Site description

The Burrishoole catchment (~100 km²) is a predominantly upland blanket peat, oligotrophic catchment located within the Nephin Beg mountain range on the west Atlantic coast of Ireland (Fig 1). Rivers and streams are generally acidic or neutral and water chemistry depends on local and regional climate as well as surrounding soil composition in each of the sub-catchments [37]. Given its coastal proximity, the catchment generally experiences a temperate climate with mild winters (mean December-February 2005–18 air temperature of 6.0° C) and summers (mean June-August 2005–18 air temperature of 14.3° C) and a diurnal sea breeze with mean wind speeds of ~5 m s^-1 [38]. Burrishoole’s primary lake basin is Lough Feeagh, with a surface area of 3.95 km², max depth of 46 m, mean depth of 14.5 m and volume of 5.9 x 10⁷ m³. The water of Feeagh is oligotrophic and humic, with a mean secchi disk depth of 1.7 m, pH of 6.6 and annual mean total nitrogen and phosphorus of 430 μg l^-1 and 6.1 μg l^-1 respectively [39]. Feeagh is located at the base of the Burrishoole catchment, draining a ~84 km² watershed into Lough Furnace through two short outflows, the Salmon Leap and (man-made) Mill Race rivers. Furnace is a deep lagoonal estuary (M₂ semi-diurnal tide dominant) with a surface area of 1.5 km², max depth of 21 m and mean depth of 7.9 m. The system comprises a main inner basin (volume of 6.75 x 10⁶ m³) connected to the open coastal waters of Clew Bay by a 1-km-long shallow and narrow channel. Tidal currents transporting coastal water must traverse this channel and two narrow topographic constrictions before reaching the inner basin [38]. The inner Furnace basin is notable for its strong saline stratification and deep anoxia, resembling a meromictic saline lake, although ventilation of the bottom water by dense tidal inflows occurs irregularly [38]. A research station operated by the Marine Institute is situated on the northern shore of Furnace and has served primarily as an international index site for migratory fish populations (Atlantic salmon (Salmo salar L.), trout (Salmo trutta L). and European eel (Anguilla Anguilla L.)) in the North Atlantic for half a century [40,41]. The upper parts of the catchment have Special Area of Conservation (SAC) status for the conservation of Atlantic salmon and otter (Lutra lutra), two species listed in Annex II of the EU Habitats Directive. Lough Furnace is part of the Clew Bay Complex Special Area of Conservation (SAC site code 001482).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Spatial distribution of rainfall over Burrishoole catchment during storm Desmond.

Measured by 13 rain gauges (RG) (note that RG 11, 12 and 13 are situated in the vicinity of the Automatic Weather Station (AWS)). Also shown are the locations of the Automatic Water Quality Monitoring Stations (AWQMS) on Lough Feeagh and Lough Furnace and the additional instrumented moorings MR01 and MR02 on Furnace, described in the text. Shapefiles used for lakes and rivers shown here are available at: http://gis.epa.ie/GetData/Download.

https://doi.org/10.1371/journal.pone.0235963.g001

Data collection and processing

This study relied on data sources routinely collected as part of the Marine Institute’s long-term environmental monitoring programme in Burrishoole. Spatial extent of rainfall over the entire Burrishoole catchment was recorded by 13 tipping bucket rain gauges (Davis Instruments Corp, Hayward, CA, USA) located throughout the catchment and lake and lagoon levels were monitored by pressure sensors (OTT, Kempton, Germany) recording every 15 minutes (Fig 1). Automatic water quality monitoring stations (AWQMS) were located on Feeagh and Furnace (Fig 1). Each AWQMS was instrumented with on-lake meteorological arrays recording wind speed and direction, air temperature, shortwave radiation and surface pressure at 2-minute intervals. A multi-parameter sonde (Hydrolab DS5, OTT, Kempton, Germany) was suspended from each AWQMS; on Feeagh this sonde was kept at 0.9 m below the surface recording every 2 minutes whilst on Furnace the sonde was attached to an undulating winch profiler, recording 4 full water column profiles daily (at 00:00, 06:00, 12:00 and 18:00 hours). Parameters measured on each sonde included temperature, conductivity, pH and dissolved oxygen. The Feeagh AWQMS also contained a surface nephelometer (Chelsea Technologies Group Ltd, Surrey, UK) and thermistors (Labfacility Ltd., Bognor Regis, UK) that spanned the full water column at depth intervals of 2.5, 5, 8, 11, 14, 16, 18, 20, 22, 27, 32, 42 m, recording every 2 minutes. An additional automatic weather station (AWS) maintained by Met Éireann since 2005 was situated on the shoreline between both loughs, which recorded relative humidity and rainfall in addition to the same over-water meteorological variables recorded at each AWQMS (Fig 1).

On Furnace two additional instrumented moorings were located at the entrance channel into the main inner basin (Fig 1). MR01 (S1 Fig) contained a bottom-mounted, upward looking 1-MHz Nortek Aquadopp three-beam current profiler (ADCP) (Nortek AS, Rud, Norway). The ADCP sensor head, with a 25° beam angle to the vertical, was mounted at 0.2 m above the bed. The ADCP recorded in 0.5 m bin intervals, with a 0.4 m blanking distance and a velocity measure averaged over a 60 second ping interval. Two temperature and conductivity sensors (T/C) were attached to the same mooring (SBE-37 MicroCATs (Sea-Bird Scientific, Bellevue, Washington, USA)). One T/C sensor was mounted on the bottom ADCP frame and the second positioned ~1 m below the surface on an L-shaped mooring. An additional T/C, also recording pressure, was located further downstream (MR02, Fig 1). All of these instruments had synchronised internal clocks and logged measurements every 30 minutes from 13:30:00 November 26^th 2015 until 11:30:00 January 21^st 2016.

CTD downcasts from the Furnace AWQMS were extrapolated onto a standard depth grid spanning the water column from 0.6 m– 12.9 m in 0.15 m depth increments. Practical salinity (derived from actual conductivity and temperature) and water density (derived from salinity and temperature) were computed according to the UNESCO algorithm [42] for all sensors recording temperature and conductivity.

ADCP velocity vectors were low-pass filtered with a 6 hour cut-off frequency to remove higher frequency tidal harmonics. Low-passed horizontal velocity vectors were then rotated into along- (v) and across-channel (u) components using principal components analysis (PCA) [43]. Whilst the standard ADCP bin sizes spanned 0.5 m in the vertical, the lowest bin was weighted to account for the bottom 1.1 m of the channel (i.e. 0.5 m bin size + 0.4 m blanking distance + 0.2 m frame height above the bed). For each individual ADCP profile, the uppermost usable bin was defined as the shallowest bin below the computed depth of sidelobe interference with contaminated shallower measurements discarded. In order to account for this portion of the upper water column not measured (~10–15%), each current profile was extrapolated to the surface level using a constant velocity from the uppermost usable bin.

Data analysis

Precipitation recorded by each of the 13 rain gauges throughout the catchment were summed to 24 hour totals. To estimate the total rainfall volume that fell in the catchment within a specific time period, the raster calculator routine in the ‘Spatial Analyst’ toolbox in the GIS Software ArcMap 10 [44] was used. The raster calculator routine allows mathematical calculations to be conducted on existing raster data sets, in this case a raster map of the topography of the catchment. The topographic raster data comprised a 10 by 10 metre grid of the catchment with an altitude above sea level value at the centre of each grid square. To estimate rainfall volumes during Storm Desmond, an altitude-rainfall regression equation was calculated for the time period 4–6 December 2015. The rainfall volumes for each grid square were summed to get the total rainfall volume. However, the altitude-rainfall volume relationship during storm Desmond (R² = 0.32) was not strong when compared to a 14-year study period, 2004–2017 (R² = 0.82). Analysis of rainfall volumes during the storm showed that lower volumes of rainfall were recorded in the rain gauges at higher altitudes in the catchment than would otherwise be expected. Presumably the breakdown of the altitude-rainfall volume relationship at higher elevations in the catchment is a result of high wind speeds causing ‘under-catch’ of rainwater during the storm. Under-catch is a common phenomenon affecting the accuracy of rain gauges, where high wind speeds during storms cause wind flow deflections and turbulence patterns around the gauge funnel which in turn cause some of the raindrops to be ejected from the funnel system [45].

To correct for this, the rainfall volumes for the three rain gauges at the greatest altitudes were revised. Using the AWS rain gauge (Fig 1A) as an index site, the rainfall volume measured at this site during Storm Desmond was used to estimate the rain volumes at the three highest altitude sites. Coefficients between the index site and each of the high altitude sites were calculated over the 14 year period for which rain gauge data was available. The rainfall volume at AWS was multiplied by each of the rainfall volume coefficients for the three highest altitude sites to adjust for under-catch. Following adjustment, the altitude-rainfall volume relationship during the specific Storm Desmond timeframe improved to give an R² value of 0.73 and a revised total rainfall during Storm Desmond was estimated using this method.

Storm effects on dynamical transfer processes of heat and momentum across the air-water interface on both Feeagh and Furnace were assessed by calculating the net surface heat flux and energy input due to wind stress. The surface heat flux (Q_shf, W m^-2) over the month of December 2015 was calculated following [46]: (1) where Q_swin (W m^-2) is the net incoming shortwave solar radiation (accounting for reflected shortwave due to surface albedo) and Q_lwin and Q_lwout are the incoming and outgoing thermal radiation (W m^-2). Q_h (W m^-2) is the sensible heat flux, Q_h = ρ_aC_PaC_HW_z(T_s−T_z), and Q_e (W m^-2) is latent heat flux, Q_e = ρ_aL_vC_EW_z(q_s−q_z), where ρ_a is the density of air at the air-sea interface (kg m^-3), C_Pa is the specific heat of air (J kg^-1°C^-1), C_H the turbulent transfer coefficient for heat, W_z the wind speed at height Z above the water (m s^-1), T_s the surface water temperature (°C), T_z the air temperature at height Z above the water surface (°C), L_v the latent heat of vaporization (J kg^-1), C_E is the turbulent transfer coefficient for humidity, q_s the saturated specific humidity (kg kg^-1) at T_s and q_z the specific humidity of unsaturated air (kg kg^-1) at height Z above the water surface. One caveat relating to the heat flux calculations presented herein is that the humidity measurements were only available from the shoreline meteorological station and not directly above either Feeagh or Furnace water surface. Positive values of each variable in Eq (1) indicate downward flux of heat; negative values indicate loss of heat to the atmosphere.

Mechanical energy transfer from the wind to the water was parameterised as the rate of work per unit area by wind stress at 10m above the water surface (W m^-2) [47]: (2) where C_D is the surface drag coefficient. Turbulent transfer coefficients for heat (C_H), momentum (C_D) and humidity (C_E) were scaled to 10m above the water surface by correcting for atmospheric stability [46,48].

An informative measure for saline Furnace was to estimate a buoyancy flux associated with the storm through the surface, B_f (m² s^-3), including a moisture flux term (e.g. [49]): (3) where g is gravity (m s^-2), ρ₀ is surface water density (kg m^-3), C_P the specific heat capacity of water (J kg^-1 K^-1), α and β are the thermal expansion (α = ρ₀⁻¹∂ρ/∂T)) and haline contraction (β = ρ₀⁻¹∂ρ/∂S) coefficients, S₀ is surface salinity and freshwater flux (m³ s^-1) is given by river discharge (Q_f) plus direct precipitation (P) minus evaporation (E) over the surface area of the lough (A_s, m²). Positive B_F indicates buoyancy gained at the water surface.

How storm events influenced full water column stability in each water body was assessed by calculating the Schmidt stability index (S_s, J m^-2) [50]: (4) where is the depth (m) at which mean density occurs, ρ(z) is density at depth z, is volume weighted mean density and A(z) is basin area at depth z. S_s estimates the amount energy needed to fully mix the water column, providing a good quantitative measure of stratification and destratification processes.

In Feeagh, which is destratified by December, the total lake heat content (H_tot, J m^-2) was identified as a potentially more informative measure of internal lake response to atmospheric forcing compared to S_s: (5)

In Furnace, storm winds would be anticipated to destabilise the stratified water column yet the large volumes of freshwater discharge would be anticipated to reinforce vertical stratification through a positive buoyancy input. An appropriate parameterisation of the wind-driven response of the semi-enclosed (i.e. finite-dimension), 2-layered Furnace basin during the storms is the dimensionless Wedderburn number (W_b) (e.g. [51]): (6) where R_i is the surface-layer Richardson number , with g′ the reduced acceleration of gravity across the pycnocline (g′ = g(Δρ/ρ₀)), h the depth of the surface layer (m) and u_* the surface water shear velocity (m s^-1) given as . Wind speeds were low-pass filtered with a cut-off frequency of ¼ the fundamental mode internal wave period [32]. l′/h is the aspect ratio of the basin, with the effective length scale l′ defined here as the length of the unbroken pycnocline along the NE-SW axis, the direction of wind stress during Desmond.

In general, large W_b values (> 10) imply a sharp, flat interface is maintained; smaller values (1 < W_b < 10) indicate partial upwelling of the pycnocline at the upwind basin end and subsequent internal seiching following wind cessation; W_b≈1 implies full pycnocline upwelling whilst still smaller values (e.g. < 0.5) indicate full surfacing of sub-pycnocline waters with formation of longitudinal density gradients and ultimately basin-scale overturns [51–53].

Estuarine exchange flow through the connecting channel between Furnace and Clew Bay controls transport dynamics in and out of the main inner Furnace basin and links the Burrishoole watershed to the coastal ocean. Modification of salt content of dense bottom inflows due to mixing with lower salinity surface outflows regulates the residence time of saline water masses in the inner basin and thus the oxygen content of deep waters [38]. The impact of the storms on diapycnal mixing in this region was investigated using ADCP profiles (MR01 Fig 1), with squared shear computed as: (7) where z is depth in the channel. As only two measures of density were available at the surface and bottom near the ADCP, we computed a full water column ‘bulk’ buoyancy frequency N as: (8)

S² was averaged over the full water column and the gradient Richardson number calculated as (e.g. [54]): (9)

It was envisaged that Ri_g would provide a simple diagnosis of the dynamic stability of flow through this important section of the system, integrating the effects of wind, river flow and tidal currents on shear and stratification. Ri_g > 0.25 is one criterion for the occurrence of flow instabilities [54].

To compute volume and salt fluxes in and out of the inner Lough Furnace basin, the full entrance cross-section at MR01 was divided into subsections in order to overcome the limited spatial coverage of the sensors (S1 Fig). An echosounder transect provided accurate bathymetry. Firstly, the entranceway was divided laterally into a deeper trough section and shallower section to the east. The deeper channel region was divided vertically in 3 sections, with section averaged salinity in the uppermost and lowermost sections set equal to the values recorded by the T/C sensors; an interpolated salinity value was used for the intermediate section. A corresponding set of along-channel, area-averaged velocity vectors were computed from the ADCP profiles. To account for the unmonitored shallower (~2m) east section we used a simplistic approach and extrapolated across the salinity and velocity from the uppermost section in the deep channel, with the caveat that the resulting volume and salt flux values represented a first-order estimate.

Thus, within each subsection i, a timeseries of area A_i, along-channel velocity v_i and salinity S_i were derived. The net salt flux (S_f) was computed as the sum of flow rate and salinity through each area of the cross-section: (10)

A_i for the top section in the deep channel and for the shallow side section increased or decreased with changing water level as measured by the ADCP pressure sensor. Angled brackets indicate averaging over a complete M₂ tidal cycle. Positive values indicate fluxes into the inner Furnace basin and negative values indicate fluxes out (seaward). Volume fluxes were estimated by omitting the salinity term in Eq (10).

Storm impacts on catchment and freshwater lake habitats

Analysis of the rain gauge data during Storm Desmond showed that the rainfall was extremely heavy, continuous and spatially distributed throughout the catchment. This was in contrast to an extreme precipitation event documented in this catchment previously, when a much shorter and intensive rainfall event exhibited a notable asymmetry and led to particularly destructive flooding on the eastern side of the catchment [17]. In Lough Feeagh, the most notable shift was the sudden increase in surface water acidity following Desmond. This was likely a consequence of the mobilisation of organic matter from the predominantly peatland catchment area [37] with a potential contribution from the direct rainfall over the lake surface. Whilst Feeagh is normally slightly acidic, post-flood values were noticeably lower than the annual mean pH (6.1 compared to 6.6). River plumes resulting from major flood events have been shown to cause decreases in lake pH relative to non-plume waters, with this effect related primarily to the soil-type surrounding the river [57,58]. These results are applicable to lake systems situated in other upland peat catchments around Ireland and the United Kingdom, that are likely to experience more of these high intensity rainfall events which are projected for these regions [35,36,59]. Significant increases in post-flood turbidity and decreased water clarity have been documented for other lakes [11,20]. Biological consequences of increased riverine input may include increased respiration [17,19,60] whilst decreased water clarity following storms may substantially decrease lake primary productivity [61]. However, as the event described here occurred in winter, when primary productivity was low, and changes in water clarity, inputs of organic material and even loss of plankton communities to bulk advection by flood waters are less likely to cause major ecological change [12,17]. Monthly measurements of water colour, total nitrogen and total phosphorus in our study did not deviate noticeably from usual winter values.

Apart from biogeochemical changes, Feeagh also underwent net warming during the storm due to positive (downward) heat fluxes (Figs 3D and 4). Our results represent a rare quantification of the direct transfer of properties from an ‘atmospheric river’ [35] directly to surface and interior lake waters. The magnitude of warming was small however and heat was rapidly redistributed throughout the water column. Thus, the winter setting was again important in terms of how effectively air-water energy transfer was circulated to interior waters, as Feeagh is fully mixed and isothermal by December. In contrast, an extreme flood event during summer 2009 when Feeagh was stratified had more drastic impacts on water column hydrodynamics, causing a decrease in stratification stability and deepening the epilimnion [17].

Storm impacts on estuarine and coastal habitats

The largest storm impacts occurred on the downstream Lough Furnace. Here, storm event timing had less impact hydrodynamically, as this system is stratified year-round [38]. Prior to Desmond, heavy rainfall throughout November had caused river discharge to increase the thickness of the freshwater surface layer, resulting in a large salinity (viz. density) gradient with depth and a doubling in stratification strength. The major consequence of a deepening halocline is that it reinforces stagnant, anoxic conditions in the bottom basin water. Whilst Furnace does not overturn vertically, occasional lateral inflows of dense, oxygenated coastal water can ventilate the bottom anoxic water [38]. Similar deep-water ventilation processes are documented for stratified coastal basins such as fjords and sea-lochs [62] and play a fundamental role in oxygen dynamics (e.g. [63]), microbial activity (e.g. [64]) and primary production (e.g. [65]). In Furnace, deep-water ventilation has been observed to occur only following prolonged dry periods with little river flow, allowing tidal inflows of coastal water to reach the bottom of the inner basin. The large freshwater plume that occurred during winter 2015 blocked communication between the deep basin water and external coastal water, the only potential supplier of oxygenated replacement water. Initially the surface freshwater layer occupied the full depth extent of the connecting channel inhibiting any landward inflows.

Consequently, this absence of stratification in the connecting channel leading up to and during Storm Desmond minimised shear, particularly below the wind-influenced surface layer (Fig 7D). Subsidence of the initial flood and onset of spring tides gave rise to an evolving and highly dynamic flow environment. Tidal currents and strong winds generated intense shear near the surface and bottom, supported by the restratification, with indication of turbulent mixing in the entrance channel (Fig 7F). However the large amount of residual freshwater and successive spring tides meant that increasingly strong stratification ultimately suppressed large-scale turbulence and gave way to a stratified exchange flow regime. Flow structure and turbulence in this section is noteworthy, as it sets the composition of fluid entering and leaving Furnace. This has implications for exchange of resident interior water [38] and influxes of nutrients or planktonic seed populations [66] and also establishes the initial density of the estuarine outflow, which can influence far field plume dynamics in the coastal ocean [67]. The development of stratified flow resulted in the dilution of incoming tidal water as it met largely unmixed low salinity surface outflow in the shallow area between MR01 and MR02 (Figs 1 and 7B); this resulted in only intermediate depth water replacement in the main Furnace basin in the months following the flood event.

More broadly, modified freshwater runoff climatology has been implicated in exacerbating dissolved oxygen depletion in estuarine and coastal systems globally [68]. This is largely due to the mechanism explored here: intensified vertical stratification and isolation of deeper waters from lateral inflows. Additionally, rivers may cause nutrient enrichment which stimulates primary productivity or directly transport organic matter to coastal zones, which both heighten respiration rates and increase oxygen demand [26]. At the largest scale, bottom oxygen dynamics in the Baltic Sea are generally controlled by deep water inflows from the contiguous North Sea; changes in precipitation intensity over North-West Europe in recent decades has been implicated in a reduction in ventilation frequency due to large freshwater outflows attenuating dense inflows [69]. In semi-enclosed coastal systems similar to Furnace with restricted connections to adjacent coastal areas, large seaward river flows can inhibit ventilation of deep waters (e.g. [70,71]). Thus, a very likely consequence of changing patterns of precipitation intensity, especially the kind of extreme flood events recounted here, could significantly reinforce stagnant bottom water conditions in deep estuarine basins, promoting anoxia.

In several studies of stratified basins during storms, upwelling of sub-pycnocline water in nearshore areas was documented [15,18,72] and in general upwelling plays a pivotal role in generating turbulent flux pathways around lateral boundaries [73,74]. Partial upwellings occurred in Furnace during individual storm events throughout December, with the stable stratification preventing complete surfacing of lower layer water (Fig 6C). Pycnocline destabilisation results in energy transfer to the internal wave field, with the ratio of h/H and range of inverse Wb in Furnace during December implying that most of the resulting standing waves were linearly damped [53]. In systems with deep anoxia, internal seiche-related upwelling can have added significance, exposing nearshore areas to drastically fluctuating oxygen levels [32,53,75].

Catchment-to-coast volume fluxes

A ratings curve developed for the Mill Race and Salmon Leap rivers (Fig 1) provided a useful reference for the anticipated freshwater outflow volumes from the Burrishoole catchment during December (although it should be noted that the ratings curve relationship almost certainly breaks down for the extreme flows that occurred during Desmond). The mean river discharge into Furnace during the month of December was 13.7 m³ s^-1. This compared reasonably well to the mean volume flux estimated from the ADCP of—15.8 m³ s^-1, which also would have integrated volumetric fluxes from direct precipitation over Furnace as well as additional surface and groundwater inflows from the surrounding land area. Thus, using the ADCP estimates, the mean volumetric flux leaving the Burrishoole system during Storm Desmond (4–6 December) was (O)43 m³ s^-1, with instantaneous fluxes potentially as high as (O)88 m³ s^-1 at peak flooding. Based on the Feeagh lake level records, this was most likely the highest volume of water leaving Burrishoole since 1976 when records began (Fig 2). Mean annual discharge from Feeagh into Furnace is only 3–5 m³ s^-1 [38]. Furthermore, the total volume of rainfall estimated from analysis of the catchment rain gauges over the 72 hour period (4–6 December) gave a volumetric rainfall flux of 46–54 m³ s^-1, which again compares reasonably well to the ADCP estimate. The mismatch may be attributed to measurement error and calculation uncertainty attributed to each method and also that not all of the rainfall would have instantaneously exited the system (for example Feeagh lake level took ~10 days to fall back to typical mean December levels after the 6^th December (Fig 2)).

Using the average volume flux of 43 m³ s^-1 as a representative transport of water through the catchment during Desmond, the renewal time of the Feeagh lake basin (T_R = Vol/Q) was reduced to 15.8 days. The average renewal time of Feeagh is ~170 days; a previous extreme flood event during summer 2009 indicated that the renewal time of Feeagh also reduced to (O)15 days [17]. However a major difference was that peak flow rates used to estimate this reduced renewal time during the summer flood lasted only ~3 hours, whereas our estimate for the Desmond volume flux is representative of a 3 day period, meaning that at least 20% of the volume of fully-mixed Feeagh was displaced over the 4–6 December, if a 100% exchange efficiency is assumed.

These values provide a first order estimate of the quantity of freshwater that ultimately reached Clew Bay and the coastal ocean from one single watershed in the aftermath of Desmond. Considering additional sources of river discharge around Clew Bay, it is likely that such elevated freshwater discharge would have affected stratification dynamics in the short term and may have delivered large quantities of terrestrial-source water properties over a very short time span (e.g. [27,28]). Numerical modelling of the impact of such extreme rainfall and flood events on the oceanography and biogeochemistry of Clew Bay could offer insight into how large coastal embayments respond to such climate-related extreme events.

Storm impacts on estuarine salt balance

In estuarine basins, the salinity budget is maintained by the competing influences of seaward and landward salt transport and depends on tidal, wind and freshwater flow [76]. In stratified estuaries such as Furnace, spring-neap oscillations in salt transport are often the dominant factor in determining the longer-term salt balance (e.g. [29]). For example, in the inner basin of Furnace, steady-state conditions in relation to salt and volume are on average achieved over a complete spring-neap cycle [77]. However large pulses of freshwater may disrupt the longer-term salt balance, especially in lagoonal estuaries with restricted horizontal connection to oceanic water. The Furnace salt balance was modified during Desmond with net export of salt which was not replenished immediately by spring tides. Perhaps most importantly, this salt flux did not appear to be imported salt from the prior spring tides, as the surface and bottom salinities at MR01 were almost zero prior to Desmond yet both increased slightly during the middle of Desmond (Fig 7B). To assess whether this exported salt came from sub-halocline water in the deep inner basin, we invoked some practical assumptions about the mean volumetric flux during Desmond (Q_mean≈43 m³ s^-1), the salinity of the water above the halocline at the mouth of the basin (S₁≈0, recorded at MR01) and salinity below the halocline (S₂≈16; Fig 5A), and estimated an entrainment flux (Q_We, m^-3 s^-1) of sub-halocline water by the outflowing plume as Q_We = Q_mean(S₁/S₂−S₁). This provided a mean estimate of Q_We during the flood of 2.8 m³ s^-1. Using the approximate basin area at the depth of the halocline gives a mean entrainment velocity of 5 x 10⁻⁶ m s^-1. Thus for S₂≈16, a simplistic estimate of mean upward salt flux across the halocline gives 45 kg s^-1; over the 4–6 December the mean salt flux through the entranceway based on Eq 10 was– 42 kg s^-1 (although the large range of values in the salt flux timeseries must be acknowledged, with max values of– 90 kg s^-1 at peak flow (Fig 8C)). Thus we tentatively conclude that salt loss from Furnace during the peak of the freshwater flood was from older resident salt reserves located beneath the halocline and below sill depth. This is an important dynamic, as in topographically constrained estuaries like Furnace, the deep basin water below sill depth is normally detached from the wind, estuarine and tidal circulation that takes place above the sill (e.g. [62]). Large flood events may be capable of freshening semi-enclosed coastal basins such as Furnace not only by limiting salt import during flood tides but also by exporting resident salt through entrainment.