Search

The classification of pollen grains using texture information in combination with shape features is presented in this paper. The surface texture of pollen is characterised by using Gabor trans- forms, the geometric shape is described by using moment invariants, and the pollen grains are clas- sifiedby anartificial neural network.In anexperiment with fivetypes of pollengrains, more than 97% of samples are correctly classified. Copyright ! 2004 John Wiley & Sons, Ltd.

This is the first of a series of papers on the theme of automated pollen analysis. The automation of pollen analysis could result in numerous advantages for the reconstruction of past environments, with larger data sets made practical, objectivity and fine resolution sampling. There are also applications in apiculture and medicine. Previous work on the classification of pollen using texture measures has been successful with small numbers of pollen taxa. However, as the number of pollen taxa to be identified increases, more features may be required to achieve a successful classi- fication. This paper describes the use of simple geometric measures to augment the texture measures. The feasibility of this new approach is tested using scanning electron microscope (SEM) images of 12 taxa of fresh pollen taken from reference material collected on Henderson Island, Polynesia. Pollen images were captured directly from a SEM connected to a PC. A threshold grey-level was set and binary images were then generated. Pollen edges were then located and the boundaries were traced using a chain coding system. A number of simple geometric variables were calculated directly from the chain code of the pollen and a variable selection procedure was used to choose the optimal subset to be used for classification. The efficiency of these variables was tested using a leave-one-out clas- sification procedure. The system successfully split the original 12 taxa sample into five sub-samples containing no more than six pollen taxa each. The further subdivision of echinate pollen types was then attempted with a subset of four pollen taxa. A set of difference codes was constructed for a range of displacements along the chain code. From these difference codes probability variables were cal- culated. A variable selection procedure was again used to choose the optimal subset of probabilities that may be used for classification. The efficiency of these variables was again tested using a leave- one-out classification procedure. The proportion of correctly classified pollen ranged from 81% to 100% depending on the subset of variables used. The best set of variables had an overall classification rate averaging at about 95%. This is comparable with the classification rates from the earlier texture analysis work for other types of pollen. Copyright 2004 John Wiley & Sons, Ltd.

The automation of palynology (the identification and counting of pollen grains and spores) will be a small step for image recognition, but a giant stride for palynology. Here we show the first successful automated identification, with 100% accuracy, of a realistic number of taxa. The technique used involves a neural network classifier applied to surface texture data from light micro- scope images. A further significance of the technique is that it could be adapted for the identification of a wide range of biological objects, both microscopic and macroscopic. Copyright 2004 John Wiley & Sons, Ltd.

The importance of research leading to the automation of pollen identification is briefly outlined. A new technique, neural network analysis, is briefly introduced, and then applied to the determination of light microscope images of pollen grains. The results are compared with some previously published statistical classifiers. Although both types of classifiers may work, the neural network is apparently superior to the statistical methods in three ways: high success rates (100% in this case), small number of samples needed for training, and simplicity of features.

Two specific problems are identified in understanding the data on the Quaternary of Montane areas in the Sunda-Sahul region. These are: (1) The Upper Montane rainforest, according to palynological evidence, appears to have been absent in the Late Pleistocene, although it is well represented today. (2) The estimates of temperature lowering in the Late Pleistocene are strikingly greater in the mountains than in the lowlands. A single possible explanation for both problems is proposed. It is argued that altitudinal zonation of vegetation is partly controlled by ultraviolet-B light which is strikingly increased at high altitudes in the tropics. This fact could account for both the above problems, not only in the Sunda-Sahul region, but also in other tropical regions.

This paper draws attention to the following needs of palynology as a technique for palaeoclimatic reconstruction: more sites, fine resolution, larger counts, speed, objectivity and finer determination. The answer to these needs is automation of both preparation and analysis. Preparation is already well on the way to automation, and is not seen as a bottleneck. Previous attempts at automation of analysis are reviewed. The most successful to date have used computerised texture analysis of SEM images, but attempts are now being made to repeat this using optical images. The prospects for furthering this work are reviewed under the following headings: realistic scale; classifier choice; prior information, reject options and misclassification costs; feature selection; application to fossil grains; and performance assessment.

Ultraviolet light occurs in three wavebands. UV-A is the longest waveband (>315 nm) which is close to visible light and is of limited biological significance. UV-B (280–315 nm) is damaging and mutagenic to living organisms. UV-C (

The biomisation method is used to reconstruct Latin American vegetation at 6000±500 and 18 000±1000 radiocarbon years before present (14C yr BP) from pollen data. Tests using modern pollen data from 381 samples derived from 287 locations broadly reproduce potential natural vegetation. The strong temperature gradient associated with the Andes is recorded by a transition from high altitude cool grass/shrubland and cool mixed forest to mid-altitude cool temperate rain forest, to tropical dry, seasonal and rain forest at low altitudes. Reconstructed biomes from a number of sites do not match the potential vegetation due to local factors such as human impact, methodological artefacts and mechanisms of pollen representivity of the parent vegetation. At 6000±500 14C yr BP 255 samples are analysed from 127 sites. Differences between the modern and the 6000±500 14C yr BP reconstruction are comparatively small. Patterns of change relative to the modern reconstruction are mainly to biomes characteristic of drier climate in the north of the region with a slight more mesic shift in the south. Cool temperate rain forest remains dominant in western South America. In northwestern South America a number of sites record transitions from tropical seasonal forest to tropical dry forest and tropical rain forest to tropical seasonal forest. Sites in Central America also show a change in biome assignment to more mesic vegetation, indicative of greater plant available moisture, e.g. on the Yucatán peninsula sites record warm evergreen forest, replacing tropical dry forest and warm mixed forest presently recorded. At 18 000±1000 14C yr BP 61 samples from 34 sites record vegetation that reflects a generally cool and dry environment. Cool grass/shrubland prevalent in southeast Brazil, Amazonian sites record tropical dry forest, warm temperate rain forest and tropical seasonal forest. Southernmost South America is dominated by cool grass/shrubland, a single site retains cool temperate rain forest indicating that forest was present at some locations at the LGM. Some sites in Central México and lowland Colombia remain unchanged in their biome assignments, although the affinities that these sites have to different biomes do change between 18 000±1000 14C yr BP and present. The "unresponsive" nature of these sites results from their location and the impact of local edaphic influence.

Easter Island occupies an exceptionally isolated position in the south Pacific Ocean. It is entirely volcanic, and is famous for its giant statues. Late Quaternary sediments have been investigated in three craters: Rano Raraku, Rano Aroi and Rano Kao, giving a continuous record over the past 30 Ka. Pollen records indicate that the island was formerly forested. Palynological and sedimentological evidence suggests a Late Pleistocene climate slightly cooler and drier than the present. Deforestation by people occurred mainly between 1200 and 800 yr BP. This may have led to an ecological disaster and to the decline of the megalithic civilization. The depauperation of the present native flora owes more to human activity than to isolation.

A suite of 23 14C age determinations, from a well-stratified rockshelter and from 3 pollen cores on Mangaia Island is reported. The rockshelter has yielded significant evidence for avifaunal extinctions during the period cal. A.D. 1000-1600. The Lake Tiriara pollen cores span a period from ca. 6500 cal. b.p. to the present, and palynological analysis of the TIR 1 core indicates major anthropogenic disturbance on the island's vegetation after ca. 1600 cal. B.P. These sites, and the radiocarbon ages associated with them, provide the first chronologically secure evidence for human impacts on the island ecosystems of the southern Cook Islands.

The location and identification of singulated objects on microscope slides is a problem that is common to many applications, including recognition of pollen. In this paper, we describe a working system to solve this problem and demonstrate that it can be used to effectively locate pollen grains on slides, focus on them, photograph them, and then identify them based on a trained neural network. Our system aims to remove the need for laborious, time-consuming, and inaccurate counting of pollen grains by humans with a low-cost machine solution. It can deal with slides obtained using different preparation techniques and media. As well as describing the system, we present positive test results, including a comparision with human experts on the classification and counting of pollen on slides.

Automation of palynology could lead to many advances: rapid results, larger data sets, objectivity, fine resolution sampling and possibly finer determinations. To test the feasibility of automation, SEM photographs of six modern pollen taxa were used. The images were digitised and samples of exine texture covering approximately 1 10 of the total pollen area were extracted from the digital images. Texture analysis was applied to 192 samples obtained in this way. First, a co-occurrence matrix of grey levels was established for each sample. Then texture measures were calculated and used as input to a classification programme. With a leave-one-out strategy and a variable selection procedure, the proportion of pollen grains correctly identified rose to 94.3%. The procedure required c.10 seconds of processing on a VAX computer for each grain. With faster computers and programs, this could be cut to 1 second.

Vegetational history can help us to predict future environments by providing data for testing AGCMs, for indicating the vegetational response to rapid warming and changing CO2 concentrations, and for mathematical modelling of vegetation. Most of the data are palynological, and there are well over 100 pollen diagrams from tropical regions. Maps are presented showing summarized pollen diagrams from the lowlands of South-East Asia and the West Pacific, Tropical Latin America and Tropical Africa. In all these regions there is some evidence suggesting that at the LGM lowland forests were somewhat restricted in area and included montane elements. This is consistent with cooler and drier climate at the LGM. From the montane and lowland areas of these three regions, the pollen evidence is summarized in altitudinal diagrams. These suggest considerable depression of altitudinal zones at the LGM, suggesting temperatures c. 5–10°C cooler than now. These results conflict with earlier oxygen isotope data from marine foraminifera, but do not conflict with more recent oxygen isotope measurements from tropical corals. It is also suggested that altitudinal movements may be partly controlled by CO2 concentration and ultraviolet light.

As usually defined, the Massenerhebung or mountain mass elevation effect means the occurrence of physiognomically and sometimes floristically similar vegetation types at higher altitudes on large mountain masses than on small isolated peaks, especially those in or near the sea. Although the effect was first reported in the European Alps (Schroeter 1908) and in North America (where it is known as the Merriam Effect; Martin 1963), it is best known in the tropics. Perhaps its clearest expression is the occurrence of tropical mountain cloud forest (TMCF) (upper montane rain forest) at lower altitudes on isolated peaks than on the main mountain masses, which are taken as the norm (Figure 1).

Stratigraphies and pollen analyses are reported from three sites within 25 km east and west from Wabag in the highlands of Papua New Guinea, namely: Sirunki, 2500 m above sea level, 32000 to 1500 yr Inferred Ages; Inim, 2500 m above sea level, 10000 to 0 yr Inferred Ages; Birip, 1900 m above sea level, 2300 to 0 yr Inferred Ages. Events evidenced by these data are described against a time scale of Inferred Ages (I.A.) based on radiocarbon dates and stratigraphic considerations. The pollen analytical data from Sirunki are presented in terms of pollen recovery (deposition) rates as grains per square centimetre per year (grains cm-2a-1) and their interpretation controlled by information about total pollen deposition rates and differential pollen production and transport at the present day. Around Sirunki, the composition of the vegetation before 27500 I.A. is enigmatic, although almost certainly it was treeless. From then until 9000 I.A. subalpine and alpine conditions dominated except during two short periods when forest taxa grew in the catchment. Final afforestation began about 9000 I.A. but the composition of the forest did not stabilize until about 3000 years later. This relative stability was shortlived; soon after 5000 I.A. fluctuations in forest composition began. These fluctuations were associated with periodic changes in the proportion of forested to unforested land. The Inim data lead to conclusions generally compatible with those drawn from Sirunki. However, data from the two areas differ in detail, particularly in the later onset of change in the local forests about 2000 I.A. and its intensification, coeval with a diminution in forest area, after 500 I.A. The short record from Birip is dominated by serai changes on the crater wall itself but the main indicators of forest disturbance and unforested areas were already there at its beginning (2300 I.A.). It seems likely that general forest destruction began, or gained greater impetus, around Birip about 450 I.A. In the most general terms, the forest taxa, recorded by pollen analysis, have behaved consistently with their present distributions and ecological relationships throughout the last 30000 years. More detailed resolution, however, exposes many deviations from this generalization. The majority of taxa are usually associated in groups which vary in their composition repeatedly during a few thousand years, yet some of the taxa occasionally behave entirely individualistically. The establishment of forest broadly comparable with that growing around Sirunki today began about 9000 I.A., when the main components entered the catchment, but took about 2500 years to achieve balance and a repeated regeneration process. About 4500 I.A., the relationship between forest canopy trees and forest ephemerals changed from one explicable in terms of the latter’s role in natural forest regeneration to one suggestive of the ephemerals’ wide spread through the forest which could only have been achieved by degradation of the canopy. It is suggested that a rise in the Sirunki basin’s water level and the destruction of the surrounding forest about 13500 I.A. may have been due to seismic activity. The failure of the forest to re-establish there until 9000 i.a. was perhaps due to continued earth movement and partially to climatic conditions. The vegetation record from Sirunki suggests that the mean annual temperature there was similar to that of today between 27000 I.A. and 25500 I.A. but fell irregularly thereafter until between 18500 I.A. and 16000 I.A. it was probably about 10 °C below present. The mean annual temperature rose rapidly after 16000 i.a. and was within 1 °C of its present level by 13500 I.A. The cold episode between 18 500 I.A. and 16000 i.a. corresponds with the last glacial maximum at higher altitudes in New Guinea. Pollen analytical evidence of the altitude of the forest limit and Climap Project Members (1976) estimates of sea surface temperature at that time suggest a temperature lapse rate of about 8.5 °C per 1000 m altitude (compared with 5.8 °C at present), with a firn line kept high, as the geomorphological evidence demands, by low precipitation at high altitudes. In this coldest period the altitudinal forest limit was about 1500 m below its present level of 3800 m. There is some evidence to suggest that the highest altitude forests of that time may have been quite different from those of today, perhaps containing components of the lower mountain forest canopy as well as the plants of the present upper mountain forest. This implies that the upper mountain forest becomes a separate entity only during comparatively short excursions up the mountains during periods of relatively warm climate. The low altitude of the forest limit during the last major cold period and its subsequent rise through 1500 m must have had substantial repercussions on the composition of the forests at lower altitudes. Although there is no archaeological evidence, the pollen analytical data suggest human interference with the forests around Sirunki from about 4300 I.A., which for 1300 years involved clearing of the forest and the enhanced growth of ephemerals of forest and open-land. Subsequently, the forest remained generally degenerate and a new wave of clearing began about 2000 I.A. near both Sirunki and Inim which continued and intensified about 500 I.A. At the lower altitude of Birip, forests were already disturbed by the beginning of the pollen analytical record at about 2300 I.A.

The relative merits of applying contemporary atmospheric diffusional models to reconstructing Quaternary vegetational patterns via modern pollen rain studies are being examined in a continuing research project to model tree-pollen dispersion mathematically in closed-canopy woodlands, and to compare it with the traditional methods of Palynologists. It has been ascertained that the forcing functions required to drive the state-of-art dispersal equations are lacking in the palynological field, and thus limits to the usefulness of trying these are proposed. Finally, applications are suggested to other problems in palynology and present-day atmospheric diffusion.

Late Pleistocene pollen records from above 2000 m on tropical mountains indicate cooler climates. Similar records of comparable age from below 1200 m in the tropics indicate more arid climates. In this paper, we investigate Sumatran and Javan sites at intermediate altitudes. These provide evidence of a greater abundance of gymnosperms and higher altitude vegetation around the sites from at least 18,200 yr B.P. to ca. 12,400 yr B.P., which suggests that forest altitudinal boundaries were much lower than today's. Changes in the forest composition after ca. 12,400 yr B.P. indicate climatic amelioration. There are no palynological indications of formerly drier conditions and the presence of everwet climatic indicators, such as Altingia , throughout the period suggests that these findings are more compatible with a cooler than a more arid late Pleistocene climate.

Easter Island is the most isolated piece of inhabited land in the world. It exhibited an unique megalithic culture1–3, involving the sculpting of giant statues (moai) especially between AD ∼1400 and ∼16804 when, for unknown reasons, the culture suddenly collapsed. The island is also of interest in relation to Pleistocene climatic change as CLIMAP5 predicted no reduction of sea-surface temperatures for this part of the Pacific at 18,000 yr BP. We have obtained fossil pollen records covering the past 37,000 yr from three craters on the island. They suggest that the late Pleistocene climate was cooler and/or drier than the present one. They also suggest the former existence of forest on the island, and its decline in the last millennium. This decline was probably due to deforestation by man and could have caused the cultural collapse.

Summary Two complementary pollen diagrams from Canna show a forest history similar to that found elsewhere in north-west Scotland, but the NAP values are unusually high. A late-glacial may possibly be present.

The former existence of palms on Easter Island has been demonstrated palynologically1,2, but the genus could not be determined from pollen morphology. We now report the discovery on the island of endocarps (shells) from palm fruits which appear to bethose of an extinct species related to the Chilean ‘wine palm’, Jubaea chilensis (Molina) Baillon. The endocarps, found in caves, have all been gnawed by rodents, which could have helped to make the species extinct. The radiocarbon age of the endocarps is 820 ± 40 yr BP, which falls within the phase during which deforestation occurred on the island2 and the giant statues (moai) were erected1,3. Decline of the palm could have contributed to the decline of the moai culture.

POLLEN of Trapa natans L. the water chestnut, has been found in two Flandrian (Holocene) deposits at Skipsea, North Humberside, providing the first in situ Flandrian records of this species in Britain. The occurrence, followed by the extinction, of T. natans in Scandinavia has been regarded as evidence of a ‘climatic optimum’ when summer temperatures were at least 2 °C higher than at the present day1–4. The new British data may indicate similar conditions but could also record human activity, in which case it would not be necessary to invoke climatic change.

Tansley1 erected the hypothesis that the grasslands of the British chalk were the product of neolithic human forest clearance although he did not discount the possibility that in some areas the grasslands might have persisted throughout the forest period. Pigott and Walters2 suggested a range of calcareous habitats where forests perhaps failed to form a closed canopy: where erosion was rapid, slopes steep, or soils too thin for trees to root. Despite these exceptions, most authors describe chalk grasslands as the product of neolithic clearance (for example, refs 3 and 4). The debate has been starved of evidence until now by the almost total lack of Flandrian botanical data from sites actually on the chalk. Here we summarize the results of a broad palaeoecological study (using pollen, and plant and animal macrofossils) of a site in the heart of the Yorkshire Wolds which has yielded the first late-glacial and early Flandrian pollen record from the British chalk5. This deposit has provided evidence of early Flandrian chalk grasslands and of their persistence throughout the Pre-Boreal and Boreal periods.

Polleniferous sediments from SE Asian lakes and swamps may reflect anthropogenic changes in their surroundings. Increases in the abundance of pollen of secondary forest trees, herbs and crop plants are possible indicators of human activity, as are evidence of soil erosion and presence of charcoal. The areas most studied so far are Sumatra and New Guinea. Sites in Sumatra show forest disturbances from 4000 BP or earlier: possibly 7000 BP. From about 2000 BP permanent clearings appear. In New Guinea there is forest clearance in the highlands from c. 5000 BP or earlier, disturbance of swamps back to c. 9000 BP, and evidence of burning from c. 10,000 BP. Man has been practising forest disturbance in SE Asia for thousands of years, and this has led to ecological degradation.

inputs vary, but still this is a common characteristic for most of the biomes about which we will be taking. When we come, however, to consider precipitation, it is a little more difficult, because regions with this high radiant energy input embrace areas ranging from those with plentiful rainfall and high humidity throughout the year to nearly rainless deserts and through many climates with complex gradations between these two extremes. When we consider the concept of 'the norm', therefore, we may well be forced to limit our definition to, say, a true core area, such as the genuine lowland tropical rain forest and Koppen's Af climates. The tropics are also a region in which chemical and biological changes are noted for their speed of reaction, and we shall have to see whether the speed and nature of these reactions differ by degree only or in fundamental principle from those obtaining elsewhere on the globe. And, finally, we must accept, initially at least, a traditional view of the lowland tropical rain forest belt, namely that it is very old and has been but little affected by the climatic and other environmental changes incident on the Quaternary glaciations. However, I suspect that Dr Flenley may have a few caveats to enter concerning that common view and we may have to consider anew the so-called stability of a core community of lowland tropical rain forest. Let us now consider the second word, that enigmatic concept of a 'norm'. The main thesis underlying the question mark of this meeting is that, perhaps, because so much biogeographical and ecological theory has been developed from work in the boreal and temperate regions of the world, this theory has turned what is essentially the abnormal?the simple and the geologically young?into the normal, and that it could be misleading to have developed most of our basic principles in regions with soils and vegetation which are, in the main, less than 10 000 years old. Should not the tropics be the norm for developing theory and basic principles and not these clearly juvenile and ill-developed areas? I shall quote, as an example of the dilemma, one extreme instance. If the wilder theories of certain writers, such as those of the Russian, Federov (1966), were proved to have substance, and speciation (the creation of new species) in lowland tropical rain forest was seen to be governed by the random processes of genetic drift and not the clearly adaptive neo-Darwinistic principles of natural selection, then there would be no obvious and essential ecological relationship between habitat and species. The fundamental tenet of most ecological thinking would, in fact, be brought into question. This extreme case illustrates well the problem which we are confronting. In this instance, the differences would not be of degree only, but of principle. Many workers, on the other hand, regard the differences as of degree rather than of principle. The

Late Quaternary pollen records are described from two swamps at c. 1500 m a.s.l. in the Sumatran Highlands. At Danau di Atas (DDA) the record extends back to c. 31,000 BP. At Telago (TEL) the record extends back to c. 9000 BP. The results suggest that, in the Late Pleistocene, vegetation zones were depressed, and a gymnospermrich rainforest, now found only above 1800 m, surrounded DDA. At times upper montane forest may have been close to the site, suggesting a vegetational depression of c. 800 m. Mean annual temperature may have been c. 1.6?C to c. 5.2?C cooler than now. There is no direct evidence of formerly drier climates, although a possible reduction of water levels in the Late Pleistocene is suggested. During the Holocene, there is evidence for disturbance of rainforests, possibly by man, from a date some time after c. 8200 BP. Extensive recent forest clearance is demonstrated.

We report the first collection of Isoetes from Sumatra. The material belongs to a new species. The discovery eliminates a previous biogeographical anomaly, the absence of the world-wide genus Isoetes from Sumatra.