leaves-ph: measuring Metro Manila's tree canopy from orbit

No one publishes how much of Metro Manila sits under trees, year by year, LGU by LGU. Quezon City reads near 22% canopy; Pasay, next to the airport, is closer to 3%. leaves-ph measures that fraction from free Sentinel-2 imagery and puts every detected tree on a map you can drag through the years.

What it measures

For each year, what fraction of an LGU’s area reads as tree canopy across the 17 LGUs of the National Capital Region (16 cities plus Pateros), 2019 to 2026. The published figures come from a gradient-boosted classifier over ten per-pixel features (NDVI, Dynamic-World tree probability, Meta’s 1m canopy height, the ESA tree class, and raw Sentinel-2 bands), trained on 656 hand-labeled high-resolution pixels. Against those labels it scores F1 0.78 and IoU 0.64, beating the old NDVI-threshold baseline (F1 0.68) and holding a steady 9 to 10% NCR snapshot instead of the year-to-year sawtooth the threshold produced. Read it as a tree-canopy estimate, not a pixel-exact census; it still carries a known grass-and-scrub margin.

Forest loss inside protected areas

The same satellite record answers a sharper question: how much forest has been cleared inside the country’s legally protected areas? Pairing the official boundaries (WDPA) with the public loss record (Hansen Global Forest Change), about 186,000 hectares of tree cover were cleared inside the 251 nationally protected areas between 2016 and 2025, roughly 4.9% of what they held in 2000. One site, the Palawan Game Refuge and Bird Sanctuary, accounts for more than half. This is satellite tree-cover loss, so it also picks up plantation harvest, fire, and storm damage. It says where to look, not who is guilty.

Does losing trees flood a barangay?

The tempting story is that less canopy means more runoff means more flooding. Metro Manila does not bear it out. The raw link is positive: greener barangays show more radar-detected flooding (r about +0.15), but that is a measurement effect, not hydrology. Radar under-sees water trapped between buildings, so open, greener barangays are simply where standing water is visible from orbit. Control for what actually drives barangay flooding (low ground, built-up fraction, distance to water) and canopy’s own effect more than halves and stops being significant. Of 892 barangays, the radar flagged observed flooding in 59, and 10 of those also sit in the lowest canopy third. Those 10 are a prioritization screen, not a causal claim.

What it is not

Not a pixel-exact census, and not a logging verdict. The flood figures are a radar lower bound, so dense urban flooding is under-counted, and the canopy-flood overlap is a co-occurrence screen, not proof one caused the other.

All inputs are public (Sentinel-2, Dynamic World, Meta canopy height, ESA WorldCover, WDPA, Hansen GFC). leaves-ph computes statistical indicators only; specific claims require independent corroboration.

Repo: github.com/xmpuspus/leaves-ph. DOI: 10.5281/zenodo.20470306.