Recent Advances in Spatial and Spatio-Temporal Modeling

Thursday, October 31, 2019
Kirwan Hall, Mathematics Department
University of Maryland, College Park

Antonio Possolo
Chief Statistician, NIST, Gaithersburg, MD.
https://www.nist.gov/people/antonio-possolo
Time: 9:00-10:30am

Title: Spatial Statistics: Marks, Maps, and Shapes
Abstract:
Spatial statistics is concerned with phenomena unfolding in space and possibly also evolving in time, expressing a system of interactions whereby an observation made at a (spatio-temporal) location is informative about observations made at other locations. In general, the interactions are best described probabilistically, rather than deterministically. The spatial scales involved range from the microscopic (for example, when describing interactions between molecules of a liquid) to planetary (for example, when studying the earth's ozone layer) or even larger; the temporal scales are similarly varied.
Marks indicate objects whose spatial locations are influenced by the presence and nature of other objects nearby: trees of the same or different species in a grove, molecules in a liquid, asteroids in the solar system, or galaxies throughout the universe. The statistical models are (marked) spatial point processes.
Maps describe the variability of the values of a property, typically observed at a set of locations that may be distributed regularly or irregularly throughout a 2D or 3D spatial domain. The Ising model of ferromagnetism describes collective properties of atoms arranged in a regular lattice. When mapping the prevalence or the incidence of a disease at the level of counties or parishes, the observations are associated with subsets of a region whose spatial relations are meaningful. Many maps describe how a property varies from point to point across a region, and are drawn based on observations made at a nite of points scattered throughout the region: for example, the mass fraction of uranium in soils surface sediments across Colorado. Gaussian random functions are a model of choice for such quantities, possibly after re-expression.
Shapes arise owing to modulated interactions between surface elements anchored to points in space { "generators" in the nomenclature of Ulf Grenander's pattern theory. Probability distributions on spaces of generators and on spaces of interactions between them can then be used to describe variations on patterns,and to 't' shape models. 
 
Bio: Antonio Possolo is a NIST Fellow and the Chief Statistician for NIST. Previously he was Chief of the Statistical Engineering Division of NIST, and has had professional engagements in academia (Princeton Univ., Univ. of Washington in Seattle, and Georgetown Univ.), and in industry (Boeing and General Electric). He has been practicing the statistical arts for about 40 years now. Antonio chairs the "Statistics and Uncertainty" Working Group of the Inter- American System of Metrology, is a member of Working Group 1 of the Joint Committee for Guides in Metrology, and also a member of the Commission on Isotopic Abundances and Atomic Weights, of the International Union of Pure and Applied Chemistry.

 

Rick Mueller
USDA/NASS

Time: 10:45-11:45am

Title: An Update on the Annual National Cropland Data Layer Program
Abstract: The Cropland Data Layer (CDL) is a national land cover product produced by the US Department of Agriculture (USDA), National Agricultural Statistics Service (NASS) to assess planted crop acreage annually. The 2018 CDL product, released in February 2019 serves as the eleventh consecutive national mapping of conterminous US agriculture. The CDL is a 30-meter agricultural-specifc national land cover product that is released into the public domain upon completion of the growing season via the CropScape portal at https://nassgeodata.gmu.edu/CropScape. The annualized CDL product provides users the opportunity to study cultivation practices, crop rotations and intensifcation, and the changing trends and localities in production agriculture. The CDL is a supervised land-cover classifcation utilizing a decision tree machine learning approach using optical satellites while leveraging ground reference data from the USDA/Farm Service Agency and the National Landcover Database (NLCD) from the US Geological Survey, and multiple cooperative industry partnerships. Medium resolution satellites such as Landsat 8, Disaster Monitoring Constellation Deimos-1 and UK2, Resourcesat-2 LISS-III, and Sentinel-2 were used to monitor agricultural production throughout the North American 2018 growing season. This talk focuses on the continuous CDL product improvement process, including leveraging cooperative industry partnerships, obtaining improved remote sensing data, improving the methods used in classifcation, and leveraging historical data to improve identifcation of specialty and locally grown crops.

Bio: Rick Mueller is Head of the USDA NASS Research and Development Division/Spatial Analysis Research Section in Washington, DC. He is responsible for agricultural monitoring of domestic crop area, yield, crop condition, and disaster monitoring programs using remote sensing and geospatial analysis methods for public dissemination. His group developed the online geospatial visualization portals called CropScape that depicts the national land cover dataset called the Cropland Data Layer, VegScape showing crop vegetative condition, and is developing a soil moisture monitoring portal. Rick received a B.S. degree in Geography from the University of Maryland and an M.S. in Business from Johns Hopkins University.

 

Victor De Oliveira
The University of Texas at San Antonio

Time: 1:00-2:00pm

Title: Gaussian Copula Models for Geostatistical Count Data
Abstract: We describe a class of random field models for geostatistical count data based on Gaussian copulas. Unlike hierarchical Poisson models often used to describe this type of data, Gaussian copula models allow a more direct modeling of the marginal distributions and association structure of the count data. We study in detail the correlation structure of these random fields when the family of marginal distributions is either negative binomial or zero-inflated Poisson; these represent two types of overdispersion often encountered in geostatistical count data. We also contrast the correlation structure of one of these Gaussian copula models with that of a hierarchical Poisson model having the same family of marginal distributions. We also describe the computation of maximum like-lihood estimators which is a computationally challenging task. Finally, a data analysis of Lansing Woods tree counts is used to illustrate the methods.

Bio: Victor De Oliveira is a professor in the Department of Management Science and Statistics, College of Business, University of Texas at San Antonio. He joined the UTSA faculty in 2006 and previously worked at the University of Arkansas and Simon Bolivar University. He holds a Ph.D. in statistics from the University of Maryland, and a master's in water resources and a bachelor's
in mathematics from the Universidad Simon Bolivar. He teaches a variety of undergraduate and graduate courses in Statistics and Applied Probability.

Claire Boryan
USDA/NASS

Time: 2:15-3:15pm

Title: Operational Agricultural Flood Monitoing with Copernicus Sentiel-1 Synthetic Aperture Radar
Abstract
: Agricultural Flood monitoring is important for food security and economic stability and is of signifcant interest for agricultural policy makers and decision support. In agricultural remote sensing applications, optical sensor data are traditionally used for acreage, yield, and crop condition assessments. However, optical data are affected by clouds, rain, and darkness, which limit their utility to monitor and estimate the extent of flooding during disaster in a timely manner. Synthetic Aperture Radar, however, can penetrate cloud cover and acquire imagery day or night, which makes it particularly useful for flood disaster monitoring in near-real time. A flood detection method was implemented in 2017 using freely available Copernicus Sentinel-1 Synthetic Aperture Radar data for operational agricultural flood monitoring in the United States. The data were used operationally in near-real time to identify and map flooding of agriculture during major hurricanes in 2017, 2018 and 2019. This presentation describes 1) the agricultural flood monitoring method that utilizes Copernicus Sentinel-1 Synthetic Aperture Radar and the United States Department of Agriculture National Agricultural Statistics Service 2018 Cultivated Layer and 2018 Cropland Data Layers and 2) inundated cropland and pasture maps and acreage estimates. This flood monitoring method based on Synthetic Aperture Radar data and National Agricultural Statistics Service geospatial data is effective, effcient, and affordable for operational disaster assessment. Further, flood assessment maps,flood inundation raster data, and a method paper are disseminated to the public on the NASS Disaster Analysis website at https://www.nass.usda.gov/ResearchandScience/DisasterAnalysis/index.php

Bio: Claire Boryan is a Senior Geographer with the Research and Development Division of the USDA/National Agricultural Statistics Service (NASS). She received a BA from the University of Virginia, a MS in Geographic and Cartographic Sciences and PhD in Earth Systems and Geoformation Sciences from George Mason University. She has extensive experience in agricultural geospatial analysis and remote sensing research. Her research interests include: using Synthetic Aperture Radar for agricultural applications, agricultural disaster analysis, remote sensing methods, geographic information science and applied research using geospatial data to improve agricultural statistics.

 

Luca Sartore
USDA/NASS

Time: 3:30-4:30

Title: Predicting Crop Yield Using Spatio-Temporal Functional Covariates
Abstract: The USDA's National Agricultural Statistics Service (NASS) produces annual yield estimates for major crops at national, state, agricultural district, and county levels. Several surveys are conducted to produce reliable estimates that are further enhanced by incorporating remote sensing measurements at the county level. These measurements consist of Moderate Resolution Imaging Spectroradiometer (MODIS) data based on multispectral composites and Land Surface Temperature (LST) based on thermal composites. The data are summarized by empirical density functions for crop regions throughout the growing season. Corn-yield predictions at the county level are then produced using non-parametric models that combine spatial coordinates with satellite data. Machine learning algorithms are compared when processing additional information based on Kullback-Leibler distances.

 

Bio: Luca Satore is a NISS postdoctoral fellow, working at the Research and Development Division, National Agricultural Statistical Service, USDA. He received his management Bachelor of Science degree in statistics and computer science, and master degree in business statistics from the Ca' Foscari University of Venice. He got his Ph.D. in statistical sciences from the University of Padua. Prior to NISS, he was a research fellow at the European Centre for Living Technology. He is a member of the American Statistical Association, and his research focuses primarily on non-standard regression techniques and spatio-temporal models.
Abstract: Spatial statistics is concerned with phenomena unfolding in space
and possibly also evolving in time, expressing a system of interactions whereby
an observation made at a (spatio-temporal) location is informative about ob-
servations made at other locations. In general, the interactions are best de-
scribed probabilistically, rather than deterministically. The spatial scales in-
volved range from the microscopic (for example, when describing interactions
between molecules of a liquid) to planetary (for example, when studying the
earth's ozone layer) or even larger; the temporal scales are similarly varied.
Marks indicate objects whose spatial locations are in uenced by the presence
and nature of other objects nearby: trees of the same or di erent species in a
grove, molecules in a liquid, asteroids in the solar system, or galaxies through-
out the universe. The statistical models are (marked) spatial point processes.
Maps describe the variability of the values of a property, typically observed
at a set of locations that may be distributed regularly or irregularly through-
out a 2D or 3D spatial domain. The Ising model of ferromagnetism describes
collective properties of atoms arranged in a regular lattice. When mapping the
prevalence or the incidence of a disease at the level of counties or parishes, the
observations are associated with subsets of a region whose spatial relations are
meaningful. Many maps describe how a property varies from point to point
across a region, and are drawn based on observations made at a nite set of
points scattered throughout the region: for example, the mass fraction of ura-
nium in soils surface sediments across Colorado. Gaussian random functions
are a model of choice for such quantities, possibly after re-expression.
Shapes arise owing to modulated interactions between surface elements anchored
to points in space { "generators" in the nomenclature of Ulf Grenander's pat-
tern theory. Probability distributions on spaces of generators and on spaces of
interactions between them can then be used to describe variations on patterns,
and to t shape models.
Bio: Antonio Possolo is a NIST Fellow and the Chief Statistician for NIST.
Previously he was Chief of the Statistical Engineering Division of NIST, and has
had professional engagements in academia (Princeton Univ., Univ. of Washing-
ton in Seattle, and Georgetown Univ.), and in industry (Boeing and General
Electric). He has been practicing the statistical arts for about 40 years now.
Antonio chairs the \Statistics and Uncertainty" Working Group of the Inter-
American System of Metrology, is a member of Working Group 1 of the Joint
Committee for Guides in Metrology, and also a member of the Commission on
Isotopic Abundances and Atomic Weights, of the International Union of Pure
and Applied Chemistry.