KG LABSKyrgyzstan · forest atlas · 1991–2025
Kyrgyzstan's forests, season by season
Nine forests, seven regions and thirty-five autumns, measured from satellite pictures. Pick a forest on the right and press play to watch its leaves turn, from the green of August to the snow of late November. Pick a forest to zoom in, or open the 3D view to see the mountains.
Measured autumn maps show the leaf stage of every tree pixel in each half-month, blended from cloud-free Sentinel-2 pictures. Pick a forest to see it up close, or open the 3D view (internet needed).
The living landscape
Seven regions, many forest worlds
Elevation, rainfall and exposure create a remarkable sequence, from dry foothill woodland to dark spruce forest near the alpine edge.
Every region has its own autumn
Each ribbon stacks a region's characteristic forest types, the most common widest, and reads left to right through the year. Batken's pistachio and almond give it a crimson October, Jalal-Abad turns walnut gold, and Issyk-Kul and Naryn stay spruce-dark. The mix describes each region; it is not a measured forest area.
The forest calendar
Typical colours of each forest type through the year. The line follows the year dial.
Colour runs downhill in autumn, uphill in spring
Three schematic slopes with typical forest belts. Within a belt, timing shifts about 3 days per 100 m of height (Hopkins' bioclimatic rule), and the snowline follows a seasonal curve. Belt heights are approximate textbook ranges, not mapped.
Autumn, measured
Thirty-five years of autumn
How golden does each forest get, how red, and how long do its leaves hold on? Eight recent autumns at half-month detail answer the first questions; the Landsat archive goes back to 1991.
How golden, how red?
The leaf colours of each forest's broadleaf trees on its most colourful half-month in a typical year. Spruce and juniper are left out because they stay green.
Where the leaves stay longest
Each bar runs from the first colour to the day half the leaves are down, in a typical year. The dot is the most colourful moment. Forests are ranked by how late their leaves hold on.
Every autumn since 1991
Each dot is one autumn: how much of the broadleaf forest was in colour at its best, as Landsat saw it. Dot colour shows the month of the peak; the line is a five-year average. Select a forest to watch its autumns below.
Read the 1990s with care. Only Landsat 5 photographed these mountains then, often once or twice in an autumn month, so its dots sit lower partly because a single picture can miss the peak. From 1999 two or more satellites overlap and the record is steadier. Monthly Landsat blends also read lower than the half-month Sentinel-2 numbers. In the high spruce gorges (Jety-Oguz, Salkyn-Tor) a November peak can be low sun and shadow rather than leaves.
Each frame is the most colourful clear month of that autumn, as Landsat saw it. Years without a clear autumn picture are greyed out.
Three windows into each autumn
The monthly lines show the share of broadleaf forest in autumn colour. Gaps are left open where the record has no usable picture. Select a point to open that year.
Nine forests up close
Pick a forest and follow its autumn
Every valley keeps its own autumn rhythm. Choose a forest, then step through the half-months to see where the colour starts, how it climbs the slopes and when the leaves come down.
How autumn moves with height
Where is autumn strongest?
KG Labs AI capabilities
AI for Kyrgyzstan's forests
KG Labs Public Foundation builds AI models and data pipelines for Kyrgyzstan's forests and tests them against field and satellite evidence before they are used. This atlas is one product of that work.
Automated tree-canopy mapping
Tree crowns and canopy patches mapped automatically from satellite imagery, forest by forest, so canopy extent can be followed year after year without hand digitising.
Canopy height and carbon stock
Canopy height estimated from high-resolution imagery and turned into carbon-stock estimates for each forest compartment.
Forest-type classification
Walnut, spruce, juniper, pistachio and other forest types classified from multi-year satellite data, and validated on forests the model had never seen in training, where it scored close to its training forests.
Multi-year change and disturbance alerting
Year-on-year comparison of satellite records flags canopy loss, dieback and other disturbance, and ranks the alerts for field checks.
Restoration-site selection
Candidate areas for assisted natural regeneration, enrichment planting and rehabilitation, screened by terrain, access and current cover against restoration targets.
Collaborate
Work with us on Kyrgyzstan's forests
We are looking for partners in forest monitoring, field checks, data sharing and research: forestry agencies, leskhozes, protected areas, universities, donors and international programmes.