# Alpine Treeline Ecotone (ATE) Detection — Method Comparison # # Compares two approaches to detecting alpine treeline ecotones: # # 1. ATEI (Alpine Treeline Ecotone Index) — Wei et al. (2020) # Index-based detection using three NDVI/elevation gradient components # combined via logistic regression. # # 2. ML Classification (Random Forest) — after Wang et al. (2026) # Supervised pixel classification using spectral bands, vegetation # indices, and topographic features. # # Validated against 268 independent field-identified treeline elevations # from Malanson & Skibbe (2023) across 66 mountain ranges in the Western # US and Canada. Each site's treeline was identified visually from Google # Earth imagery — fully independent of both ATEI and RF. # # The each: block fans out ATEI computation across all 268 validation # sites, estimating treeline elevation at each one. Results are compared # against the field observations to produce accuracy metrics (r², RMSE). # # References: # Wei, C., Karger, D.N., Wilson, A.M. (2020). Spatial detection of # alpine treeline ecotones in the Western US. RSE, 240, 111672. # Wang, G.-G. et al. (2026). ML-based ATE detection on Xue Mountain # in Taiwan. Biogeosciences, 23, 623-638. # Malanson, G.P. & Skibbe, A.M. (2023). Potential alpine habitat in # the western USA based on treeline elevation [Dataset]. Dryad. # doi:10.5061/dryad.sqv9s4n9m # Wei, C. & Wilson, A.M. (2024). Global Alpine Treeline Elevational # Transects [Dataset]. Zenodo. doi:10.5281/zenodo.10047965 # # Pipeline: # # data/weiss_268_treeline.csv ── validation/ground-truth # │ # For EACH of 268 sites: │ # Landsat 8 → NDVI ──┐ │ # USGS NED DEM ───────┤ │ # C1 (gradient mag) ┤ │ # C2 (Gaussian NDVI)┤ │ # C3 (covariation) ┤ │ # → ATEI → treeline elev estimate │ # │ │ # └───── validation/comparison # (predicted vs observed, r², RMSE) name: ate-analysis version: "2.0" description: > Alpine treeline ecotone detection validated against 268 independent field sites. Computes ATEI (Wei et al. 2020) at each site using folia's each: fan-out, then compares predicted treeline elevation against Malanson & Skibbe (2023) ground truth. # ============================================================ # Reproduction metadata (ADR-0032) # # This block declares provenance, quality tier, and the registry # operations used to reproduce the paper's methodology. It does # not affect pipeline execution — it is consumed by the catalog # ingestion layer when publishing to data.folia.sh. # ============================================================ reproduction: paper_doi: "10.1016/j.rse.2020.111672" paper_title: "Wei et al. 2020 — Spatial detection of alpine treeline ecotones in the Western US" quality_tier: human # recipe | agent | human | peer | canonical validation_status: validated method_ops: - buffer_point - landsat8_ndvi - load_dem - atei_components - atei_index - atei_weighted_elevation - table_join validation_sites: 268 validation_source: "Malanson & Skibbe (2023), doi:10.5061/dryad.sqv9s4n9m" additional_papers: - doi: "10.5194/bg-23-623-2026" title: "Wang et al. 2026 — ML-based ATE detection (RF comparison)" - doi: "10.5061/dryad.sqv9s4n9m" title: "Malanson & Skibbe 2023 — Ground truth dataset (Dryad)" - doi: "10.5281/zenodo.10047965" title: "Wei & Wilson 2024 — Global alpine treeline transects (Zenodo)" settings: default_crs: EPSG:4326 year_start: 2013 year_end: 2022 layers: # ============================================================ # GROUND TRUTH — Malanson & Skibbe (2023) # # 268 treeline elevations across 66 mountain ranges in the # Western US + Canada. Identified from Google Earth imagery # by visual interpretation. Independent of ATEI/NDVI methods. # ============================================================ validation/ground-truth: type: table uri: data/weiss_268_treeline.csv description: > 268 field-identified treeline elevations from Malanson & Skibbe (2023, Dryad doi:10.5061/dryad.sqv9s4n9m). Each point is the treeline position on one aspect (N/E/S/W) of a mountain range, identified visually from Google Earth high-resolution imagery. Columns: site_id, lat, lon, treeline_elev (m), IDW, kriging. # ============================================================ # ATEI AT EACH VALIDATION SITE (each: fan-out) # # For each of the 268 sites, we: # 1. Buffer the point by 5km to create a local study area # 2. Load Landsat 8 NDVI and USGS NED DEM # 3. Compute ATEI (C1 + C2 + C3 → logistic regression) # 4. Estimate treeline elevation as the ATEI-weighted mean # elevation (Eq. 6 in Wei et al. 2020) # # This produces one estimated treeline elevation per site, # directly comparable to the ground truth. # ============================================================ atei/per-site: type: table each: source: type: csv path: data/weiss_268_treeline.csv key: site_id compute: engine: gee steps: # Buffer site point → 5km study area - op: buffer_point params: lat: $each.lat lon: $each.lon radius_km: 5 # Landsat 8 C2 L2 — greenest pixel composite - op: landsat8_ndvi params: year_start: "${year_start}" year_end: "${year_end}" smooth_kernel: { type: circle, radius: 10, units: pixels } # USGS NED elevation - op: load_dem params: source: USGS/NED smooth_kernel: { type: circle, radius: 10, units: pixels } # ATEI components (Wei et al. 2020) - op: atei_components params: gaussian_b: 0.44 # C2 peak NDVI gaussian_c: 0.06 # C2 width covariation_n: 10 # C3 exponent theta_smooth: 3 # C3 direction smoothing radius # ATEI logistic regression - op: atei_index params: beta_0: -1.47 beta_1: 0.44 beta_2: 0.58 beta_3: 0.56 threshold: 0.357 # Estimate treeline elevation (Wei et al. 2020, Eq. 6) # ATEE_j = Σ(E_ij × ATEI_ij) / Σ(ATEI_ij) # ATEI-weighted average elevation within the 5km buffer - op: atei_weighted_elevation reduce: mode: concat # ============================================================ # VALIDATION — Predicted vs Observed # ============================================================ validation/comparison: type: table description: > Join ATEI-estimated treeline elevation with ground truth. Metrics: Pearson's r, RMSE, mean absolute error. Wei et al. (2020) achieved r = 0.98 and RMSE ≈ 170m at 22 sites from Weiss et al. (2015). This extends validation to 268 sites from Malanson & Skibbe (2023). compute: op: table_join inputs: left: { layer: atei/per-site } right: { layer: validation/ground-truth } params: on: site_id metrics: [pearson_r, rmse, mae, bias] style: chart: type: scatter x: treeline_elev # observed (ground truth) y: atei_elev_estimate # predicted (ATEI) identity_line: true xLabel: "Field-Observed Treeline Elevation (m)" yLabel: "ATEI-Estimated Treeline Elevation (m)" title: "ATEI Validation: 268 Sites (Malanson & Skibbe 2023)" # ============================================================ # SINGLE-SITE DETAIL VIEW (Glacier NP) # # For interactive exploration of the ATEI components and # RF comparison at one representative location. # ============================================================ detail/ndvi: type: raster description: > Landsat 8 smoothed annual max NDVI at Glacier NP for interactive inspection of the ATEI input data. compute: engine: gee steps: - op: landsat8_ndvi params: bbox: [-113.8, 48.55, -113.2, 48.85] year_start: "${year_start}" year_end: "${year_end}" smooth_kernel: { type: circle, radius: 10, units: pixels } style: colorRamp: - { value: 0.0, color: "#d7191c" } - { value: 0.2, color: "#fdae61" } - { value: 0.4, color: "#ffffbf" } - { value: 0.6, color: "#a6d96a" } - { value: 0.8, color: "#1a9641" } detail/atei: type: raster description: > ATEI probability surface at Glacier NP. Treeline should appear as a band of high ATEI (>0.357) around 1960-2040m, matching the Malanson ground truth points in this area. compute: engine: gee steps: - op: atei_full params: bbox: [-113.8, 48.55, -113.2, 48.85] year_start: "${year_start}" year_end: "${year_end}" style: colorRamp: - { value: 0.0, color: "#2166ac" } - { value: 0.2, color: "#67a9cf" } - { value: 0.357, color: "#fddbc7" } - { value: 0.5, color: "#ef8a62" } - { value: 0.8, color: "#b2182b" }