Learning UAV-Based Above-Ground Biomass Regression Models in Sparse Training Data Environments
This study aims at recovering above-ground biomass information from ultra-high resolution UAV RGB-NIR orthophotos. We focus on a realistic scenario where a limited number of training samples for a landscape with heterogeneous herbaceous vegetation is given. Consequently, we explore different machine learning methods explicitly addressing the limitations of small training samples and compare their predictions quantitatively and qualitatively. Our results show that random forest models perform similarly well to deep learning models. While simpler machine learning models may, therefore, still be preferable, our study also points the way to promising architectures and regularisation techniques for deep learning approaches. Beyond vegetation cover, accurate regression of other variables, including vegetation height, volume and biomass remains a difficult task regardless of the model choice.
(pp. 3322-3325, 16/07/2023)
LETG - Rennes, UBO EPE, UR2, LETG, UBO EPE, UR2, CNRS, Nantes Univ - IGARUN, EPOC, EPHE, PSL, UB, INSU - CNRS, CNRS, OBELIX, IRISA-D6, IRISA, UR, INSA Rennes, UBS, ENS Rennes, Inria, CNRS, IMT Atlantique
Contrasting vegetation response during Heinrich stadials in Western Europe at the end of the Last Glacial period
Deep-sea and terrestrial records allow to document the amplitude, timing and duration of the oceanic and vegetation responses to orbital and millennial-scale changes, in particular during North Atlantic cooling events (Heinrich events, HEs) in Western Europe during the Last Glacial Period (~115- 27 ka). We propose a multiproxy study based on two deep-sea cores retrieved in the Bay of Biscay (MD04-2845) and the Gulf of Lion (MD99-2343) to reconstruct vegetation and climatic changes in southwestern and southeastern France during Heinrich stadials 4, 5 and 6. These records are well chronologically constrained by numerical dating (new IRSL ages for the MD04-2845 deep-sea core) and new age-depth models, based on Bayesian statistics and stratigraphic constrains using ChronoModel software and ArchaeoPhases R-package. The comparison of both deep-sea cores with other terrestrial and marine records from NE Atlantic and NW Mediterranean regions show different magnitudes in the semi-desert expansions following Heinrich Stadials (HS, 4, 5 and 6) in Western Europe. Although in southwestern France, the development of the semi-desert is more pronounced during HS6 compared to HS 4 and 5, in the southeast, this stadial is marked by forest development. These contrasted responses during these events appear to be the result of different intensities of the thermohaline circulation and local oceanic processes associated with the instability of the Laurentian ice sheet.
(14/07/2023)
UB, UBM, CNRS, EPOC, EPHE, PSL, UB, INSU - CNRS, CNRS, EPHE, PSL, LCE, CNRS, UFC, UBFC, UQAT, LMJL, CNRS, Nantes univ - UFR ST, Nantes Univ
How to combine dental wear and pollen proxies for a better knowledge of past local environments?
(14/07/2023)
LAMPEA, AMU, CNRS, MC, IPHES-CERCA, URV, ICREA, UNIROMA, EPHE, PSL, EPOC, EPHE, PSL, UB, INSU - CNRS, CNRS, EHESS
The regional imprints of Dansgaard-Oeschger events on land and marine environments
One of the most important challenges of the IPCC is documenting the regional impact of the present-day global warming. Yet, identifying the regional expression of global changes is the first and necessary step to understand the mechanisms behind them. The best examples of past global warming events, comparable in velocity and magnitude to those expected in the 21st-century, are those associated with the Dansgaard–Oeschger (D–O) cycles that punctuated the last glacial period, ~115,000-27,000 years ago (115-27 ka). Reasonably well-chronologically constrained deep-sea and terrestrial records of D-O cycles provide an excellent opportunity for documenting the nature (shape, amplitude, timing and duration) of the vegetation, fire and oceanic regional responses to past rapid global warming events. Building on the ACER (https://doi.org/10.1594/PANGAEA.870867) and PaleoJump (https://www.tipes.dk/paleojump-a-tipes-database-for-research-on-rapid-climate-transitions/) databases, I will present a new compilation of D-O records from geochemical, sedimentological and micropaleontological data. I will pay particular attention to one of the best dated event, the D-O 8, centered at ~38,000 years ago (38 ka). It followed the Heinrich Stadial 4 cold phase (~40 -38 years ago), and occurred during a period of minima in precession and intermediate ice volume (Marine Isotope Stage 3). Compared to the other D-O events, D-O 8 was marked by strong Greenland warming, ~10°C, and strong increase in atmospheric CH4 and CO2 concentrations, by at around 150 ppb and 20 ppm, respectively, and was associated with a vigorous resumption of the Atlantic Meridional Overturning Circulation. D-O 8 warming has also the advantage to be relatively recent and, therefore, one of the best recorded all around the world.
(14/07/2023)
EPHE, PSL, EPOC, EPHE, PSL, UB, INSU - CNRS, CNRS, UB, UBM, CNRS
Testing the synchrony between cultural and paleoenvironmental changes in southern France during the Middle-to-Upper Palaeolithic transition
Determining the impact of climatic variations on past human cultural changes is a difficult task due to the chronological uncertainties inherent to the dating methods applied to archaeological and paleoclimatic archives, and by the different temporal resolution of both archives. Here, we present two high-resolution pollen-based palaeoenvironmental sequences from the Bay of Biscay (45°21'N, 5°13'W) and the Gulf of Lion (40°29'N, 4°01'E) plotted against up-to-date chronologies. These sequences unambiguously identify millennial-scale vegetation and climatic changes in southern France in response to Greenland warming and cooling events, i.e. Dansgaard-Oeschger (D-O) cycles, and to the North Atlantic major iceberg discharges called Heinrich events (HEs). The chronologies are well constrained by numerical dating (new IRSL ages for the Bay of Biscay deep-sea core) and new age-depth models, based on Bayesian statistics and stratigraphic constrains using ChronoModel software and R-package ArchaeoPhases. The construction and updating of archaeological databases for the Middle-to-Upper Palaeolithic transition in southwestern and southeastern France has allowed the development of age models based on ChronoModel. These age models provide more reliable chronological windows for the observed cultural changes in Neanderthals and Anatomically Modern Humans (AMH) in Western Europe. Despite the improved paleoclimatic and archeological chronologies, the identification of a potential synchrony between climate and cultural changes still remains difficult due to new uncertainties. Nevertheless, this study suggests that the progressive opening of landscape since the D-O 12 (~47 ka) favoured the arrival of AMH in Western Europe, leading to competition with Neanderthals for the same ecological niches, and thus to the disappearance of the latter at ~40 ka.
(14/07/2023)
UB, UBM, CNRS, EPOC, EPHE, PSL, UB, INSU - CNRS, CNRS, EPHE, PSL, LCE, CNRS, UFC, UBFC, UQAT, LMJL, CNRS, Nantes univ - UFR ST, Nantes Univ
Informing past fire regimes using microscopic charcoal preserved in deep-sea sediments, INQUA, Roma, Italy, 13-20 July, poster
Large uncertainties remain in understanding the evolution of fire activity under projected warming scenarios because fire is a complex process to integrate into global modelling. Empirical models used for projections lack potential changes in the interaction between climate, vegetation and fire. Process-based models of the coupled vegetation-fire system provide new tools to address this issue. Evaluating those models against benchmark datasets from charcoal sediment records, outside of the modern climate conditions range, is necessary. Long marine charcoal records capture regional-scale biomass burning over a large range of natural climate variability. However, the development of comprehensive data-model comparisons is limited by the lack of common physical units between data and model output: changes in ‘fire activity’ are not directly comparable to simulated fire regime characteristics (such as fire intensity or burnt areas). The ANR BRAISE intends to develop a calibration which, applied on paleofire records, should provide new datasets of regional fire regimes for key periods in the past. Here, we present the analyses of abundance and morphology of microscopic charcoal preserved in surface sediment samples collected off Iberia and Africa. The spatial distribution patterns of microcharcoal parameters were compared with satellite derived fire regime characteristics on land (fire number, size, intensity and type of burnt vegetation). Results suggest the abundance and morphology of charcoal in marine sediments detect specific fire regimes, in particular fires of high intensity in mixed vegetation in Iberia and in graminoid-mixed ecosystems in Africa. These results are a springboard for converting marine charcoal records into past fire regime history.
(13/07/2023)
EPOC, EPHE, PSL, UB, INSU - CNRS, CNRS
Relations entre granularité des sédiments, pente des plages et déferlement des vagues
La pente de plage est le plus souvent considérée comme constante dans le temps, et reliée à une taille de grain de sable homogène. L’étude couplée des pentes et des sédiments des plages sableuses exposées à l'action des vagues est complexe car leur modélisation et les techniques d’analyse à mettre en œuvre sont longues et fastidieuses, et par conséquent limitées en temps et en espace. Dans cette thèse, en s’appuyant sur une large base de données topographique sur la plage du Truc Vert et une approche de modélisation à complexité réduite, nous démontrons que la pente de plage répond à un comportement d’équilibre dynamique, avec des valeurs maximales et minimales respectivement en fin d’été et fin d’hiver. Nous avons réalisé sur trois ans des suivis intensifs in situ biannuels (fin d’été et fin d’hiver sur les périodes de pentes extrêmes) de la granularité 4D sur cinq plages représentatives de la diversité naturelle des plages en France métropolitaine, associés à des suivis topographiques. L'ensemble des données recueillies révèle une variabilité spatiale et temporelle importante de la granulométrie (en surface et en profondeur) et de la morphologie des plages sableuses concernées. Les motifs temporels et spatiaux observés dépendent fortement de l’état de plage (réflectif, intermédiaire, dissipatif). Ces recherches contribueront à affiner les modèles prédictifs utilisés en océanographie, en particulier sur les aspects morphodynamiques et sédimentaires.
(11/07/2023)
EPOC, EPHE, PSL, UB, INSU - CNRS, CNRS
A combined retrospective and spatial analysis of PFAS in river sediments demonstrates changes in both levels and patterns over the last 40 years (Rhône River, France).
Per- and polyfluoroalkyl substances (PFAS) are commonly detected in aquatic systems and can be transported across long distances. Concentration levels and patterns of PFAS depend on various factors, including the type of matrix analyzed and the relative contribution of direct and indirect sources. Long-chain perfluoroalky carboxylic acids (PFCA) and sulphonic acids (PFSA) are persistent and more hydrophobic than short-chain PFAS, which may lead to their accumulation in specific compartments such as sediments. Sediment can act as a secondary source of contamination, making this a potential long- term problem. Bed sediments and dated sediment cores were collected in the Rhone River, from both upstream and downstream locations of Lyon (France) and its industrial corridor, to evaluate the spatial and temporal trends of PFAS contamination. Total concentrations of PFAS in upstream sediments were low, ranging between 0.19 and 2.6 ng g-1 dry weight (dw), with a high proportion of perfluorooctane sulfonate (PFOS). Downstream, concentrations reached up to 48.7 ng g-1 dw which is likely due to the presence of a fluoropolymer manufacturing plant in the area. Further downstream, concentrations gradually decreased. The temporal evolution of PFAS emissions was analyzed using three sediment cores sampled from secondary channels located 6 to 42 km downstream from the hotspot. Before 1987, PFAS concentrations were low (2 ng g-1 dw), peaking at 51 ng g-1 dw in the 1990s and high until 2007. Since 2010, trends have been stable with total PFAS concentrations of 4.2 ±1.4 ng g-1 dw near the production site and 2.1 ±1.0 ng g-1 at the most downstream site. The proportion of PFSA remained constant since the 1980s (~10%), while PFCA showed large variations. Long chain- (C > 8) PFCA with an even number of perfluorinated carbons represented 74% of the total PFAS load until 2006. From 2006 to 2020, the relative contribution of long chain- (C > 8) PFCA with an odd number of perfluorinated carbons increased to 80%. The changes observed in the PFAS profiles over time indicate a significant modification in the industrial production process. This study emphasizes the necessity of an improved monitoring of these substances in aquatic systems.
(09/07/2023)
LEHNA, UCBL, ENTPE, CNRS, ONIRIS, EPOC, EPHE, PSL, UB, INSU - CNRS, CNRS
Assessing the accuracy of paired and random sampling for quantifying plant–plant interactions in natural communities
AbstractPlant interactions in extreme environments are often inferred from spatial associations and quantified by means of paired sampling. Yet, this method might be confounded by habitat‐sharing effects. Here, we address whether paired and random sampling methods provide similar results at varying levels of environmental heterogeneity. We quantified spatial associations with the two methods at three sites that encompass different micro‐environmental heterogeneity and stress levels: Mediterranean environments in Canary Islands, Spain, and Sardinia, Italy, and a cold alpine environment in Hokkaido, Japan. Then, we simulated plant communities with different levels of species micro‐habitat preferences, environmental heterogeneity, and stress levels. We found that differences in species associations between paired and random sampling were indistinguishable from zero in a homogeneous space. When simulating codispersion over a decreasing abundance gradient, both sampling methods correctly identified facilitation and distinguished it from codispersion. Yet, the pairwise method provided higher facilitation estimates than the random one. At each site, there were strong differences between beneficiary species in their spatial association with nurse species, and associations became more positive with increasing stress in Spain. Most importantly, there were no differences in results yielded by the two methods at any of the different stress levels at the Spanish and Japanese sites. At the Italian site, although micro‐environmental heterogeneity was low, we found weakly significant differences between methods that were unlikely due to habitat‐sharing effects. Thus, the paired sampling method can provide significant insights into net and long‐term effects of plant interactions in spatially conspicuous environments.
(Population Ecology. vol. 66, n° 1438-3896, pp. 39-52, 05/07/2023)
EPOC, EPHE, PSL, UB, INSU - CNRS, CNRS
La sécheresse fluviale et ses impacts socio-environnementaux : cas de la vallée de Manombo, sud-ouest de Madagascar
(03/07/2023)
LAM, IEP Bordeaux, IRD, UBM, CNRS, UBM, EPOC, EPHE, PSL, UB, INSU - CNRS, CNRS