FORMSpoT: A Decade of Tree-Level, Country-Scale Forest Monitoring

TL;DR

This work tackles the need for fine-grained, time-resolved forest monitoring by developing FORMSpoT, a 1.5 m canopy height mapping framework for France using SPOT-6/7 imagery and an ALS-trained hierarchical transformer. It further derives FORMSpoT-Delta, high-resolution disturbance polygons that detect tree-scale changes by analyzing annual height differences, paired with a robust co-registration and TV denoising pipeline. Validation against dense LiDAR and repeated NFI measurements demonstrates that FORMSpoT-Delta outperforms existing disturbance products, especially for small and fragmented disturbances in mountainous regions. The approach enables detailed analyses of management practices, early signals of decline, and improved carbon-loss quantification, highlighting the value of very high-resolution satellite data and open data initiatives for climate-smart forest monitoring.

Abstract

The recent decline of the European forest carbon sink highlights the need for spatially explicit and frequently updated forest monitoring tools. Yet, existing satellite-based disturbance products remain too coarse to detect changes at the scale of individual trees, typically below 100 m$^{2}$. Here, we introduce FORMSpoT (Forest Mapping with SPOT Time series), a decade-long (2014-2024) nationwide mapping of forest canopy height at 1.5 m resolution, together with annual disturbance polygons (FORMSpoT-$Δ$) covering mainland France. Canopy heights were derived from annual SPOT-6/7 composites using a hierarchical transformer model (PVTv2) trained on high-resolution airborne laser scanning (ALS) data. To enable robust change detection across heterogeneous acquisitions, we developed a dedicated post-processing pipeline combining co-registration and spatio-temporal total variation denoising. Validation against ALS revisits across 19 sites and 5,087 National Forest Inventory plots shows that FORMSpoT-$Δ$ substantially outperforms existing disturbance products. In mountainous forests, where disturbances are small and spatially fragmented, FORMSpoT-$Δ$ achieves an F1-score of 0.44, representing an order of magnitude higher than existing benchmarks. By enabling tree-level monitoring of forest dynamics at national scale, FORMSpoT-$Δ$ provides a unique tool to analyze management practices, detect early signals of forest decline, and better quantify carbon losses from subtle disturbances such as thinning or selective logging. These results underscore the critical importance of sustaining very high-resolution satellite missions like SPOT and open-data initiatives such as DINAMIS for monitoring forests under climate change.

Figures (11)

• Figure 1: Overview of the workflow used to generate FORMSpoT and FORMSpoT-Delta. SPOT-6/7 images over mainland France at 1.5 m resolution (2014–2024) were processed with a pretrained PVTv2 model Fogel_2025_CVPR to produce annual forest canopy height maps. These height maps were then co-registered and spatio-temporally regularized to ensure consistency across the time series. From the resulting 1.5 m annual canopy height time series (FORMSpoT), disturbance polygons (FORMSpoT-Delta) were extracted by applying a 5 m height-loss threshold and a morphological filter.
• Figure 2: FORMSpoT-Delta validation sites. Locations of the 19 ALS sites used to validate FORMSpoT-Delta polygons. The six sites presented in the main part of this study are highlighted in bold black font. Additional validation sites, whose metrics are presented in the supplementary materials (Fig. S2), are indicated with smaller orange dots. Greenness represents tree cover density clmsTreeCoverDensity2023.
• Figure 3: Computing disturbance polygons. We first computed the difference between two ALS-derived 1 m CHMs and applied a 5 m height-loss threshold to generate a binary disturbance mask. We then applied an opening morphological filter serraImageAnalysisMathematical1982 to clean the mask. Finally, we retained all resulting polygons larger than 10 m2 as reference disturbance polygons.
• Figure 4: Qualitative evaluation of FORMSpoT height maps and FORMSpoT-Delta polygons. a) The FORMSpoT maps at 1.5 resolution derived from SPOT-6/7 imagery with the PVTv2 transformer model trained on ALS data. b) Closer view of a SPOT-6/7 image (2022) over the Jura mountains. c) FORMSpoT predicted height (2022) at 1.5 m resolution over this area. Brighter colors indicate higher heights. d) FORMSpoT-Delta disturbance polygons derived from the FORMSpoT height time series over this area. Brighter colors indicate more recent forest disturbances. e) Comparison of ALS-derived disturbance polygons from successive acquisitions (2019-2022) with FORMSpoT-Delta polygons spanning these years.
• Figure 5: Validation of FORMSpoT canopy height estimates using NFI field plots across 11 years. (a) Density plot of FORMSpoT-predicted height versus NFI reference height (see \ref{['sec:NFI_data']}), with brighter colors indicating higher point density. The dashed line represents the 1:1 relationship. (b) Histogram with boxplots showing the distribution of height differences within 5 m reference height bins. The red line denotes the median, the box edges indicate the lower and upper quartiles, and the whiskers represent the 5th and 95th percentiles. (c) Table reporting the mean absolute error (MAE) and the coefficient of determination (r2) for all years (2014–2024). The standard deviation (SD) across years is shown in the last row.