Reforms to agricultural and land management practices are needed to combat a future of soil exhaustion and food insecurity forecasted by scientific organizations worldwide, as the world’s populations continue to rise and issues of food security and nutrition are exacerbated.
Motivated by this call, we have been developing technologies to support agroecological analyses on a variety of different spatio-temporal scales.
Working as a partner in the Africa Soil Information Service (AfSIS) project, QED has focused on developing end-to-end data processing workflows to support the full lifecycle of data. This workflow includes field data acquisition, barcoding, database management, and map visualization. These technologies have been introduced in several African countries in collaboration with their governments, with the aim of closing information gaps and positively influencing public policy and industrial practices, if executed in coordination with partners. Below we describe some of these technologies in further detail.
Predictive Maps of Croplands, Human Settlements, and Soil Chemistry
We have been developing technologies to allow scientists to more efficiently and reliably catalogue soil and landscape resources by using some of today’s most influential technologies, including web and mobile, crowdsourcing, machine learning, cloud, and UAVs. Our hope is that the technologies we build will be of use to scientists and industry practitioners aiming to collect and analyze updated field data.
As an example, we can generate digital maps such as those found on the right by collecting crowdsourced assessments of recent satellite imagery backed by expert verification, and then building ML models to execute predictions throughout the region. These predictions can also supplemented with uncertainty measures.
Readers are encouraged to browse more examples of applications below.
End-to-End Data Processing Workflow
QED aims to build suites of interlocking software providing for end-to-end support. The components should be freely interchangeable, as our design is inspired by several key tenets of ESR’s rules of UNIX, reproduced below:
- Rule of Modularity: Write simple parts connected by clean interfaces.
- Rule of Composition: Design programs to be connected with other programs.
- Rule of Simplicity: Design for simplicity; add complexity only where you must.
- Rule of Transparency: Design for visibility to make inspection and debugging easier.
- Rule of Diversity: Distrust all claims for one true way.
Data Collection
- Crowdsourced surveying of aerial imagery (satellites + drones) to efficiently generate land cover maps and determine regions of interest such as cropland masks.
- Randomized survey design: multi-stage sampling algorithms for balancing broad coverage with collection efficiency.
- Mobile data collection apps for field data entry and real-time data collection.
- Sampling protocols tested in twelve countries throughout in Africa.
Data Management
- Customizable QR codes and barcodes for tagging samples in the field.
- Secure and cost-effective data management systems stored in the cloud.
- Supports relational data, binary data, and geospatial data on scalable cloud infrastructure.
- Visual web interfaces for easily browsing data, uploading data, and conducting geospatial queries.
Data Computation
- Generate maps of agroecological properties by leveraging crowdsourced data and scalable ML computation in the cloud.
- Predict soil chemistry concentrations of critical nutrients and trace elements in the soil from cost-efficient spectroscopy measurements, using the best calibrated models to date and backed by uncertainty measurements.
- Augment predictions with machine learning models tempered through international programming competitions.
Data Visualization
- Interactive visualizations of agricultural metrics, presented as surfaces interpolated from point data and re-rendered in real-time when queried against geospatial and chemical constraints.
- Building decision analysis tools for optimizing budget allocation across laboratory methodologies, fertilizer blends, and crop varieties.
- + Data Collection
Data Collection
- Crowdsourced surveying of aerial imagery (satellites + drones) to efficiently generate land cover maps and determine regions of interest such as cropland masks.
- Randomized survey design: multi-stage sampling algorithms for balancing broad coverage with collection efficiency.
- Mobile data collection apps for field data entry and real-time data collection.
- Sampling protocols tested in twelve countries throughout in Africa.
- + Data Management
Data Management
- Customizable QR codes and barcodes for tagging samples in the field.
- Secure and cost-effective data management systems stored in the cloud.
- Supports relational data, binary data, and geospatial data on scalable cloud infrastructure.
- Visual web interfaces for easily browsing data, uploading data, and conducting geospatial queries.
- + Data Computation
Data Computation
- Generate maps of agroecological properties by leveraging crowdsourced data and scalable ML computation in the cloud.
- Predict soil chemistry concentrations of critical nutrients and trace elements in the soil from cost-efficient spectroscopy measurements, using the best calibrated models to date and backed by uncertainty measurements.
- Augment predictions with machine learning models tempered through international programming competitions.
- + Data Visualization
Data Visualization
- Interactive visualizations of agricultural metrics, presented as surfaces interpolated from point data and re-rendered in real-time when queried against geospatial and chemical constraints.
- Building decision analysis tools for optimizing budget allocation across laboratory methodologies, fertilizer blends, and crop varieties.
Aerial Imagery
Leverage networks of satellites and UAVs to quickly scan regions of interest and generate maps of ecological indicators and DEMs.
Access Anywhere
Data is reliably stored on cloud infrastructure and replicated in multiple availability zones. Automated input validation of laboratory data.
Predict
Predictions of soil chemistry and land usage using machine learning executed on cloud infrastructure.
Acknowledgments
We express our thanks to the Africa Soil Information Service for supporting us in the development of many of the tools described above.
Get In Touch Now
For more inquiries about our services or to request a demo, please e-mail: info@qed.ai