Statistical & Geostatistical Forecasting Projects
Space-Time Geostatistical Analysis
INTERA conducted a space-time geostatistics analysis on rainfall time series data from historic NOAA rain gauges (1876 – 2004) and monthly NEXRAD rainfall imagery (1995-2004). The objective of this project was to develop a set of monthly rainfall simulations for the period 2007-2027 for use in transient groundwater modeling. The first step in this project involved conducting a rigorous statistical analysis of the consistency between the rain gauge and radar rainfall data. Given the change in support, the data agreed very closely. The wet and dry seasonality in the data was removed using a autocorrelation model for each of the 22,000 NEXRAD pixels and the residuals were modeled using directional variogram analysis and sequential Gaussian simulation. Long-term temporal trends in the rainfall (as identified in the rain gauge data) driven by AMO were incorporated into the 20 year rainfall simulations.
Statistical Modeling of Recovery
INTERA was an integral part of a team that conducted a hydrologic recovery forecasting project to support reducing groundwater production in order to reverse impacts on a surficial aquifer and wetlands and lakes. The project focused on identifying temporal and spatial statistical methods that had the potential to distinguish between the effects of pumpage reduction from meteorological effects (e.g., rainfall and evapotranspiration) and from anthropogenic effects (e.g., land use and drainage modifications). The variables examined included pumping rates and volumes, Surficial Aquifer System (SAS), Upper Floridan Aquifer System (UFAS), and wetland water levels, lake levels, stream flow, and meteorological data (precipitation and evapotranspiration). Anthropogenic effects other than pumping were also considered for inclusion. Given the immense amount of water level data to be analyzed as part of this project, data mining tools (e.g., Principal Component Analysis (PCA) and entropy analysis) were applied to all the observed SAS water level hydrographs in the study area to identify representative clusters in the data set. Transfer function noise (TFN) modeling was selected for the temporal statistical analysis given the correlation between the time series being modeled (e.g., rainfall, pumpage, ET). After completing the TFN modeling, the measured water levels were confirmed to lie within the 95% confidence interval of the estimate.
Forensic Statistical Reconstruction
INTERA developed explanatory statistical models to predict daily spring discharge time series for Apopka and Bugg springs from an assortment of auxiliary data that included: previously recorded spring discharge rates at the spring of interest and at adjacent springs, groundwater level measurements from adjacent monitoring wells, lake level measurements from nearby lake gages, and rainfall data from nearby gauging stations. Stepwise regression analysis was used to build multivariate linear input-output models between the response variable (spring discharge) and the independent variables (moving averages of spring discharge, water level measurements, lake levels and precipitation) at the springs of interest. Daily discharge predictions were then made for Apopka and Bugg springs as far back as 1949 and 1973 respectively, with reasonable accuracy. Flow duration curves were also generated for the two springs along with high- and low-frequency analyses for set durations (1-, 2-, 3-, 4-, 6- and 12-months) from the simulated daily spring discharge.
