Presented by Francesco Fava (ILRI), Nathan Jensen (ILRI), Lucas de Oto (Uni-Twente) and Andrew Mude (ILRI) at the CGIAR Platform for Big Data in Agriculture Convention, Nairobi, 3-5 October 2018
Disentangling the origin of chemical differences using GHOST
Integrating local crowdsourced and remotely sensed data to characterize rangeland resource use in extensive pasturelands
1. Integrating local
crowdsourced and
remotely sensed data to
characterize rangeland
resource use in extensive
pasturelands
Francesco Fava (ILRI)
Nathan Jensen (ILRI)
Lucas de Oto (Uni-Twente)
Andrew Mude (ILRI)
2. The Problem
Complex Socio-ecological Systems
In pastoral regions, household welfare and resilience is
tightly tied to the availability and quality of forage
resources.
Remotely sensed (RS) data is currently used to map
rangeland cover types and forage condition.
Grazing resource use and accessibility cannot be mapped RS
data, while they are critical aspects of pastoralist mobility
and management decision making.
3. The Idea
Mobile tech. are deeply
penetrating even in remote
areas
Pastoralists can provide critical
information to understand
land cover dynamics,
migration patterns, and
management challenges
4. The Setup
Screena Screenb Screenc
Screen0
Screens1
Screen2
Screen3
Screen 4
Screen 5
Screen 6
• Dedicated smartphone app (offline)
• Short, image & audio based survey on their
interpretation of the immediate vegetation conditions,
water availability and stocking rate
• Reward system to incentivize data collection
5. The Setup
• Study Area: semi-arid rangelands in
Laikipia and Samburu
• Period: March-August 2015. Long
Rain and Long Dry seasons
• 113 local pastoralists across 5 sites
trained and provided with
smartphones, internet bundles, and
solar changers
• Participation: 112K submissions,
~95K valid
6. Spatial sampling challenges
• Submissions were clustered under
baseline with spatially uniform
rewards (Left)
• Survey region is divided into 96
reward sub-regions (Center) to adjust
the distribution of submissions
• Each 10 days, the rewards are
updated to reflect the distribution of
submissions to date
7. Big data, low quality?
1
Internal quality
check-based on
data consistency
rules
2
Using the crowd
to validate the
crowd
3
‘Scientific’
validation/cleaning
using pictures,
imagery, geotag
8. WET SEASON
Integration with remote sensing
Unsupervised classification from MODIS NDVI seasonal profiles
DRY SEASON
AGRO-ECOLOGY
WHAT CAN WE LEARN FROM THE CROWD?
USE OF RESOURCES – SPATIAL DISTRIBUTION
DURING THE DRY AND WET SEASONS
1,2,3
4,5,6
7,8,9
9. Integration with remote sensing
LAND USE
CLASS 1 : Wet season / Low stocking rate / Moderate to low
water accessibility / Moderate intensity of use
CLASS 2 : Annual / Moderate-Low stocking rate / Moderate
water accessibility / High intensity of use
CLASS 3 : Annual /high stocking rate. Good water accessibility /
Moderate Intensity of use
CLASS 5 : Wet season / High-Very High Stocking rate /
Good water accessibility. Important dry season. Very good
water accessibility wet season – limited in the dry / Low
intensity of use.
CLASS 4 : Annual /moderate carrying capacity / Moderate to low
water accessibility / Moderate intensity of use particularly
during the dry season.
CLASS 6 : Wet season / High-Very High Stocking rate /
Good water accessibility. Important dry season. Very limited
water accessibility / Low intensity of use.
CLASS 7, 8, 9 : Very Low intensity of use.
Production
- limited
Dry season
- refugees
Water
- limited
Poorly
accessible