Declining lake ice in response to climate change can impact spending for local communities

of multiple research teams estimating the biophysical responses of climate change [32]. Detailed methodology and adjustments of all ISIMIP models, including the Global Lake Sec- tor, are available online (https://www.isimip.org/impactmodels). These models have also been previously compared to in situ observations and found to have significant agreement between the two [33]. Historical simulations (1901–2005) represent a historical climate, whereas future projec- tions (2006–2099) consider three Representative Concentration Pathways (RCPs) representing low greenhouse gas emissions (RCP 2.6), moderate emissions (RCP 6.0), and high emissions (RCP 8.5) [34]. Climate input data to the lake models included projections of air temperatures at 2 m, wind speed at 10 m, surface solar and thermal radiation, relative humidity, total precip- itation, surface air pressure, and snowfall, all of which were available at a daily resolution. Fol- lowing the ISIMIP2b global lake sector protocol, each of these lake models were used to simulate daily lake ice cover at a 0.5˚-by-0.5˚ grid resolution, based on the mean depth and surface area of all known lakes within a 0.5˚ grid. Notably, for each 0.5˚ longitude-latitude grid, the lake models simulated the ice cover of a “representative lake”, with its size and depth determined by the distribution of all known lakes within that grid. The locations, depths, and grid-scale fractions of lakes within each 0.5˚ grid were determined by the Global Lake Data Base version 1, originally provided at a 30 arc-second resolution [35–37].

Post-processing of daily modeled lake ice cover was performed to attain homogenized ice onset, break-up, and duration values following Grant et al., [33]. Specifically, ice-cover indices were calculated with hydrological years, defined as year-long periods that contain ice onset or break-up dates. For ice onset and break-up dates, we select the October to September and Sep- tember to August hydrological years, respectively. Ice duration was computed as the sum of all ice cover days across the October to September hydrological year. Furthermore, we converted the ice off and ice on dates to be centered around January 1st as 0. Hence, dates before January 1st are negative and dates after January 1st are positive. In total, we had 16 unique lake-climate models represented by combinations of four GCMs and four lake models under three different RCPs to estimate the timing of ice-on (n = 45) and ice-off (n = 45). We excluded three model combinations with anomalously high values: Had- GEM2-ES-CLM4.5-RCP8.5, MIROC5-LAKE-RCP8.5, and GFDL-ESM2M-LAKE-RCP8.5. Estimates of lake ice duration and area We calculated the mean ice duration, change in the area of ice cover, and change in the dura- tion of ice cover, for each pixel across all of the lake models and climate scenarios. We then cal- culated the average ice duration and total area of ice cover per year across the Northern Hemisphere for each of the lake models and climate scenarios. To quantify the relative change in ice patterns, we calculated the contemporary annual average (between 1970 and 1999) of ice

area and ice duration and subtracted each yearly value from this historic value. Furthermore, we calculated a 31-year rolling mean (function rollapply, package zoo in the R computing lan- guage) to observe future patterns of ice area controlling for interannual variability. We selected a 31-year rolling average to capture variability in natural climate cycles such as the El Niño- Southern Oscillation, Pacific Decadal Oscillation, and solar cycles. Estimating the spending generated by lake ice We used two types of searches to find examples of spending on activities associated with lake ice. In the first search, we chose countries in the Northern Hemisphere. Using the search terms (lake ice) AND (revenue OR econom* OR spend*) and the name of a country in the Northern Hemisphere (e.g., Russia, Canada) to identify articles and sources that contained PLOS ONE Financial loss of declining lake ice PLOS ONE | https://doi.org/10.1371/journal.pone.0299937 July 5, 2024 3 / 15