ecological-agent-skills 3.1.0

package/skills/landscape-connectivity/examples/jaguar_mesoamerica_corridor_example.md

@@ -0,0 +1,271 @@
+---
+skill_id: landscape-connectivity
+example_type: full_walkthrough
+taxon: Jaguar (Panthera onca)
+region: Mesoamerican Biological Corridor
+---
+
+# Jaguar Corridor Analysis — Mesoamerican Biological Corridor Walkthrough
+
+## Study Context
+
+**Location:** Guatemala–Honduras–Nicaragua corridor, Mesoamerican Biological Corridor (MBC)
+**Objective:** Identify critical habitat patches and movement corridors for jaguars along the MBC; prioritise areas for protection and restoration
+**Input data:**
+- Forest cover 2023 (30-m resolution, derived from Global Forest Watch)
+- Land cover map (11 classes, 30-m resolution)
+- DEM (SRTM 30 m)
+- Road network (OSM)
+- 124 camera trap stations with jaguar detections (occupancy data)
+
+---
+
+## Step 1 — Build Resistance Surface
+
+### Resistance table (jaguar-specific)
+
+```
+data/jaguar_resistance.csv:
+lc_code, resistance, description
+1,  1,   Dense tropical forest
+2,  3,   Secondary forest
+3,  5,   Riparian gallery forest
+4,  8,   Cerrado/savanna
+5, 15,   Shrubland
+6, 25,   Pasture (low density)
+7, 50,   Pasture (high density)
+8, 80,   Paved road
+9,200,   Highway
+10,500,  Urban
+11, 10,  Wetland/swamp
+```
+
+```bash
+Rscript resistance_surface.R \
+  data/landcover_2023_mbc.tif \
+  data/jaguar_resistance.csv \
+  outputs/resistance/ \
+  data/dem_srtm_30m.tif \
+  data/roads_osm.shp
+```
+
+**Output statistics:**
+
+| Statistic | Value |
+|-----------|-------|
+| Min resistance | 1.0 |
+| Median | 8.3 |
+| Mean | 24.7 |
+| Q95 | 95.4 |
+| Max (capped) | 1000.0 |
+
+**Decision:** Highway cells assigned resistance = 200 (not 1000) for this analysis, because jaguars have been documented crossing highways via culverts at night. Decision logged with citation (Quigley & Crawshaw 1992).
+
+---
+
+## Step 2 — Delineate Forest Patches
+
+```r
+suppressPackageStartupMessages(library(terra))
+suppressPackageStartupMessages(library(sf))
+
+forest <- rast("data/forest_cover_2023_mbc.tif")
+# Patches: connected forest cells ≥ 500 ha
+patches_r <- patches(forest == 1, directions = 8)
+# Compute patch sizes
+patch_sizes <- freq(patches_r)
+large_patches <- patch_sizes[patch_sizes$count * 0.09 >= 500, ]  # 30m² = 0.0009 ha
+forest_patches <- as.polygons(patches_r %in% large_patches$value)
+forest_patches_sf <- st_as_sf(forest_patches)
+forest_patches_sf$area_ha <- as.numeric(st_area(forest_patches_sf)) / 1e4
+```
+
+**Result:** 214 forest patches ≥ 500 ha identified.
+- Median patch area: 1,240 ha
+- Largest patch: 312,000 ha (Maya Biosphere Reserve)
+- Range: 502–312,000 ha
+
+---
+
+## Step 3 — Graph Connectivity Analysis
+
+### Sensitivity analysis across dispersal distances
+
+```bash
+# Jaguar max dispersal: 50–500 km reported; use range 20–80 km
+for dmax in 20000 50000 80000; do
+  Rscript connectivity_metrics.R \
+    data/forest_patches_500ha.shp \
+    outputs/connectivity/dmax_${dmax}/ \
+    $dmax \
+    area_ha
+done
+```
+
+**Sensitivity results:**
+
+| dmax (m) | IIC | PC | n_components | Largest component |
+|----------|-----|----|----|---|
+| 20,000 | 0.0412 | 0.0389 | 18 | 71 patches |
+| 50,000 | 0.0847 | 0.0814 | 7  | 142 patches |
+| 80,000 | 0.1183 | 0.1156 | 3  | 198 patches |
+
+**Interpretation:** At all dispersal distances, landscape connectivity is < 0.12 — indicating high fragmentation. The MBC is not functionally connected for jaguars across most of its length.
+
+**Decision:** Analysis proceeds with dmax = 50,000 m (50 km), representing conservative mid-range jaguar dispersal (Rabinowitz & Zeller 2010).
+
+### Top patches by dPC
+
+```r
+metrics <- read.csv("outputs/connectivity/dmax_50000/patch_metrics.csv")
+head(metrics[, c("patch_id", "area_ha", "dPC_pct", "BC_norm")], 10)
+```
+
+| patch_id | area_ha | dPC (%) | BC (norm) |
+|----------|---------|---------|-----------|
+| P001 | 312,000 | 18.4 | 0.42 |
+| P017 | 85,200  | 12.1 | 0.38 |
+| P003 | 142,100 | 9.7  | 0.29 |
+| P089 | 8,400   | **7.2** | **0.81** |
+| P122 | 3,200   | 5.1  | 0.76 |
+| P071 | 12,800  | 4.8  | 0.67 |
+
+**Key finding:** Patch P089 (8,400 ha) has disproportionately high dPC (7.2%) and betweenness centrality (0.81) relative to its size. It is a **stepping stone** connecting the two largest forest blocks. High BC = critical corridor function.
+
+---
+
+## Step 4 — Circuitscape Current Flow
+
+```r
+source("scripts/run_circuitscape.R")
+
+# Use top 20 patches by dPC as focal nodes
+focal_patches <- metrics[1:20, "patch_id"]
+focal_shp <- forest_patches_sf[forest_patches_sf$patch_id %in% focal_patches, ]
+st_write(focal_shp, "data/focal_nodes.shp", delete_dsn = TRUE)
+
+cur_map <- run_circuitscape(
+  resistance_tif  = "outputs/resistance/resistance_combined.tif",
+  focal_points_shp = "data/focal_nodes.shp",
+  output_dir      = "outputs/circuitscape/",
+  scenario        = "pairwise"
+)
+```
+
+**Pinch point extraction:**
+
+```r
+cur <- rast("outputs/circuitscape/cs_output_cum_curmap.asc")
+q95 <- quantile(values(cur), 0.95, na.rm = TRUE)
+pinch <- cur >= q95
+writeRaster(pinch, "outputs/circuitscape/pinchpoints.tif", overwrite = TRUE)
+# Extract pinch point polygons for reporting
+pinch_poly <- as.polygons(pinch)
+```
+
+**Results:** 23 pinch point clusters identified, totalling ~180 km². Top 5:
+
+| Pinch point | Area (km²) | Location | Threat |
+|-------------|-----------|----------|--------|
+| PP01 | 42 | Petén–Alta Verapaz junction | Active deforestation |
+| PP02 | 31 | Río Motagua corridor | Highway CA-9 |
+| PP03 | 18 | Honduras–Nicaragua border | Cattle expansion |
+| PP04 | 12 | Caribbean lowlands Guatemala | Oil palm |
+| PP05 | 8  | Bay Islands bridge | Tourism development |
+
+---
+
+## Step 5 — Scenario Analysis: Corridor Protection
+
+**Question:** If PP01 (42 km²) were protected and restored to forest, how much would PC increase?
+
+```r
+# Create scenario patch: add PP01 as new forest patch
+pp01 <- data.frame(
+  patch_id = "PP01_restored",
+  x = centroid_x, y = centroid_y,
+  area_ha = 4200
+)
+patches_scenario <- rbind(patches_original, pp01)
+
+# Recompute connectivity
+python_cmd <- sprintf(
+  "python connectivity_analysis.py data/patches_scenario.csv outputs/scenario/ --dmax 50000"
+)
+system(python_cmd)
+
+pc_scenario <- read.csv("outputs/scenario/landscape_summary.csv")
+pc_baseline <- 0.0814
+pc_new <- pc_scenario$value[pc_scenario$metric == "PC"]
+cat(sprintf("PC improvement: %.4f → %.4f (+%.1f%%)\n",
+            pc_baseline, pc_new, (pc_new - pc_baseline) / pc_baseline * 100))
+```
+
+**Result:** Restoring PP01 increases PC from 0.0814 to 0.1021 (+25.4%). This is the single highest-leverage restoration action in the landscape.
+
+---
+
+## Step 6 — Final Outputs
+
+```
+outputs/
+├── resistance/
+│   ├── resistance_lc.tif
+│   ├── resistance_combined.tif    # LC + slope + road proximity
+│   └── resistance_stats.csv
+├── connectivity/
+│   └── dmax_50000/
+│       ├── patch_metrics.csv      # 214 patches, ranked by dPC
+│       ├── landscape_summary.csv
+│       └── connectivity_graph.png
+├── circuitscape/
+│   ├── cs_output_cum_curmap.asc   # Cumulative current flow
+│   ├── pinchpoints.tif            # Top 5% current cells
+│   └── pinchpoints.shp            # Polygon pinch points
+└── scenario/
+    └── landscape_summary.csv      # PC after PP01 restoration
+```
+
+---
+
+## Summary Table
+
+| Metric | Baseline | PP01 restored | Change |
+|--------|---------|--------------|--------|
+| IIC | 0.0847 | 0.1094 | +29.2% |
+| PC | 0.0814 | 0.1021 | +25.4% |
+| n components | 7 | 6 | −1 |
+| Largest component (patches) | 142 | 158 | +16 |
+
+---
+
+## Decision Log
+
+```yaml
+- date: 2025-01-20
+  skill_id: landscape-connectivity
+  decision: "Highway resistance set to 200, not 1000"
+  rationale: "Culvert crossings documented; Quigley & Crawshaw (1992)"
+  outputs: ["outputs/resistance/resistance_combined.tif"]
+
+- date: 2025-01-21
+  skill_id: landscape-connectivity
+  decision: "Dispersal distance set to 50 km for main analysis"
+  rationale: "Conservative mid-range; sensitivity shows IIC robust ±20% across 20-80km"
+  outputs: ["outputs/connectivity/dmax_50000/landscape_summary.csv"]
+
+- date: 2025-01-22
+  skill_id: landscape-connectivity
+  decision: "Patch P089 flagged as critical stepping stone"
+  rationale: "High BC (0.81) despite small area; bridges components 3 and 4"
+  outputs: ["outputs/connectivity/dmax_50000/patch_metrics.csv"]
+```
+
+---
+
+## References
+
+- Rabinowitz, A. & Zeller, K.A. (2010). A range-wide model of landscape connectivity and conservation for the jaguar. *Biological Conservation*, 143(4), 939–945. DOI: 10.1016/j.biocon.2010.01.002
+- Pascual-Hortal, L. & Saura, S. (2006). Comparison and development of new graph-based landscape connectivity indices. *Landscape Ecology*, 21(7), 959–967. DOI: 10.1007/s10980-006-0013-z
+- McRae, B.H. et al. (2008). Using circuit theory to model connectivity in ecology. *Ecology*, 89(10), 2712–2724. DOI: 10.1890/07-1861.1
+- Zeller, K.A. et al. (2012). Estimating landscape resistance to movement. *Landscape Ecology*, 27(6), 777–797. DOI: 10.1007/s10980-012-9737-0

package/skills/landscape-connectivity/resources/circuitscape-parameter-guide.md

@@ -0,0 +1,155 @@
+---
+resource_id: circuitscape-parameter-guide
+skill_id: landscape-connectivity
+---
+
+# Circuitscape Parameter Guide
+
+## What Circuitscape Models
+
+Circuitscape applies circuit theory to landscape connectivity. Each raster cell is a resistor; current flows from source to ground nodes. **High current = high movement probability.** Unlike least-cost paths (single path), Circuitscape identifies all possible movement paths simultaneously, making it robust to barrier uncertainty.
+
+**When to use Circuitscape:**
+- Species with exploratory or random-walk dispersal (amphibians, many invertebrates)
+- Gene flow modelling (isolation by resistance)
+- Identifying multiple corridors and bottlenecks simultaneously
+
+**When least-cost path may be preferred:**
+- Species with directed, optimal-path movement (large carnivores following trails)
+- Fine-resolution analysis where single corridors are sufficient
+
+---
+
+## Key Parameters
+
+| Parameter | Options | Recommended default | Effect |
+|-----------|---------|---------------------|--------|
+| `scenario` | `pairwise`, `one-to-all`, `all-to-one` | `pairwise` | Analysis mode |
+| `data_type` | `raster`, `network` | `raster` | Input format |
+| `resistance_file` | path to .asc/.tif | — | Resistance surface |
+| `point_file` | path to focal nodes .shp | — | Source/destination points |
+| `write_cum_cur_map_only` | True/False | True (pairwise sums) | Output cumulative vs all pairwise maps |
+| `log_transform_maps` | True/False | True | Log-scale current maps easier to visualise |
+| `compress_grids` | True/False | True | Reduce output file size |
+| `solver` | `cg+amg`, `cholmod` | `cg+amg` | Large rasters; cholmod for small/medium |
+| `max_parallel` | int | n_cores - 1 | Parallelisation |
+
+---
+
+## Running Circuitscape from R (via System Call)
+
+```r
+# Usage: source("run_circuitscape.R")
+suppressPackageStartupMessages(library(terra))
+
+run_circuitscape <- function(resistance_tif, focal_points_shp, output_dir,
+                              scenario = "pairwise", solver = "cg+amg") {
+  dir.create(output_dir, recursive = TRUE, showWarnings = FALSE)
+
+  # Export resistance as .asc (Circuitscape requires ASCII grid or GeoTIFF)
+  res <- rast(resistance_tif)
+  asc_path <- file.path(output_dir, "resistance.asc")
+  writeRaster(res, asc_path, filetype = "AAIGrid", overwrite = TRUE)
+
+  # Write Circuitscape INI configuration
+  ini_path <- file.path(output_dir, "cs_config.ini")
+  ini_content <- sprintf(
+    "[circuitscape options]
+scenario = %s
+data_type = raster
+resistance_file = %s
+point_file = %s
+output_file = %s
+write_cum_cur_map_only = True
+log_transform_maps = True
+compress_grids = True
+solver = %s
+max_parallel = %d
+",
+    scenario,
+    normalizePath(asc_path),
+    normalizePath(focal_points_shp),
+    normalizePath(file.path(output_dir, "cs_output")),
+    solver,
+    max(1L, parallel::detectCores() - 1L)
+  )
+  writeLines(ini_content, ini_path)
+
+  # Call Circuitscape (must be installed and on PATH)
+  cmd <- sprintf("cs_run.py %s", shQuote(ini_path))
+  ret <- system(cmd)
+  if (ret != 0) warning("Circuitscape returned non-zero exit code: ", ret)
+  return(file.path(output_dir, "cs_output_cum_curmap.asc"))
+}
+```
+
+---
+
+## Interpreting Circuitscape Output Maps
+
+| Output file | What it shows |
+|-------------|--------------|
+| `*_cum_curmap.asc` | Summed current across all pairwise flows — identifies corridors |
+| `*_resistances.out` | Effective resistance between each pair of nodes |
+| `*_curmap_X_Y.asc` | Current map for specific node pair X→Y |
+
+### Post-processing current map in R
+
+```r
+suppressPackageStartupMessages(library(terra))
+
+cur <- rast("outputs/circuitscape/cs_output_cum_curmap.asc")
+
+# Log-scale for visualisation
+cur_log <- log1p(cur)
+
+# Identify pinch points: cells with high current in narrow flow paths
+# Threshold: top 5% of current values
+q95 <- quantile(values(cur_log), 0.95, na.rm = TRUE)
+pinchpoints <- cur_log >= q95
+writeRaster(pinchpoints, "outputs/circuitscape/pinchpoints.tif", overwrite = TRUE)
+```
+
+---
+
+## Choosing Focal Nodes
+
+| Scenario | Recommended focal nodes |
+|----------|------------------------|
+| Cross-landscape connectivity | Large core habitat patches (> 100 ha) |
+| Corridor between two areas | Source and destination patches only |
+| Full landscape assessment | All patches > minimum size threshold |
+| Genetic sampling sites | Population sampling locations |
+
+**Rule:** Keep focal node count ≤ 50 for pairwise scenario on standard hardware. More nodes → O(n²) compute time.
+
+---
+
+## Computational Considerations
+
+| Raster resolution | Focal nodes | Typical runtime | Solver recommendation |
+|------------------|-------------|----------------|----------------------|
+| 30 m, 10×10 km extent | 20 | 2–5 min | cg+amg |
+| 30 m, 100×100 km extent | 50 | 30–90 min | cg+amg |
+| 10 m, 50×50 km extent | 30 | 2–8 h | cholmod or HPC |
+| 1 m, any extent | — | Not feasible | Resample to ≥ 10 m |
+
+**Tip:** Resample resistance raster to the coarsest resolution that still captures key barriers (e.g., roads, rivers). Test with 100 m resolution first; refine if necessary.
+
+---
+
+## Pitfalls
+
+- **Current maps conflate corridors with high-permeability areas:** High current can result from many paths OR from a single wide, low-resistance path. Always check raster context (road maps, imagery) around high-current cells.
+- **Focal nodes inside high-resistance cells:** Circuitscape places current sources at node locations. If a node is on a road (high resistance), effective resistance will be artificially inflated. Snap nodes to nearest low-resistance cell.
+- **Ignoring NoData handling:** NoData cells are treated as barriers. Ensure NoData correctly represents barriers (not missing data). Fill missing cells with average neighbourhood resistance before analysis.
+- **Comparing Circuitscape between landscapes without standardisation:** Cumulative current scales with number of focal nodes and landscape size. Only compare effective resistance values, not raw current maps, across landscapes.
+- **Using Circuitscape for directed movement:** Circuitscape is symmetric. For strongly directional dispersal (stream-mediated, prevailing winds), use directional least-cost models instead.
+
+---
+
+## References
+
+- McRae, B.H., Dickson, B.G., Keitt, T.H. & Shah, V.B. (2008). Using circuit theory to model connectivity in ecology, evolution, and conservation. *Ecology*, 89(10), 2712–2724. DOI: 10.1890/07-1861.1
+- Shah, V.B. & McRae, B.H. (2008). Circuitscape: a tool for landscape ecology. *Proceedings of the 7th Python in Science Conference*, 62–65.
+- Dickson, B.G. et al. (2019). Informing strategic efforts to expand and connect protected areas using a model of ecological flow, with application to the western United States. *Conservation Letters*, 12(3), e12647. DOI: 10.1111/conl.12647

package/skills/landscape-connectivity/resources/graph-theory-for-ecology.md

@@ -0,0 +1,134 @@
+---
+resource_id: graph-theory-for-ecology
+skill_id: landscape-connectivity
+---
+
+# Graph Theory for Landscape Ecology
+
+## Core Connectivity Metrics
+
+| Metric | Formula | Scale | Interpretation | Requires |
+|--------|---------|-------|----------------|---------|
+| **IIC** (Integral Index of Connectivity) | (1/A²) Σ (aᵢ × aⱼ) / (1 + nᵢⱼ) | 0–1 | Landscape-level binary connectivity | Dispersal threshold |
+| **PC** (Probability of Connectivity) | (1/A²) Σ pᵢⱼ* × aᵢ × aⱼ | 0–1 | Landscape-level probabilistic connectivity | Dispersal function |
+| **dIIC** (patch contribution to IIC) | [(IIC − IICremove) / IIC] × 100 | % | Importance of patch i to overall IIC | — |
+| **dPC** (patch contribution to PC) | [(PC − PCremove) / PC] × 100 | % | Importance of patch i to overall PC | — |
+| **BC** (Betweenness Centrality) | fraction of shortest paths through node i | 0–1 | Stepping stone / corridor value | — |
+| **EC** (Eigenvalue Centrality) | dominant eigenvector component | 0–∞ | Well-connected patches in dense cluster | — |
+
+**Landscape area A:** total study area (same units as patch areas).
+
+---
+
+## When to Use IIC vs PC
+
+| Condition | Recommended metric | Reason |
+|-----------|--------------------|--------|
+| Dispersal threshold unknown | IIC | Binary; only requires yes/no connection |
+| Dispersal probability decay known | PC | Incorporates full movement function |
+| Comparing two landscapes | dIIC, dPC | Relative patch importance is comparable |
+| Identifying stepping stones | BC | Patches with high BC are corridor nodes |
+| Data on gene flow or mark-recapture | PC | Can calibrate pᵢⱼ* with empirical data |
+
+---
+
+## Computing IIC and dPC in R (igraph)
+
+```r
+# Usage: source("connectivity_metrics.R")
+suppressPackageStartupMessages(library(igraph))
+suppressPackageStartupMessages(library(sf))
+
+# Patches: sf polygon layer with attribute 'area_ha'
+# Dispersal threshold: dmax in map units (e.g., metres)
+
+compute_iic <- function(patches, dmax) {
+  n   <- nrow(patches)
+  A   <- sum(patches$area_ha)
+  coords <- st_centroid(patches) %>% st_coordinates()
+  dist_mat <- as.matrix(dist(coords))
+
+  # Binary adjacency: connected if distance < dmax
+  adj <- (dist_mat < dmax) * 1
+  diag(adj) <- 0
+  g <- graph_from_adjacency_matrix(adj, mode = "undirected")
+
+  # Shortest paths (hop count)
+  sp <- distances(g, algorithm = "bfs")
+
+  IIC_num <- 0
+  for (i in seq_len(n)) {
+    for (j in seq_len(n)) {
+      nij <- sp[i, j]
+      if (is.finite(nij)) {
+        IIC_num <- IIC_num +
+          (patches$area_ha[i] * patches$area_ha[j]) / (1 + nij)
+      }
+    }
+  }
+  IIC <- IIC_num / A^2
+  return(IIC)
+}
+
+# Patch importance: dIIC
+compute_diic <- function(patches, dmax) {
+  IIC_full <- compute_iic(patches, dmax)
+  diic_vec  <- numeric(nrow(patches))
+  for (i in seq_len(nrow(patches))) {
+    IIC_i     <- compute_iic(patches[-i, ], dmax)
+    diic_vec[i] <- (IIC_full - IIC_i) / IIC_full * 100
+  }
+  patches$dIIC <- diic_vec
+  return(patches)
+}
+```
+
+---
+
+## Interpreting dPC Values
+
+| dPC (%) | Patch importance | Recommended action |
+|---------|-----------------|-------------------|
+| > 10    | Critical | Highest protection priority; any loss severely disrupts connectivity |
+| 5–10    | High | Priority for protection or restoration buffer |
+| 1–5     | Moderate | Monitor; stepping-stone function likely |
+| < 1     | Low | May contribute if adjacent patches are lost |
+| 0       | Isolated | Not connected within dispersal distance; restoration needed first |
+
+---
+
+## Dispersal Distance Sensitivity Analysis
+
+When the species dispersal distance is uncertain, run connectivity across a range:
+
+```r
+dmax_values <- c(500, 1000, 2000, 5000)  # metres
+results <- lapply(dmax_values, function(d) {
+  iic <- compute_iic(patches, dmax = d)
+  data.frame(dmax = d, IIC = iic)
+})
+sens_df <- dplyr::bind_rows(results)
+plot(sens_df$dmax, sens_df$IIC, type = "b",
+     xlab = "Dispersal distance (m)", ylab = "IIC",
+     main = "Connectivity sensitivity to dispersal distance")
+```
+
+**Rule:** If IIC changes < 10% across the plausible dmax range → results are robust to dispersal distance uncertainty.
+
+---
+
+## Pitfalls
+
+- **Euclidean vs cost distance:** Graph edges based on straight-line distance ignore barriers (roads, rivers). Use cost distance for species with limited matrix permeability.
+- **Patch definition:** Patches defined at too coarse a resolution inflate IIC by merging sub-patches that are not functionally connected.
+- **Island patches:** Isolated patches (not connected to any other patch) contribute only their intra-patch value to IIC/PC. Report isolation separately.
+- **Directed vs undirected graphs:** Most landscape connectivity models assume undirected movement. For rivers or wind-pollinated species, use directed graphs.
+- **Comparing across study areas:** IIC and PC scale with A². Always normalise by landscape area when comparing sites.
+
+---
+
+## References
+
+- Pascual-Hortal, L. & Saura, S. (2006). Comparison and development of new graph-based landscape connectivity indices. *Landscape Ecology*, 21(7), 959–967. DOI: 10.1007/s10980-006-0013-z
+- 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
+- Urban, D. & Keitt, T. (2001). Landscape connectivity: a graph-theoretic perspective. *Ecology*, 82(5), 1205–1218. DOI: 10.1890/0012-9658(2001)082[1205:LCAGTP]2.0.CO;2