ecological-agent-skills 3.1.0

package/docs/repository-statistics.md

@@ -0,0 +1,101 @@
+# Repository Statistics
+
+Generated: 2026-03-06
+Version: 3.0.0
+
+---
+
+## Content Summary
+
+| Category | Count |
+|----------|-------|
+| Skills | 17 |
+| Workflows | 9 |
+| R scripts | 33 |
+| Python scripts | 25 |
+| Worked examples | 14 |
+| Resource documents | 53 |
+| Documentation files (docs/) | 4 |
+| Test datasets (CSV) | 11 |
+| Total files (excluding .git) | 323 |
+
+## Skill Breakdown
+
+| Phase | Skills | Count |
+|-------|--------|-------|
+| Phase 1 (Foundation) | ecological-data-foundation, geoprocessing-for-ecology, biostatistics-workbench, predictive-modeling-best-practices, reproducible-ecology-pipeline | 5 |
+| Phase 2 (Modeling) | model-validation-and-uncertainty, species-distribution-modeling, ecological-impact-assessment, environmental-time-series | 4 |
+| Phase 3 (Specialist) | occupancy-and-detection, community-ecology-ordination, ecosystem-services-assessment | 3 |
+| Phase 4 (Advanced) | camera-trap-processing, acoustic-monitoring, landscape-connectivity, population-viability-analysis, spatial-prioritization | 5 |
+
+## Geographic Coverage
+
+| Continent | Examples | Representative |
+|-----------|---------|---------------|
+| South America | 6 | Jaguar (Amazon), Anteater (Cerrado), Birds (Atlantic Forest), Phytoplankton (Amazon reservoirs), Puma (Atlantic Forest), BACI road (Atlantic Forest), Ecosystem services (São Paulo) |
+| Europe | 2 | Grey Wolf (Western Europe), Arctic tundra (Greenland) |
+| Asia | 3 | Reef fish (Indo-Pacific), Snow leopard (Himalayas), Forest loss (Borneo) |
+| Oceania | 2 | Koala (Australia), Reef fish (Indo-Pacific) |
+| North America | 2 | Red fox (Holarctic), Arctic tundra (Canada) |
+| Africa | 1 | Red fox (Holarctic — North Africa margin) |
+| **Total continents** | **6/6** | |
+
+## Taxonomic Coverage
+
+| Group | Species / Systems |
+|-------|------------------|
+| Mammals | Jaguar, anteater, wolf, koala, red fox, puma, snow leopard (7) |
+| Birds | Atlantic Forest bird community (127 spp.) |
+| Fish | Indo-Pacific reef fish (987 spp.) |
+| Plants | Arctic tundra vegetation |
+| Phytoplankton | Amazon reservoir assemblages |
+| Ecosystems | Borneo forest, Atlantic Forest remnants |
+
+## Analysis Types Covered
+
+| Analysis | Example count |
+|----------|-------------|
+| SDM (species distribution modeling) | 5 |
+| Community ecology (ordination, diversity) | 4 |
+| Occupancy modeling | 2 |
+| BACI impact assessment | 2 |
+| Time series (NDVI, BFAST) | 2 |
+| Ecosystem services | 1 |
+| Climate change projection | 1 |
+| Landscape fragmentation | 1 |
+| Conflict analysis | 1 |
+| Fully reproducible pipeline | 1 |
+
+## CI Check Results
+
+| Section | Checks |
+|---------|--------|
+| Structure checks | 585/585 passed |
+| Skills verified | 17 |
+| Workflows verified | 9 |
+| Global coverage | 6/6 continents |
+| Empty files | 0 (in tracked content) |
+| JSON validation | SKILL_INDEX.json valid, smoke_test_cases.json valid |
+
+## Test Suite
+
+| Test type | Count |
+|-----------|-------|
+| CI structural checks | 585 |
+| Python unit tests (pytest) | 176+ |
+| R unit tests (testthat) | 28+ |
+| Agent smoke test cases | 15 |
+| Regression test infrastructure | Ready (reference outputs to be generated) |
+
+## Release History
+
+| Version | Date | Highlights |
+|---------|------|-----------|
+| 1.0.0 | 2026-02-28 | Initial: 12 skills, 8 workflows, 7 examples |
+| 1.1.0 | 2026-03-01 | Rename, MaxEnt calibration, GBIF download, 9th workflow |
+| 1.2.0 | 2026-03-02 | Agent infrastructure (AGENT_CONTEXT, SKILL_INDEX, CI) |
+| 2.0.0 | 2026-03-03 | 5 advanced skills (camera trap, acoustic, connectivity, PVA, prioritization) |
+| 2.1.0 | 2026-03-04 | Structured logging in all 43 scripts, SDM prediction, BACI power analysis |
+| 2.2.0 | 2026-03-05 | Global data download scripts (iNat, eBird, OBIS, IUCN), predictor sources |
+| 2.3.0 | 2026-03-06 | Scientific documentation, 7 global examples, comparison with alternatives |
+| 3.0.0 | 2026-03-06 | Quality infrastructure: regression tests, smoke tests, expanded CI, release process |

package/docs/theoretical-foundations.md

@@ -0,0 +1,188 @@
+# Theoretical Foundations
+
+Reference document justifying the methodological decisions embedded in the ecological-agent-skills library. Intended as a citable companion for the Methods section of publications that use this skill set.
+
+---
+
+## 1. Spatial Cross-Validation over Random Cross-Validation
+
+**Problem**: Spatial autocorrelation inflates model performance metrics when training and test sets share spatially proximate records. Standard random k-fold CV assumes independence between folds — an assumption violated by virtually all georeferenced ecological data.
+
+**Solution**: Spatial block cross-validation partitions data into spatially disjoint folds whose block size equals or exceeds the species' spatial autocorrelation (SAC) range. This forces the model to predict into genuinely novel geographic space, producing honest estimates of transferability.
+
+**Primary reference**: Roberts, D.R., Bahn, V., Ciuti, S., et al. (2017). Cross-validation strategies for data with temporal, spatial, hierarchical, or phylogenetic structure. *Ecography*, 40(8), 913–929. doi:10.1111/ecog.02881
+
+**Supporting reference**: Valavi, R., Elith, J., Lahoz-Monfort, J.J. & Guillera-Arroita, G. (2019). blockCV: an R package for generating spatially or environmentally separated folds for k-fold cross-validation of species distribution models. *Methods in Ecology and Evolution*, 10, 225–232. doi:10.1111/2041-210X.13107
+
+**When random CV is still acceptable**: When occurrence data are spatially uniform (e.g., systematic grid surveys) and SAC range is negligible relative to study area extent. In practice, this is rare for opportunistic occurrence data.
+
+**Repository implementation**: `skills/predictive-modeling-best-practices/scripts/spatial_cv.py` and `spatial_cv.R`; block size is set to the SAC range estimated via variogram or Moran's I.
+
+---
+
+## 2. Ensemble of Models over Single Best Algorithm
+
+**Problem**: Model uncertainty — the variation in predictions among algorithms fitted to the same data — is the largest source of uncertainty in range projections, often exceeding climate scenario uncertainty.
+
+**Solution**: Combining predictions from multiple algorithms (e.g., MaxEnt, BRT, Random Forest, GLM) reduces variance and produces more robust suitability surfaces. The ensemble can be weighted by performance (TSS or AUC) or use equal weights.
+
+**Primary reference**: Araújo, M.B. & New, M. (2007). Ensemble forecasting of species distributions. *Trends in Ecology & Evolution*, 22(1), 42–47. doi:10.1016/j.tree.2006.09.010
+
+**Supporting reference**: Thuiller, W., Lafourcade, B., Engler, R. & Araújo, M.B. (2009). BIOMOD — a platform for ensemble forecasting of species distributions. *Ecography*, 32, 369–373. doi:10.1111/j.1600-0587.2008.05742.x
+
+**Weighting guidance**: TSS-weighted ensemble outperforms equal-weight when model performance varies substantially (TSS range > 0.15 among algorithms). Equal-weight is safer when all models perform similarly and sample size is small (n < 50 presences).
+
+**Repository implementation**: `skills/species-distribution-modeling/SKILL.md` Step 7; ensemble weights are computed from spatial CV metrics stored in validation outputs.
+
+---
+
+## 3. Partial ROC over Full AUC
+
+**Problem**: AUC integrates model performance across the entire ROC curve, including specificity regions that are ecologically irrelevant (very high commission rates no ecologist would accept). A model that performs poorly in the decision-relevant region can still achieve a high AUC.
+
+**Solution**: Partial ROC restricts evaluation to the portion of the ROC curve where omission rate is below a specified tolerance (e.g., E = 0.05 or 0.10), matching real-world acceptable error thresholds.
+
+**Primary reference**: Peterson, A.T., Papeş, M. & Soberón, J. (2008). Rethinking receiver operating characteristic analysis applications in ecological niche modeling. *Ecological Modelling*, 213(1), 63–72. doi:10.1016/j.ecolmodel.2007.11.008
+
+**Supporting reference**: Cobos, M.E., Peterson, A.T., Barve, N. & Osorio-Olvera, L. (2019). kuenm: an R package for detailed development of ecological niche models using Maxent. *PeerJ*, 7, e6281. doi:10.7717/peerj.6281
+
+**How to calculate**: Implemented via `ntbox::pROC()`, `enmSdmX::evalSDM()`, or `ENMeval::calc.pROC()`. Ratio of partial AUC (model) to partial AUC (random) yields the AUC ratio; values > 1.0 indicate better-than-random prediction within the tolerance region.
+
+**Repository implementation**: `skills/model-validation-and-uncertainty/SKILL.md` validation metrics section.
+
+---
+
+## 4. Calibration Area (M) Delimitation
+
+**Problem**: Background points sampled outside the species' accessible area (M) produce an artificially broad environmental niche estimate. The model learns to distinguish presence environments from environments the species has never encountered, inflating apparent discrimination.
+
+**Solution**: M should reflect the area historically accessible to the species via dispersal. Common M approximations: minimum convex polygon of occurrences buffered by estimated dispersal distance, biogeographic regions overlapping with occurrences, or distance-based buffers calibrated by home range.
+
+**Primary reference**: Barve, N., Barve, V., Jiménez-Valverde, A., et al. (2011). The crucial role of the accessible area in ecological niche modeling and species distribution modeling. *Ecological Modelling*, 222(11), 1810–1819. doi:10.1016/j.ecolmodel.2011.02.011
+
+**Supporting reference**: Soberón, J. & Peterson, A.T. (2005). Interpretation of models of fundamental ecological niches and species' distributional areas. *Biodiversity Informatics*, 2, 1–10. doi:10.17161/bi.v2i0.4
+
+**Repository implementation**: `skills/species-distribution-modeling/SKILL.md` Step 2 (study area delimitation); `skills/geoprocessing-for-ecology/resources/` provides spatial operation guides for M construction.
+
+---
+
+## 5. Bray-Curtis Dissimilarity for Community Data
+
+**Problem**: Euclidean distance is sensitive to "double zeros" — species absent from both sites contribute to perceived similarity, which is ecologically meaningless (shared absences ≠ ecological similarity).
+
+**Solution**: Bray-Curtis dissimilarity ignores joint absences, weighting only shared and unshared presences/abundances. It is the default metric for NMDS ordination and PERMANOVA in community ecology.
+
+**Primary reference**: Faith, D.P., Minchin, P.R. & Belbin, L. (1987). Compositional dissimilarity as a robust measure of ecological distance. *Vegetatio*, 69, 57–68. doi:10.1007/BF00045575
+
+**Supporting reference**: Legendre, P. & Legendre, L. (2012). *Numerical Ecology*, 3rd edn. Elsevier. Chapter 7.
+
+**When to use Jaccard**: Presence/absence data where abundance is unreliable or unavailable.
+**When to use Aitchison**: Compositional data (proportions summing to 1) after CLR transformation, common in microbiome and eDNA metabarcoding studies.
+
+**Repository implementation**: `skills/community-ecology-ordination/SKILL.md`; distance matrix computation in `scripts/ordination.R`.
+
+---
+
+## 6. Effective Mesh Size (MESH) for Fragmentation
+
+**Problem**: Simple fragmentation metrics such as Number of Patches (NP) or Total Class Area (CA) do not capture connectivity and are scale-dependent. A landscape with many small patches can have the same NP as one with a few large patches plus many tiny fragments.
+
+**Solution**: MESH (Effective Mesh Size) is the probability that two randomly placed points in the landscape fall within the same patch, expressed in area units. It is interpretable (expected patch size experienced by a randomly placed organism) and scale-invariant under the cross-boundary connectivity (CBC) approach.
+
+**Primary reference**: Jaeger, J.A.G. (2000). Landscape division, splitting index, and effective mesh size: new measures of landscape fragmentation. *Landscape Ecology*, 15(2), 115–130. doi:10.1023/A:1008129329289
+
+**Repository implementation**: `skills/ecological-impact-assessment/scripts/fragmentation_analysis.R` and `.py`.
+
+---
+
+## 7. IIC/PC over MESH for Functional Connectivity
+
+**Problem**: MESH is a structural metric — it measures fragmentation geometry but does not model species-specific dispersal capacity or movement between patches.
+
+**Solution**: Integral Index of Connectivity (IIC) and Probability of Connectivity (PC) model the landscape as a graph where patches are nodes and links are weighted by inter-patch dispersal probability. PC additionally accounts for stepping-stone connectivity (indirect paths through intermediate patches).
+
+**Primary reference**: Saura, S. & Pascual-Hortal, L. (2007). A new habitat availability index to integrate connectivity in landscape conservation planning: comparison with existing indices and application to a case study. *Landscape and Urban Planning*, 83(2–3), 91–103. doi:10.1016/j.landurbplan.2007.03.005
+
+**Supporting reference**: Saura, S. & Torné, J. (2009). Conefor Sensinode 2.2: a software package for quantifying the importance of habitat patches for landscape connectivity. *Environmental Modelling & Software*, 24(1), 135–139. doi:10.1016/j.envsoft.2008.05.005
+
+**Repository implementation**: `skills/landscape-connectivity/SKILL.md`; `scripts/connectivity_analysis.R` computes IIC, PC, and dPC for patch prioritization.
+
+---
+
+## 8. 10% Omission Rate (OR10) as MaxEnt Threshold
+
+**Problem**: The minimum training presence threshold (OR0, 0% omission) assumes zero error in occurrence coordinates — unrealistic for museum and citizen-science data with typical georeferencing errors of 1–10 km.
+
+**Solution**: OR10 (10th percentile training presence) tolerates 10% omission, accommodating coordinate imprecision and environmental marginality. It produces more ecologically realistic range maps than minimum training presence.
+
+**Primary reference**: Phillips, S.J., Anderson, R.P. & Schapire, R.E. (2006). Maximum entropy modeling of species geographic distributions. *Ecological Modelling*, 190(3–4), 231–259. doi:10.1016/j.ecolmodel.2005.03.026
+
+**Supporting reference**: Warren, D.L. & Seifert, S.N. (2011). Ecological niche modeling in Maxent: the importance of model complexity and the performance of model selection criteria. *Ecological Applications*, 21(2), 335–342. doi:10.1890/10-1171.1
+
+**Repository implementation**: `skills/species-distribution-modeling/SKILL.md` threshold selection step; model outputs include OR10 and MTP thresholds for comparison.
+
+---
+
+## 9. Regularization Multiplier Calibration in MaxEnt
+
+**Problem**: MaxEnt's default regularization multiplier (RM = 1.0) often produces overfitted models, especially for species with few occurrence records (n < 50). Overfitting manifests as artificially narrow predicted ranges that fail to transfer to novel environments.
+
+**Solution**: Calibrate RM across a range (e.g., 0.5–4.0 in steps of 0.5) and select the value that minimizes AICc or optimizes the trade-off between omission rate and model complexity. This process is automated by ENMeval and kuenm.
+
+**Primary reference**: Warren, D.L. & Seifert, S.N. (2011). Ecological niche modeling in Maxent: the importance of model complexity and the performance of model selection criteria. *Ecological Applications*, 21(2), 335–342. doi:10.1890/10-1171.1
+
+**Supporting reference**: Muscarella, R., Galante, P.J., Soley-Guardia, M., et al. (2014). ENMeval: an R package for conducting spatially independent evaluations and estimating optimal model complexity for Maxent ecological niche models. *Methods in Ecology and Evolution*, 5, 1198–1205. doi:10.1111/2041-210X.12261
+
+**Repository implementation**: `skills/species-distribution-modeling/resources/maxent-calibration-guide.md`; calibration script in `scripts/`.
+
+---
+
+## 10. Random Effects for Site in BACI Designs
+
+**Problem**: In Before-After-Control-Impact (BACI) designs, repeated measurements at the same site are not independent. Ignoring this intra-site correlation (pseudoreplication) inflates Type I error rates and produces unreliable p-values for the BA×CI interaction.
+
+**Solution**: Fit a mixed-effects model with `(1|site)` random intercept to absorb between-site variation. The fixed-effect BA×CI interaction then tests the true impact effect with correct degrees of freedom.
+
+**Primary reference**: Underwood, A.J. (1994). On beyond BACI: sampling designs that might reliably detect environmental disturbances. *Ecological Applications*, 4(1), 3–15. doi:10.2307/1942110
+
+**Supporting reference**: Schwarz, C.J. (2015). Analysis of BACI experiments. In *Course Notes for Beginning and Intermediate Statistics*. Available: http://www.stat.sfu.ca/~cschwarz/CourseNotes
+
+**Repository implementation**: `skills/ecological-impact-assessment/scripts/baci_analysis.R`; model specification uses `lmer(response ~ period * treatment + (1|site))`.
+
+---
+
+## Recommended Citation
+
+When using ecological-agent-skills in a publication, cite as:
+
+```
+[Your Name] (2026). ecological-agent-skills: A modular skill library for
+quantitative ecology with AI agent integration. Version [X.Y.Z].
+https://github.com/[user]/ecological-agent-skills
+```
+
+In the Methods section, reference specific methodological justifications:
+
+> "Spatial cross-validation block size was set to exceed the species' SAC range following Roberts et al. (2017); model ensemble was TSS-weighted following Araújo & New (2007). Full methodological justifications are documented in the theoretical-foundations companion document (https://github.com/[user]/ecological-agent-skills/blob/main/docs/theoretical-foundations.md)."
+
+---
+
+## References (consolidated)
+
+- Araújo, M.B. & New, M. (2007). *Trends in Ecology & Evolution*, 22, 42–47. doi:10.1016/j.tree.2006.09.010
+- Barve, N. et al. (2011). *Ecological Modelling*, 222, 1810–1819. doi:10.1016/j.ecolmodel.2011.02.011
+- Cobos, M.E. et al. (2019). *PeerJ*, 7, e6281. doi:10.7717/peerj.6281
+- Faith, D.P. et al. (1987). *Vegetatio*, 69, 57–68. doi:10.1007/BF00045575
+- Jaeger, J.A.G. (2000). *Landscape Ecology*, 15, 115–130. doi:10.1023/A:1008129329289
+- Legendre, P. & Legendre, L. (2012). *Numerical Ecology*, 3rd edn. Elsevier.
+- Muscarella, R. et al. (2014). *Methods in Ecology and Evolution*, 5, 1198–1205. doi:10.1111/2041-210X.12261
+- Peterson, A.T. et al. (2008). *Ecological Modelling*, 213, 63–72. doi:10.1016/j.ecolmodel.2007.11.008
+- Phillips, S.J. et al. (2006). *Ecological Modelling*, 190, 231–259. doi:10.1016/j.ecolmodel.2005.03.026
+- Roberts, D.R. et al. (2017). *Ecography*, 40, 913–929. doi:10.1111/ecog.02881
+- Saura, S. & Pascual-Hortal, L. (2007). *Landscape and Urban Planning*, 83, 91–103. doi:10.1016/j.landurbplan.2007.03.005
+- Saura, S. & Torné, J. (2009). *Environmental Modelling & Software*, 24, 135–139. doi:10.1016/j.envsoft.2008.05.005
+- Soberón, J. & Peterson, A.T. (2005). *Biodiversity Informatics*, 2, 1–10. doi:10.17161/bi.v2i0.4
+- Thuiller, W. et al. (2009). *Ecography*, 32, 369–373. doi:10.1111/j.1600-0587.2008.05742.x
+- Underwood, A.J. (1994). *Ecological Applications*, 4, 3–15. doi:10.2307/1942110
+- Valavi, R. et al. (2019). *Methods in Ecology and Evolution*, 10, 225–232. doi:10.1111/2041-210X.13107
+- Warren, D.L. & Seifert, S.N. (2011). *Ecological Applications*, 21, 335–342. doi:10.1890/10-1171.1

package/environment.yaml

@@ -0,0 +1,106 @@
+name: ecological-agent-skills
+# SPDX-License-Identifier: GPL-3.0-or-later
+channels:
+  - conda-forge
+  - defaults
+
+dependencies:
+  - python=3.11
+  - r-base=4.4.1
+
+  # ── Core Python ────────────────────────────────────────────────────────
+  - numpy=1.26.4
+  - pandas=2.2.2
+  - scipy=1.13.1
+  - matplotlib=3.9.0
+  - seaborn=0.13.2
+
+  # ── Geospatial Python ─────────────────────────────────────────────────
+  - geopandas=0.14.4
+  - rasterio=1.3.10
+  - shapely=2.0.4
+  - pyproj=3.6.1
+  - fiona=1.9.6
+  - rasterstats=0.19.0
+
+  # ── Statistics / ML Python ────────────────────────────────────────────
+  - scikit-learn=1.5.0
+  - statsmodels=0.14.2
+
+  # ── Ecology Python ────────────────────────────────────────────────────
+  - pip
+
+  # ── Acoustic monitoring Python ────────────────────────────────────────
+  - librosa=0.10.2
+  - soundfile=0.12.1
+
+  # ── Landscape connectivity Python ─────────────────────────────────────
+  - networkx=3.3
+  - scikit-image=0.23.2
+
+  # ── Workflow Python ───────────────────────────────────────────────────
+  - pyyaml=6.0.1
+  - pathlib2=2.3.7
+
+  # ── R core packages (via conda-forge) ─────────────────────────────────
+  - r-terra=1.7_78
+  - r-sf=1.0_16
+  - r-dplyr=1.1.4
+  - r-ggplot2=3.5.1
+  - r-tidyr=1.3.1
+  - r-readr=2.1.5
+  - r-yaml=2.3.8
+  - r-janitor=2.2.0
+  - r-lubridate=1.9.3
+  - r-zoo=1.8_12
+  - r-broom=1.0.6
+
+  # ── R statistics ──────────────────────────────────────────────────────
+  - r-lme4=1.1_35.3
+  - r-emmeans=1.10.2
+  - r-mumin=1.47.5
+  - r-effectsize=0.8.9
+  - r-corrplot=0.92
+
+  # ── R ecology ─────────────────────────────────────────────────────────
+  - r-vegan=2.6_6.1
+  - r-unmarked=1.4.1
+  - r-gbm=2.2.2
+  - r-randomforest=4.7_1.1
+  - r-trend=1.1.6
+
+  # ── R advanced skills (v2.0.0) ────────────────────────────────────────
+  - r-igraph=2.0.3
+  - r-ggrepel=0.9.5
+
+  # ── R occurrence data download (v2.2.0) ───────────────────────────────
+  - r-rgbif=3.7.9
+  - r-geodata=0.6_2
+
+  pip:
+    # Python ecology (not on conda-forge)
+    - pymannkendall==1.4.3
+    - scikit-bio==0.6.2
+    - elapid==1.1.2
+
+    # ── Occurrence data download (v2.2.0) ─────────────────────────────────
+    - pygbif==0.6.3
+    - pyinaturalist==0.19.0
+    - requests==2.32.3
+    - cdsapi==0.7.2
+
+    # ── Spatial predictor download (v2.2.0) ───────────────────────────────
+    - pystac-client==0.8.3
+    - stackstac==0.5.1
+    - planetary-computer==1.0.0
+
+    # R packages not on conda-forge (install via renv after environment creation)
+    # See renv.lock for: CoordinateCleaner, blockCV, maxnet, biomod2,
+    #                    bfast, landscapemetrics, DHARMa, glmmTMB, betapart, SPEI,
+    #                    camtrapR, overlap, circular, soundecology, tuneR,
+    #                    popbio, prioritizr, highs,
+    #                    rinat, auk, robis, rredlist (v2.2.0)
+
+variables:
+  R_LIBS_USER: "${CONDA_PREFIX}/lib/R/library"
+  OMP_NUM_THREADS: "4"

package/examples/community/arctic_tundra_vegetation_example.md

@@ -0,0 +1,247 @@
+# Worked Example: Arctic Tundra Vegetation — NDVI Greening, Community Shifts and Temperature Anomalies
+
+**Workflow:** analyze-community-structure + environmental-time-series
+**System:** Arctic tundra vegetation, Greenland and northern Canada
+**Sites:** 5 study regions spanning High Arctic (CAVM zones A--C) and Low Arctic (zones D--E)
+**Temporal extent:** 2000--2023 (24 growing seasons)
+**Analysis:** NDVI time series trend detection, BFAST breakpoint analysis, temperature anomaly correlation, community composition shift synthesis
+
+---
+
+## Step 1 — Data Sources
+
+| Item | Detail |
+|------|--------|
+| NDVI | MODIS MOD13Q1, 250 m spatial resolution, 16-day composite, Collection 6.1 |
+| NDVI temporal range | 2000--2023 (24 years, growing season June--August) |
+| Temperature | ERA5-Land 2 m air temperature, monthly means, 0.1 degree resolution |
+| Temperature temporal range | 2000--2023, summer months (JJA) |
+| Vegetation plots | Arctic Vegetation Archive (AVA), community plot data (Walker et al. 2005) |
+| Vegetation map | Circumpolar Arctic Vegetation Map (CAVM Team 2003), bioclimatic zones A--E |
+| Pixel reliability | MOD13Q1 QA band; only pixels flagged as "good" or "marginal" retained |
+
+---
+
+## Step 2 — Study Design
+
+### Study Regions
+
+| Region | Latitude | Longitude | CAVM zone | Arctic class | Extraction window |
+|--------|----------|-----------|-----------|-------------|-------------------|
+| Zackenberg | 74.5 N | 21.0 W | B | High Arctic | 100 x 100 km |
+| Disko Island | 69.3 N | 53.5 W | D | Low Arctic | 100 x 100 km |
+| Cambridge Bay | 69.1 N | 105.1 W | C | High Arctic | 100 x 100 km |
+| Churchill | 58.8 N | 94.2 W | E | Low Arctic | 100 x 100 km |
+| Bylot Island | 73.2 N | 80.0 W | B | High Arctic | 100 x 100 km |
+
+- Each region represented by a 100 x 100 km MODIS extraction window centred on the study area
+- Pixels classified as water, permanent ice, or barren rock masked using the CAVM land cover layer
+- Vegetation zones assigned per CAVM bioclimatic classification: High Arctic (zones A--C, mean July temperature < 9 C) and Low Arctic (zones D--E, mean July temperature 9--12 C)
+
+---
+
+## Step 3 — NDVI Time Series Analysis
+
+Growing season NDVI calculated as the mean of all 16-day MOD13Q1 composites falling within June 1 -- August 31 for each region and year (24 annual values per region).
+
+### Mann-Kendall Trend Test
+
+| Region | CAVM zone | Trend direction | Sen's slope (NDVI/decade) | tau | p-value | Classification |
+|--------|-----------|-----------------|--------------------------|-----|---------|----------------|
+| Zackenberg | B | Positive | +0.008 | 0.34 | 0.012 | Greening |
+| Disko Island | D | Positive | +0.014 | 0.48 | 0.002 | Strong greening |
+| Cambridge Bay | C | Positive | +0.011 | 0.41 | 0.008 | Greening |
+| Churchill | E | Positive | +0.006 | 0.21 | 0.087 | Non-significant |
+| Bylot Island | B | Negative | -0.003 | -0.11 | 0.341 | Stable |
+
+- Three of five regions show statistically significant greening trends (p < 0.05)
+- Disko Island exhibits the strongest greening signal (Sen's slope = +0.014 NDVI units/decade), consistent with pronounced shrub expansion documented in Low Arctic western Greenland
+- Churchill shows a weak positive trend that does not reach significance (p = 0.087), potentially reflecting competing browning from permafrost thaw disturbance
+- Bylot Island shows no significant trend; High Arctic polar desert sites remain moisture- and substrate-limited
+
+---
+
+## Step 4 — Temperature Anomaly Correlation
+
+Summer (JJA) temperature anomalies computed relative to the 2000--2010 baseline mean for each region. Growing-season NDVI correlated with JJA temperature anomaly using Pearson's r.
+
+### Pearson Correlation: JJA Temperature Anomaly vs Growing-Season NDVI
+
+| Region | CAVM zone | Pearson r | p-value | 95% CI |
+|--------|-----------|-----------|---------|--------|
+| Zackenberg | B | 0.54 | 0.006 | 0.18--0.78 |
+| Disko Island | D | 0.71 | < 0.001 | 0.44--0.87 |
+| Cambridge Bay | C | 0.63 | 0.001 | 0.31--0.82 |
+| Churchill | E | 0.42 | 0.041 | 0.02--0.71 |
+| Bylot Island | B | 0.47 | 0.021 | 0.08--0.74 |
+
+- All correlations positive, indicating warmer summers are consistently associated with higher NDVI
+- Strongest correlation at Disko Island (r = 0.71), where Low Arctic shrub tundra responds rapidly to thermal amelioration
+- Even Bylot Island, which shows no long-term NDVI trend, displays significant interannual covariation with temperature (r = 0.47)
+
+### Cross-Correlation Analysis
+
+| Region | Lag with maximum correlation | r at optimal lag |
+|--------|------------------------------|-----------------|
+| Zackenberg | 0 years | 0.54 |
+| Disko Island | 0 years | 0.71 |
+| Cambridge Bay | 0 years | 0.63 |
+| Churchill | 1 year | 0.48 |
+| Bylot Island | 0 years | 0.47 |
+
+- NDVI response to temperature is predominantly contemporaneous (lag 0) across most regions
+- Churchill shows a marginally stronger correlation at lag 1 (r = 0.48 vs 0.42 at lag 0), possibly reflecting delayed vegetation response mediated by permafrost active-layer dynamics
+
+---
+
+## Step 5 — BFAST Breakpoint Detection
+
+Breaks For Additive Season and Trend (BFAST) applied to the full 16-day NDVI time series (2000--2023) per region to detect structural breaks in the trend component.
+
+### Detected Breakpoints
+
+| Region | Breakpoint year | JJA temp anomaly at break (C) | Anomaly in SD units | Post-break NDVI shift | Interpretation |
+|--------|----------------|-------------------------------|---------------------|----------------------|----------------|
+| Disko Island | 2012 | +2.4 | +2.3 SD | +0.031 (persistent) | Regime shift |
+| Cambridge Bay | 2010 | +2.1 | +2.1 SD | +0.024 (persistent) | Regime shift |
+| Zackenberg | None | -- | -- | -- | Gradual trend |
+| Churchill | None | -- | -- | -- | No clear trend |
+| Bylot Island | None | -- | -- | -- | Stable |
+
+- Significant structural breaks detected at Disko Island (2012) and Cambridge Bay (2010)
+- Both breakpoints correspond to anomalously warm summers exceeding +2 SD above the 2000--2010 baseline
+- Post-breakpoint NDVI remains consistently elevated relative to pre-breakpoint values, indicating a regime shift rather than a transient spike
+- Zackenberg shows steady greening without abrupt transitions, suggesting a more gradual response
+
+```
+# BFAST implementation (R)
+library(bfast)
+ndvi_ts <- ts(disko_ndvi, start = c(2000, 1), frequency = 23)  # 23 composites/year
+bfast_result <- bfast(ndvi_ts, h = 0.15, season = "harmonic", max.iter = 10)
+```
+
+---
+
+## Step 6 — Community Composition Shift (Literature Synthesis)
+
+Vegetation change documented at long-term Arctic monitoring sites, synthesised from published plot-level resurvey data.
+
+### Documented Vegetation Shifts at Study Regions
+
+| Region | Period | Dominant change | Magnitude | Source |
+|--------|--------|----------------|-----------|--------|
+| Zackenberg | 1996--2018 | Graminoid expansion into barren microsites | +8% cover | Schmidt et al. 2012 |
+| Disko Island | 2002--2019 | Deciduous shrub cover increase (*Betula nana*, *Salix glauca*) | +34% cover | Myers-Smith et al. 2011; local resurvey |
+| Cambridge Bay | 2000--2020 | Prostrate shrub expansion into frost-boil complexes | +18% cover | Walker et al. 2005; resurvey data |
+| Churchill | 1998--2018 | Spruce treeline advance; shrub densification | +15% shrub cover | Elmendorf et al. 2012 |
+| Bylot Island | 2004--2019 | Moss/lichen communities stable; minor graminoid increase | +3% cover | Gauthier et al. 2013 |
+
+### Functional Group Trends Across All Sites
+
+| Functional group | High Arctic trend | Low Arctic trend |
+|-----------------|-------------------|-----------------|
+| Deciduous shrubs (*Betula*, *Salix*) | Stable to slight increase | Strong increase (+15--40%) |
+| Evergreen shrubs (*Cassiope*, *Dryas*) | Stable | Slight decline in shrub-invaded areas |
+| Graminoids (*Carex*, *Eriophorum*) | Expanding into barren microsites | Stable to slight increase |
+| Mosses | Stable to slight decline | Decline where shaded by shrubs (-12%) |
+| Lichens (*Cladonia*, *Cetraria*) | Stable | Decline where shaded by shrubs (-18%) |
+| Forbs | Stable | Stable |
+
+### PERMANOVA on AVA Plot Data (2005 vs 2020 Resurvey)
+
+```
+adonis2(bray_dist ~ period * arctic_zone, data = ava_meta, permutations = 999)
+
+                   Df  SumOfSqs    R2      F    Pr(>F)
+period              1    1.847  0.140  14.21   0.001 ***
+arctic_zone         1    2.914  0.221  22.42   0.001 ***
+period:zone         1    0.412  0.031   3.17   0.008 **
+Residual           76    8.012  0.608
+Total              79   13.185  1.000
+```
+
+- Period effect (R2 = 0.14, p = 0.001): significant community compositional change between 2005 and 2020 resurveys
+- Arctic zone effect (R2 = 0.22, p = 0.001): High and Low Arctic plots remain compositionally distinct
+- Significant interaction (R2 = 0.03, p = 0.008): Low Arctic plots shifted more than High Arctic plots, consistent with accelerated shrub expansion in warmer tundra
+
+---
+
+## Step 7 — Synthesis: NDVI Greening vs Community Change
+
+### Cross-Validation of Remote Sensing and Ground-Truth Data
+
+| Region | NDVI trend | Ground-truth change | Consistency |
+|--------|-----------|-------------------|-------------|
+| Zackenberg | Greening (+0.008/decade) | Graminoid expansion (+8% cover) | Consistent |
+| Disko Island | Strong greening (+0.014/decade) | Shrub increase (+34% cover) | Strongly consistent |
+| Cambridge Bay | Greening (+0.011/decade) | Shrub expansion (+18% cover) | Consistent |
+| Churchill | Non-significant (+0.006/decade) | Mixed: shrub increase + permafrost thaw | Partially consistent |
+| Bylot Island | Stable (-0.003/decade) | Minimal change (+3% cover) | Consistent |
+
+Key patterns:
+
+1. **Greening is spatially heterogeneous** — not a uniform pan-Arctic signal. Low Arctic sites show substantially stronger greening than High Arctic sites.
+2. **Low Arctic greening is driven by shrub expansion.** Deciduous shrub cover increases of +15--40% at Low Arctic sites correspond directly to the strongest NDVI trends.
+3. **High Arctic sites show variable responses.** Where greening occurs, it is associated with graminoid expansion into formerly barren microsites, modulated by moisture availability and substrate stability.
+4. **Churchill's non-significant trend reflects competing processes.** Localised permafrost thaw and thermokarst-driven browning offset temperature-driven greening, yielding a net non-significant trend despite documented shrub densification.
+5. **BFAST breakpoints align with community regime shifts.** The structural breaks at Disko Island (2012) and Cambridge Bay (2010) coincide with years of anomalous warmth that appear to have crossed ecological thresholds for shrub establishment.
+
+---
+
+## Ecological Interpretation
+
+1. **Climate sensitivity.** Arctic tundra is among the most climate-sensitive biomes globally. Summer warming of +1--2 C over two decades has produced measurable increases in vegetation greenness and biomass across most Low Arctic sites.
+
+2. **Shrub expansion as a primary greening mechanism.** NDVI greening is predominantly driven by the expansion of deciduous shrubs, particularly *Betula nana* and *Salix* spp. These species respond rapidly to thermal amelioration through increased lateral growth and infilling of open tundra.
+
+3. **Positive feedback dynamics.** Shrub expansion initiates a positive feedback loop: taller shrubs trap snow during winter, insulating the underlying soil and raising winter soil temperatures. Warmer soils accelerate decomposition and nutrient mineralisation, increasing nitrogen availability, which further promotes shrub growth. This snow-shrub feedback is a key driver of non-linear tundra change.
+
+4. **Threshold dynamics.** BFAST breakpoints at Disko Island and Cambridge Bay indicate that vegetation change in some regions proceeds through abrupt transitions rather than gradual trends. Anomalously warm summers exceeding +2 SD above baseline appear to cross establishment thresholds for shrub cohorts, producing persistent regime shifts in NDVI.
+
+5. **Spatial heterogeneity reflects local controls.** The divergence among study regions is not simply a function of latitude or mean temperature. Local moisture availability, permafrost condition, soil substrate, and topographic exposure interact to modulate the vegetation response to warming. Churchill exemplifies how permafrost degradation can counteract temperature-driven greening.
+
+6. **Moss and lichen decline.** In areas with vigorous shrub expansion, understory mosses and lichens decline due to shading. This has implications for caribou/reindeer foraging habitat (lichen dependence) and for soil thermal regimes (moss insulation effects).
+
+---
+
+## Recommendations
+
+1. **Combine remote sensing with ground-truth surveys.** NDVI captures aggregate greenness but cannot distinguish among functional groups. Paired MODIS analysis and plot-level resurveys are essential for ecological interpretation.
+
+2. **BFAST is effective for regime shift detection.** The method successfully identified structural breaks corresponding to documented ecological transitions. Apply BFAST to the full 16-day NDVI time series rather than annual aggregates to maximise detection power.
+
+3. **Include moisture variables.** Temperature alone explains 18--50% of interannual NDVI variance. Incorporating precipitation, soil moisture (ERA5-Land volumetric soil water), and snow-cover duration will improve trend attribution.
+
+4. **Monitor browning signals.** Sites with permafrost degradation (e.g., Churchill) may exhibit browning that offsets greening. Active-layer depth monitoring and thermokarst mapping should accompany NDVI trend analysis.
+
+5. **Account for NDVI saturation.** In dense Low Arctic shrub tundra, NDVI saturates at moderate-to-high leaf area index. Consider supplementing with Enhanced Vegetation Index (EVI) or Solar-Induced Fluorescence (SIF) for sites where shrub cover exceeds approximately 60%.
+
+6. **Long-term monitoring continuity.** The 24-year MODIS record is approaching the minimum length for robust trend detection in high-latitude systems with strong interannual variability. Continuity via VIIRS (Suomi NPP, NOAA-20) and Sentinel-2 is essential.
+
+---
+
+## References
+
+| Reference | DOI |
+|-----------|-----|
+| Myers-Smith, I.H. et al. (2011). Shrub expansion in tundra ecosystems: dynamics, impacts and research priorities. *Environmental Research Letters*, 6(4), 045509. | doi:10.1088/1748-9326/6/4/045509 |
+| Walker, D.A. et al. (2005). The Circumpolar Arctic Vegetation Map. *Journal of Vegetation Science*, 16(3), 267--282. | doi:10.1111/j.1654-1103.2005.tb02365.x |
+| Verbesselt, J. et al. (2010). Detecting trend and seasonal changes in satellite image time series. *Remote Sensing of Environment*, 114(1), 106--115. | doi:10.1016/j.rse.2009.08.014 |
+| Elmendorf, S.C. et al. (2012). Plot-scale evidence of tundra vegetation change and links to recent summer warming. *Nature Climate Change*, 2, 453--457. | doi:10.1038/nclimate1465 |
+| CAVM Team (2003). Circumpolar Arctic Vegetation Map (1:7,500,000 scale). Conservation of Arctic Flora and Fauna (CAFF) Map No. 1. U.S. Fish and Wildlife Service, Anchorage, AK. | -- |
+| Didan, K. (2021). MODIS/Terra Vegetation Indices 16-Day L3 Global 250m SIN Grid V061. NASA EOSDIS Land Processes DAAC. | doi:10.5067/MODIS/MOD13Q1.061 |
+| Hersbach, H. et al. (2020). The ERA5 global reanalysis. *Quarterly Journal of the Royal Meteorological Society*, 146(730), 1999--2049. | doi:10.1002/qj.3803 |
+
+### R and Python Packages Used
+
+| Package | Language | Purpose |
+|---------|----------|---------|
+| `bfast` | R | Breaks For Additive Season and Trend decomposition |
+| `trend` | R | Mann-Kendall test and Sen's slope estimator |
+| `vegan` | R | PERMANOVA (adonis2), Bray-Curtis dissimilarity |
+| `terra` | R | MODIS raster processing and extraction |
+| `MODIStsp` | R | Automated MODIS time series download and preprocessing |
+| `xarray` | Python | ERA5-Land NetCDF processing and anomaly computation |
+| `scipy.stats` | Python | Pearson correlation and cross-correlation analysis |
+| `matplotlib` | Python | Time series and trend visualisation |
+| `pandas` | Python | Tabular data wrangling |