Transcript PPT

Data Analysis
What is unique about phenology?
• Data is sparse
• Definition of many phenological events is
fuzzy
• More dependence on visual interpretation
• Need long term data for accurate analysis
• Models may need to more scale sensitive
– Include more parameters
– Data driven modeling
CS Research Issues
• Spatio-temporal data mining
• Sensor networks
– Image analysis
– Trigger event (send alerts)
– Cheap sensors
• Visual, temp, precip, soil moisture (build)
• Locate near already existing automated networks
– Upto 2 miles line of sight
– Cost a few thousand dollars
• Visualization
Dataset Requirements
• What data do we need?
– Can we get European datasets?
– Quality of Data, Quality of Information, Quality
of Knowledge
• Pedigree, provenance
• Track the sources
• Tracability (where did the data from, e.g. meat
labeling, Gerber baby food)
What might be possible with 20
years (or less) of phenological
data?
• Facilitate understanding of plant phenological
cycles and their relationship to climate
– Exploratory data analysis
• Data Mining Tools
• Spatial, temporal, spatio-temporal, integrated (plant,
insect)
– Extending these to spatio-temporal will be innovative
– Event detection in temporal datasets
• Case Based Reasoning
– Simulation tools
• What tools are out there?
• Do they have computational bottlenecks
– Visualization tools
What might be possible with 20
years (or less) of phenological
data?
• Comprehensive evaluation of satellitederived measurements
– Detecting hidden signals
• May be use data mining techniques
– Large Data Volume management and
manipulation
• High performance storage and computing
– Change Detection
– Inter-sensor calibration issues
What might be possible with 20
years (or less) of phenological
data?
• Detection of long-term phenological trends
in response to climate variability/global
warming
– Much of the work uses linear regression
models
– Assumes stationarity over time
– Change point detection (e.g El Nino became
more frequent in1980s)
– Need to break up the time into smaller slices
What might be possible with 20
years (or less) of phenological
data?
• Evaluate impacts of longer growing
seasons on pollinators, cattle, crop and
forest pests, wildfires, carbon storage, and
water use
– Regression, spatial autocorrelation
– Has space and time components
– Early spring is arriving earlier faster (second
order analysis)
Two sample problems
• Alfalfa and lady beetle
– When do we harvest alfalfa
– Need to model and match phenology of both
• Critical climate for crops
– Phenology events as critical triggers in crop
yield
Notes
• Need closer interaction between CS and
Phenology
– Need to know more about the models
• How quickly do we need answers?
– Seconds, days, months
• How do we leverage the NADSS effort
Signature Project