**Lake Bathymetry Mapping**

The bathymetry of the lakes was recorded with a purpose-built lightweight, low-power echosounder unit with integrated GPS and AHRS (attitude and heading reference system), towed behind a kayak. (Data courtesy of Herwig Stibor, Thomas Stieglitz.)

**Lake Bathymetry Analysis**

Raw bathymetry data were down-sampled to include one sample every 2 m distance along the survey track. Data were visually inspected and false soundings removed. In some lakes, GPS reception was partially compromised along the steep shorelines. Where required to not compromise data density, the survey track was reconstructed from field notes and AHRS data (less than 5% of data where applicable). During the bathymetry surveys, tidal water level was not recorded. Therefore, tidal water level variations were corrected for by modeling tidal water level in each lake. The water level for each lake was correlated with tidal water level measured at a reference station at CRRF or predicted tide (x-tide database), by determining tidal lag time and tidal efficiency (see below) from data previously collected concurrently in the respective lake and at this reference station. Subsequently, bathymetry data was reduced to a grid with 2m resolution by kriging of a further down-sampled subset of data using every third data point. Lake-specific variograms were applied, and contour lines and total lake volumes were calculated in a GIS. (Data courtesy of Thomas Stieglitz.)

**Distance from lake to ocean & lake surface area**

Lake shorelines and island coastlines were manually extracted from satellite data (Microsoft Bing). The nearest, mean and median distance of each lake to the respective island’s coastline as well as lake surface area was calculated in a GIS. The perimeter was measured in meters using the ‘Measure Line’ tool in QGIS 3.4.

Habitable surface area for benthic organisms was estimated either as (1) a multiple of the perimeter and depth of the chemocline, i.e. assuming a cylindrical model for the lake, or (2) as the area of a frustum, i.e. a truncated cone, using the perimeter of the lake at the surface, the average length of transects, and the average angle of the transect from the vertical (estimated using the sine of maximum depth / transect length).

Tidal lag time and tidal efficiency were calculated from concurrently measured tidal water level in a lake and the adjacent ocean. The tidal lag time — the time between high tide in the adjacent ocean and high tide in the lake — was determined by least-square fit between lake and ocean tide measured using HOBO 30-foot depth Titanium water level data loggers (Part # U20-001-01-Ti). Tidal efficiency was calculated as the ratio between amplitude of ocean tide and lake tide (e.g. Ayers, JF and Vacher, HL, 1986. Hydrogeology of an Atoll Island: A Conceptual Model from a Detailed Study of a Micronesian Example. Groundwater 24(2) 185-198.). Larger tidal lag time and smaller tidal efficiency respectively indicate a less efficient hydrological connection between ocean and lake.

Error-checking: Estimates of basic dimensions (depth, distance, length, surface area) were double-checked manually for a subset of measurements by a second person using using Google Earth.