Identification of land-use and land-cover changes in East-Asia

Masayuki Tamura, Jin Chen, Hiroya Yamano, and Hiroto Shimazaki

National Institute for Environmental Studies

Objectives

To develop a robust and reliable algorithm for detecting land use/cover changes using coarse spatial resolution data (MODIS, NOAA/AVHRR, SPOT/VEGETATION).

To analyze land use/cover changes in China during 1982-1999 using the Pathfinder 8km NDVI and climate data.

1) Satellite data: Pathfinder 8km NDVI data Spatial resolution: 8 x 8 km Temporal resolution: 10-day. 20 years of data (198

1-2000). Preprocessing

2) Climate data: China National Meteorological Bureau. 620 meteorological stations 10-day mean temperatures and precipitations

from 1980-1999. Preprocessing

Data Sources

NDVI Data Preprocessing

NDVINoises caused by cloud

BISE(Best Index Slope Extraction)

Kriging Interpolation

Climate Data Interpolation

Meteorological Stations

Temperature

Precipitation

NDVI profile differences are used to detect land cover changes between two years.

Normalization and correction of NDVI data Calibration of sensor degradation. Atmospheric correction Normalization of climate conditions (T, P)

Method

otherwTo NDVINDVI *

An observed NDVI (NDVIo) for a pixel can be expressed as:

where NDVI * is a potential NDVI in an optimum climate condition. ɛT, and ɛW account for the effects of temperature and precipitation differences from the optimum conditions respectively. ɛother accounts for the effects of sensor degradation and atmospheric condition changes.

Land cover change detection should be performed by comparing NDVI* differences between two years rather than NDVIo directly.

Normalization for Climate Conditions

otherwToNDVINDVI

111*

ɛother can be moved off through pre-processing of original NDVI data, which includes sensor calibration, atmospheric correction and cloud filter. ɛT, ɛW can be estimated according to the relationship between vegetation growth and seasonal climate condition.

ɛT Estimation

ɛT reflects the concept that plant growth is depressed when plant is growing at a temperature displaced from its optimum temperature. According to existing study (Potter, 1993; Hamlyn G. Jones, 1992) , ɛT has an asymmetric bell shape that falls off more quickly at high than at low temperature.

-0.2

0

0.2

0.4

0.6

0.8

1

0 5 10 15 20 25 30 35 40

Temperature

Epis

on T

Topt

4

422

)()()()(2

BTBTBTBT

opt

optT

(Hamlyn G. Jones, 1992)

Topt is optimum temperature, defined as the air temperature when the NDVI reaches its maximum for a long period.

ɛw Estimation

ɛw 　 describes the effect of water stress to plant growth. By considering the lag effect of precipitation, it is calculated by

When Sum (PPT) < Sum (PET)

When Sum (PPT) > Sum (PET)

 

where PET is potential evapotranspiration and determined by Thornthwaite method, PPT is precipitation for calculating period.

22

/5.04.0j

jmm

j

jmmw PETPPT

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2Ep

ison

W1w

12 3

NDVI dataset in 1983

Change Vector Calculation

NDVI(n)NDVI(2)•••

NDVI(1)

NDVI(n)NDVI(n)NDVI(n)NDVI(2)•••NDVI(2)•••NDVI(2)••••••••

NDVI(1)NDVI(1)

NDVI(n)NDVI(2)•••

NDVI(1)

NDVI(n)NDVI(n)NDVI(n)NDVI(2)•••NDVI(2)•••NDVI(2)••••••••

NDVI(1)NDVI(1)

NDVI(n)NDVI(2)•••

NDVI(1)

NDVI(n)NDVI(n)NDVI(n)NDVI(2)•••NDVI(2)•••NDVI(2)••••••••

NDVI(1)NDVI(1)

NDVI dataset in 1984

NDVI dataset in 1999

……Threshold Applying

Time Series Filtering

Change Pixels

Change Pixel Detection Flow

Base Dataset

Change pixels during 1984-1988

Change pixels during 1989-1993

Change pixels during 1994-1997

Chang Pixels in Different Periods

NDVI Decreasing Trend

NDVI Increasing Trend

Change Pixel Distributions with Different Trends

Special Modification by Forest Fire

No Trend

Grassland MonitoringGrassland Monitoring

Xilinhot

Haibei

Wetland MonitoringHabitats of Red-Crowned Cranes and Oriental White Storks

Circles show the sites where birds stayed more than 10 days.

Thank you!