Remote sensing –Beyond images
Mexico 14-15 December 2013
The workshop was organized by CIMMYT Global Conservation Agriculture Program (GCAP) and funded by the Bill & Melinda Gates Foundation (BMGF), the Mexican Secretariat of Agriculture, Livestock, Rural Development, Fisheries and Food (SAGARPA), the International Maize and Wheat Improvement Center (CIMMYT), CGIAR Research Program on Maize, the Cereal System Initiative for South Asia (CSISA) and the Sustainable Modernization of the Traditional Agriculture (MasAgro)
Forecasting Wheat Yield and Production for Punjab Province, Pakistan from Satellite Image Time Series
1. Forecasting Wheat Yield and
Production for Punjab Province,
Pakistan from Satellite Image
Time Series
Jan Dempewolf, Inbal Becker-Reshef, Bernard
Adusei, Matt Hansen, Peter Potapov, Brian Barker,
Chris Justice
Department of Geographical Sciences
University of Maryland, United States
Beyond Diagnostics: Insights and Recommendations from Remote Sensing
Workshop at CIMMYT 2013 in Texcoco, Mexico 14-15 December 2013
2. Pakistan: Strengthening Provincial Capacity
(USDA funded, collaboration between USDA, FAO, SUPARCO, CRS Pakistan, & UMD)
Training Workshops
4. Food Crop Production in Pakistan
Winter Season (Rabi) % of Total
Vegetables
5%
Other
5%
Fruits
9%
Potatoe
11%
Wheat
70%
Data source: Crop Reporting Service of the Government of Punjab,
Pakistan, www.agripunjab.gov.pk
5. Total wheat dry matter and NDVI in Maryland, USA
(Tucker et al., 1981)
Tucker, C. J., B. N. Holben, J.
H. Elgin Jr, and J. E. McMurtrey
III. “Remote Sensing of Total
Dry-matter Accumulation in
Winter Wheat.” Remote
Sensing of Environment 11
(1981): 171–189.
6. Wheat yield and AVHRR-NDVI integrated over the
growing season in Montana, USA (Labus et al., 2002)
Labus, M. P., G. A. Nielsen,
R. L. Lawrence, R. Engel,
and D. S. Long. “Wheat Yield
Estimates Using Multitemporal NDVI Satellite
Imagery.” International
Journal of Remote Sensing
23, no. 20 (January 2002):
4169–4180.
7. Reported wheat yield and predicted yield from
MODIS-NDVI in Shandong, China (Ren et al., 2008)
Ren, J., Z. Chen, Q. Zhou, and
H. Tang. “Regional Yield
Estimation for Winter Wheat with
MODIS-NDVI Data in Shandong,
China.” International Journal of
Applied Earth Observation and
Geoinformation 10, no. 4
(December 2008): 403–413.
8. MODIS-NDVI and Wheat Yield in Kansas, USA
(Becker-Reshef et al., 2010)
Daily Normalized Difference Vegetation Index (NDVI
from MODIS) 2000-2008, Harper County
Blue numbers are Yield (MT/Ha)
Winter Wheat emergence
NDVI peak
2.35
Winter Wheat seasonal
NDVI peak
2.69
3.36
2.54
2.49
2.49
2.21
1.61
1.4
8
Year
Strong correlation between NDVI Peak and yield
Becker-Reshef, I., E. Vermote, M. Lindeman, and C. Justice. “A Generalized Regression-based Model for Forecasting Winter Wheat Yields in Kansas
and Ukraine Using MODIS Data.” Remote Sensing of Environment 114, no. 6 (2010): 1312–1323.
9. Wheat Mask and Area from 250 m MODIS
Multi-Temporal Landsat
1. Early growing season
2. Height of growing season
3. After harvest
Classify Landsat
•
•
Select training data visually
Bagged decision trees
10. Visual Interpretation of Wheat Areas
Early Season
(8. Feb. 2012)
Landsat-7 ETM
scene for
Punjab
Band
combination 45-3 (green
vegetation
appears red)
11. Visual Interpretation of Wheat Areas
Near Peak
(24. Feb. 2012)
Landsat-7 ETM
scene for
Punjab
Band
combination
4-5-3 (green
vegetation
appears red)
12. Visual Interpretation of Wheat Areas
Harvest
(4. Apr. 2012)
Landsat-7 ETM
scene for
Punjab
Band
combination 45-3 (green
vegetation
appears red)
13. Select Wheat Training Areas
Training
(12. Apr. 2012)
Landsat-7 ETM
scene for
Punjab
Band
combination
4-5-3 (green
vegetation
appears red)
14. Classify for Wheat Areas
Classification
(12. Apr. 2012)
Landsat-7 ETM
scene for
Punjab
Band
combination
4-5-3 (green
vegetation
appears red)
16. Landsat Training Scenes for Wheat Area
Pakistan
Landsat
training
scenes
Sindh
WRS2
Path/Row
Grid
17. Wheat Mask and Area from 250 m MODIS
Multi-Temporal Landsat
1. Early growing season
2. Height of growing season
3. After harvest
Classify Landsat
•
•
Select training data visually
Bagged decision trees
Aggregate to 250 m
resolution
18. Wheat Mask and Area from 250 m MODIS
Multi-Temporal Landsat
1. Early growing season
2. Height of growing season
3. After harvest
Classify Landsat
•
•
Select training data visually
Bagged decision trees
Aggregate to 250 m
resolution
MODIS 250 m surface reflectance 8day composites time series bands 1,
2, 5, 7 (red, nir, swir, therm)
1. 1. Dec. – 26th Feb.
2. QA Filter (clouds, etc.)
3. Calculate NDVI
19. Wheat Mask and Area from 250 m MODIS
Multi-Temporal Landsat
1. Early growing season
2. Height of growing season
3. After harvest
MODIS 250 m surface reflectance 8day composites time series bands 1,
2, 5, 7 (red, nir, swir, therm)
1. 1. Dec. – 26th Feb.
2. QA Filter (clouds, etc.)
3. Calculate NDVI
Classify Landsat
•
•
Select training data visually
Bagged decision trees
Aggregate to 250 m
resolution
Convert to 588 metrics per season
•
•
•
0th, 10th, 25th, 50th, 75th, 90th, 100th
percentiles
Means of sequential percentiles and
their differences
Band values ranked by other bands
20. Wheat Mask and Area from 250 m MODIS
Multi-Temporal Landsat
1. Early growing season
2. Height of growing season
3. After harvest
MODIS 250 m surface reflectance 8day composites time series bands 1,
2, 5, 7 (red, nir, swir, therm)
1. 1. Dec. – 26th Feb.
2. QA Filter (clouds, etc.)
3. Calculate NDVI
Classify Landsat
•
•
Select training data visually
Bagged decision trees
Aggregate to 250 m
resolution
Classify MODIS time series
•
Bagged decision trees
Convert to 228 metrics per season
•
•
•
0th, 10th, 25th, 50th, 75th, 90th, 100th
percentiles
Means of sequential percentiles and
their differences
Band values ranked by other bands
Percent wheat per 250 m
pixel for Punjab Province
22. Wheat Yield and Production Forecast
Percent wheat
per pixel
MODIS 8-day
composites
Select 20%
highest density
wheat pixels
Calculate spatial
average of NDVI,
weighted by
percent wheat
Regression
estimator of pixel
counts against
reported area
Multiply area
forecast with yield
forecast to obtain
production forecast
Historic reported
yield
Regression-based
wheat model yield
against 95th NDVI
percentile
23. Timing of Forecast and Number of Training Years for
Punjab Province, Pakistan, 2010/11 Rabi Season
R2, RMSE at the district level and deviation (D) at the province level of forecast
versus reported yield for the 2010/11 Rabi season.
Left: Changes through the cropping season. Right: Number of training years.
24. Performance of Vegetation Indices for Forecasting
Wheat Yield for the 2010/11 and 2011/12 Rabi Seasons
NDVI
VCI
WDRVI
SANDVI
25. Forecast Wheat Production per District for
Punjab Province, Pakistan, Seasons 2008/09 to 2011/12
2008/09
2010/11
2009/10
2011/12
26. Remote Sensing Applications for
Smallholder Farming Systems in Tanzania
(Proposed Project)
Explore feasible pathways to use remote sensing
tools for smallholder agriculture:
Improve crop condition monitoring by the National Food
Security Office (NFSO).
Produce current cropland extent core dataset.
Support agricultural extension through Sokoine University.
Monitor crop condition of smallholder agricultural areas.
Assess distribution of smallholder cropping systems and crop
types.
27. Primary Use-Case Challenges
1.
2.
3.
4.
5.
Whether, how, and with which datasets can we
produce national-scale cropland layers for
smallholder agriculture?
How can smallholder agricultural fields be
sampled and monitored through remote sensing?
How can agricultural areas be monitored at the
national scale in near-realtime?
How can we inform decision makers?
What are the pathways to reach smallholder
farmers?
28. Remote Sensing Systems
MODIS
Satellite Time
Series Pipeline
and Archive
Landsat
RapidEye/
PlanetLabs
UAV
Field Data
Test
Sites
(
)
Time Series
(one season)
Groundtruth landcover and
land-cover
dynamics
Rela ve NDVI /
Crop Condi on at
MODIS and Landsat
resolu on
Prototype of
Agricultural Areas Base
Map (Cropland Mask)
Methodologies for
classifying
• Cropland
• Maize produc on
systems