Quantification of water-level variability effect on plant species populations using paleoecological and hydrological time series data

Abstract

Soil cores provide valuable data on historical changes in vegetation and hydrologic conditions. Empirical models were developed to quantify the effect of meteorological and hydrologic forcing on plant species distributions over a 110-year period in Water Conservation Area 1 (WCA1) in the Florida Everglades, also known as the Arthur R. Marshall Loxahatchee National Wildlife Refuge. Empirical models that predict plant species distributions at sites within WCA1 were developed by linking temporally sparse seed bank data from soil cores with continuous multi-decadal daily meteorological and hydrologic time series data. The meteorological data included rainfall and maximum daily temperatures that spanned the entire study period of 110 years. The hydrologic data included stage data from two gages in WCA1 established in 1954. These stage data were hindcasted to be concurrent with the meteorological data by using correlation models that fit measured stages as a function of the meteorological parameters. The historical plant species data came from seven peat cores from WCA1. Different depths from each core were carbon-dated and assayed for relative percentages of 83 plant species using pollen counts. The oldest dates were more than 1,000 years old; however, only core data that overlapped the study period were used, for a total of 67 assays among the seven cores. Twenty-three of the species had ratios of at least 5 percent for one or more of the 67 assays, hereafter referred to as the "top23". Using the assays as input vectors, the top23 were grouped using the k-means clustering into four plant classes that represented the extent to which the various species have historically appeared together. This reduced the modeling problem to one of predicting the relative ratios of the four plant classes from the hindcasted stage time-series data. A separate empirical model was developed for each class using a multi-layer perceptron artificial neural network, which provides multivariate, nonlinear curve fitting. The models predicted the relative ratios of the classes, and the sums of the predictions are near 1. The coefficient of determination (R2) of the models varied from 0.87 to 0.96, indicating that the relative ratios of the plant classes are predictable, and therefore controllable, from stage forcing. Similar soil cores are available for the Coastal Plain of North Carolina and are planned for the Congaree National Park in South Carolina.