ABSTRACT
During past glacial periods, the land cover of Northern Eurasia and North America repeatedly shifted between open steppe tundra and boreal/temperate forest. Tracking these changes and estimating the coverage of open versus forested vegetation in past glacial and interglacial landscapes is notoriously difficult because the characteristic dwarf birches of the tundra and the tree birches of the boreal and temperate forests produce similar pollen grains that are difficult to distinguish in the pollen record. One objective approach to separating dwarf birch pollen from tree birch pollen is to use grain size statistics. However, the required grain size measurements are time-consuming and, therefore, rarely produced. Here, we present an approach to automatic size measurement based on image recognition with convolutional neural networks and machine learning. It includes three main steps. First, the TOFSI algorithm is applied to detect and classify pollen, including birch pollen, in lake sediment samples. Second, a Resnet-18 neural network is applied to select the birch pollen suitable for measurement. Third, semantic segmentation is applied to detect the outline and the area and mean width of each detected birch pollen grain. Test applications with two pollen records from Northern Germany, one covering the Lateglacial-Early Holocene transition and the other covering the Mid to Late Pleistocene transition, show that the new technical approach is well suited to measure the area and mean width of birch pollen rapidly (>1000 per hour) and with high accuracy. Our new network-based tool facilitates more regular size measurements of birch pollen. Expanded analysis of modern birch pollen will help to better understand size variations in birch pollen between birch species and in response to environmental factors as well as differential sample preparation. Analysis of fossil samples will allow better quantification of dwarf birch versus tree birch in past environments.
ABSTRACT
Plant roots influence many ecological and biogeochemical processes, such as carbon, water and nutrient cycling. Because of difficult accessibility, knowledge on plant root growth dynamics in field conditions, however, is fragmentary at best. Minirhizotrons, i.e. transparent tubes placed in the substrate into which specialized cameras or circular scanners are inserted, facilitate the capture of high-resolution images of root dynamics at the soil-tube interface with little to no disturbance after the initial installation. Their use, especially in field studies with multiple species and heterogeneous substrates, though, is limited by the amount of work that subsequent manual tracing of roots in the images requires. Furthermore, the reproducibility and objectivity of manual root detection is questionable. Here, we use a Convolutional Neural Network (CNN) for the automatic detection of roots in minirhizotron images and compare the performance of our RootDetector with human analysts with different levels of expertise. Our minirhizotron data come from various wetlands on organic soils, i.e. highly heterogeneous substrates consisting of dead plant material, often times mainly roots, in various degrees of decomposition. This may be seen as one of the most challenging soil types for root segmentation in minirhizotron images. RootDetector showed a high capability to correctly segment root pixels in minirhizotron images from field observations (F1 = 0.6044; r2 compared to a human expert = 0.99). Reproducibility among humans, however, depended strongly on expertise level, with novices showing drastic variation among individual analysts and annotating on average more than 13-times higher root length/cm2 per image compared to expert analysts. CNNs such as RootDetector provide a reliable and efficient method for the detection of roots and root length in minirhizotron images even from challenging field conditions. Analyses with RootDetector thus save resources, are reproducible and objective, and are as accurate as manual analyses performed by human experts.
Subject(s)
Neural Networks, Computer , Plant Roots , Humans , Reproducibility of Results , Carbon , SoilABSTRACT
The recent developments in artificial intelligence have the potential to facilitate new research methods in ecology. Especially Deep Convolutional Neural Networks (DCNNs) have been shown to outperform other approaches in automatic image analyses. Here we apply a DCNN to facilitate quantitative wood anatomical (QWA) analyses, where the main challenges reside in the detection of a high number of cells, in the intrinsic variability of wood anatomical features, and in the sample quality. To properly classify and interpret features within the images, DCNNs need to undergo a training stage. We performed the training with images from transversal wood anatomical sections, together with manually created optimal outputs of the target cell areas. The target species included an example for the most common wood anatomical structures: four conifer species; a diffuse-porous species, black alder (Alnus glutinosa L.); a diffuse to semi-diffuse-porous species, European beech (Fagus sylvatica L.); and a ring-porous species, sessile oak (Quercus petraea Liebl.). The DCNN was created in Python with Pytorch, and relies on a Mask-RCNN architecture. The developed algorithm detects and segments cells, and provides information on the measurement accuracy. To evaluate the performance of this tool we compared our Mask-RCNN outputs with U-Net, a model architecture employed in a similar study, and with ROXAS, a program based on traditional image analysis techniques. First, we evaluated how many target cells were correctly recognized. Next, we assessed the cell measurement accuracy by evaluating the number of pixels that were correctly assigned to each target cell. Overall, the "learning process" defining artificial intelligence plays a key role in overcoming the issues that are usually manually solved in QWA analyses. Mask-RCNN is the model that better detects which are the features characterizing a target cell when these issues occur. In general, U-Net did not attain the other algorithms' performance, while ROXAS performed best for conifers, and Mask-RCNN showed the highest accuracy in detecting target cells and segmenting lumen areas of angiosperms. Our research demonstrates that future software tools for QWA analyses would greatly benefit from using DCNNs, saving time during the analysis phase, and providing a flexible approach that allows model retraining.
