RS-SSKD: Self-Supervision Equipped with Knowledge Distillation for Few-Shot Remote Sensing Scene Classification

While growing instruments generate more and more airborne or satellite images, the bottleneck in remote sensing (RS) scene classification has shifted from data limits toward a lack of ground truth samples. There are still many challenges when we are facing unknown environments, especially those with insufficient training data. Few-shot classification offers a different picture under the umbrella of meta-learning: digging rich knowledge from a few data are possible. In this work, we propose a method named RS-SSKD for few-shot RS scene classification from a perspective of generating powerful representation for the downstream meta-learner. Firstly, we propose a novel two-branch network that takes three pairs of original-transformed images as inputs and incorporates Class Activation Maps (CAMs) to drive the network mining, the most relevant category-specific region. This strategy ensures that the network generates discriminative embeddings. Secondly, we set a round of self-knowledge distillation to prevent overfitting and boost the performance. Our experiments show that the proposed method surpasses current state-of-the-art approaches on two challenging RS scene datasets: NWPU-RESISC45 and RSD46-WHU. Finally, we conduct various ablation experiments to investigate the effect of each component of the proposed method and analyze the training time of state-of-the-art methods and ours.

Bio-optical evidence for increasing Phaeocystis dominance in the Barents Sea

Increasing contributions of prymnesiophytes such as Phaeocystis pouchetii and Emiliania huxleyi to Barents Sea (BS) phytoplankton production have been suggested based on in situ observations of phytoplankton community composition, but the scattered and discontinuous nature of these records confounds simple inference of community change or its relationship to salient environmental variables. However, provided that meaningful assessments of phytoplankton community composition can be inferred based on their optical characteristics, ocean-colour records offer a potential means to develop a synthesis between sporadic in situ observations. Existing remote-sensing algorithms to retrieve phytoplankton functional types based on chlorophyll-a (chl-a) concentration or indices of pigment packaging may, however, fail to distinguish Phaeocystis from other blooms of phytoplankton with high pigment packaging, such as diatoms. We develop a novel algorithm to distinguish major phytoplankton functional types in the BS and apply it to the MODIS-Aqua ocean-colour record, to study changes in the composition of BS phytoplankton blooms in July, between 2002 and 2018, creating time series of the spatial distribution and intensity of coccolithophore, diatom and Phaeocystis blooms. We confirm a north-eastward expansion in coccolithophore bloom distribution, identified in previous studies, and suggest an inferred increase in chl-a concentrations, reported by previous researchers, may be partly explained by increasing frequencies of Phaeocystis blooms. This article is part of the theme issue 'The changing Arctic Ocean: consequences for biological communities, biogeochemical processes and ecosystem functioning'.

One View Per City for Buildings Segmentation in Remote-Sensing Images via Fully Convolutional Networks: A Proof-of-Concept Study

The segmentation of buildings in remote-sensing (RS) images plays an important role in monitoring landscape changes. Quantification of these changes can be used to balance economic and environmental benefits and most importantly, to support the sustainable urban development. Deep learning has been upgrading the techniques for RS image analysis. However, it requires a large-scale data set for hyper-parameter optimization. To address this issue, the concept of “one view per city” is proposed and it explores the use of one RS image for parameter settings with the purpose of handling the rest images of the same city by the trained model. The proposal of this concept comes from the observation that buildings of a same city in single-source RS images demonstrate similar intensity distributions. To verify the feasibility, a proof-of-concept study is conducted and five fully convolutional networks are evaluated on five cities in the Inria Aerial Image Labeling database. Experimental results suggest that the concept can be explored to decrease the number of images for model training and it enables us to achieve competitive performance in buildings segmentation with decreased time consumption. Based on model optimization and universal image representation, it is full of potential to improve the segmentation performance, to enhance the generalization capacity, and to extend the application of the concept in RS image analysis.

Greater focus on water pools may improve our ability to understand and anticipate drought-induced mortality in plants

Drought-induced tree mortality has major impacts on ecosystem carbon and water cycles, and is expected to increase in forests across the globe with climate change. A large body of research in the past decade has advanced our understanding of plant water and carbon relations under drought. However, despite intense research, we still lack generalizable, cross-scale indicators of mortality risk. In this Viewpoint, we propose that a more explicit consideration of water pools could improve our ability to monitor and anticipate mortality risk. Specifically, we focus on the relative water content (RWC), a classic metric in plant water relations, as a potential indicator of mortality risk that is physiologically relevant and integrates different aspects related to hydraulics, stomatal responses and carbon economy under drought. Measures of plant water content are likely to have a strong mechanistic link with mortality and to be integrative, threshold-prone and relatively easy to measure and monitor at large spatial scales, and may complement current mortality metrics based on water potential, loss of hydraulic conductivity and nonstructural carbohydrates. We discuss some of the potential advantages and limitations of these metrics to improve our capacity to monitor and predict drought-induced tree mortality.

High Spatial Resolution Simulation of Sunshine Duration over the Complex Terrain of Ghana

In this paper, we propose a remote sensing model based on a 1 × 1 km spatial resolution to estimate the spatio-temporal distribution of sunshine percentage (SSP) and sunshine duration (SD), taking into account terrain features and atmospheric factors. To account for the influence of topography and atmospheric conditions in the model, a digital elevation model (DEM) and cloud products from the moderate-resolution imaging spectroradiometer (MODIS) for 2010 were incorporated into the model and subsequently validated against in situ observation data. The annual and monthly average daily total SSP and SD have been estimated based on the proposed model. The error analysis results indicate that the proposed modelled SD is in good agreement with ground-based observations. The model performance is evaluated against two classical interpolation techniques (kriging and inverse distance weighting (IDW)) based on the mean absolute error (MAE), the mean relative error (MRE) and the root-mean-square error (RMSE). The results reveal that the SD obtained from the proposed model performs better than those obtained from the two classical interpolators. This results indicate that the proposed model can reliably reflect the contribution of terrain and cloud cover in SD estimation in Ghana, and the model performance is expected to perform well in similar environmental conditions.

Use of Geospatial Surveillance and Response Systems for Vector-Borne Diseases in the Elimination Phase

The distribution of diseases caused by vector-borne viruses and parasites are restricted by the environmental requirements of their vectors, but also by the ambient temperature inside the host as it influences the speed of maturation of the infectious agent transferred. The launch of the Soil Moisture Active Passive (SMAP) satellite in 2015, and the new ECOSTRESS instrument onboard the International Space Station (ISS) in 2018, established the leadership of the National Aeronautics Space Administration (NASA) in ecology and climate research by allowing the structural and functional classification of ecosystems that govern vector sustainability. These advances, and the availability of sub-meter resolution data from commercial satellites, contribute to seamless mapping and modelling of diseases, not only at continental scales (1 km²) and local community or agricultural field scales (15–30 m²), but for the first time, also at the habitat–household scale (<1 m²). This communication presents current capabilities that are related to data collection by Earth-observing satellites, and draws attention to the usefulness of geographical information systems (GIS) and modelling for the study of important parasitic diseases.

Phenotyping Whole Forests Will Help to Track Genetic Performance

Phenotyping is the accurate and precise physical description of organisms. Accurate and quantitative phenotyping underpins the delivery of benefits from genetic improvement programs in agriculture. In forest trees, phenotyping at an equivalent precision has been impossible because trees and forests are large, long-lived, and highly variable. These facts have restricted the delivery of genetic gains in forestry compared to other agricultural sectors. We describe a landscape-scale phenotyping platform that integrates remote sensing, spatial information systems, and genomics to facilitate the delivery of greater gains enabling forestry to catch up with other sectors. Combining remote sensing at a range of spatial and temporal scales with genomics will ultimately impact on tree breeding globally.

Estimating activity of Japanese macaques (Macaca fuscata) using accelerometers

Accelerometers have been used to study both terrestrial and aquatic wildlife, mainly for mammal and bird species. In terrestrial mammals, there is a bias toward ungulates and carnivores, with fewer studies on nonhuman primates. In this study, we tested the use of accelerometers for studying the activity of Japanese macaques (Macaca fuscata). We modeled the activity of a male and a female subject by matching continuous focal observations from video recordings to sensor parameters derived from collar-mounted accelerometers. Models achieved classification performance (AUC) of greater than 90% for both subjects, with similar results when subjects were cross-validated. Accelerometer-based estimates of activity had comparable accuracies to estimates from instantaneous sampling at 1 min and 5 min intervals. We further demonstrated the use of model estimates for analyzing circadian rhythm and night time activity of M. fuscata. Our results add support to the feasibility of using accelerometers for studying activity of nonhuman primates. We discussed the limitations, benefits and potential applications of remote-sensing technology like accelerometers for advancing primalotogical studies.

Emerging Technologies to Conserve Biodiversity

Technologies to identify individual animals, follow their movements, identify and locate animal and plant species, and assess the status of their habitats remotely have become better, faster, and cheaper as threats to the survival of species are increasing. New technologies alone do not save species, and new data create new problems. For example, improving technologies alone cannot prevent poaching: solutions require providing appropriate tools to the right people. Habitat loss is another driver: the challenge here is to connect existing sophisticated remote sensing with species occurrence data to predict where species remain. Other challenges include assembling a wider public to crowdsource data, managing the massive quantities of data generated, and developing solutions to rapidly emerging threats.

Change in terrestrial ecosystem water-use efficiency over the last three decades

Defined as the ratio between gross primary productivity (GPP) and evapotranspiration (ET), ecosystem-scale water-use efficiency (EWUE) is an indicator of the adjustment of vegetation photosynthesis to water loss. The processes controlling EWUE are complex and reflect both a slow evolution of plants and plant communities as well as fast adjustments of ecosystem functioning to changes of limiting resources. In this study, we investigated EWUE trends from 1982 to 2008 using data-driven models derived from satellite observations and process-oriented carbon cycle models. Our findings suggest positive EWUE trends of 0.0056, 0.0007 and 0.0001 g C m(-2)  mm(-1)  yr(-1) under the single effect of rising CO2 ('CO2 '), climate change ('CLIM') and nitrogen deposition ('NDEP'), respectively. Global patterns of EWUE trends under different scenarios suggest that (i) EWUE-CO2 shows global increases, (ii) EWUE-CLIM increases in mainly high latitudes and decreases at middle and low latitudes, (iii) EWUE-NDEP displays slight increasing trends except in west Siberia, eastern Europe, parts of North America and central Amazonia. The data-driven MTE model, however, shows a slight decline of EWUE during the same period (-0.0005 g C m(-2)  mm(-1)  yr(-1) ), which differs from process-model (0.0064 g C m(-2)  mm(-1)  yr(-1) ) simulations with all drivers taken into account. We attribute this discrepancy to the fact that the nonmodeled physiological effects of elevated CO2 reducing stomatal conductance and transpiration (TR) in the MTE model. Partial correlation analysis between EWUE and climate drivers shows similar responses to climatic variables with the data-driven model and the process-oriented models across different ecosystems. Change in water-use efficiency defined from transpiration-based WUEt (GPP/TR) and inherent water-use efficiency (IWUEt , GPP×VPD/TR) in response to rising CO2 , climate change, and nitrogen deposition are also discussed. Our analyses will facilitate mechanistic understanding of the carbon-water interactions over terrestrial ecosystems under global change.

MODISTools - downloading and processing MODIS remotely sensed data in R

Remotely sensed data – available at medium to high resolution across global spatial and temporal scales – are a valuable resource for ecologists. In particular, products from NASA's MODerate-resolution Imaging Spectroradiometer (MODIS), providing twice-daily global coverage, have been widely used for ecological applications. We present MODISTools, an R package designed to improve the accessing, downloading, and processing of remotely sensed MODIS data. MODISTools automates the process of data downloading and processing from any number of locations, time periods, and MODIS products. This automation reduces the risk of human error, and the researcher effort required compared to manual per-location downloads. The package will be particularly useful for ecological studies that include multiple sites, such as meta-analyses, observation networks, and globally distributed experiments. We give examples of the simple, reproducible workflow that MODISTools provides and of the checks that are carried out in the process. The end product is in a format that is amenable to statistical modeling. We analyzed the relationship between species richness across multiple higher taxa observed at 526 sites in temperate forests and vegetation indices, measures of aboveground net primary productivity. We downloaded MODIS derived vegetation index time series for each location where the species richness had been sampled, and summarized the data into three measures: maximum time-series value, temporal mean, and temporal variability. On average, species richness covaried positively with our vegetation index measures. Different higher taxa show different positive relationships with vegetation indices. Models had high R² values, suggesting higher taxon identity and a gradient of vegetation index together explain most of the variation in species richness in our data. MODISTools can be used on Windows, Mac, and Linux platforms, and is available from CRAN and GitHub (https://github.com/seantuck12/MODISTools).