Data to Support Analyses of Widening and Narrowing of Beach Segments in California

The first processing step consisted of preprocessing of the raw CoastSat transect data derived from the version 1.5 database provided by Vos (2025). For each transect, we removed extreme outliers, if they occurred, using a standard Hampel filter run across 5 percent subsets of the data and removing all shoreline position values that were greater than 3 standard deviations from the local mean. Monthly mean time series of the shoreline positions were then calculated from these filtered values for each CoastSat transect over the range of January 1984 to December 2024. These monthly time series records were demeaned by calculating the average shoreline position of the entire monthly record and subtracting this value from each time series value.

The second processing step consisted of integrating the demeaned monthly shoreline positions from the CoastSat transects into 329 beach segments of the California coast. Each beach segment represents a unique geographical setting along the California study area, and the geographic bounds for these beach segments were defined by the locations of features including headlands, breakwaters and groins, harbors, river mouths, piers, changes in shoreline orientation, or changes in shoreline change trends. We excluded all transect segments that included shoreline position irregularities caused by land-cast shadows, offshore rocks, coastal structures such as piers, and the opening and closing of river mouth inlets. These removed transects are identical to those removed by the seasonal analyses of Warrick and others (2025). For each beach segment, mean monthly shoreline positions were calculated from the demeaned monthly shoreline positions for all transects within the segment.

The third processing step consisted of “deseasonalizing” the monthly shoreline position records for each beach segment. This followed the seasonal, trend decomposition with LOESS (STL) techniques proposed by Cleveland and others (1990) and implemented by Warrick and others (2025) to measure shoreline seasonal cycles for the California coast. First, a locally estimated scatterplot smoothing (LOESS) fit through the monthly shoreline time series was computed to determine the time-dependent trend of each beach segment. This LOESS low-pass fit used the parameterization of Warrick and others (2025), which included a single-pass of the standard second order polynomial fit with a tri-cube weight function and a LOESS bandwidth (or “smoothing factor”) of 4 years. The LOESS low-pass was subtracted from each monthly shoreline time-series, resulting in detrended monthly shoreline time-series for each beach segment. The regular seasonal cycles of the shoreline within each beach segment were calculated by averaging all of the detrended shoreline values for each month of the year, following Warrick and others (2025). To “deseasonalize” the original shoreline positions, these computed seasonal cycles were replicated year-after-year and subtracted from the mean monthly shoreline positions. Thus, the resulting deseasonalize records removed seasonal effects but presented long-term trends.

The fourth processing step consisted of calculating mean annual deseasonalized shoreline position time series for each beach segment. The wave year was defined as November 1 to October 31, and annual mean values of the deseasonalized monthly values were computed over each of these wave years (for example, wave year 2014 is equivalent to November 1, 2013, to October 31, 2014). The standard deviations for each wave year were also computed from the available monthly deseasonalized records, and the total uncertainty for each wave year was calculated as the quadrature sum of the computed standard deviation and the mean measurement uncertainty at the annual scale, which was computed to be 2.9 meters.