Pollen identification through convolutional neural networks: First application on a full fossil pollen sequence.

The automation of pollen identification has seen vast improvements in the past years, with Convolutional Neural Networks coming out as the preferred tool to train models. Still, only a small portion of works published on the matter address the identification of fossil pollen. Fossil pollen is commonly extracted from organic sediment cores and are used by paleoecologists to reconstruct past environments, flora, vegetation, and their evolution through time. The automation of fossil pollen identification would allow paleoecologists to save both time and money while reducing bias and uncertainty. However, Convolutional Neural Networks require a large amount of data for training and databases of fossilized pollen are rare and often incomplete. Since machine learning models are usually trained using labelled fresh pollen associated with many different species, there exists a gap between the training data and target data. We propose a method for a large-scale fossil pollen identification workflow. Our proposed method employs an accelerated fossil pollen extraction protocol and Convolutional Neural Networks trained on the labelled fresh pollen of the species most commonly found in Northeastern American organic sediments. We first test our model on fresh pollen and then on a full fossil pollen sequence totalling 196,526 images. Our model achieved an average per class accuracy of 91.2% when tested against fresh pollen. However, we find that our model does not perform as well when tested on fossil data. While our model is overconfident in its predictions, the general abundance patterns remain consistent with the traditional palynologist IDs. Although not yet capable of accurately classifying a whole fossil pollen sequence, our model serves as a proof of concept towards creating a full large-scale identification workflow.

Human-induced fire regime shifts during 19th century industrialization: A robust fire regime reconstruction using northern Polish lake sediments.

Fire regime shifts are driven by climate and natural vegetation changes, but can be strongly affected by human land management. Yet, it is poorly known how humans have influenced fire regimes prior to active wildfire suppression. Among the last 250 years, the human contribution to the global increase in fire occurrence during the mid-19th century is especially unclear, as data sources are limited. Here, we test the extent to which forest management has driven fire regime shifts in a temperate forest landscape. We combine multiple fire proxies (macroscopic charcoal and fire-related biomarkers) derived from highly resolved lake sediments (i.e., 3-5 years per sample), and apply a new statistical approach to classify source area- and temperature-specific fire regimes (biomass burnt, fire episodes). We compare these records with independent climate and vegetation reconstructions. We find two prominent fire regime shifts during the 19th and 20th centuries, driven by an adaptive socio-ecological cycle in human forest management. Although individual fire episodes were triggered mainly by arson (as described in historical documents) during dry summers, the biomass burnt increased unintentionally during the mid-19th century due to the plantation of flammable, fast-growing pine tree monocultures needed for industrialization. State forest management reacted with active fire management and suppression during the 20th century. However, pine cover has been increasing since the 1990s and climate projections predict increasingly dry conditions, suggesting a renewed need for adaptations to reduce the increasing fire risk.

Palaeo-trajectories of forest savannization in the southern Congo.

Tropical savannah and forest are thought to represent alternative stable states in ecosystem structure in some climates. The implication is that biomes are maintained by positive feedbacks, e.g. with fire, and that historical distributions could play a role in determining modern ones. In this context, climate alone does not govern transitions between biomes, and understanding the causes and pathways of such transitions becomes crucial. Here, we use a multi-proxy analysis of a 2000-year core to evaluate modes of transition in vegetation structure and fire regimes. We demonstrate a first transition ca 1540 BP, when a cyclic fire regime entered a forested landscape, eventually resulting, by ca 1060 BP, in a transition to a more open savannah-like or mosaicked structure. This pattern may parallel currently accelerating fire regimes in tropical forests suggesting that fires can savannize forests, but perhaps more slowly than feared. Finally, ca 540 BP, a drought combined with anthropogenic influences resulted in a conclusive transition to savannah, probably resembling the modern landscape in the region. We show here that fire interacted with drought to transition forest to savannah, suggesting that disturbance by fire can be a major driver of biome change.

The climate, the fuel and the land use: Long-term regional variability of biomass burning in boreal forests.

The influence of different drivers on changes in North American and European boreal forests biomass burning (BB) during the Holocene was investigated based on the following hypotheses: land use was important only in the southernmost regions, while elsewhere climate was the main driver modulated by changes in fuel type. BB was reconstructed by means of 88 sedimentary charcoal records divided into six different site clusters. A statistical approach was used to explore the relative contribution of (a) pollen-based mean July/summer temperature and mean annual precipitation reconstructions, (b) an independent model-based scenario of past land use (LU), and (c) pollen-based reconstructions of plant functional types (PFTs) on BB. Our hypotheses were tested with: (a) a west-east northern boreal sector with changing climatic conditions and a homogeneous vegetation, and (b) a north-south European boreal sector characterized by gradual variation in both climate and vegetation composition. The processes driving BB in boreal forests varied from one region to another during the Holocene. However, general trends in boreal biomass burning were primarily controlled by changes in climate (mean annual precipitation in Alaska, northern Quebec, and northern Fennoscandia, and mean July/summer temperature in central Canada and central Fennoscandia) and, secondarily, by fuel composition (BB positively correlated with the presence of boreal needleleaf evergreen trees in Alaska and in central and southern Fennoscandia). Land use played only a marginal role. A modification towards less flammable tree species (by promoting deciduous stands over fire-prone conifers) could contribute to reduce circumboreal wildfire risk in future warmer periods.

Fire ecology of a tree glacial refugium on a nunatak with a view on Alpine glaciers.

In paleoecology, the function of biomass as a fire driver has become a focus of attention in cold ecosystems, and concerns have been raised about climate in this context. Little is known about the fire frequency and fire-plant relationships during glaciation when woodlands were limited and the climate was cold. Fire history and tree biomass were reconstructed from sedimentary charcoal and macroremains, respectively, archived in lake sediments from the western Alps. Two nunataks were investigated, both with lacustrine sediments covering the last 21 000 yr at least. During the Last Glacial Maximum (LGM) and the Lateglacial, fires occurred only on the nunatak sheltering woody plants. Cembra pine (Pinus cembra) and larch (Larix decidua) survived above glaciers during the LGM, thus evidencing a biological refugium and supporting the nunatak theory. We highlighted a long-term relationship between fires and dominant trees over the last 21 000 yr, where fire frequencies track the global climate and the local changes in tree biomass. Glacial climate (dry, cold) does not rule out fires. Fuel load and composition were significant fire drivers, with cembra pine dominating during colder periods with rare fires, and larch during the warmer Holocene with frequent fires. These findings increase knowledge of fire ecology in cold environments, and open perspectives in tree population genetics by considering new areas of tree glacial refugia in Europe.

Tree cover in Central Africa: determinants and sensitivity under contrasted scenarios of global change.

Tree cover is a key variable for ecosystem functioning, and is widely used to study tropical ecosystems. But its determinants and their relative importance are still a matter of debate, especially because most regional and global analyses have not considered the influence of agricultural practices. More information is urgently needed regarding how human practices influence vegetation structure. Here we focused in Central Africa, a region still subjected to traditional agricultural practices with a clear vegetation gradient. Using remote sensing data and global databases, we calibrated a Random Forest model to correlatively link tree cover with climatic, edaphic, fire and agricultural practices data. We showed that annual rainfall and accumulated water deficit were the main drivers of the distribution of tree cover and vegetation classes (defined by the modes of tree cover density), but agricultural practices, especially pastoralism, were also important in determining tree cover. We simulated future tree cover with our model using different scenarios of climate and land-use (agriculture and population) changes. Our simulations suggest that tree cover may respond differently regarding the type of scenarios, but land-use change was an important driver of vegetation change even able to counterbalance the effect of climate change in Central Africa.

Land-use change outweighs projected effects of changing rainfall on tree cover in sub-Saharan Africa.

Global change will likely affect savanna and forest structure and distributions, with implications for diversity within both biomes. Few studies have examined the impacts of both expected precipitation and land use changes on vegetation structure in the future, despite their likely severity. Here, we modeled tree cover in sub-Saharan Africa, as a proxy for vegetation structure and land cover change, using climatic, edaphic, and anthropic data (R(2)  = 0.97). Projected tree cover for the year 2070, simulated using scenarios that include climate and land use projections, generally decreased, both in forest and savanna, although the directionality of changes varied locally. The main driver of tree cover changes was land use change; the effects of precipitation change were minor by comparison. Interestingly, carbon emissions mitigation via increasing biofuels production resulted in decreases in tree cover, more severe than scenarios with more intense precipitation change, especially within savannas. Evaluation of tree cover change against protected area extent at the WWF Ecoregion scale suggested areas of high biodiversity and ecosystem services concern. Those forests most vulnerable to large decreases in tree cover were also highly protected, potentially buffering the effects of global change. Meanwhile, savannas, especially where they immediately bordered forests (e.g. West and Central Africa), were characterized by a dearth of protected areas, making them highly vulnerable. Savanna must become an explicit policy priority in the face of climate and land use change if conservation and livelihoods are to remain viable into the next century.

Regional paleofire regimes affected by non-uniform climate, vegetation and human drivers.

Climate, vegetation and humans act on biomass burning at different spatial and temporal scales. In this study, we used a dense network of sedimentary charcoal records from eastern Canada to reconstruct regional biomass burning history over the last 7000 years at the scale of four potential vegetation types: open coniferous forest/tundra, boreal coniferous forest, boreal mixedwood forest and temperate forest. The biomass burning trajectories were compared with regional climate trends reconstructed from general circulation models, tree biomass reconstructed from pollen series, and human population densities. We found that non-uniform climate, vegetation and human drivers acted on regional biomass burning history. In the open coniferous forest/tundra and dense coniferous forest, the regional biomass burning was primarily shaped by gradual establishment of less climate-conducive burning conditions over 5000 years. In the mixed boreal forest an increasing relative proportion of flammable conifers in landscapes since 2000 BP contributed to maintaining biomass burning constant despite climatic conditions less favourable to fires. In the temperate forest, biomass burning was uncoupled with climatic conditions and the main driver was seemingly vegetation until European colonization, i.e. 300 BP. Tree biomass and thus fuel accumulation modulated fire activity, an indication that biomass burning is fuel-dependent and notably upon long-term co-dominance shifts between conifers and broadleaf trees.

Assessing paleo-biodiversity using low proxy influx.

We developed an algorithm to improve richness assessment based on paleoecological series, considering sample features such as their temporal resolutions or their volumes. Our new method can be applied to both high- and low-count size proxies, i.e. pollen and plant macroremain records, respectively. While pollen generally abounds in sediments, plant macroremains are generally rare, thus leading to difficulties to compute paleo-biodiversity indices. Our approach uses resampled macroremain influxes that enable the computation of the rarefaction index for the low influx records. The raw counts are resampled to a constant resolution and sample volume by interpolating initial sample ages at a constant time interval using the age∼depth model. Then, the contribution of initial counts and volume to each interpolated sample is determined by calculating a proportion matrix that is in turn used to obtain regularly spaced time series of pollen and macroremain influx. We applied this algorithm to sedimentary data from a subalpine lake situated in the European Alps. The reconstructed total floristic richness at the study site increased gradually when both pollen and macroremain records indicated a decrease in relative abundances of shrubs and an increase in trees from 11,000 to 7,000 cal BP. This points to an ecosystem change that favored trees against shrubs, whereas herb abundance remained stable. Since 6,000 cal BP, local richness decreased based on plant macroremains, while pollen-based richness was stable. The reconstructed richness and evenness are interrelated confirming the difficulty to distinguish these two aspects for the studies in paleo-biodiversity. The present study shows that low-influx bio-proxy records (here macroremains) can be used to reconstruct stand diversity and address ecological issues. These developments on macroremain and pollen records may contribute to bridge the gap between paleoecology and biodiversity studies.

Vegetation limits the impact of a warm climate on boreal wildfires.

Strategic introduction of less flammable broadleaf vegetation into landscapes was suggested as a management strategy for decreasing the risk of boreal wildfires projected under climatic change. However, the realization and strength of this offsetting effect in an actual environment remain to be demonstrated. Here we combined paleoecological data, global climate models and wildfire modelling to assess regional fire frequency (RegFF, i.e. the number of fires through time) in boreal forests as it relates to tree species composition and climate over millennial time-scales. Lacustrine charcoals from northern landscapes of eastern boreal Canada indicate that RegFF during the mid-Holocene (6000-3000 yr ago) was significantly higher than pre-industrial RegFF (AD c. 1750). In southern landscapes, RegFF was not significantly higher than the pre-industrial RegFF in spite of the declining drought severity. The modelling experiment indicates that the high fire risk brought about by a warmer and drier climate in the south during the mid-Holocene was offset by a higher broadleaf component. Our data highlight an important function for broadleaf vegetation in determining boreal RegFF in a warmer climate. We estimate that its feedback may be large enough to offset the projected climate change impacts on drought conditions.

Control of the multimillennial wildfire size in boreal North America by spring climatic conditions.

Wildfire activity in North American boreal forests increased during the last decades of the 20th century, partly owing to ongoing human-caused climatic changes. How these changes affect regional fire regimes (annual area burned, seasonality, and number, size, and severity of fires) remains uncertain as data available to explore fire-climate-vegetation interactions have limited temporal depth. Here we present a Holocene reconstruction of fire regime, combining lacustrine charcoal analyses with past drought and fire-season length simulations to elucidate the mechanisms linking long-term fire regime and climatic changes. We decomposed fire regime into fire frequency (FF) and biomass burned (BB) and recombined these into a new index to assess fire size (FS) fluctuations. Results indicated that an earlier termination of the fire season, due to decreasing summer radiative insolation and increasing precipitation over the last 7.0 ky, induced a sharp decrease in FF and BB ca. 3.0 kyBP toward the present. In contrast, a progressive increase of FS was recorded, which is most likely related to a gradual increase in temperatures during the spring fire season. Continuing climatic warming could lead to a change in the fire regime toward larger spring wildfires in eastern boreal North America.

Fire, fuel composition and resilience threshold in subalpine ecosystem.

BACKGROUND: Forecasting the effects of global changes on high altitude ecosystems requires an understanding of the long-term relationships between biota and forcing factors to identify resilience thresholds. Fire is a crucial forcing factor: both fuel build-up from land-abandonment in European mountains, and more droughts linked to global warming are likely to increase fire risks. METHODS: To assess the vegetation response to fire on a millennium time-scale, we analyzed evidence of stand-to-local vegetation dynamics derived from sedimentary plant macroremains from two subalpine lakes. Paleobotanical reconstructions at high temporal resolution, together with a fire frequency reconstruction inferred from sedimentary charcoal, were analyzed by Superposed Epoch Analysis to model plant behavior before, during and after fire events. PRINCIPAL FINDINGS: We show that fuel build-up from arolla pine (Pinus cembra) always precedes fires, which is immediately followed by a rapid increase of birch (Betula sp.), then by ericaceous species after 25-75 years, and by herbs after 50-100 years. European larch (Larix decidua), which is the natural co-dominant species of subalpine forests with Pinus cembra, is not sensitive to fire, while the abundance of Pinus cembra is altered within a 150-year period after fires. A long-term trend in vegetation dynamics is apparent, wherein species that abound later in succession are the functional drivers, loading the environment with fuel for fires. This system can only be functional if fires are mainly driven by external factors (e.g. climate), with the mean interval between fires being longer than the minimum time required to reach the late successional stage, here 150 years. CONCLUSION: Current global warming conditions which increase drought occurrences, combined with the abandonment of land in European mountain areas, creates ideal ecological conditions for the ignition and the spread of fire. A fire return interval of less than 150 years would threaten the dominant species and might override the resilience of subalpine forests.