Understanding Global Change from Data

(1)

Understanding Global Change

from Data

Karsten Steinhaeuser

University of Minnesota

[email protected]

www.umn.edu/~ksteinha

(2)

Global Change:

A Defining Issue of our Era

(3)

Global Change:

A Defining Issue of our Era

(4)

Global Change:

A Defining Issue of our Era

Industrialization & Modernization

Population Growth & Demographic Shifts

(5)

Global Change:

A Defining Issue of our Era

Land Use Change

Urbanization Deforestation

Industrialization & Modernization Population Growth &

(6)

Global Change:

A Defining Issue of our Era

Population Growth & Demographic Shifts Industrialization & Modernization Climate Change Land Use Change

(7)

Global Change:

A Defining Issue of our Era

Biodiversity Loss & Natural Disasters

Monoculture Ocean _{Acidification}

Demographic Shifts Land Use Change Climate Change Fires Cyclones

(8)

Global Change:

A Defining Issue of our Era

Natural Disasters

THIS IS GLOBAL CHANGE

Biodiversity

Loss

Demographic Shifts

Land Use Change

Climate Change

(9)

Responding to Societal Needs

• Where is population growth putting pressure on urban

infrastructure and natural resources?

• What is the interplay between the global climate system,

local ecosystems and natural disasters?

• How does increased biofuel production impact crop

patterns and food availability?

• How do changing oceans affect the atmosphere and

land climate?

• What are the major feedback mechanisms among

eco-climatic processes?

(10)

• Climate Models • Reanalysis Data • River Discharge • Agricultural Statistics • Population Data • Air Quality • …

Transformation: Data-Poor to Data-Rich

• Satellite Data – Spectral Reflectance – Elevation Models – Nighttime Lights – Aerosols • Oceanographic Data – Temperature – Salinity – Circulation

“The future of science depends […] on cleverness being

applied to data for their own sake, complementing

scientific hypotheses as a basis for exploring today’s information cornucopia.”

(11)

Global Change is a Big Data Problem

 Scale and nature of the data offer numerous

challenges and opportunities for research in the computational analysis of large datasets.

 Data-driven discovery methods hold great

promise for advancing our understanding of the climate and ecosystem processes contributing to global change.

 Advances are of scientific importance and

societal relevance.

"data-intensive science [is] so different that

it is worth distinguishing [it] … as a new,

(12)

Computational & Data Challenges

• Spatio-temporal, non-i.i.d., non-stationary, heteroskedastic

• High-dimensionality, massive dataset size, increasing complexity • Noise, measurement error, missing values

• Multi-scale, multi-resolution processes • Long-range spatial dependencies

• Long memory temporal dependencies • Low-frequency variability

• Heterogeneity, multiple data sources • . . .

(13)

Active Research Projects

• GOPHER: Global Observatory for Planetary Health

and Resources

Project Aim: Monitoring of global ecosystem for

changes in land cover, land use, etc.

• NSF Expeditions: Understanding Climate Change –

A Data Driven Approach

Project Aim: Develop novel data analysis

methods to help improve understanding

and prediction of climate change

NSF Expeditions

(14)

GOPHER: Ecosystem Monitoring

What is the current state of the global

forest ecosystems and how are they

changing as a result of logging and natural disasters?

How are the demands of a growing population affecting agriculture, e.g., creation of new farmland, changings in cropping patterns, conversion to

biofuels, etc.?

How is urbanization affecting the surrounding ecosystem resources and water supply?

(15)

Traditional Approach for Change Detection

• Requires high-quality imagery

– Available infrequently

• Requires high resolution

– No global coverage

• Requires training data

– Must be created manually

– Labor-intensive, time-consuming, expensive

Image-to-Image Comparison

 Studies are limited to small regions and unable to identify change point or rate of change

(16)

Alternate Approach: Spatio-Temporal

Multi-Spectral Data

• Provides global coverage daily • (Relatively) coarse resolution • Sometimes poor quality

– Noisy

– Missing Data

Trade-Off

lower spatial vs. higher frequency, resolution increased coverage

 opportunities and challenges for

spatio-temporal data mining

(17)

Time Series Change Detection

What happened

here?

Sudden vegetation

loss could indicate

forest fire, logging, flooding, …? Recovery period Gradual recovery suggest a fire is likely possibility.

(18)

Time Series Change Detection

Did a change happen here?

Data can be noisy (measurement error, interference, etc.) or

missing (e.g., due to

cloud cover or snow). How much information is required to identify a change?

(19)

Time Series Change Detection

Is there a decrease in vegetation?

Changes can also be

gradual, occurring

over multiple years… How to distinguish this

subtle change from

the dominant signal of seasonal phenology?

(20)

Time Series Change Detection

And what if there is

high variability

instead of a clear seasonal cycle? How does data

quality affect

change detection? Change or

natural variability?

(21)

Time Series Change Detection

(22)

Time Series Change Detection

(23)

Time Series Change Detection

(24)

Novel Change Detection Techniques

Current methods are

not adequate

to address these challenges. We focus

on developing algorithms that are:

• Robust to missing data, noise and

outliers

• Able to automatically characterize

different types of changes

• Capable of incremental update

and (near) real-time detection

• Aware of spatial context

Segmentation Approaches:

Divide time series into pieces and determine if a change occurred

Before After

Model + Predict

Prediction-Based Methods:

Build model of the “normal” behavior and predict, measure deviation

(25)

ALERTS: Automated Land change Evaluation,

Reporting and Tracking System

• Planetary Information System

for interactive investigation of ecosystem disturbances discovered by GOPHER – Forest Fires – Deforestation – Droughts – Urbanization – …

• Helps quantify carbon impact of changes, understand the relationship between climate variability and human activity • Provides ubiquitous web-

based access to changes

occurring across the globe, creating public awareness

(26)

(27)

(28)

Illustrative Examples

Large forest fires in Canada have converted the forests from a sink into source of carbon in the atmosphere.

Logging is legal

in some parts of Canada, further reducing carbon sequestration

Brazil Accounts for almost 50% of all

humid tropical forest clearing, nearly 4 times that of the next highest country.

(29)

Illustrative Examples

Examples of afforestation can be seen in several areas around the world, including this region near Beijing (China) where new trees have been planted to prevent dust storms and erosion.

One winter the Ob River caused massive flooding due to freezing of the Bay of Ob / Kara Sea.

Hurricane Katrina caused significant damage

and vegetation loss along the US Gulf Coast.

Political conflict and the ensuing “land reform” resulted in wide-spread farm abandonment and loss of

(30)

Impact on REDD+

“The [Peru] government needs to spend more than $100m a year on high-resolution satellite pictures of its billions of trees.

But … a computing facility

developed by the Planetary Skin Institute (PSI) … might help cut that budget.”

“ALERTS, which was launched at Cancún, uses … data-mining algorithms developed at the

University of Minnesota and a

lot of computing power … to spot places where land use changed.”

(31)

Understanding Climate Change:

A Data Driven Approach

• 5-year / $10M NSF Expeditions in Computing

• Team led by UMN, consists of 15 senior personnel

and ~50 students and post-docs

• Developing state of the art computational methods

(32)

Climate Change is Real…

•

The planet is warming

•

Multiple lines of evidence

•

Credible link to human GHG (green house gas) emissions

•

Consequences can be dire

•

The potential societal cost of both action and inaction is large

•

There is an urgency to act

•

Adaptation: “Manage the unavoidable”

(33)

…but Understanding is Limited

Cell

Clouds Land

Ocean

Much of what we know is derived from computer simulations of general circulation models

(mathematical equations describing the physical processes involved in climate)

“The sad truth of climate science is that the most crucial information is the

least reliable” (Nature, 2010)

Physics-based models are essential but not adequate

• Relatively reliable for projections at global scale for

smooth fields such as temperature, pressure

• Less reliable for variables that are crucial for impact

(34)

Expeditions Project Highlights

High-Performance Data Analytics Hurricane Intensity Prediction

and Land-Fall Modeling Climate Extremes and Uncertainty

Teleconnections & Sparse Predictive Modeling

-0.4 -0.2 0 0.2 0.4 0.6 0.8 lat itud e

Correlation Between ANOM 1+2 and Land Temp (>0.2)

90 60 30 0 -30 El Nino Events Nino 1+2 Index

(35)

Dipoles & Teleconnections

Crucial for understanding the climate system…

Correlation of Land temperature

anomalies with NAO Correlation of Land temperature _{anomalies with SOI}

SOI dominates tropical climate variability,

affecting rainfall patterns over East Asia, Australia & South America, as well as droughts over North America and Africa, among others.

NAO influences sea level pressure (SLP) over

most of the Northern Hemisphere. Strong positive phases (Islandic Low/Azores High) are associated with above-avg. temperatures in the Eastern US.

(36)

Automatic Dipole Discovery

Nodes in the graph correspond to grid points, edges weighted

with correlation between anomaly time series.

Graph-based representation of the global climate system

Shared Reciprocal Nearest Neighbors (SRNN) Density

Overcomes limitations of traditional approaches such as EOF analysis

• Capable of identifying dynamic dipoles • Discovers dipoles of different strength

(not dominated by leading signals) • Comprehensive global data analysis,

(37)

Focus on Climate Extremes: Temp.

The World Climate Research Program (WCRP) ranked

“Prediction of Extremes” as the top Grand Science Challenge.

Shifted Mean

Increased Variability

Changed Shape

• Expect statistically

significantly higher trends in warming with trend of increasing heat waves

• Larger uncertainty and

regional variability than previously thought

• Even with a general warming

trend, intensity and duration of cold extremes will persist

• Certain regions exhibit

consistent cold extremes despite uncertainties

(38)

The uncertainty in projections grows with increasing precision and particularly over the tropics.

Precipitation Extremes

No increasing trends

in rainfall extremes in India during last half-century

Trend of 30-year return levels Trend of 100-year return levels

Credit: Ganguly, Ghosh, Kao, Kodra, et al.

Significant growth in intensity and frequency of rainfall at aggregate scales in the extratropics

(39)

Drought Detection

•

Major Droughts

•

Persistent over space and time

•

Catastrophic consequences

•

Examples

•

1930s North American Dust Bowl

•

Late 1960s Sahel drought

•

Dataset: Climate Research Unit

•

http://data.giss.nasa.gov/precip_cru/

•

Monthly precipitation over land

•

Duration: 1901 to 2006

•

Resolution: 0.5◦_{latitude x 0.5}◦_longitude

(40)

(Sparse) Predictive Modeling

Sparse High-Dimensional Regression

•

Understanding extremes: atmospheric covariates to rainfall extremes

•

Understanding teleconnections:

long-range dependencies in space/time

•

Understanding hurricanes: ocean properties, trends

•

Multi-model ensembles: skill selection, regional processes vs global average

•

Statistical downscaling: coarse to fine scale, capture dependencies

OLS: High RMSE, High Complexity

(41)

Hurricane Track Forecasting

Physics-based Models

What if the error gets

interpolated to 10-15 day in advance forecast?

~500 km

Improving but have mean error >185km beyond 48 h

(A)

HURDAT Historic Data

(42)

Extreme Event Forecasting via

Contrast-based Network Motive Discovery

SLP SST VWS

(B)

(C) (D)

Phase: Land Phase: Curve

(E)

(F)

Intuition: If an extreme event (e.g. hurricane track) is in one of its key phases (e.g. land-hitting), then there exist network motifs (recurrent patterns in climate networks) that are

specific to that phase.

(43)

Scaling I/O and Analytics

• Global Cloud Resolving Model (GCRM)

– Simulate circulation associated with large convective clouds

– Developed by David Randall (Colorado State U) and Karen Schuchardt (PNNL)

• Geodesic grid model

• 1.4 PB data per simulation

– 4 km resolution, 3 hourly, 1 simulated year – 1.5 TB per checkpoint

• Parallel NetCDF I/O library

I/O was previously a major bottleneck: The only reason execution at this scale became possible was due to I/O scaling.

(44)

Illustrative Result: Parallel NetCDF

• Improved I/O throughput

– Using PnetCDF optimizations, massive scalability

– For 3.5 km grid resolution, grid size is 41.9M cells with 256 vertical layers – Data analysis read and simulation checkpoint

Credit: Choudhary, et al.

(45)

Thank You! Questions?

UMN Team Members

Vipin Kumar, Arindam Banerjee, Shyam Boriah, Snigdhansu Chatterjee, Jonathan Foley, Joseph Knight, Stefan Liess, Shashi Shekhar, Peter Snyder, Michael Steinbach, Karsten Steinhaeuser UMN Students

Graduate: Ivan Brugere, Yashu Chamber, Xi Chen, James Faghmous, Jaya Kawale, Arjun Kumar, Michael Lau, Varun Mithal; Undergraduate: Kelly Cutler, Ryan Haasken, Zachary O’Connor, Dominick Ormsby

PSI Collaborators NSF Expeditions Collaborators

NASA Ames: Christopher Potter NCSU: Nagiza Samatova, Fredrick Semazzi Cal State Monterey Bay: Stephen Klooster Northeastern: Auroop Ganguly

Michigan State: Pang-Ning Tan Northwestern: Alok Choudhary, Wei-keng Liao PSI: Juan Carlos Castilla-Rubio North Carolina A&T: Abdollah Homaifar

website: gopher.cs.umn.edu website: climatechange.cs.umn.edu