Skip to content

labels.md

Latest commit

 

History

History
76 lines (58 loc) · 3.18 KB

labels.md

File metadata and controls

76 lines (58 loc) · 3.18 KB
Raw

Proxy labels

The three labels are operational indicators, not measurements of ecological resilience. They are built in verdi/labels/.

Label What it is Built from Availability
R*_A Relative cross-environment condition Census health, within-species NDVI and DBH percentiles NYC, Melbourne (Paris has no census health field)
R*_B Short-term NDVI retention around qualifying heat events ERA5-Land daily maximum temperature, Sentinel-2 pre/post pairs Where a qualifying event and a cloud-free pair exist
R*_C Relative deviation from a species baseline Sentinel-2 NDVI All cells in all three cities; the primary reported target

R*_C is the only label complete across cities and is the primary target. Other label combinations are implemented in verdi/labels/fusion.py.

R*_A

P_i  = 0.4 * health_i + 0.4 * ndvi_percentile_i + 0.2 * dbh_percentile_i
R*_A = rank(mean P over the cell) / N_s
  • Only species with at least 500 individuals are scored; a tree of a smaller species stays in the frame with eligible = False.
  • Multi-species cells are averaged by canopy area.
  • Health grades are harmonised from the census classes: NYC Good / Fair / Poor = 1.0 / 0.6 / 0.2; Melbourne ULE > 20 yr = 1.0 and 10-20 yr = 0.6.
  • R*_A shares the within-species NDVI component with R*_C, so agreement between the two labels is partly built in; the census health grade is the component that does not come from NDVI.

Settings, exposed as arguments:

  • N_s is the number of scored cells, and the rank is ascending with average ties.
  • The percentile scale is pandas.rank(pct=True) within species.
  • The Melbourne ULE class below 10 years has no grade; such trees raise a KeyError rather than being silently dropped.
  • Tree-level NDVI extraction (point or buffer) and the canopy-area source are caller-supplied through canopy_area_col.

R*_B

R*_B = 1 - |NDVI_post - NDVI_pre| / max(NDVI_pre, 0.2)   clipped to [0, 1]
  • A qualifying heat event is at least three consecutive days above the city-specific 95th percentile of the ERA5-Land record for 2018-2023.
  • Pre/post Sentinel-2 pairs must fall within 14 days and have cloud cover below 10 %.
  • Cells without a qualifying event or a usable pair are NaN.

The absolute value means a greening response also lowers the value.

Settings: the temperature variable and the hourly-to-daily aggregation (a calendar-day maximum), the base period of the percentile (the series passed in), the anchor of the 14-day window (event start), whether the cloud threshold is per scene or per pixel (per record), and how several events in one cell are combined (case="mean" or "last").

R*_C

R*_C = sigmoid( (NDVI_grid - baseline(species)) / spread(species) )

The baseline statistic is a parameter, statistic="median" or "mean", and is recorded with the run. Cells whose species has a single individual or zero spread return NaN. apply_minmax=True adds a min-max rescaling to [0, 1] after the sigmoid.

What the labels do not measure

Recovery capacity, persistence, functional stability and post-disturbance regeneration. The labels are city-relative rather than on a common absolute scale.