Figure 4 . Relative sea level change computed using all reservoirs in the Zarfl et al. (2015) database, assumed to be completed by 2040, with seepage included as described in the text. (a ) Global GRD fingerprint signal. Colors as in Fig. 3. (b ) Detail of frame (a ) in South America, showing sea level rise of a few mm along the west coast of Ecuador and Peru and around the mouth of the Amazon River in Brazil. Cool colors show sea level fall; warm colors show sea level rise. (c ) Detail of frame (a ) in Southeast Asia, indicating sea level rise of a few mm along the low-lying coasts of Bangladesh and Myanmar, and 1–2 mm along the Vietnamese coast along the Gulf of Tonkin.
4 Discussion
4.1 Tide Gauge Observations
Our time series provides a globally resolved estimate of the yearly change in sea level from 1900 – 2011 due to the construction of reservoirs. Tide gauges record local changes in sea level, which may show the effect of reservoir construction. Because of the variability in sea level recorded in tide gauge data, only the largest, most rapid changes will be visible. These are likely to be not the result of sustained building of reservoirs over time, but rather a tide gauge’s proximity to a single large reservoir. To determine whether any of these changes are resolvable in the tide gauge record, we first calculate the maximum yearly rate of change predicted at each tide gauge site in the Permanent Service for Mean Sea Level (PSMSL) revised local reference (RLR) database (http://www.psmsl.org/data/obtaining/; Holgate et al ., 2013), shown in Fig. 5a. The largest predicted changes in sea level due to water impoundment are in northern Europe, Ghana, Malaysia, and the Saint Lawrence Seaway in Canada, all of which have tide gauges in close proximity (< 200 km) to the construction of a single large reservoir.
We begin by correcting the tide gauge data using two barotropic ocean circulation models (Piecuch et al ., 2019). These models predict sea surface height from surface wind stress and air pressure records from two forcing datasets, ERA-20C (Poli et al ., 2013) and NOAA-20CRv2 (Compo et al ., 2011), and extend throughout the 20th century. At each tide gauge location, we subtract the annual average sea surface height derived in these models at each candidate tide gauge location from the raw tide gauge data to remove the atmospheric contribution from the sea level records. We then compare the atmospheric-corrected sea level time series to our predicted changes due to historical impoundment, focusing on locations where we predict that the largest water impoundment-induced signals occurred.
The largest predicted rates of change due to impoundment occur at two tide gauges in the St. Lawrence River in Canada, Baie-Comeau and Point-au-Père/Rimouski. The filling of the Manicouagan reservoir from 1968 – 1974 is predicted to have caused a sea level rise at these sites of almost 40 mm over six years (more than 6 mm yr-1, which is more than 7 times the GMSL rise over this period, 0.8 mm yr-1; Dangendorf et al ., 2019) The expected, observed, and corrected tide gauge records are shown in Fig. 5b–c. There is too much variability in the observations to confidently detect the signal associated with the construction of the Manicouagan reservoir in the tide gauge data. In any case, the St. Lawrence River is a controlled waterway, so the impoundment signal in the record may be muted or entirely absent. Other sites were either not in operation at the right time (e.g ., Stenungsund, Sweden; Tema, Ghana; and Cendering, Malaysia). Candidate tide gauge predictions and observations are shown in Fig. S3. A more complete characterization of sea level behavior, whether through using a variety of ocean reanalysis models (Chepurin et al ., 2014) or more sophisticated modeling techniques (Piecuch et al ., 2017), may resolve an impoundment signal in tide gauge data sets.