Niruthi provides data analytics and technology solutions like satellite imagery, drones, weather stations and mobile apps to monitor crops, assess yields and damage, and provide location-specific climate data and expert advisories to help insurers, agencies and farmers. Their CropSnap mobile app uses photos of crops translated through machine learning algorithms into crop yield estimates, providing a low-cost and scalable way to sample fields and reduce the need for in-person crop cutting experiments. Their experience in India includes creating historical climate and crop yield data, testing claims settlement at the village level, and reducing sampling costs for crop insurance programs in Maharashtra.
3. 3
TechnologySense
Satellites | Drones
Weather Stn | Mobiles
IoT
Predict
Machine Learning - photo-to-feature
AI - agro-advisories
Crop modeling (nowcasts, forecasts)
Data assimilation (point-to-space)
Data mining for anomalies
Cloud Compute
Scalable Data Stores & fast query
Elastic compute & rapid storage
Cost optimized
Collaborate
Weather advisories
Crop advisories
Sampling schemes
Crop statistics
Communicate
Viz | Newspaper | TV
IAV| web | APIs
An end-to-end intelligent information system to serve the
many needs of insurance providers, agencies and
farmers.
Technology
4. 4
Maharashtra Pune District Indapur Taluk/Circles Indapur Taluk/Villages
Agro-weather Advisories Crop Insurance
Tens of farms/village
Crop yields
Phase I (Sponsored by SwissRe)
(Niruthi/AIC/CRIDA/GoMH)
Objective:
• To create village-level historical
climate and crop yields to assess
basis risk for insurance
What was done:
• Implemented TOPS, integrating
ground, satellite observations
Outcome:
• Created historical daily climate data
for 40,000 villages covering 1981-2013
• Created crop yield models for Bajra,
Cotton, Gram, Jowar and Soybean
Phase II (PPP/IAD)
(GoMH/Niruthi/CRIDA/SwissRe/AIC)
Objective:
• To reduce the number of CCEs
through smart sampling and geo-
spatial data
What was done:
• Conducted over 12,000 CCEs
• Created dynamic sampling
schemes near harvest season
Outcome:
• Smart sampling can save 40-70% of
resources and stay within
10%accuracy at the Circle level
Phase III (PPP/IAD)
(GoMH/Niruthi/CRIDA/SwissRe/AIC)
Objective:
• To test the ability to settle claims at
village-level
What was done:
• Deployed CropSnap (mobile
technology) with smart sampling to
assess yields in 108 villages
Outcome:
• CropSnap yields are within 15% in
Jowar, Bajra, Gram, and 22% in
Soybean
• Costs less than 25% and meets the
village-level insurance guidelines
Our experience spns the entire range
Experience: High resolution climate and crop yields for insurance in MH
5. Data – Model Assimilation for Crop Yield Forecasts/Estimates Forecasting (mid-season)
Method:
• Based on modeled
photosynthesis integrating
satellite data, ground data,
models
Outcome:
• Categorical yields (normal,
below, above)
Final Estimation
Method:
• Based on modeled
photosynthesis integrating
satellite data, models and
CCE data from the State
using smart sampling
Outcome
• Probabilistic yield surface
at field scale, can be
aggregated to farm, village,
mandal or district
Probabilistic yield surface
SmartSampling
6. CropYield Recognition System (CYRS), Virtual Crop Cutting Experiments
Goodresults but cumbersome to deploy,
difficult to train,significant time-lag, and
non-scalable
Automated segmentation results for wheat crop
FielddeploymentinSoybean,2014–2015
7. Mobile App for field-level data collection
Allows for rapidly capturing the spatial variability that traditional CCEs may not
CropSnap allows users to record a
variety of information about crops
grown, crop condition, canopy
density and expected yields.
Information is automatically
uploaded onto the cloud where it is
integrated with satellite data to
create maps of farm-village-regional
assessments.
8.
9. Soybean (40 villages)
y = 0.79x + 0.81
R² = 0.59
RMSE=1.7 (22%)
0
2
4
6
8
10
12
14
16
0 2 4 6 8 10 12 14 16
MeasuredYield,q/ha
CropSnap Yield, q/ha
Low Med High
<6 6 - 12 >12
Village 1 21 6 3
Village 2 10 19 1
Village 3 3 10 17
Minimum 30 locations per village
Example distribution of yields
High Yield
MediumYield Low Yield
Scale-invariant deep learning algorithm trained on 2060 samples with coincident CCE and photos
Input = CropSnap photos
Output = low, med, high
10. BAJRA (14 villages)
y = 1.0x - 0.092
R² = 0.75
RMSE = 0.52 (14%)
0
1
2
3
4
5
6
0 1 2 3 4 5 6
MeasuredYield,q/ha
CropSnap Yield, q/ha
Scale-invariant deep learning algorithm
trained on 1666 samples with coincident CCE
and photos
Measured yields are derived from 4 CCEs per
village
11. Summary
Niruthi technology provides a transparent, scalable,
cost-effective approach needed to expand access to
crop insurance.
CropSnap Mobile app pictures translated to crop yields
save time, money and improve accuracy through large
number of samples
Innovations in mobile computing may allow on-board
estimation of yield categories, reducing the need for
bandwidth and connectivity.
Further training may allow additional yield categories,
ultimately actual yields.
12. Contact us
Contact Us
Mr. Mallikarjun Kukunuri
mkukunuri@niruthi.com
www.niruthi.com
402 Man Bhum Jade Towers
Somajiguda, Hyderabad
Telangana, 500 082