Air Pollution in the Southern San Joaquin

The basic data are ARB measurements of multiple pollutants over time at about 10 monitoring stations in the Southern San Joaquin. Data are rather sparse, because some stations are only on line for short periods, others provide measurements over 20 years. And some monitoring stations only look at a few pollutants, while others measure many more. We will try imputation techniques to estimate missing information, use filtering and interpolation techniques, and look at other space-time methods that have been used in the literature.

Ambulance Response Times in Southern California

The data are ambulance response times to 911 calls in Santa Barbara, Kern, Los Angeles, and Ventura Counties. There is both location and
time information in the data sets because we know where and when the call originates. We will try to predict ambulance response times using temporal and spatial information, as well as call-characteristics, and present the results in GIS maps. Geographically Weighted Regression and point process models will be used.

Car and Truck Traffic in the I-5 Corridor

We study car and truck traffic in the I-5 corridor from the intersection with SR-14 in northern Los Angeles County to the I-5/SR-99 split in southern Kern County. Aggregated and smoothed data from Caltrans are used (average annual daily traffic and average annual
daily truck traffic) at about 20 intersections over up to 50 years. Hourly loop counts are available over one full year at 4 locations.
Imputation, time series, and repeated measures techniques will be used to analyze the data. Graphical representations of various sorts will be presented.

Data Fusion for Satellite Remote Sensing

A fundamental problem in the analysis of satellite remote sensing data is that of “merging” two data sets which provide complementary information. Usually, the two data sets will have different observation grids: the pixel footprints are non-nested, of different resolutions, and are oriented differently. The most common ad-hoc solution is to either aggregate the data up to a coarse, common resolution, or to “match-up” pixel centers. In this project, we seek to develop a principled methodology for fusing (merging) the data into a statistically consistent data set suitable for both exploratory and inferential science analysis.

Extreme Waves

Predicting extreme waves is vital for issuing warnings and saving people’s lives. For this study, we will be using remote sensing data from the satellite Jason-1. The goals of the project are to: 1) determine the locations of unusually high waves either in the ocean or errors in the data and; 2) validate the data received from Jason-1 with other satellites.

Fire Hazard Estimation Using Point Process Methods

The goal of this project is to develop and assess spatial-temporal models for Los Angeles County wildfires. Currently, fire department officials rely heavily on the Burning Index (BI), which uses weather data recorded at various Remote Automatic Weather Stations, and blends these records together using a complex non-linear dynamic model in order to output a number that is often interpreted as a measure of overall wildfire hazard. However, the BI turns out not to be an effective predictor of wildfire activity in Los Angeles County. We are exploring alternatives, including simple point process models using the same weather variables as those used in the computation of the BI, as well as the history of prior local wildfire activity.


This project concerns the use of multispectral remote sensing observations obtained from two different aircraft instruments flying linear transects through hurricanes, HAMSR (the High MIMC Sounding Radiometer) views a swath some 30km wide along the flight path in 25 spectral channels. EDOP (the ER-2 Doppler Radar) sees a narrow path directly under the aircraft in 15 channels. EDOP’s observations are strongly related to rainfall and atmospheric turbulence, and our goal is to infer what EDOP would have seen had it seen the wider HAMSR swath. We are using non-linear multivariate analysis methods based on Gifi (1990) to model the relationship between HAMSR and EDOP directly under the aircraft, and spatial-statistical techniques to transfer that relationship away from the nadir (directly downward looking) view.

Statistical Evaluation of Earthquake Occurrance Data using Point Process Techniques

This project focuses on space-time marked point process models for California seismicity, such as the widely-used Epidemic-Type Aftershock Sequence (ETAS) models developed by Yosihiko Ogata. Our aim is to develop methods for evaluating the fit of such models to data, and to suggest improvements and refinements to the models currently in use. In particular, detailed data on seismic moment tensors have been available for most earthquakes observed in California for several years, yet other than the scalar seismic moment, which is a crude summary essentially ignoring all directional information in the tensor, the moment tensor is typically not used in ETAS models. We are exploring ways of incorporating this potentially crucial information in order to improve models and forecasts of future seismicity.

Statistics for Planetary Science

Multispectral imagery of Jupiter and Saturn provides a wealth of information about the structure of the atmospheres of these gas giants. Different spectral channels penetrate to different depths of their atmospheres and capture various compositional features. Principle component analysis has been a workhorse of Planetary Science because it can reveal latent structures not visible in the original images. We seek to generalize and extend this idea by applying other linear and non-linear techniques. Data come from the orbiters Galileo (Jupiter) and Cassini (Saturn), from the Hubble Space Telescope, and from ground based telescopes.