Webinar – Guide for Collecting and Sharing Ground Reference Data for Machine Learning Applications

Webinar organized by the Radiant Earth Foundation and the CGIAR Platform for Big Data in Agriculture.

In this webinar, the five best criteria for making ground reference data ready for Earth observation machine learning models are discussed.

Click here to read Radiant Earth Foundation’s short write up on the proposed criteria.

Click here to review the Guide for Collecting and Sharing Ground Reference Data.

April 22, 2020

Radiant Earth Foundation & CGIAR Platform for Big Data in Agriculture

Latest videos

Video: Beyond the Census of Things

Video: Demystifying ontologies for agriculture

Q&A with Georgia Barrie from Farm.Ink

Q&A with James Legg from NuruAI

Q&A with Sander Janssen from Wageningen University

Search the website

Discover agricultural data and publications

Powered by GARDIAN

Become a youth in data partner

Submit an initiative!

AgroFIMS: Your new companion for easy standardization of data collection and description

The Agronomy Field Information Management System (AgroFIMS) allows users to create fieldbooks to collect agronomic data that is already tied to a metadata standard (the CG Core Metadata Schema, aligned with the standard Dublin Core), and semantic standards like the Agronomy Ontology (AgrO), generating data that is Findable, Accessible, Interoperable, and Reusable (FAIR) at collection. AgroFIMS therefore standardizes data collection and description for easy aggregation and inter-linking across disparate datasets. The fieldbooks you create can be exported to the Android-based KDSmart data collection application, and collected data imported back to AgroFIMS for statistical analysis and reports. In 2021 AgroFIMS will allow you to set up agronomic survey questionnaires, for data collection via ODK. It will also allow easy upload of your “born FAIR” data to Dataverse repository platforms with Dublin Core-compliant metadata schemas. Funding for AgroFIMS was provided by the Bill and Melinda Gates Foundation’s Open Access, Open Data Initiative, and the CGIAR Platform for Big Data in Agriculture. AgroFIMS is under GPL license. Go to AGROFIMS →

Responsible Data Management Guidelines to protect privacy

CGIAR Platform for Big Data in Agriculture advocates open data for agricultural research for development. It considers that opening up research data for scrutiny and reuse confers significant benefits to society.

However, the Platform appreciates that not all research data can be open and that a broad range of legitimate circumstances may require data to be restricted.

As an integral component of its advocacy for open data, the Platform promotes responsible data management through the entire research data lifecycle from planning, collecting, storing, disclosing or publishing, transferring, discovery and archiving.

These guidelines were created from information collected from: review on best and emerging practices across various sectors in the fast changing landscape of privacy and ethics (130 external resources); privacy and ethic materials sourced from seven CGIAR centers; first draft was circulated for input and feedback across CGIAR and incorporated into this edition. It’s important to note that this is an evolving document, the next stage is to consult externally for further input.

These Guidelines are intended to assist agricultural researchers handle privacy and personally identifiable information (PII) in the research project data lifecycle.

Check the guidelines →

REUSE / TRANSFER

Ensure consistency with the DMP-PII and the purpose for which prior informed consent has been obtained
Revaluate likelihood of (re-)identification and risk of harm, particularly if it involves a public data-set containing PII (as above)
Ensure PII is stored securely to protect privacy (as above)
Minimize use of PII and risk of disclosure through pro-privacy access controls and analytical tools (as above)

Don’t transfer data containing PII unless have explicit consent
Don’t transfer data containing PII in the absence of a data sharing agreement identifying aspects such as purpose and scope of use, privacy protections measures, confidentiality and any limitations)
Don’t reuse or transfer PII until any inconsistencies with the DMP-PII and/or purpose compatibility have been resolved (e.g. through updated ethics review or consent from participant)

ARCHIVING / DISCARDING

Plan for archiving or data destruction early in the process. Destroying data can be more secure, however, archiving can be beneficial if the data has ongoing evidentiary, scientific or cultural value. If archiving, identify where and how, the budget require
Ensure DMP-PII and purpose compatibility (as above)
Ensure adequate security measures to protect privacy (as above)

Don’t wait until the end of the project to assess archiving needs when time and resources may be limited
Don’t assume the longevity of a particular format, future-proof your archives data
Don’t forget to budget for archiving data, this should be done as part of your Data Management Plan

PUBLISHING AND DISCOVERY

Ensure DMP-PII and purpose compatibility (as above)
Revaluate likelihood of (re-)identification and risk of harm, particularly if it involves a public data-set containing PII
Indicate in metadata the availability of raw data or minimized data containing PII, if available bilaterally
Minimize use of PII and risk of disclosure through pro-privacy access controls and analytical tools

Don’t include PII in public datasets unless absolutely necessary to preserve the data’s analytic potential, scientific utility or benefit to the participant (and subject to participants informed consent and a rigorous risk assessment)

STORAGE AND ANALYSIS

Ensure compatibility with the DMP-PII (as above) and also the purpose for which prior informed consent has been obtained

Ensure PII is stored securely to protect privacy, through organizational or project specific safeguards to prevent unauthorized access, accidental disclosure or breach of data (physical & technical)

encryption for the storage and transmission of PII
access control measures to limited access to PII
two-factor or multifactor authentication
cloud services & back-end security

Don’t store data in unsecured locations or on unsecured devices or servers

Don’t store encrypted data and encryption keys in locations where they can be easily accessed simultaneously

Don’t underestimate the importance and value of administrative safeguards to standardize practices (i.e. organizational policies, procedures and maintenance of security measures that are designed to protect private information, data and access)

COLLECTION

Ensure compatibility with the DMP-PII
De-identify data to anonymize by default unless it will impair the data’s analytic potential, scientific utility or benefit to the participant,
If you cannot anonymize, minimize the PII and pseudonymize to reduce the disclosure risk
Provide research participants sufficient information to use reasoned judgment to decide whether or not they wish to participate in the project
Ensure informed consent is designed to address the following elements:
- competence, comprehension, full disclosure, voluntariness
- legitimate scientific purpose for which the PII is collected and scope of use (e.g. stored, transferred, published and whether as anonymized, minimized or raw data)
- foreseeable risk of privacy loss and consequences
- meaningful alternatives including opt-in protection/anonymization
- safeguards to protect privacy, conditions on which PII may be shared and any limitations on reuse or third- party access and use of PII
- permission to follow-up or contact the participant and for what purpose (including by third- parties)
- participant’s right to withdraw and rights regarding their data (e.g. to be informed; to access; to rectify; to object; to erase)
- inclusion of physical, phone and/or electronic contact (at least two forms of contact) that participant can reach to exert her/rights
- explicit consent and participant’s acknowledgement of understanding
- if written, provide the participant a copy of processed informed consent
Use plain language and adapt informed consent to meet the needs of vulnerable populations (e.g. obtain orally or in local language)

Don’t collect PII unless you have a Data Management Plan and any necessary approvals in place, including the recorded approval of the potential participant
Don’t collect PII unless you absolutely need it
Don’t assume that removal of direct identifiers is sufficient to anonymize data or that all de-identification techniques will result in anonymized data. Consider the risk of re-identification of a research participant, particularly if datasets are combined. If there is a reasonable risk of re-identification the information should be handled as PII (i.e. undertake risk analysis, evaluate stronger anonymization techniques, seek informed consent for the disclosure of data and explain its possible consequences)
Don’t include vulnerable participants or communities if their ability or capacity to provide voluntary informed consent is genuinely in question
Don’t underestimate the potential of quasi or indirect identifiers to identify an individual, particularly the inherent ability of location-based data to identify participants and their communities, and the increased risk of harm this may pose to potentially vulnerable individuals/communities
Avoid seeking overly broad consent that may call into question transparency or a research participant’s understanding regarding the use of their PII, be specific regarding the activities, purpose and limitations associated with PII so that the participant can make a genuinely informed decision and downstream users can evaluate purpose compatibility and seek fresh consent if needed

PLANNING AND APPROVAL

Develop a Data Management Plan which governs the handling of PII in the research project and beyond (DMP-PII). It should address:
- the type and nature of PII
- compliance requirements (including necessary forms for obtaining consent, and ethics clearance, if applicable)
- legitimate research objectives that will be advanced by the PII
- foreseeable risks and consequences if participants are identified from the data
- privacy protection measures (or lack thereof) for collection, storage, transfer and publishing
- process for obtaining informed consent
- timeframe or trigger for archiving or deletion of PII
Employ stricter standards for research involving vulnerable populations such as children or illiterate participants or sensitive data such as ethnicity or religious beliefs
Undertake due-diligence of datasets previously collected by you or third parties to ensure you are entitled/permitted to use for your research project
Consult the legal, IRB or ethics clearance committee or any other relevant institutional group for specific institutional, local, regional or national policies and regulatory frameworks that may apply to PII in the context of your work

Don’t leave the handling of PII and privacy protection as an after-thought, plan ahead!
Don’t forget to check local laws and donor or third-party requirements in addition to institutional policies governing research ethics and privacy protection (seek expert support if unsure!)
Don’t ignore ethical practices/standards, if your institution does not have an ethics framework or clearance process in place self-assess!
In assessing whether information is capable of identifying someone (i.e. PII) don’t limit your focus to direct identifiers, also consider indirect/quasi identifiers. Appreciate this will depend on the context of the research project, the data in question and external data which is or may become otherwise available (i.e. there is no exhaustive list).
In assessing risk of harm don’t forget to consider potential harm to the participant’s community or groups of individuals that can otherwise be identified or associated with the participant

Webinar – Guide for Collecting and Sharing Ground Reference Data for Machine Learning Applications