Problem: Three wetlands in Bure and Womberma, NW Ethiopia lost ~48.8% area to farming over 36 years (1985–2021), per a July 2, 2026 PLOS ONE study. If your team studies land-use change with mixed data (Landsat maps, n=396 household surveys, 75 experts, 12 KIIs), this post shows how to run a reproducible qualitative analysis of wetland conversion and turn transcripts and surveys into stakeholder-ready themes with AI. Learn a short, practical workflow and where Evidano (www.evidano.com) speeds annotation, thematic synthesis, cross-segment comparisons and secure sharing.
Key Takeaways
Evidano is an AI-powered qualitative data analysis platform that speeds transcript workflows, supports AI-assisted coding, enables cross-segment analysis, and provides secure handling for sensitive research data.
The PLOS ONE study (published July 2, 2026) documents about 3, 355 ha converted and a 48.8% wetland loss (1985–2021), and finds drivers including distribution to landless youth and institutional incentives.
- High-confidence Landsat classification in the study: overall accuracy 97.27% and Kappa 87.19% for 2021, enabling reliable mapping of wetland-to-cropland change.
- Social data scale: household survey n = 396 (Mar–May 2022), experts n = 75, and 12 KIIs provide the qualitative evidence that complements LULC change.
- Use a reproducible pipeline (transcription → AI-assisted coding → cross-segment analysis → exports) to shorten synthesis time while preserving audit trails for peer review.
Fast take + source
Fast take: The PLOS ONE paper (published July 2, 2026) found 3, 355 ha of wetlands converted to farmland in three sites (Kotlan, Foket, Wadera) and reports drivers including distribution of wetlands to landless youth and institutional incentives.
- Why it matters: high-accuracy LULC (overall accuracy ≥95%, 2021 Kappa ~87%) plus rich qualitative data equals policy-ready recommendations if synthesis is rigorous.
- Payoff for researchers: replicate the study’s mixed-method inference faster and with reproducible coding, quote management, and cross-segment testing.
- Full study: PLOS ONE.
Findings snapshot
| Date / Metric | Value | Source | Implication |
|---|---|---|---|
| Published | July 2, 2026 | PLOS ONE | Use exact citation/date in briefs |
| Wetland loss (1985–2021) | 48.8% (≈3, 355 ha) | Landsat classification | Substantial conversion to cropland |
| Household survey | n = 396 (Mar–May 2022) | Questionnaires | Perceptions and attitudes dataset |
| Experts & heads | n = 75 | Self-administered survey | Institutional perspective captured |
| Key informant interviews | 12 KIIs (elders, admins) | Audio recordings | Narratives for thematic coding |
| LULC accuracy (2021) | Overall accuracy 97.27% / Kappa 87.19% | Confusion matrix | High confidence in map-derived change |
How the study combined methods
This section explains how the authors combined high-quality Landsat composites and a multi-source social dataset to infer drivers of wetland conversion.
The authors merged Landsat composites (1985, 1995, 2000/2010, 2021) and supervised classification with a multi-source social dataset: a face-to-face household survey (n = 396), a 75-person experts survey, and 12 audio-recorded KIIs.
- Remote sensing: Google Earth Engine plus ArcGIS, maximum likelihood classification, accuracy assessment via confusion matrix.
- Surveys: translated Amharic questionnaires, SPSS for descriptives; STATA v18 for ordered probit.
- Qualitative: audio to transcription to thematic coding with smallest meaningful units (manual thematic analysis).
What this means for qualitative researchers
Primary takeaway
Primary takeaway: Mixed datasets with strong geospatial signals (Landsat) plus structured social data enable causal framing, in this study institutional allocation and youth landlessness explain conversion better than immediate subsistence needs.
For UX / field teams
For UX and field teams: Treat transcripts and survey responses as linked layers, and code by respondent attributes (age, landholding, livelihood) to surface segment-specific themes and quote exemplars for reports.
For policy analysts
For policy analysts: Use co-registered LULC change and coded interview themes to show where institutional actions (for example, 1997 redistribution and 2010 allocations) align with on-the-ground responses, giving stronger evidence for intervention design.
How Evidano helps run a reproducible qualitative analysis of wetland conversion
Ingest & prep
Ingest and prep: Upload KIIs, FGDs and survey spreadsheets and auto-transcribe audio with a custom dictionary (Amharic support), PII redaction, and parallel translation so multilingual teams work from the same text.
Coding & thematic synthesis
Coding and synthesis: Auto-generate candidate themes, import or build a codebook, apply AI-assisted coding across transcripts and free-text survey responses, then refine with manual review to preserve nuance.
Cross-segment analysis
Cross-segment analysis: Run frequency and cross-segment comparisons (for example, age cohorts, landholding size, office affiliation) and test associations that map to ordered probit outputs, all exportable as tables for STATA or SPSS.
Traceability & visual reports
Traceability and reports: Produce clickable quotes linked to codes, co-occurrence networks and hierarchical code-to-subcode visuals to save hours during peer review and stakeholder briefing.
Security & compliance
Security and compliance: Evidano provides end-to-end encryption; proprietary LLMs tuned for qualitative research; Evidano does not use your data to train third-party models, important for sensitive ecological and human-subject data.
This week’s 7-step workflow (apply to the Ethiopia study)
This section lists a compact, reproducible 7-step pipeline to get from raw files to policy-ready findings.
- 1) Collect and version files: Landsat outputs (geotiffs), survey spreadsheets, KII audio. Store originals and working copies.
- 2) Transcribe and translate KIIs in Evidano, using a custom dictionary for local terms (for example, kebele, cheffee).
- 3) Import survey sheet, harmonize respondent IDs to link transcripts to survey rows.
- 4) Auto-suggest codes in Evidano, then create a verified codebook reflecting provisioning, regulating and cultural services and drivers (land redistribution, youth employment).
- 5) Run thematic frequency and co-occurrence analyses, slice by segment (age, landholding, office).
- 6) Export coded quotations and cross-tab tables for ordered probit replication in STATA; produce a visual brief (map plus top themes plus representative quotes).
- 7) Package an executive brief and a reproducible appendix (codebook, exportable CSVs) for stakeholders.
FAQ: qualitative analysis of wetland conversion
How do I link Landsat change to interview themes?
Link Landsat change to interview themes by using georeferenced parcel IDs or village identifiers and attaching map metadata to respondent IDs so themes can be aggregated by change class (converted versus retained).
How reliable are AI-generated codes for policy reports?
AI-generated codes are useful as candidate suggestions but require researcher validation: use Evidano to get candidate themes, then sample-check and refine to preserve rigor and maintain audit trails for reviewers.
Is this approach ethical for human-subject data?
This approach is ethical if you follow consent procedures and IRB guidance, and use PII redaction and secure storage: Evidano supports PII redaction and secure storage for research-focused projects.
Wrapping up & next steps
Wrapping up: You can cut synthesis time while keeping rigor by combining reproducible transcript workflows, AI-assisted coding, and cross-segment analytics for mixed-methods land-use studies like the Bure/Womberma paper.
- Ready to try it: upload a small batch (3–5 KIIs plus one survey CSV) into Evidano to auto-transcribe, generate candidate themes, and produce a 2-page stakeholder brief in under a day.
- Try Evidano for free to run a pilot, keep your data encrypted and research-only, link qualitative themes to LULC change, and make policy recommendations that stand up to peer review.