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)
1. Scalable yield gap analysis
David B. Lobell
Associate Professor, Department of Environmental Earth System Science
Associate Director, Center on Food Security and the Environment
dlobell@stanford.edu
2. •
In 2004, Ivan and I gave a talk in El Batan about how remote
sensing could be really useful
•
A lot has changed since, but these resources are still
generally underutilized
•
Technology has advanced a little slower than many of us
expected, but the pace of progress seems to be picking up
3. When is remote sensing useful?
Type of cropping system
Applications that are likely useful
1) Low input, subsistence
systems
•Providing basic statistics on area and
production
•Early warning of shortfalls
•Tracking impacts of interventions
2) High input, low input
use efficiency
3) High input, high input
use efficiency
•Real-time management assistance
•Yield gap analysis
4. The goals of yield gap analysis
To answer questions such as:
•How big are exploitable yield gaps?
•What key factors cause yield gaps?
•On what practices should research
and extension efforts focus for
biggest yield gain?
(From Van Ittersum et al. 2013)
•Which fields are especially good or
bad for a particular crop or variety?
5. Scalable yield gap analysis
What do I mean by scalable?
•Can be rapidly applied in a new area
•Does not rely on field samples to calibrate yield estimates
Our current approach is a 4-step process
1. Yield estimation for individual fields for 3+ years
2. Analysis of the temporal consistency of spatial patterns
3. Comparison of average yields with other readily available
spatial datasets, such as on soil types, roads, and irrigation
infrastructure
4. Targeted field surveys that focus on areas with highest and
lowest average yields
6. 1. Automated yield estimation
100+ sets of crop
model parameters
(sow date, density,
fertilizer, etc.)
Daily
Weather
Data
Available
satellite
images for
year T
Daily time step
crop model
(e.g. APSIM,
Hybrid-Maize)
Surface
Reflectance
Data inputs
Crop Model
Prescribed Parameters
Outputs / Intermediate Variables
Simulated
yields and
veg. indices
(N > 100)
Veg. Indices
(WDRVI)
Crop
Classification
Maps
Regressions
that link VIs on
image date(s)
to final yields
Annual maps
of crop yields
7. Wheat yield estimates derived from
Landsat in Yaqui Valley, Mexico
~7.0 ton/Ha
~4.0 ton/Ha
-1
M e a n S a te llite -B a s e d Y ie ld (to n h a )
1 km
8
2
R = 0 .7 8
-1
rm s = 0 .3 7 to n h a
7
2002-03
6
5
1 :1 lin e
4
4
5
6
7
8
-1
F a rm e r R e p o rte d Y ie ld (to n h a )
2001-02
2000-01
~7.0 ton/Ha
1999-00
1993-94
~4.0 ton/Ha
8. 2002 Wheat Yield in Punjab (Mg/ha)
Faridkot
Moga
Faridkot
Sangrur
Mukstar
Sangrur
Bhatinda
Mansa
Mansa
25
Moga
Mukstar
Bhatinda
0
2002 Planting Date in Punjab
50km
5.5
Dec 25
4.0
2.5
5 km
Nov 19
Oct 15
11. Outline
Our current approach is a 4-step process
1. Yield estimation for individual fields for 3+ years
2. Analysis of the temporal consistency of spatial patterns
3. Comparison of average yields with other readily available
spatial datasets, such as on soil types, roads, and irrigation
infrastructure
4. Targeted field surveys that focus on areas with highest and
lowest average yields
13. Measures of yield persistence can help identify how
much of overall yield gap is driven by persistent factors
14. Measures of yield persistence can help identify how
much of overall yield gap is driven by persistent factors
15. Can also readily do things like look at between vs.
within field yield variation
Estimated maize yields (t/ha) in part of Madison, Nebraska
2002
2004
2003
Estimated maize yields (t/ha) in part of Madison, Nebraska
2007
2008
2011
2012
2009
2005
2010
16. Outline
Our current approach is a 4-step process
1. Yield estimation for individual fields for 3+ years
2. Analysis of the temporal consistency of spatial patterns
3. Comparison of average yields with other readily available
spatial datasets, such as on soil types, roads, irrigation
infrastructure, crop rotation, etc.
4. Targeted field surveys that focus on areas with highest and
lowest average yields
17. Some factors typically emerge as important,
others not
Wheat yields in Indian Punjab vs. distance to roads or canals
Lobell et al. 2010, Field Crops Research
18. Outline
Our current approach is a 4-step process
1. Yield estimation for individual fields for 3+ years
2. Analysis of the temporal consistency of spatial patterns
3. Comparison of average yields with other readily available
spatial datasets, such as on soil types, roads, and irrigation
infrastructure
4. Targeted field surveys that focus on areas with highest and
lowest average yields
Note: this is the most time consuming step, but it comes last
and is guided by the first three.
20. My questions:
The next five years should be very exciting, but there is finite time
and resources. So…
•
How much interest exists at CIMMYT for yield gap analysis, or
is there more interest on other uses, like real-time
management, estimating crop areas, and impact evaluation?
•
Is it better to continue in ‘research mode’ where estimates are
made for areas with specific questions and projects in mind,
or in “public good mode” where we simply try to map all
wheat and maize systems of the world and make it available to
researchers (and farmers)?
21. Acknowledgements
Most of this work was inspired by and done in collaboration with
Ivan Ortiz-Monasterio
Funding from NASA, Fundacion Sonora, Stanford University
Students/ Research Assistants: Adam Sibley, Yi Zhao, Nancy
Thomas, Christopher Seifert
Thanks for your attention!