Geostatistical Monitoring of Radiological Contamination
Geostatistical Monitoring of Radiological Contamination at the Idaho National Laboratory C. P. Oertel J. R. Giles
Background g •
INL is a US Department of Energy (DOE) facility with an operational history of reactor development and testing, fuel reprocessing, waste generation and storage, and nuclear research.
•
Main site is 890 mi2, and located 45 miles west of Idaho Falls, Idaho.
•
US DOE requires an ongoing environmental monitoring program which includes yearly monitoring of the radionuclide Cs-137.
•
Cs-137 is a fission product distributed in/on site soils due to atmospheric nuclear weapons fallout and operational activities.
Soil Radiological g Monitoring g • This is a yearly effort managed by the E i Environmental t l Monitoring M it i organization i ti • Rigorous data collection began in 2006 1984 2005, but was less • Some data exists for years 1984-2005 rigorously collected and analyzed • Field collection occurs during summer months when h roads d are clear l and d locations l ti are accessible ibl
Study Goals and General Method GOALS: 1. Identify areas of Cs-137 > 0.23 pCi/g --risk value of 1.0 E-06, the risk based concentration for a 30 year residential scenario. This is an INL specific p limit 2. Build multiyear database in order to closely monitor any trends and provide measurement efficiency METHOD: 1. Measure Cs-137 in field 2. Exploratory p y Data Analysis y 3. Use measured data to generate predicted Cs-137 at unmeasured locations 4 Use geostatistical predictions and probabilities 4. along with meteorological information to refine measurement locations
Radiological g Properties p of Cs-137 Cs-137 is radioactive, with a half-life of 30 years. It decays to Ba-137m with release of gamma radiation
Equipment q p Description p •
Standard field configuration g – N-Type HPGe detector on tripod – 60-ft. diameter uncollimated field of view – Data analysis performed using Environmental Measurements Laboratory M-1 Protocol (HASL-300)
INL POINTS
ARA Measurement Points
0
0.125
0.25
0.5 Miles
INTEC Measurement Points
Spatial p Interpolation p • It is impossible to measure enough points to construct t t full f ll surface f model d l for f Cs-137 C 137 • Spatial interpolation is the process of using measured data to predict the Cs-137 values at unmeasured locations across the site • Technique used here is kriging
Kriging g g Models • Kriging models used to get predictions of Cs-137 at unmeasured d locations l ti based b d on measured d values l • Kriging models used to get probabilities of Cs-137 Cs 137 concentrations >0.23 pCi/g • Kriging models easily generated and compared in Geostatistical Analyst
Use of kriging g g • Model chosen for this data set is a spherical f function: ti
Where h= lag dist, a = range, c=partial sill, Co =nugget
Kriging g g Basics •
Divided into 2 tasks: 1. Quantify the spatial structure of the data using variography This fits a spatial model to the data variography. 2 Use model from variography along with values 2. of measured sample points around prediction locations to generate a surface
Summary y of Past Results: 2006 • 2006: 290 locations measured. – Cs-137 mean = 0.61+/- 0.10 pCi/g • Geostatistical analysis showed large uncertainties in Cs-137 Cs 137 prediction surfaces and showed large spatial data gaps • Locations highly grouped or clustered
Summary of Past Results: 2007 • 2007: 342 locations measured – Cs-137 mean = 0.68 +/- .11 pCi/g – Geostatistical analysis included more types of kriging but still generated Cs Cs-137 137 prediction surfaces with high uncertainties. – Explored data using declustering techniques – Results still showed spatial gaps in data
Summary y of Past Results: 2008 • 2008: 342 locations measured – Cs-137 mean = 0.74 +/- 0.11 pCi/g – Clustered sampling has made it difficult to develop spatial dependence models for the purpose of conducting geostatistical estimation (mapping) of Cs-137 concentrations across the entire site site. – Advanced declustering technique used to improve variogram performance for 2006 and 2007 d data t sets. t
2008 Data Analysis y • Used nearest neighbor declustering (Olea, 2007) technique to develop better variogram parameters for use in Geostatistical Analyst • Technique based on MathCAD analysis developed by Dr. S.M. Miller, Univ. of Idaho (2008) • Compared these results to cell declustering method which is one of GA defaults. Also compared to ordinary kriging using GA defaults
2008 Data Analysis y • Using nearest neighbor declustering on 342 points resulted lt d in i subset b t off 50 unclustered l t d points i t • These points used to create variogram • Variogram parameters (sill (sill, nugget nugget, range) used in Geostatistical Analyst to generate kriged prediction surface from original 342 points • Cross validation results then compared to other surface predictions based on entire 342 points.
MODEL COMPARISON-2008 TYPE
ORDIN. ORDIN DEFAULTS
DEFAULTS
ORDIN. Nearest ORDIN Neighbor Decluster
TRANSFORM
NONE
NORMAL SCORE
NONE
MEAN PRED ERROR
0 002 0.002
0 04 0.04
0 009 0.009
0
RMS
0.54 0.99
0.65 0.64
0.44 0.03
SMALL =RMS
Mean Stand Pred Error
.0009
.063
0.22
0
RMS STD
.59
1.0
47
1
AVG SE
Disj Krig. Krig
IDEAL
Summary y •
Existing Cs-137 data was combined with geostatistical tests and models to generate prediction INL site
•
Year 2008 data was declustered using nearest neighbor declustering in order to try to improve variogram
•
Year 2008 prediction surface reflects ground truth
•
Model comparison shows that this method has high standard RMS. This means the model underestimates variability of predicted values
Background g •
INL is a US Department of Energy (DOE) facility with an operational history of reactor development and testing, fuel reprocessing, waste generation and storage, and nuclear research.
•
Main site is 890 mi2, and located 45 miles west of Idaho Falls, Idaho.
•
US DOE requires an ongoing environmental monitoring program which includes yearly monitoring of the radionuclide Cs-137.
•
Cs-137 is a fission product distributed in/on site soils due to atmospheric nuclear weapons fallout and operational activities.
Soil Radiological g Monitoring g • This is a yearly effort managed by the E i Environmental t l Monitoring M it i organization i ti • Rigorous data collection began in 2006 1984 2005, but was less • Some data exists for years 1984-2005 rigorously collected and analyzed • Field collection occurs during summer months when h roads d are clear l and d locations l ti are accessible ibl
Study Goals and General Method GOALS: 1. Identify areas of Cs-137 > 0.23 pCi/g --risk value of 1.0 E-06, the risk based concentration for a 30 year residential scenario. This is an INL specific p limit 2. Build multiyear database in order to closely monitor any trends and provide measurement efficiency METHOD: 1. Measure Cs-137 in field 2. Exploratory p y Data Analysis y 3. Use measured data to generate predicted Cs-137 at unmeasured locations 4 Use geostatistical predictions and probabilities 4. along with meteorological information to refine measurement locations
Radiological g Properties p of Cs-137 Cs-137 is radioactive, with a half-life of 30 years. It decays to Ba-137m with release of gamma radiation
Equipment q p Description p •
Standard field configuration g – N-Type HPGe detector on tripod – 60-ft. diameter uncollimated field of view – Data analysis performed using Environmental Measurements Laboratory M-1 Protocol (HASL-300)
INL POINTS
ARA Measurement Points
0
0.125
0.25
0.5 Miles
INTEC Measurement Points
Spatial p Interpolation p • It is impossible to measure enough points to construct t t full f ll surface f model d l for f Cs-137 C 137 • Spatial interpolation is the process of using measured data to predict the Cs-137 values at unmeasured locations across the site • Technique used here is kriging
Kriging g g Models • Kriging models used to get predictions of Cs-137 at unmeasured d locations l ti based b d on measured d values l • Kriging models used to get probabilities of Cs-137 Cs 137 concentrations >0.23 pCi/g • Kriging models easily generated and compared in Geostatistical Analyst
Use of kriging g g • Model chosen for this data set is a spherical f function: ti
Where h= lag dist, a = range, c=partial sill, Co =nugget
Kriging g g Basics •
Divided into 2 tasks: 1. Quantify the spatial structure of the data using variography This fits a spatial model to the data variography. 2 Use model from variography along with values 2. of measured sample points around prediction locations to generate a surface
Summary y of Past Results: 2006 • 2006: 290 locations measured. – Cs-137 mean = 0.61+/- 0.10 pCi/g • Geostatistical analysis showed large uncertainties in Cs-137 Cs 137 prediction surfaces and showed large spatial data gaps • Locations highly grouped or clustered
Summary of Past Results: 2007 • 2007: 342 locations measured – Cs-137 mean = 0.68 +/- .11 pCi/g – Geostatistical analysis included more types of kriging but still generated Cs Cs-137 137 prediction surfaces with high uncertainties. – Explored data using declustering techniques – Results still showed spatial gaps in data
Summary y of Past Results: 2008 • 2008: 342 locations measured – Cs-137 mean = 0.74 +/- 0.11 pCi/g – Clustered sampling has made it difficult to develop spatial dependence models for the purpose of conducting geostatistical estimation (mapping) of Cs-137 concentrations across the entire site site. – Advanced declustering technique used to improve variogram performance for 2006 and 2007 d data t sets. t
2008 Data Analysis y • Used nearest neighbor declustering (Olea, 2007) technique to develop better variogram parameters for use in Geostatistical Analyst • Technique based on MathCAD analysis developed by Dr. S.M. Miller, Univ. of Idaho (2008) • Compared these results to cell declustering method which is one of GA defaults. Also compared to ordinary kriging using GA defaults
2008 Data Analysis y • Using nearest neighbor declustering on 342 points resulted lt d in i subset b t off 50 unclustered l t d points i t • These points used to create variogram • Variogram parameters (sill (sill, nugget nugget, range) used in Geostatistical Analyst to generate kriged prediction surface from original 342 points • Cross validation results then compared to other surface predictions based on entire 342 points.
MODEL COMPARISON-2008 TYPE
ORDIN. ORDIN DEFAULTS
DEFAULTS
ORDIN. Nearest ORDIN Neighbor Decluster
TRANSFORM
NONE
NORMAL SCORE
NONE
MEAN PRED ERROR
0 002 0.002
0 04 0.04
0 009 0.009
0
RMS
0.54 0.99
0.65 0.64
0.44 0.03
SMALL =RMS
Mean Stand Pred Error
.0009
.063
0.22
0
RMS STD
.59
1.0
47
1
AVG SE
Disj Krig. Krig
IDEAL
Summary y •
Existing Cs-137 data was combined with geostatistical tests and models to generate prediction INL site
•
Year 2008 data was declustered using nearest neighbor declustering in order to try to improve variogram
•
Year 2008 prediction surface reflects ground truth
•
Model comparison shows that this method has high standard RMS. This means the model underestimates variability of predicted values
