Datasets for "Combining spatial information sources while accounting for systematic errors in proxies"

Contact information:
Christopher Paciorek
paciorek@alumni.cmu.edu

=======================
=== simulations.zip ===
=======================

The file simulations.zip contains the data files for the simulations. There
are two files for each of the six scenarios with the number in the
file name indicating the scenario.

The files named "obs" contain the site information for the gold
standard data, including the grid cell in which each site resides,
plus the 10 replicates of the observations.

The files named "proxy" contain grid cell information, including
covariate values, plus the 10 replicates of the proxy values and of
the true surface values. The 'land' field indicates if the cell is
land (1) or water (0).

Note that in the fitting I replaced some of the covariate values with
upper or lower bounds to reduce extrapolation issues. For the distance
to road terms, values less than 10 or greater than 500 were replaced
with those bounds.  For population density, values less than 1 were
replaced. For elevation, values less than 500 were replaced. For road
density, values less than 100 were replaced. For area emissions,
values less than 0.1 were replaced.

In the data generation process, distance to road values less than 15
were replaced, and elevation, population density, and road density
values less than 1 were replaced.

=======================
=== midAtlantic.zip ===
=======================

The file midAtlantic.zip contains the data files for the mid-Atlantic
analyses. There are two files for each of the 12 months, with the
number in the file name indicating the month.

The files named "obs" contain the site information for the PM2.5 data,
including the grid cell in which each site resides, and the monthly
average PM2.5 for each monitor. The nDays field indicates the number
of observations in the month, the pairs field indicates those
observations coming from monitors at the same site (0 for monitors
that do not share their site with another monitor), the improve field
indicates whether the monitor is from the IMPROVE network (1) or not
(0), and the cvFold field indicates which of 10 cross-validation folds
the monitor was part of.

The files named "proxy" contain grid cell information, including
covariate values, monthly average raw AOD and calibrated AOD (missing
values are coded -9999), and number of days with an AOD retrieval. The
'clearProp' field is the monthly average proportion of cloud-free
retrievals in the cell, based on the GOES cloud retrieval algorithm,
which is used as the cloud cover variable.  The 'land' field indicates
if the cell is land (1) or water (0).

In addition, the "pointEmit" file contains the locations and point
emissions for sources emitting more than 5 tons annually based on the
2002 NEI. These data are not month-specific, so there is only one
file.

The x and y coordinates in both sets of files, as well as in the
simulated data, are based on an Albers equal area projection.

===========================
=== eastern US analyses ===
===========================

The real data for the eastern US analyses, provided to me by AER with
funding from EPRI, are not available as they are proprietary.