Remote sensing data supported by ground observations are considered as the key for the monitoring of the dynamics of both canopy height and gaps at several scales in space and time, which is known to be crucial to improve both monitoring and modelling of forest ecosystems and thus contribute to their sustainable management. In particular, gaps play a major role in forest ecology because of their influence on microclimate and habitat quality and thus on biodiversity. The gradient in gap sizes is known to influence many parts of biodiversity and many references indicate an enrichment of biodiversity shortly after harvesting (Wermelinger et al. 1995). Temporal effects have also been reported. Through the succession that occurs after gap formation, the effect of gaps can shift and even be reversed. Especially, for vascular plants, silviculture based on large cuttings has been found to have an adverse impact on floristic biodiversity, but decades after gap formation (Duguid and Ashton 2013). Recent developments in digital photogrammetry have provided the ability to improve canopy surface reconstruction from optical imagery, which explains the renewed interest towards this technology for forest applications and forest structure characterisation. Large databases of aerial photographs, dating back over 60 years, exist in several countries (Véga and St-Onge 2008). Therefore digital photogrammetry applied to long time series of aerial photographs offers an exceptional opportunity to study the link between the past disturbance regime and the current biodiversity level. Several studies already analyzed forest dynamics based on CHM (Canopy Height Model) time series obtained by subtracting a DTM (Digital terrain Model) to photogrammetric DSMs (Digital Surface Model). Several approaches were used to produce the required DTM: photogrammetry on leaf-off aerial photographs (Tanaka and Nakashizuka 1997), field surveys (Fujita et al. 2003) or lidar data processing (Véga and St-Onge 2008). Implementing thresholding approaches on CHMs, using either fixed or context-specific threshold, proved efficient to identify gaps (Vepakomma et al. 2008; Zhang 2008). The GNB project (http://gnb.irstea.fr/) seeks at studying the link between biodiversity, forest exploitation and naturalness by comparing exploited and natural forests in the French context. To that aim the study is based on a national network of integral biological and nature reserves excluded from exploitation and their managed counterparts and on sampling 7 taxonomic groups: vascular plants, mosses, mushrooms, bats, birds, and carabid and saproxylic beetles (Debaive et al. 2013). As part of this project the present study aims at developing a method to provide indicators of the past disturbance regime at several scales in the vicinity of the plots for which biodiversity surveys were realized. The method was developed under two main constraints: (i) the lack of accurate DTM on most study sites, which led us to work by analysing the changes in DSMs rather than in CHMs, (ii) the unequal quality between DSMs combined to slight geometrical discrepancies, which prevented using pixel to pixel comparisons to analyze the changes in elevation. To develop the method we focused on a lowland hardwood forest site, the Fontainebleau Forest, a (48°25’ N and 2°40' E). Among the IGN (French National Institute of Geographic and Forest Information) archive photographs, 6 dates were chosen that span over a 54 year period, from 1949 to 2003. MICMAC (Multi Image Matches for Auto Correlation Methods), an open source software combining photogrammetric and newly-developed Structure from Motion approaches, was used to retrieve high resolution surface models from either scanned old analog photographs or more recent digital ones. Then several maps were produced to characterize the disturbance regime over the analysed period. The process included the following steps: - A grid with 44 x 44 m (~ 0,2 ha) squared cells was superimposed to the DSMs, and elevation frequencies were computed for each grid cell (2,5 m elevation classes). - Differences between histograms of each DSM pair of consecutive dates were computed for each cell and aggregated into three height classes: [Hmin ; Hmin + 5 m] ; ]Hmin + 5 m; Hmin + 17,5 m]; ]Hmin + 17,5 m ; Hmax]. Hmin and Hmax were the minimal, respectively the maximal, elevation value considering all the elevation data sets in a given grid cell, and the 5 and 17,5 m thresholds were chosen considering the growth curve of the main tree species present on the site. - Expert-based classification rules were defined to identify several classes of forest evolution from the successive histogram differences and were validated by visual interpretation for a set of plots. Among the many possible classes we focused on the four following ones: (1) area without disturbance, (2) area affected by a disturbance that led to deep gaps, i.e. reaching the ground, (3) area affected by a disturbance that led to either shallow or deep gaps (note that this class includes the previous one), (4) other. - From the maps obtained for all the pair of successive DSMs and for each one of the three first classes, synthetic indicators were built and computed for buffers of different sizes, from 5 to 78 ha, centered on each field plot. These indicators included total class surface and cumulated surfaces of clumps of various sizes (i.e., > 5 ha; > 2 ha; > 0,5 and < 2 ha, and < 0,5 ha). The resulting indicators will be further tested in Bayesian statistical models developed to explain the level of biodiversity from a set of variables (Zilliox and Gosselin 2013) in order to assess their ability to explain the current biodiversity level and to compare exploited and natural forests. According to the results additional indicators might be computed and the age of past perturbations better taken into account. The method, which was developed for lowland forests, will have to be further adapted to process relief areas. Proceeding this way we expect to bring to light relationships between past disturbance regime and current biodiversity that were impossible to identify based on field measurements.