Amanda S. Hering

Powering Up With Space-Time Wind Forecasting

The technology to harvest electricity from wind energy is now advanced enough to make entire cities powered by it a reality. High-quality, short-term forecasts of wind speed are vital to making this a more reliable energy source. Gneiting et al. (2006) have introduced a model for the average wind speed two hours ahead based on both spatial and temporal information. The forecasts produced by this model are accurate, and subject to accuracy, the predictive distribution is sharp, that is, highly concentrated around its center. However, this model is split into nonunique regimes based on the wind direction at an offsite location. This paper both generalizes and improves upon this model by treating wind direction as a circular variable and including it in the model. It is robust in many experiments, such as predicting wind at other locations. We compare this with the more common approach of modeling wind speeds and directions in the Cartesian space and use a skew-t distribution for the errors. The quality of the predictions from all of these models can be more realistically assessed with a loss measure that depends upon the power curve relating wind speed to power output. This proposed loss measure yields more insight into the true value of each model’s predictions.

Variability of trace organic chemical concentrations in raw wastewater at three distinct sewershed scales

The site-specific daily fluctuations and scale-dependent variability of influent water quality, particularly concentrations of trace organic chemicals (TOrCs), have not yet been well described. In this study, raw wastewater from three distinct sewershed scales was sampled including a centralized wastewater treatment facility in Boulder, Colorado (population ~125,000) and two decentralized wastewater catchments in Golden, Colorado (clustered system population 400, and septic system population 32). Each site was sampled hourly for 26 h and samples were subsequently analyzed in triplicate for 32 TOrCs using liquid chromatography with tandem mass spectrometry and stable isotope dilution. Detection frequency (DF) of the various TOrCs was positively correlated with sewershed size with the greatest DF of the targeted TOrCs at the Boulder site and with decreasing DF with decreasing sewershed size. Site-specific fluctuations were both scale and compound-specific. The 11 TOrCs detected greater than 75% of the time across all three sites were used to further investigate and quantify variability and to develop a statistical model to investigate the flow-dependence and time-dependence of TOrC variability. Sewershed scale was inversely correlated to variability with coefficients of variation ranging from 0.24 to 0.96, 0.39 to 2.22, and 0.32 to 3.93 for the Boulder, cluster, and septic sites, respectively. A significant linear relationship was observed between concentration and flow and concentration and the concentration at prior time points for most TOrCs at the Boulder site. This suggests less variable influent concentrations result from dispersion and mixing in the conveyance system and a larger number of discrete inputs. A notable exception was the chlorinated flame retardant TCPP, which is likely associated with a high concentration, low-flow industrial input. A significant linear relationship between flow and concentration and sequential time points was not common at the decentralized sites. Scientists and engineers developing decentralized treatment systems must consider a larger range of influent qualities, particularly with respect to TOrCs.

Modeling spatio-temporal wildfire ignition point patterns.

We analyze and model the structure of spatio-temporal wildfire ignitions in the St. Johns River Water Management District in northeastern Florida. Previous studies, based on the K-function and an assumption of homogeneity, have shown that wildfire events occur in clusters. We revisit this analysis based on an inhomogeneous K-function and argue that clustering is less important than initially thought. We also use K-cross functions to study multitype point patterns, both under homogeneity and inhomogeneity assumptions, and reach similar conclusions as above regarding the amount of clustering. Of particular interest is our finding that prescribed burns seem not to reduce significantly the occurrence of wildfires in the current or subsequent year over this large geographical region. Finally, we describe various point pattern models for the location of wildfires and investigate their adequacy by means of recent residual diagnostics.

Comparing Spatial Predictions.

Under a general loss function, we develop a hypothesis test to determine whether a significant difference in the spatial predictions produced by two competing models exists on average across the entire spatial domain of interest. The null hypothesis is that of no difference, and a spatial loss differential is created based on the observed data, the two sets of predictions, and the loss function chosen by the researcher. The test assumes only isotropy and short-range spatial dependence of the loss differential but does allow it to be non-Gaussian, non-zero-mean, and spatially correlated. Constant and nonconstant spatial trends in the loss differential are treated in two separate cases. Monte Carlo simulations illustrate the size and power properties of this test, and an example based on daily average wind speeds in Oklahoma is used for illustration. Supplemental results are available online.

Spatio-temporal hydro forecasting of multireservoir inflows for hydro-thermal scheduling

Hydro-thermal scheduling is the problem of finding an optimal dispatch of power plants in a system containing both hydro and thermal plants. Since hydro plants are able to store water over long time periods, and since future inflows are uncertain due to precipitation, the resulting multi-stage stochastic optimization problem becomes challenging to solve. Several solution methods have been developed over the past few decades to compute practically useful operation policies. One of these methods is stochastic dual dynamic programming (SDDP). SDDP poses strong restrictions on the forecasting method generating the necessary inflow scenarios. In this context, the current state-of-the-art in forecasting are periodic autoregressive (PAR) models. We present a new forecasting model for hydro inflows that incorporates spatial information, i.e., inflow information from neighboring reservoirs of the system, and that also satisfies the restrictions posed by SDDP. We benchmark our model against a PAR model that is similar to the one currently used in Brazil. Three multi-reservoir basins in Brazil serve as a case study for the comparison. We show that our approach outperforms the benchmark PAR model and present the root mean squared error (RMSE) as well as the seasonally-adjusted coefficient of efficiency (SACE) for each reservoir modeled. The overall decrease in RMSE is 8.29 percent using our approach for one month-ahead forecasts. The decrease in RMSE is achieved without additional data collection while only adding 11.8 percent more state variables for the SDDP algorithm.

Comparison of linear and nonlinear dimension reduction techniques for automated process monitoring of a decentralized wastewater treatment facility

Multivariate statistical methods for online process monitoring have been widely applied to chemical, biological, and engineered systems. While methods based on principal component analysis (PCA) are popular, more recently kernel PCA (KPCA) and locally linear embedding (LLE) have been utilized to better model nonlinear process data. Additionally, various forms of dynamic and adaptive monitoring schemes have been proposed to address time-varying features in these processes. In this analysis, we extend a common simulation study in order to account for autocorrelation and nonstationarity in process data and comprehensively compare the monitoring performances of static, dynamic, adaptive, and adaptive–dynamic versions of PCA, KPCA, and LLE. Furthermore, we evaluate a nonparametric method to set thresholds for monitoring statistics and compare results with the standard parametric approaches. We then apply these methods to real-world data collected from a decentralized wastewater treatment system during normal and abnormal operations. From the simulation study, adaptive–dynamic versions of all three methods generally improve results when the process is autocorrelated and nonstationary. In the case study, adaptive–dynamic versions of PCA, KPCA, and LLE all flag a strong system fault, but nonparametric thresholds considerably reduce the number of false alarms for all three methods under normal operating conditions.

SCaMF–RM: A Fused High-Resolution Land Cover Product of the Rocky Mountains

Land cover (LC) products, derived primarily from satellite spectral imagery, are essential inputs for environmental studies because LC is a critical driver of processes involved in hydrology, ecology, and climatology, among others. However, existing LC products each have different temporal and spatial resolutions and different LC classes that rarely provide the detail required by these studies. Using multiple existing LC products, we implement our Spatiotemporal Categorical Map Fusion (SCaMF) methodology over a large region of the Rocky Mountains (RM), encompassing sections of six states, to create a new LC product, SCaMF–RM. To do this, we must adapt SCaMF to address the prediction of LC in large space–time regions that present nonstationarities, and we add more flexibility in the LC classifications of the predicted product. SCaMF–RM is produced at two high spatial resolutions, 30 and 50 m, and a yearly frequency for the 30-year period 1983–2012. When multiple products are available in time, we illustrate how SCaMF–RM captures relevant information from the different LC products and improves upon flaws observed in other products. Future work needed includes an exhaustive validation not only of SCaMF–RM but also of all input LC products.

Building predictive models of counterinsurgent deaths using robust clustering and regression

Counterinsurgencies are conflicts where an insurgent organization conducts violence to replace or influence a recognized government. Furthering our understanding of the conditions that influence violence in different types of counterinsurgencies is important to government leaders who must deploy scarce resources efficiently. Subject matter experts (SMEs) have developed classification schemes that divide counterinsurgencies into similar groups, but no data-driven methods have ever been developed. Using the robust partitioning around medoids (PAM) algorithm, we cluster counterinsurgencies based on distances among independent variables measured on each counterinsurgency. We apply several criteria for choosing the optimal number of clusters, and then we take these groups of counterinsurgencies and build regression models for counterinsurgent deaths, an annual measure of conflict status. We evaluate these schemes using cross-validation to select the grouping whose regression models best predict counterinsurgent deaths. This approach produces a set of data-driven clusters whose predictive ability is similar to the best existing SME classification scheme, but reduces error in the assignment of a new counterinsurgency to a cluster.