SPATIOTEMPORAL WEIGHTED REGRESSION

DEPARTMENT OF COMPUTER SCIENCE

UNIVERSITY OF IDAHO

Chao Ma, Xiang Que, Marshall Xiaogang Ma

IDEA Lab

OUTLINE-STWR

• MOTIVATION

• METHOD

• CASE STUDY• Simulated Data

• Real-world Data

• CONCLUSIONS

1

MOTIVATION

Geographic processes of temporal and spatial data Interpolation / out-of-sample prediction NO Ordinary Least Squares regression (OLS)

2

2010 2015 2019

METHODSpatiotemporal Weighted RegressionWeight ∝ time and space distance

3

β0 β1 β2 Y_true

4

CASE STUDY

SIMULATED DATA

X1 X2

RESULTS-1:

R2, ESTIMATED STANDARD ERROR

5

RESULTS-2 ERROR OF PREDICTION

6

EXPERIMENT WITH REAL-WORLD DATAprecipitation δ2H isotope map

0 1 2 3i iy ppt tmean height = + + + +

Daily average δ2H from three days (Oct.29 – 31, 2012)

Data source of water isotopes δ2Hhttp://wateriso.utah.edu/waterisotopes/pages/spatial_db/SPATIAL_DB.html

http://www.prism.oregonstate.eduhttps://topotools.cr.usgs.gov/gmted_viewer/viewer.htm

7

CASE STUDY: REAL-WORLD DATAResults of model performance with real-world dataprecipitation δ2H isotope map

8

CASE STUDY: REAL-WORLD DATAComparison Results of precipitation δ2H

9

(a) (b)

CASE STUDY: REAL-WORLD DATAComparison Results of precipitation δ2H

10

CONCLUSIONS

➢ STWR is a powerful tool to analyze geographic processes and interpolations of spatiotemporal data

➢ To do 1: Predications for future time stages

➢ To do 2: Apply STWR to different disciplines. E.g. house prices, environmental science, biology, geology …

11

Parallel computing for large scale spatiotemporal data

12House prices of 204611 observations from 19 years

Marshall Mamax@uidaho.edu

Xiang Quexiangq@uidaho.eduquexiang@fafu.edu.cn

Chao Machao@uidaho.edu

CONTACTS

13

IDEA Lab