Deep learning and citizen science enable automated plant trait predictions from photographs

Crop Identification Using Deep Learning on LUCAS Crop Cover Photos

Massive and high-quality in situ data are essential for Earth-observation-based agricultural monitoring. However, field surveying requires considerable organizational effort and money. Using computer vision to recognize crop types on geo-tagged photos could be a game changer allowing for the provision of timely and accurate crop-specific information. This study presents the first use of the largest multi-year set of labelled close-up in situ photos systematically collected across the European Union from the Land Use Cover Area frame Survey (LUCAS). Benefiting from this unique in situ dataset, this study aims to benchmark and test computer vision models to recognize major crops on close-up photos statistically distributed spatially and through time between 2006 and 2018 in a practical agricultural policy relevant context. The methodology makes use of crop calendars from various sources to ascertain the mature stage of the crop, of an extensive paradigm for the hyper-parameterization of MobileNet from random parameter initialization, and of various techniques from information theory in order to carry out more accurate post-processing filtering on results. The work has produced a dataset of 169,460 images of mature crops for the 12 classes, out of which 15,876 were manually selected as representing a clean sample without any foreign objects or unfavorable conditions. The best-performing model achieved a macro F1 (M-F1) of 0.75 on an imbalanced test dataset of 8642 photos. Using metrics from information theory, namely the equivalence reference probability, resulted in an increase of 6%. The most unfavorable conditions for taking such images, across all crop classes, were found to be too early or late in the season. The proposed methodology shows the possibility of using minimal auxiliary data outside the images themselves in order to achieve an M-F1 of 0.82 for labelling between 12 major European crops.

Temporal stability of the hybrid zone between Calocitta magpie-jays revealed through comparison of museum specimens and iNaturalist photos

Hybrid zones are natural experiments for the study of avian evolution. Hybrid zones can be dynamic, moving as species adjust to new climates and habitats, with unknown implications for species and speciation. There are relatively few studies that have comparable modern and historic sampling to assess change in hybrid zone location and width over time, and those studies have generally found mixed results, with many hybrid zones showing change over time, but others showing stability. The white-throated magpie-jay (Calocitta formosa) and black-throated magpie-jay (Calocitta colliei) occur along the western coast of Mexico and Central America. The two species differ markedly in throat color and tail length, and prior observation suggests a narrow hybrid zone in southern Jalisco where individuals have mixed throat color. This study aims to assess the existence and temporal stability of this putative hybrid zone by comparing throat color between georeferenced historical museum specimens and modern photos from iNaturalist with precise locality information. Our results confirm the existence of a narrow hybrid zone in Jalisco, with modern throat scores gradually increasing from the parental ends of the cline toward the cline center in a sigmoidal curve characteristic of hybrid zones. Our temporal comparison suggests that the hybrid zone has not shifted its position between historical (pre-1973) and modern (post-2005) time periods-a surprising result given the grand scale of habitat change to the western Mexican lowlands during this time. An anomalous pocket of white-throated individuals in the northern range of the black-throated magpie-jay hints at the possibility of prehistorical long-distance introduction. Future genomic data will help disentangle the evolutionary history of these lineages and better characterize how secondary contact is affecting both the DNA and the phenotype of these species.

Applicability of Point- and Polygon-Based Vegetation Monitoring Data to Identify Soil, Hydrological and Climatic Driving Forces of Biological Invasions-A Case Study of Ailanthus altissima, Elaeagnus angustifolia and Robinia pseudoacacia

Invasive tree species are a significant threat to native flora. They modify the environment with their allelopathic substances and inhibit the growth of native species by shading, thus reducing diversity. The most effective way to control invasive plants is to prevent their spread which requires identifying the environmental parameters promoting it. Since there are several types of invasive plant databases available, determining which database type is the most relevant for investigating the occurrence of alien plants is of great importance. In this study, we compared the efficiency and reliability of point-based (EUROSTAT Land Use and Coverage Area Frame Survey (LUCAS)) and polygon-based (National Forestry Database (NFD)) databases using geostatistical methods in ArcGIS software. We also investigated the occurrence of three invasive tree species (Ailanthus altissima, Elaeagnus angustifolia, and Robinia pseudoacacia) and their relationships with soil, hydrological, and climatic parameters such as soil organic matter content, pH, calcium carbonate content, rooting depth, water-holding capacity, distance from the nearest surface water, groundwater depth, mean annual temperature, and mean annual precipitation with generalized linear models in R-studio software. Our results show that the invasion levels of the tree species under study are generally over-represented in the LUCAS point-based vegetation maps, and the point-based database requires a dataset with a larger number of samples to be reliable. Regarding the polygon-based database, we found that the occurrence of the invasive species is generally related to the investigated soil and hydrological and climatic factors.

Citizen science plant observations encode global trait patterns

Global maps of plant functional traits are essential for studying the dynamics of the terrestrial biosphere, yet the spatial distribution of trait measurements remains sparse. With the increasing popularity of species identification apps, citizen scientists contribute to growing vegetation data collections. The question emerges whether such opportunistic citizen science data can help map plant functional traits globally. Here we show that we can map global trait patterns by complementing vascular plant observations from the global citizen science project iNaturalist with measurements from the plant trait database TRY. We evaluate these maps using sPlotOpen, a global collection of vegetation plot data. Our results show high correlations between the iNaturalist- and sPlotOpen-based maps of up to 0.69 (r) and higher correlations than to previously published trait maps. As citizen science data collections continue to grow, we can expect them to play a significant role in further improving maps of plant functional traits.

An Outlook for Deep Learning in Ecosystem Science

Rapid advances in hardware and software, accompanied by public- and private-sector investment, have led to a new generation of data-driven computational tools. Recently, there has been a particular focus on deep learning—a class of machine learning algorithms that uses deep neural networks to identify patterns in large and heterogeneous datasets. These developments have been accompanied by both hype and scepticism by ecologists and others. This review describes the context in which deep learning methods have emerged, the deep learning methods most relevant to ecosystem ecologists, and some of the problem domains they have been applied to. Deep learning methods have high predictive performance in a range of ecological contexts, leveraging the large data resources now available. Furthermore, deep learning tools offer ecosystem ecologists new ways to learn about ecosystem dynamics. In particular, recent advances in interpretable machine learning and in developing hybrid approaches combining deep learning and mechanistic models provide a bridge between pure prediction and causal explanation. We conclude by looking at the opportunities that deep learning tools offer ecosystem ecologists and assess the challenges in interpretability that deep learning applications pose.