented with additional closures to their east, i.e., upstream under summer NE winds. The same study recommends redistributing clusters of total bans into an evenly spaced network at 10–60 km intervals, on the premise that multiple small reserves spaced at dispersal distances are more effective than single larger closures in current-driven systems [2401.09513].

Quota-setting is also linked to ocean forcing. Strong upwelling, represented by high IUI, predicts higher CPUE four years later, and the suggested operational use is to integrate seasonal IUI into TAC decisions and rotational closure timing. Conversely, low or interrupted upwelling years imply reduced recruitment and support conservative TACs and longer recovery periods for heavily fished rocks [2401.09513].

The monitoring agenda implied by the Biscay studies is broader than fisheries alone. For larval dispersal, the stated limitations are a single ADCP mooring, only two seasons, 2D horizontal transport, simple vertical allocation, a uniformly sticky coast, and a single maternal locus for connectivity inference. Recommended extensions include arrays of nearshore ADCPs and HF radar, coupling to 3D hydrodynamic models such as ROMS, inclusion of diel vertical migrations and explicit mortality, habitat-quality mapping, and augmenting COI with genome-wide SNPs and nuclear markers [2401.09513].

For deep-basin internal-wave dynamics, the main open problem is direct corroboration of the inferred mixing balance. The simple advection model captures spectral shapes statistically, but it omits backward cascades, diffusion, and rotation in the interaction term, and direct turbulence observations are lacking [2203.00475]. For offshore wave prediction, proposed enhancements include multi-site training, adding bathymetry and currents, transfer learning with buoy data, and extension to predict direction and period in addition to \(H_s\) [2205.13325].

A plausible implication is that Biscay functions as a natural laboratory for testing how shelf geometry, remote forcing, and sampling design interact across ecological, physical, and engineering timescales. The sources do not claim a unified framework, but they repeatedly show that coarse resolution—whether in current sampling, turbulence observation, or wave-forcing representation—limits interpretation.

## 6. BISCAY as a mobile-network systems acronym

In a separate technical usage, **BISCAY** stands for **“Practical Radio KPI Driven Congestion Control for Mobile Networks”**. This BISCAY is unrelated to the marine region except by name. It is a device-centric congestion control system that uses fine-grained radio KPI measurements from the phone’s modem to compute instantaneous cellular-link bandwidth and set the sender’s congestion window accordingly [2509.02806].

Its core enabler is **OpenDiag**, an in-kernel KPI extraction tool coupled to the Qualcomm Diag driver. OpenDiag drains the modem’s internal diagnostic buffer every 1 ms, shorter than the most frequent diag message period of approximately 10 ms, and achieves approximately 10.9 ms per-sample granularity for KPIs such as RSRP and PRB/TBS, compared with approximately 1000 ms for MobileInsight and the Android Telephony API. Processing delays are tens of microseconds per subscribed KPI packet, versus hundreds of milliseconds for MobileInsight [2509.02806].

BISCAY estimates bandwidth from radio grants using standardized TBS lookup. For carrier \(c\) at time \(t\),

$$
\hat{B}(t) = \sum_{c \in \mathcal{C}(t)} N_{\mathrm{layers}}(c,t)\cdot \frac{\mathrm{TBS}(\mathrm{PRB}(c,t), \mathrm{TBS\text{-}idx}(c,t))}{T_{\mathrm{TTI}}},
$$

and sets the congestion window from the bandwidth–delay product, split across active flows:

$$
\mathrm{cwnd}(t) = \frac{\hat{B}(t)}{N_{\mathrm{flows}}(t)} \cdot \min\!\big(\mathrm{RTT}(t), \mathrm{RTT}_{\min}\big).
$$

It uses a three-state machine—STARTUP, BISCAY, and FALLBACK—and reverts to wired CCA, specifically BBR in the implementation, when the wired segment rather than the cellular link is the bottleneck [2509.02806].

Evaluation in emulation and real networks reports large delay reductions with similar or better throughput. Relative to BBR in Pantheon experiments, BISCAY achieved approximately \(1.03\times\) throughput, 58.51% average delay reduction, and 41.18% tail delay reduction. Relative to CUBIC, it achieved approximately \(0.96\times\) throughput, 98.74% average delay reduction, and 99.03% tail delay reduction. In real-world experiments on an unrooted Google Pixel 5, BISCAY delivered approximately 4.6% higher throughput than both BBR and CUBIC while reducing average delay by approximately 46% and tail delay by approximately 44% compared to BBR [2509.02806].

The acronymic BISCAY also has explicit limitations: KPI availability depends on modem state, vendor support is uneven because Diag is proprietary, mapping KPI timing to exact TTI boundaries can require care, and equal per-flow sharing can be gamed by applications that open many sockets. In this sense, “Biscay” has acquired a second, purely technical meaning in systems research, distinct from the Bay of Biscay but still centered on measurement resolution, transport asymmetry, and control under rapidly varying conditions [2509.02806].

Source: https://www.emergentmind.com/topics/biscay