Tracking agricultural outcomes at field scales:
methods and applications
Abstract:
Improved management of agriculture is needed to meet many sustainability goals, such as reducing hunger, slowing climate change, conserving water, and protecting biodiversity. Better data is, in turn, needed to improve management, both for operational decisions and for improved knowledge of how agro-ecosystems function. I will present recent research to map both crop type and crop yield across entire continents, at a resolution of individual fields. This field-level detail provides useful information on yield variability that can be used to understand various phenomena. I will discuss two examples for the United States: understanding the response of yields to drought and to local air pollution.
Bio:
David Lobell is a Professor at Stanford University in the Department of Earth System Science and the Gloria and Richard Kushel Director of the Center on Food Security and the Environment. He is also the William Wrigley Senior Fellow at the Stanford Woods Institute for the Environment, and a senior fellow at the Freeman Spogli Institute for International Studies (FSI) and the Stanford Institute for Economic Policy and Research (SIEPR). His research focuses on agriculture and food security, specifically on generating and using unique datasets to study rural areas throughout the world. He has been recognized with a Macarthur Fellowship in 2013, a McMaster Fellowship from CSIRO in 2014, and the Macelwane Medal from the American Geophysical Union in 2010. He also served as lead author for the food chapter and core writing team member for the Summary for Policymakers in the recent Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report.
Prior to his current appointment, Dr. Lobell was a Senior Research Scholar at FSE from 2008-2009 and a Lawrence Post-doctoral Fellow at Lawrence Livermore National Laboratory from 2005-2007. He received a PhD in Geological and Environmental Sciences from Stanford University in 2005, and a Sc.B. in Applied Mathematics, Magna Cum Laude from Brown University in 2000.
Summary:
Motivation: Make farming more efficient to achieve Sustainable Development Goals Farming is
A complex system
Slow to evolve (information sharing is very local and anecdotal)
Technique: Mapping crop types and their yields across large scales
Spatial granularity: individual fields (US: 10k sq m, others: 500-1000 sq m)
Training data is highly localized and varies within countries
Challenge: not enough training pairs
Crowdsourced phone data
e.g. Plantix
Photos of plants, with labels of type and location (so we know the crop type in each field at a given time)
Question: can we use these photos to train a model to predict crop type from satellite images?
Satellites: Sentinel 1 and 2 (10m spatial resolution, operating for several years)
Achieves 80% accuracy relative to highly reliable labels
GEDI satellite: Space-borne LiDAR
Active sensor, sends laser pings to ground and waits for return
25m resolution to map forest canopy at multiple levels (e.g. tree top, bushes, ground) precisely
Cannot see through clouds
Complementary to Synthetic Aperture Radar and Light sensors
Train model to estimate maize/non-maize from GEDI LiDAR
Available at lines of individual points
At irregular time intervals (so focus on peak growing season)
Then train same using Sentinel 2 (available over broad areas, regular samples)
Apply Sentinel 2 model on new regions
Crop yield maps
Use simulations to infer the yield based on what they look like from satellites
Simulations
Accurately capture effects of water and nutrient stress
Not good at late season infestations and other late season effects
Need more data
Late season moisture
Pest populations
Application: Impact of drought on crops
We have county-level yield records from USDA
Too coarse to track impact of droughts
But can use field-granularity data to estimate impact of droughts live
Major input: water storage capacity of soil
They focused on this instead of soil moisture because it is exogenous (instrumental variable enables causal inference)
Soil moisture was also used, produced similar results
Crop yield predictions
They’ve observed that over the past few decades management techniques have made yield more sensitive to water availability (can make better use of water)
Application: Impact of air pollution on crops
Using EPA pollution monitoring stations
Use pollution gradients from power plans to establish how pollution levels drop at different distances from source
Observed strong yield gradients away from power plans (yields rise with distance)
Estimated impact of each pollutant on yield
Impacts have been declining over the years with lower pollution levels