Dr. Achilleas Psomas

Land-use intensification in grassland ecosystems (i.e. increased mowing frequency, intensified grazing) has a strong negative effect on biodiversity and ecosystem services. However, accurate information on grassland-use intensity is difficult to acquire and restricted to the local or regional level. Recent studies have shown that mowing events can be mapped for large areas using satellite image time series. The transferability of such approaches, especially to mountain areas, has been little explored, however, and the relevance for ecological applications in biodiversity and conservation has hardly been investigated. Here, we used a rule-based algorithm to produce annual maps for 2018–2021 of grassland-management events, that is, mowing and/or grazing, for Switzerland using Sentinel-2 and Landsat 8 satellite data. We assessed the detection of management events based on independent reference data, which we acquired from daily time series of publicly available webcams that are widely distributed across Switzerland. We further examined the relationships between the generated grassland-use intensity measures and plant species richness and ecological indicator values derived from a nationwide field survey. The webcam-based verification for 2020 and 2021 revealed that most detected management events were actual mowing/grazing events (≥78%), but that a substantial number of events were not detected (up to 57%), particularly grazing events at higher elevations. We found lower plant species richness and higher mean ecological indicator values for nutrients and mowing tolerance with more frequent management events and those starting earlier in the year. A large proportion of the variance was explained by our use-intensity measures. Our findings therefore highlight that remotely assessed management events can characterise land-use intensity at fine spatial and temporal resolutions across broad scales and can explain plant biodiversity patterns in grasslands.

Urban green spaces with healthy vegetation play a key role in improving the quality of life in cities. However, urban soils, the basis of the urban greenery, are under strong anthropogenic influence and can considerably differ from natural soils. In this study, we observed short-term greening and browning of lawns during one vegetation period in urban parks of Geneva (Switzerland). We related the temporal trajectory of seasonal Normalized Difference Vegetation Index (NDVI) (8 days median) as a proxy of vegetation condition at different test sites to the physical soil properties (soil depth, coarse material, bulk density, conditioned air and water content) and Soil Organic Carbon (SOC). Strong drops of NDVI during dry periods in summer were related to shallow soil depths (<40 cm) and a high amount of coarse material (>10%) as well as lower SOC. Bulk density of the fine earth and the soil structure quality (expressed by air and water content of soil cores conditioned at a soil water potential of -100 hPa) had a significant influence on grass growth in spring but not in summer. Dense soils with conditioned air content closer to the trigger value of degraded soil structure resulted in lower NDVI values in spring. Our approach of using Earth Observation (EO) data for observing short-term greening and browning patterns, in this case the rise and decline of NDVI values, revealed that the role of the soil properties changed with the season. This approach may contribute to digital soil mapping and the assessment of soil ecosystem services in urban contexts. Urban planners are advised to save natural soils from over-building and keep them for green spaces. If soil has to be restored to create new green spaces, it should be deep and should not contain much coarse material, even for grassy vegetation.

Essential forest ecosystem services can be assessed by better understanding the diversity of vegetation, specifically those of Mediterranean region. A species level classification of maquis would be useful in understanding vegetation structure and dynamics, which would be an indicator of degradation or succession in the region. Although remote sensing was regularly used for classification in the region, maquis are simply represented as one to three categories based on density or height. To fill this gap, we test the capability of Sentinel-2 imagery, together with selected ancillary variables, for an accurate mapping of the dominant maquis formations. We applied Recursive Feature Selection procedure and used a Random Forest classifier. The algorithm is tested using ground truth collected from site and reached 78% and 93% overall accuracy at species level and physiognomic level, respectively. Our results suggest species level characterization of dominant maquis is possible with Sentinel-2 spatial resolution.

Increasing urbanization degrades quantity, quality, and the functionality of spatial cohesion of natural areas essential to biodiversity and ecosystem functioning worldwide. The uncontrolled pace of building activity and the erosion of blue (i.e., aquatic) and green (i.e., terrestrial) landscape elements threaten existing habitat ranges and movability of wildlife. Local scale measures, such as nature-inspired engineered Blue-Green Infrastructure (BGI) are emerging mitigation solutions. Originally planned to promote sustainable stormwater management, adaptation to climate change and improved human livability in cities, such instruments offer interesting synergies for biodiversity in support of existing ecological infrastructure. BGI are especially appealing for globally declining amphibians, a rich and diverse vertebrate assemblage sensitive to urbanization. We integrated biological and highly resolved urban-rural land-cover data, ensemble models of habitat suitability, and connectivity models based on circuit theory to improve multi-scale and multi-species protection of core habitats and ecological corridors in the Swiss lowlands. Considering a broad spectrum of amphibian biodiversity, we identified distributions of amphibian biodiversity hotspots and four landscape elements essential to amphibian movability at the regional scale, namely i) forest edges, ii) wet-forest habitats, iii) soils with variable moisture and iv) riparian zones. Our work shows that cities can make a substantial contribution (e.g., up to 15% of urban space in the study area) to wider landscape habitat connectivity. We highlight the importance of planning BGI locally in strategic locations across urban and peri-urban areas to promote the permeability and availability of ‘stepping stone’ habitats in densely populated landscapes, essential to the maintenance of regional habitat connectivity and thereby enhancing biodiversity and ecosystem functioning.

Central Europe has been experiencing unprecedented droughts during the last decades, stressing the decrease in tree water availability. However, the assessment of physiological drought stress is challenging, and feedback between soil and vegetation is often omitted because of scarce belowground data. Here we aimed to model Swiss forests' water availability during the 2015 and 2018 droughts by implementing the mechanistic soil-vegetation-atmosphere-transport (SVAT) model LWF-Brook90 taking advantage of regionalized depth-resolved soil information. We calibrated the model against soil matric potential data measured from 2014 to 2018 at 44 sites along a Swiss climatic and edaphic drought gradient. Swiss forest soils' storage capacity of plant-available water ranged from 53 mm to 341 mm, with a median of 137 ± 42 mm down to the mean potential rooting depth of 1.2 m. Topsoil was the primary water source. However, trees switched to deeper soil water sources during drought. This effect was less pronounced for coniferous trees with a shallower rooting system than for deciduous trees, which resulted in a higher reduction of actual transpiration (transpiration deficit) in coniferous trees. Across Switzerland, forest trees reduced the transpiration by 23% (compared to potential transpiration) in 2015 and 2018, maintaining annual actual transpiration comparable to other years. Together with lower evaporative fluxes, the Swiss forests did not amplify the blue water deficit. The 2018 drought, characterized by a higher and more persistent transpiration deficit than in 2015, triggered widespread early wilting across Swiss forests that was better predicted by the SVAT-derived mean soil matric potential in the rooting zone than by climatic predictors. Such feedback-driven quantification of ecosystem water fluxes in the soil–plant-atmosphere continuum will be crucial to predicting physiological drought stress under future climate extremes.

Pour la champignonneuse amatrice ou le joggeur du dimanche, les changements se produisant dans les écosystèmes forestiers sont visible en quelques coups d'œil depuis déjà quelques temps. L'étonnement n'est desormais plus de mise lorsque l'on lit dans les journaux quotidiens que des agriculteurs du jura ont perdu l'intégralité de leurs récoltes, ou que les vaches, manquant d'herbe grasse en altitude, redescendent de l'alpage au milieu de l'été. Outre les faits innombrables quant à la diminution autant quantitative que qualitative des écosystèmes forestiers de part le monde dû aux impacts de l'homme, une motivation particulièrement émotionnelle à motiver l'équipe d'ExoLabs de se joindre au WSL et au laboratoire EcoVision de l'ETH pour proposer une solution complémentaire a la surveillance des forêts. II est néanmoins utile de rappeler que les forêts sont soumises à la déforestation, à l'expansion des zones urbaines, à des changements climatiques et souvent à des asséchements locaux, ainsi qu'aux incendies, à des maladies et parasites; de plus, ces pressions sont souvent combinées voire en synergies sur la plupart des parcelles forestières tout autour du globe. Etant spécialistes de solution informatique basées sur de l'imagerie satellitaire passive, nos efforts portent ici sur le développement d'une solution basée sur cette technologie: ExoSilva.

Seit einigen Jahren sind auch für die passionierte Pilzsammlerin oder den Sonntagsjogger die Veränderungen in den Wäldern deutlich sichtbar. Deshalb ist es keine Überraschung, wenn in den Medien über die Ernteverluste der Bauerh im Jura oder dem frühzeitigen Abstieg der Kühe von der Alp wegen Futtermangel berichtet wird. Neben den klimatisch bedingten Veränderungen nimmt aber auch der anthropogene Einfluss auf die Waldökosysteme immer mehr zu. Dies war die Motivation eines Konsortiums, bestehend aus der ExoLabs GmbH, dem WSL und der EcoVision Gruppe der ETHZ, um eine Plattform für ein regelmässiges Monitoring der Schweizer Wälder umzusetzen. Dieses Monitoring-System sollte nicht nur die menschlichen Aktivitäten erfassen, sondern auch die Auswirkungen von Trockenheit, Krankheiten, Waldbrände oder den Befall mit Parasiten frühzeitig erkennbar machen. Aus der Kombination von Satellitendaten und umfangreichen Geländeinformationen sowie der Erfahrung im Bereich Dateninfrastruktur und Bildverarbeitung wurde ein neues, interaktives Tool für das Waldmonitoring entwickelt: ExoSilva.

Since the opening of Earth Observation (EO) archives (USGS/NASA Landsat and EC/ESA Sentinels), large collections of EO data are freely available, offering scientists new possibilities to better understand and quantify environmental changes. Fully exploiting these satellite EO data will require new approaches for their acquisition, management, distribution, and analysis. Given rapid environmental changes and the emergence of big data, innovative solutions are needed to support policy frameworks and related actions toward sustainable development. Here we present the Swiss Data Cube (SDC), unleashing the information power of Big Earth Data for monitoring the environment, providing Analysis Ready Data over the geographic extent of Switzerland since 1984, which is updated on a daily basis. Based on a cloud-computing platform allowing to access, visualize and analyse optical (Sentinel-2; Landsat 5, 7, 8) and radar (Sentinel-1) imagery, the SDC minimizes the time and knowledge required for environmental analyses, by offering consistent calibrated and spatially co-registered satellite observations. SDC derived analysis ready data supports generation of environmental information, allowing to inform a variety of environmental policies with unprecedented timeliness and quality.

Intensiv genutzte Landschaften und Lebensraumfragmentierung sind mitverantwortlich für den Rückgang der Biodiversität, weil Lebensraumqualität und grundlegende biologische Prozesse wie z.B. die Ausbreitung gestört werden. Stetig wachsende Städte als überregionale Barrieren für die Vernetzung natürlicher Lebensräume stellen eine besondere Herausforderung dar (Abb. 1). Wir zeigen am Beispiel von Amphibien auf, wie eine gezielte Platzierung blaugrüner Infrastrukturen die Durchlässigkeit von Städten verbessern könnte und dadurch Lebensräume und Laichgebiete von regionaler Bedeutung besser vernetzt werden. Die Einrichtung blaugrüner Infrastrukturen ist eine vielversprechende Massnahme, um die Anpassung an den Klimawandel zu fördern und die Lebensqualität zu verbessern. Synergien zur Vernetzung der Biodiversität bieten sich auf lokaler und regionaler Ebene an.

L'utilisation intensive des paysages et la fragmentation des habitats sont en partie co-responsables du déclin de la biodiversité, dans la mesure où la qualité des habitats et des processus biologiques fondamentaux tels que la dispersion sont perturbés. L'urbanisation croissante constitue un défi particulier dans ce contexte, notamment en tant qu'obstacle transrégional à la connectivité des habitats naturels. En prenant l'exemple des amphibiens, nous montrons comment la planification ciblée d'infrastructures bleuesvertes pourrait améliorer la perméabilité des villes en reliant mieux les habitats naturels et les sites de reproduction d'importance régionale. L'installation d'infrastructures bleues-vertes est une mesure prometteuse qui est actuellement planifiée dans de nombreuses villes surtout pour promouvoir l'adaptation au changement climatique ainsi que pour l'amélioration de la qualité de vie. Cependant, des synergies pour la mise en réseau de la biodiversité sont aussi disponibles non seulement à niveaux local mais aussi régional.

Countrywide winter and summer Sentinel-1 (S1) backscatter data, cloud-free summer Sentinel-2 (S2) images, an Airborne Laser Scanning (ALS)-based Digital Terrain Model (DTM) and a forest mask were used to model and subsequently map Dominant Leaf Type (DLT) with the thematic classes broadleaved and coniferous trees for the whole of Switzerland. A novel workflow was developed that is robust, cost-efficient and highly automated using reference data from aerial image interpretation. Two machine learning approaches based on Random Forest (RF) and deep learning (UNET) for the whole country with three sets of predictor variables were applied. 24 subareas based on aspect and slope categories were applied to explore effects of the complex mountainous topography on model performances. The reference data split into training, validation and test data sets was spatially stratified using a 25 km regular grid. Model accuracies of both RF and UNET were generally highest with Kappa (K) around 0.95 when predictors were included from both S1/S2 and the topographic variables aspect, elevation and slope from the DTM. While only slightly lower accuracies were obtained when using S2 and DTM data, lowest accuracies were obtained when only predictors from S1 and DTM were included, with RF performing worse than UNET. While on countrywide level RF and UNET performed overall similarly, substantial differences in model performances, i.e. higher variances and lower accuracies, were found in subareas with northwest to northeast orientations. The combined use of S1/S2 and DTM predictors mitigated these problems related to topography and shadows and was therefore superior to the single use of S1 and DTM or S2 and DTM data. The comparison with independent National Forest Inventory (NFI) plot data demonstrated precisions of K around 0.6 in the predictions of DLT and indicated a trend of increasing deviations in mixed forests. A comparison with the Copernicus High Resolution Layer (HRL) DLT 2018 revealed overall higher map accuracies with the exception of pure broadleaved forest. Although, spatial patterns of DTL were overall similar, UNET performed better than RF in areas with a distinct DLT on forest stand level, with the largest differences occurring when only S1 and DTM data was used. In contrast, predictions obtained from RF were more accurate in mixed stands. This study goes beyond the case study level and meets the requirements of countrywide data sets, in particular regarding repeatability, updating, costs and characteristics of training data sets. The 10 m countrywide DLT maps add complementary and spatially explicit information to the existing NFI estimates and are thus highly relevant for forestry practice and other related fields.

Multispektrale Satellitenbilder in hoher zeitlicher Wiederholung wurden genutzt, um die Gebiete in der Schweiz zu ermitteln, in denen die Buchen, aber auch die Nadelbäume besonders stark von der im Sommer 2018 herrschenden Trockenheit betroffen waren. Hierfür wurden aus den Satellitenbildern Vegetationsindizes berechnet und Schwellenwerte für Trockenheitssymptome wie die vorzeitige Verbraunung abgeleitet. Durch die freie Verfügbarkeit der Sentinel-2-Daten des Copernicus-Programms und von Cloud-Plattformen wie Google Earth Engine konnten die Analysen rasch durchgeführt werden.

The combination of drought and heat affects forest ecosystems by deteriorating the health of trees, which can lead to large‐scale die‐offs with consequences on biodiversity, the carbon cycle, and wood production. It is thus crucial to understand how drought events affect tree health and which factors determine forest susceptibility and resilience. We analyze the response of Central European forests to the 2018 summer drought with 10 × 10 m satellite observations. By associating time‐series statistics of the Normalized Difference Vegetation Index (NDVI) with visually classified observations of early wilting, we show that the drought led to early leaf‐shedding across 21,500 ± 2,800 km², in particular in central and eastern Germany and in the Czech Republic. High temperatures and low precipitation, especially in August, mostly explained these large‐scale patterns, with small‐ to medium‐sized trees, steep slopes, and shallow soils being important regional risk factors. Early wilting revealed a lasting impact on forest productivity, with affected trees showing reduced greenness in the following spring. Our approach reliably detects early wilting at the resolution of large individual crowns and links it to key environmental drivers. It provides a sound basis to monitor and forecast early‐wilting responses that may follow the droughts of the coming decades.

Pressures on natural resources are increasing and a number of challenges need to be overcome to meet the needs of a growing population in a period of environmental variability. The key to sustainable development is achieving a balance between the exploitation of natural resources for socioeconomic development and maintaining ecosystem services that are critical to human's wellbeing and livelihoods. Some of these environmental issues can be monitored using remotely sensed Earth Observations (EO) data that are increasingly available from freely and openly accessible repositories. Hereafter, we present the Swiss Data Cube, a unique Analysis Ready Data archive of satellite imagery and some use cases to monitor Sustainable Development Goals.

Befall durch den Buchdrucker (Ips typographus) ist in Fichtenwäldern der hochmontanen und subalpinen Stufe eine grosse Herausforderung für die Waldbewirtschaftung. Es ist unmöglich, im oft steilen Gelände einzig mit terrestrischen Beobachtungen alle anfälligen beziehungsweise bereits befallenen Fichten (Picea abies) rechtzeitig zu identifizieren. Um rechtzeitig eingreifen zu können, ist eine zeitsparende, kostengünstige und effektive Methode zum Auffinden dieser Fichten notwendig. Vielversprechend scheint die Technik der Fernerkundung. Deshalb untersuchten wir das Potenzial von hyperspektralen Fernerkundungsdaten für die Früherkennung gestresster oder befallener Fichten an einem Fallbeispiel im Lehr- und Forschungswald der Eidgenössischen Technischen Hochschule (ETH) Zürich in Sedrun im Kanton Graubünden. Der verwendete Ansatz beruht auf einer Kombination von Feldaufnahmen und hyperspektralen Daten, daraus berechneten spektralen Vegetationsindizes und einer Klassifizierung der Fichten in tote, gestresste und gesunde Bäume durch Random-Forest-Modelle, eine Methode des maschinellen Lernens. Es konnte gezeigt werden, dass gestresste Fichten mit diesem Ansatz zwar aufgespürt werden können, die Methode für den operativen Einsatz aber noch nicht ausgereift ist. Insbesondere wäre eine hangspezifische Kalibrierung nötig, was die praktische Anwendung verunmöglicht.

Mapping and monitoring agricultural land-use intensity (LUI) changes are essential for understanding their effects on biodiversity. Current land-use models provide a rather coarse spatial resolution, while in-situ measurements of LUI cover only a limited extent and are time-consuming and expensive. The purpose of this study is to evaluate the feasibility of using habitat type, topo-climatic, economic output, and remote-sensing data to map LUI at a high spatial resolution. To accomplish this, we first rated the habitat types across the agricultural landscape in terms of the amount and frequency of fertiliser input, pesticide input, ploughing, grazing, mowing, harvesting, and biomass output. We consolidated these ratings into one LUI index per habitat type that we then related to topo-climatic, economic output, and remote-sensing predictors. The results showed that the LUI index was strongly related to plant indicator values for mowing tolerance and soil nutrient content and to aerial nitrogen deposition, and thus, is an adequate index. Topo-climatic, and, to a smaller extent, economic output and remote-sensing predictors, proved suitable for mapping LUI. Large-to medium-scale patterns are explained by topo-climatic predictors, while economic output predictors explain medium-scale patterns and remote-sensing predictors explain local-scale patterns. With the fine-scale LUI map produced from this study, it is now possible to estimate within unvarying land-use classes, the effect on agrobiodiversity of an increase in LUI on fertile and accessible lands and of a decrease of LUI by the abandonment of marginal agricultural lands, and thus, provide a valuable base for understanding the effects of LUI on biodiversity. Due to the worldwide availability of remote-sensing and climate data, our methodology can be easily applied to other countries where habitat-type data are available. Given their low explanatory power, economic output variables may be omitted if not available.

Nur gerade 20 mm² beträgt im Monitoringnetzwerk TreeNet die Messfläche eines Dendrometers auf der Oberfläche eines Baumstammes, während die Satelliten Sentinel-2 Tausende von Quadratkilometern mit einer Auflösung von 100 m² erfassen. Die Daten aus dem Jahr 2018 zeigen, dass die durch die herrschende Trockenheit verursachten, zum Teil extrem hohen Baumwasserdefizite von Einzelbäumen zumindest zum Teil aus dem All zu erkennen waren und unterschiedlich grosse Auswirkungen auf das Stammwachstum hatten. Während das Wachstum der TreeNet-Fichten fast durchwegs einbrach, war dasjenige der im Netzwerk enthaltenen Buchen und Waldföhren uneinheitlich und dasjenige der Eichen eher besser.

Livestock farmers rely on a high and stable grassland productivity for fodder production to sustain their livelihoods. Future drought events related to climate change, however, threaten grassland functionality in many regions across the globe. The introduction of sustainable grassland management could buffer these negative effects. According to the biodiversity-productivity hypothesis, productivity positively associates with local biodiversity. The biodiversity-insurance hypothesis states that higher biodiversity enhances the temporal stability of productivity. To date, these hypotheses have mostly been tested through experimental studies under restricted environmental conditions, hereby neglecting climatic variations at a landscape-scale. Here, we provide a landscape-scale assessment of the contribution of species richness, functional composition, temperature, and precipitation on grassland productivity. We found that the variation in grassland productivity during the growing season was best explained by functional trait composition. The community mean of plant preference for nutrients explained 24.8% of the variation in productivity and the community mean of specific leaf area explained 18.6%, while species richness explained only 2.4%. Temperature and precipitation explained an additional 22.1% of the variation in productivity. Our results indicate that functional trait composition is an important predictor of landscape-scale grassland productivity.

Methods and workflows for creating area-wide products using remote sensing methods have been developed for various reasons. For the sample-based estimates in the Swiss National Forest Inventory (NFI), area-wide data sets are used for the two-phase estimations, which have substantially lower estimation errors than one-phase inventories. Using area-wide data sets, it is possible to skip dense manual interpretation of the stereo-images and thus save resources. Further, many functions of the forest, such as biodiversity and protection against natural hazards, can be described and quantified better with area-wide spatial data than with field plot data.

 

Studies have shown that spatial information on abandoned land could play an important role in urban land management, as land abandonment was proven capable of revealing future trajectories of change. However, mapping land abandonment with traditional methods (e.g., field work, digitization of aerial images) can be time consuming, expensive, and require considerable man-power. In this context, Landsat imagery proves to be a reliable source of data. To our knowledge, the potential of Landsat imagery for mapping abandoned land has not been tested in heterogeneous and highly fragmented urban settings. The aim of our paper is to propose a resource-efficient (i.e., in terms of time and manpower) method for the assessment of land abandonment in areas affected by urban sprawl, by using seasonal time series of Landsat data. Bucharest, Romania was chosen as case study area. Landsat scenes from the year 2013 are grouped based on vegetation phenology into four PGSs (periods of the growing season). NDVI values corresponding to land abandonment are analyzed with Classification and Regression Trees. A total of 23 models – representing combinations of PGSs – are tested in order to determine what period of the vegetation growing season fits best for mapping abandoned land, and for the purpose of deriving such a map for Bucharest. Finally, results are validated against independent data and a resource estimation for the entire mapping process is performed. Results show that abandoned land can be mapped with Landsat imagery with accuracies above 80%. Higher accuracies are obtained when scenes encompassing the beginning of the vegetation growing season are included in the models. We also observed that accuracy tends to decrease from models which include PGSs representing the beginning of the vegetation growing season towards those representing the end. Estimation showed that mapping land abandonment with Landsat data could reduce time and workforce resources by almost half compared with aerial imagery and field work. As our method is rapid, easy to implement, and based on freely available data, it can be used by local authorities that cannot allocate significant resources for land change monitoring. Furthermore, the approach could provide objective information to municipalities where official statistics on land abandonment are unreliable or of low quality.

Mapping changes in vegetation cover is essential for understanding the consequences of climate change on Arctic ecosystems. Classification of ultra-high spatial-resolution (UHR, < 1 cm) imagery can provide estimates of vegetation cover across space and time. The challenge of this approach is to assure comparability of classification across many images taken at different illumination conditions and locations. With warming, vegetation at higher elevation is expected to resemble current vegetation at lower elevation. To investigate the value of classification of UHR imagery for monitoring vegetation change, we collected visible and near-infrared images from 108 plots with hand-held cameras along an altitudinal gradient in Greenland and examined the classification accuracy of shrub cover on independent images (i.e. classification transferability). We implemented several models to examine if colour calibration improves transferability based on an in-image calibration target. The classifier was trained on different number of images to find the minimum training subset size. With a training set of ~ 20% of the images the overall accuracy levelled off at about 81% and 68% on the non-calibrated training and validation images, respectively. Colour calibration improved the accuracy on training images (1–4%) while it only improved the classifier transferability significantly for training sets < 20%. Linear calibration only based on the target’s grey series improved transferability most. Reasonable transferability of Arctic shrub cover classification can be obtained based only on spectral data and about 20% of all images. This is promising for vegetation monitoring through multi-image classification of UHR imagery acquired with hand-held cameras or Unmanned Aerial Systems.
 

Despite their importance to ocean productivity, global patterns of marine phytoplankton diversity remain poorly characterized. Although temperature is considered a key driver of general marine biodiversity, its specific role in phytoplankton diversity has remained unclear. We determined monthly phytoplankton species richness by using niche modeling and >540,000 global phytoplankton observations to predict biogeographic patterns of 536 phytoplankton species. Consistent with metabolic theory, phytoplankton richness in the tropics is about three times that in higher latitudes, with temperature being the most important driver. However, below 19°C, richness is lower than expected, with ~8°– 14°C waters (~35° to 60° latitude) showing the greatest divergence from theoretical predictions. Regions of reduced richness are characterized by maximal species turnover and environmental variability, suggesting that the latter reduces species richness directly, or through enhancing competitive exclusion. The nonmonotonic relationship between phytoplankton richness and temperature suggests unanticipated complexity in responses of marine biodiversity to ocean warming.