BEGIN:VCALENDAR VERSION:2.0 PRODID:-//pretalx//talks.osgeo.org//foss4g-2026//talk//WBBBA8 BEGIN:VTIMEZONE TZID:JST BEGIN:STANDARD DTSTART:20000101T000000 RRULE:FREQ=YEARLY;BYMONTH=1 TZNAME:JST TZOFFSETFROM:+0900 TZOFFSETTO:+0900 END:STANDARD END:VTIMEZONE BEGIN:VEVENT UID:pretalx-foss4g-2026-WBBBA8@talks.osgeo.org DTSTART;TZID=JST:20260903T150000 DTEND;TZID=JST:20260903T153000 DESCRIPTION:Forest vertical structural diversity is a key indicator of ecos ystem complexity\, habitat heterogeneity\, and biodiversity potential. Fol iage height diversity (FHD)\, derived from vertical vegetation profiles\, is widely used to quantify this structural heterogeneity. Airborne laser s canning (ALS) provides accurate three-dimensional forest structure informa tion\, but its limited spatial coverage and high cost hinder large-scale m onitoring. The Global Ecosystem Dynamics Investigation (GEDI) mission has enabled global sampling of forest vertical structure using spaceborne LiDA R. However\, its footprint-based sampling produces spatially discontinuous observations. As a result\, continuous regional-scale mapping of forest v ertical structural diversity remains a major challenge. Previous studies h ave focused primarily on height-related metrics or canopy cover estimation . Regional-scale mapping of foliage height diversity remains limited\, esp ecially in cool-temperate and boreal forest ecosystems like northern Japan \, where complex terrain\, climate gradients\, and forest management regim es interact.\nThis study proposes a large-scale framework for estimating f orest vertical structural diversity across Hokkaido\, Japan\, by integrati ng GEDI-derived FHD with multi-source satellite remote sensing data and ma chine learning. The novelty of this work lies in two key contributions: (i ) integrating a wide range of complementary satellite data sources includi ng multi-frequency SAR\, optical imagery\, climate variables\, land cover products\, topography\, and nighttime light data\, for FHD estimation\, an d (ii) explicitly interpreting model behavior using Shapley additive expla nations (SHAP) to identify physically meaningful drivers of forest vertica l structural diversity.\nWe used diverse openly available geospatial data. GEDI Level 2B observations acquired in 2023 served as reference data. Aft er quality filtering\, GEDI footprints were spatially matched with satelli te-derived features\, including Sentinel-1 SAR\, Sentinel-2 spectral bands \, ALOS L-band SAR metrics\, a digital elevation model (DEM) from JAXA\, T erraClimate climate variables\, forest type classes from Copernicus Global Land Service (CGLS)\, Dynamic World-derived forest probability\, and VIIR S nighttime light intensity. These variables were selected to represent co mplementary aspects of canopy structure\, vegetation condition\, terrain-d riven environmental gradients\, climatic constraints\, and anthropogenic d isturbance. A Light Gradient Boosting Machine (LightGBM) regression model was trained to predict GEDI-derived FHD from the multi-source feature set. Model evaluation was conducted using out-of-fold (OOF) predictions to red uce optimistic bias and to provide a robust estimate of generalization per formance. The resulting predictions were used to produce a spatially conti nuous wall-to-wall map of FHD for the entire region.\nThe model achieved a n RMSE of 0.360 and an R² of 0.306 in predicting GEDI-derived FHD. This i ndicates that multi-source satellite observations capture part of the spat ial variability in forest vertical structural diversity across heterogeneo us landscapes.\nThe predicted wall-to-wall FHD map reveals spatially coher ent patterns across Hokkaido. Higher values appear in mountainous forested regions\, while lower values occur in flatter or more human-influenced ar eas. Extremely high-elevation areas show lower FHD values\, likely reflect ing harsher climatic conditions and simplified forest structures near the treeline.\nThese spatial patterns align with known ecological gradients in forest composition and management intensity. This suggests the model capt ures meaningful large-scale structural variability rather than random nois e. To interpret the model beyond simple prediction\, we used SHAP analysis to understand how individual features contribute to predicted FHD values. \nSHAP-based global feature importance showed that tree cover probability (from land cover products) was the most influential predictor. This highli ghts the fundamental role of forest presence and canopy continuity in dete rmining vertical structural diversity. Optical spectral bands from Sentine l-2\, particularly visible and red-edge bands\, contributed strongly\, ref lecting how spectral responses vary with canopy density and vegetation con dition. SAR backscatter from ALOS L-band and Sentinel-1 also showed high i mportance\, indicating that longer-wavelength radar signals capture canopy structure and woody biomass information relevant to vertical heterogeneit y. Topographic variables (DEM) contributed significantly\, suggesting that elevation-related environmental gradients influence forest structure thro ugh climate and disturbance patterns. Nighttime light intensity and its di stance-based metrics showed measurable but secondary contributions\, imply ing that human pressure is associated with reduced vertical structural div ersity in more developed areas. SHAP summary plots further revealed nonlin ear and asymmetric relationships between key predictors and FHD. Higher tr ee cover probability and stronger L-band HV backscatter were associated wi th positive contributions to predicted FHD\, while increasing nighttime li ght intensity tended to reduce it. Elevation showed both positive and nega tive contributions depending on context\, reflecting complex interactions between topography\, forest type\, and management practices. These results demonstrate that the machine learning model captures ecologically meaning ful relationships rather than purely statistical correlations.\nThis study 's key contribution is demonstrating that foliage height diversity can be estimated at regional scale by combining spaceborne LiDAR data with multi- modal satellite observations. The resulting model can be meaningfully inte rpreted in ecological terms using SHAP analysis. This interpretability dis tinguishes our framework from purely predictive approaches and reveals how canopy cover\, radar-derived structure\, topography\, and human influence shape forest vertical structure.\nWhile predictive accuracy remains moder ate\, reflecting the inherent challenge of inferring three-dimensional for est structure from two-dimensional satellite observations\, the results sh ow that broad regional patterns of forest vertical structural diversity ca n be captured consistently. DTSTAMP:20260717T225747Z LOCATION:Cosmos1 SUMMARY:Regional 10-m Mapping of Forest Foliage Height Diversity in Hokkaid o\, Japan\, Using GEDI and Foundation-Model Satellite Embeddings - Narumas a Tsutsumida\, Tetsu Sasaki URL:https://talks.osgeo.org/foss4g-2026/talk/WBBBA8/ END:VEVENT END:VCALENDAR