2010 Pakistani flood case study
Ishtiaq Mahsud- AP
Pedro Ugarte _AFP/Getting Images
Sheikh Saleem Raza-AP
Emergency driven remote sensing estimate of national crop production losses: 2010 Pakistani flood case study Tatiana Nawrocki Joseph Fortier Christi Ludlow Townsend Guy Serbin Dath Mita
Problem
August 2010 – heavy rains caused widespread flooding along the Indus River, Pakistan Alarming emergency reports from localities, but lack of complete nationwide data Urgent need to evaluate crop losses nationwide
Daniel Berehulak – Getty Images
Solution
Assess data sources and availability Determine flood extent from remote sensing data Identify specific crops within the flood zone Estimate potential crop losses SPOT 4: 27th Aug 2010 K190 J294
2
3
Remote sensing data sources Sensor Data / Resolution
Sample
Source
MODIS (daily) 250 m Landsat 5 30 m
MODIS Rapid Response / USDA Crop Explorer (Free)
SPOT 4/5 10 - 20 m
SPOT Image
US Geological Survey Earth Explorer (Free)
4
WHERE?
Daily MODIS products provided dynamic outlines of the flooded zone along the entire Indus River MODIS imagery had resolution 250-m
August 21th, 2010
August 27th, 2010
5
Flood mapping: assemble MODIS temporal-spatial mosaic
6
Flood mapping: classify MODIS imagery and delineate water Incremental flood advances day by day
7
Landsat classification
Upland Depression Flooding
Main Channel Flooding
Landsat 5: 12st August 2010 Supervised Classification
8
SPOT classification Main Channel Flooding
Upland Depression Flooding Supervised Classification SPOT 4: 21st August 2010
9
Verify MODIS flood vs. Landsat & SPOT Data Source
Sq. km
% Agreement
MODIS vs. Landsat
10,397
93.5%
MODIS vs. SPOT
3,262
97.6%
10
WHICH CROPS?
Fall harvest crops (rice and cotton) were examined with imagery classification (Landsat 5), agricultural statistics, NDVI time-series, and geotagged photos available on Google Earth. Flood extent was compared with known crop locations to estimate flood-affected agricultural areas and potential crop losses.
11
Crop allocation review at regional level: rice and cotton
12
Examine detail of spatial allocation of specific crops based on remote sensing cropland classification in most vulnerable areas
13
Crop Mapping: Classifying Agriculture 1
3
2
Landsat Spectral Stack
Classified Image
Unsupervised Classification
5
Google Truthing
4
Final Classification Agriculture
Identify Agricultural Classes 14
Verification of cropland: Google Earth
Geotagged photos 15
Crop mapping: classifying rice
Using multiple dates of imagery specific crop types may become evident based on unique characteristics,
field flooding + green-up = rice paddies
May 12 July 15
July 31 September 17
16
Evaluate area of flooded crops
8 Landsat Path/Rows 23 Districts
17
Evaluate area of flooded rice
2 Landsat Path/Rows 9 Districts
18
Review crop seasonal progress Crop abundance 0.6
Flood in Punjab
Flood in Sindh
NDVI
0.5 0.4 0.3 0.2
Punjab
9/ 22 /1 0 10 /1 2/ 10 11 /1 /1 0
9/ 2/ 10
7/ 24 /1 0 8/ 13 /1 0
7/ 4/ 10
5/ 25 /1 0 6/ 14 /1 0
0.1
Sindh
Heavier losses occurred in Sindh as the flood coincided with the peak phase of crop growth
19
Assess crop losses caused by flood
20
Review
Turnball, Greig Box. ‘Pakistan flood disaster ‘could be worse than the Boxing Day tsunami, Kashmir earthquake and Haiti earthquake combined. 8 Oct 2010. The Mirror. (http://www.mirro.co.uk)
Multiple data sources were used Flood extent was determined using MODIS, SPOT 4, and Landsat 5 Crops were identified using Landsat 5 Ultimately, a method for flood damage assessment to crops was applied in a dynamic, emergency driven situation SPOT 4: 27th Aug 2010 K190 J294
21
Thank You Tatiana Nawrocki Joseph Fortier Christi Ludlow Townsend Guy Serbin Dath Mita ASRC Research and Technology Solutions (ARTS) E-Mail: [email protected] 22
Pedro Ugarte _AFP/Getting Images
Sheikh Saleem Raza-AP
Emergency driven remote sensing estimate of national crop production losses: 2010 Pakistani flood case study Tatiana Nawrocki Joseph Fortier Christi Ludlow Townsend Guy Serbin Dath Mita
Problem
August 2010 – heavy rains caused widespread flooding along the Indus River, Pakistan Alarming emergency reports from localities, but lack of complete nationwide data Urgent need to evaluate crop losses nationwide
Daniel Berehulak – Getty Images
Solution
Assess data sources and availability Determine flood extent from remote sensing data Identify specific crops within the flood zone Estimate potential crop losses SPOT 4: 27th Aug 2010 K190 J294
2
3
Remote sensing data sources Sensor Data / Resolution
Sample
Source
MODIS (daily) 250 m Landsat 5 30 m
MODIS Rapid Response / USDA Crop Explorer (Free)
SPOT 4/5 10 - 20 m
SPOT Image
US Geological Survey Earth Explorer (Free)
4
WHERE?
Daily MODIS products provided dynamic outlines of the flooded zone along the entire Indus River MODIS imagery had resolution 250-m
August 21th, 2010
August 27th, 2010
5
Flood mapping: assemble MODIS temporal-spatial mosaic
6
Flood mapping: classify MODIS imagery and delineate water Incremental flood advances day by day
7
Landsat classification
Upland Depression Flooding
Main Channel Flooding
Landsat 5: 12st August 2010 Supervised Classification
8
SPOT classification Main Channel Flooding
Upland Depression Flooding Supervised Classification SPOT 4: 21st August 2010
9
Verify MODIS flood vs. Landsat & SPOT Data Source
Sq. km
% Agreement
MODIS vs. Landsat
10,397
93.5%
MODIS vs. SPOT
3,262
97.6%
10
WHICH CROPS?
Fall harvest crops (rice and cotton) were examined with imagery classification (Landsat 5), agricultural statistics, NDVI time-series, and geotagged photos available on Google Earth. Flood extent was compared with known crop locations to estimate flood-affected agricultural areas and potential crop losses.
11
Crop allocation review at regional level: rice and cotton
12
Examine detail of spatial allocation of specific crops based on remote sensing cropland classification in most vulnerable areas
13
Crop Mapping: Classifying Agriculture 1
3
2
Landsat Spectral Stack
Classified Image
Unsupervised Classification
5
Google Truthing
4
Final Classification Agriculture
Identify Agricultural Classes 14
Verification of cropland: Google Earth
Geotagged photos 15
Crop mapping: classifying rice
Using multiple dates of imagery specific crop types may become evident based on unique characteristics,
field flooding + green-up = rice paddies
May 12 July 15
July 31 September 17
16
Evaluate area of flooded crops
8 Landsat Path/Rows 23 Districts
17
Evaluate area of flooded rice
2 Landsat Path/Rows 9 Districts
18
Review crop seasonal progress Crop abundance 0.6
Flood in Punjab
Flood in Sindh
NDVI
0.5 0.4 0.3 0.2
Punjab
9/ 22 /1 0 10 /1 2/ 10 11 /1 /1 0
9/ 2/ 10
7/ 24 /1 0 8/ 13 /1 0
7/ 4/ 10
5/ 25 /1 0 6/ 14 /1 0
0.1
Sindh
Heavier losses occurred in Sindh as the flood coincided with the peak phase of crop growth
19
Assess crop losses caused by flood
20
Review
Turnball, Greig Box. ‘Pakistan flood disaster ‘could be worse than the Boxing Day tsunami, Kashmir earthquake and Haiti earthquake combined. 8 Oct 2010. The Mirror. (http://www.mirro.co.uk)
Multiple data sources were used Flood extent was determined using MODIS, SPOT 4, and Landsat 5 Crops were identified using Landsat 5 Ultimately, a method for flood damage assessment to crops was applied in a dynamic, emergency driven situation SPOT 4: 27th Aug 2010 K190 J294
21
Thank You Tatiana Nawrocki Joseph Fortier Christi Ludlow Townsend Guy Serbin Dath Mita ASRC Research and Technology Solutions (ARTS) E-Mail: [email protected] 22
