The Collaborative Research Centre (CRC) ECOSENSE will quantify, understand and predict forest ecosystem processes in the context of climate change across scales, spatially from root/leaf to forest stands and temporally from seconds to years, capturing fluxes from the hydrosphere, pedosphere, biosphere and atmosphere (Fig. 1).

Key challenges of forest ecosystems are their heterogeneity and unknown non-linear upscaling laws of processes from the leaf to a whole forest ecosystem. ECOSENSE aims to derive these upscaling laws, which requires sensing, analysis and modelling of the heterogeneity with a unique spatio-temporal resolution.

We will enable this holistic approach by new dedicated robust sensing mechanisms and their implementation into a novel and smart sensor network. Our multi-scale models will integrate data across scales, derive and establish new scaling laws, optimise sensing and predict impacts of climate extremes.

ECOSENSE brings together experts from environmental sciences and microsystems engineering. Both research fields are especially strong at the University of Freiburg and unique in this combination. Joining their forces fits excellently into the vision of the University, the green city of Freiburg and the region of the Black Forest. The Faculty for Environmental Sciences and Natural Resources – UNR – embraces a 100-year-tradition of forest science and stands for a broad view of ecosystem research across all scales. The Faculty of Engineering was founded in 1995 and is growing rapidly. With the Department of Microsystems Engineering – IMTEK – the Faculty of Engineering comprises the largest microsystems research facility in Germany.

Our research proposal targets forest ecosystems, as they provide important climate regulatory functions, sequestering roughly 20-25% of global anthropogenic CO₂ emissions each year (Le Quéré et al. 2018, Pugh et al. 2019). Global climate change endangers the sustainability of forest ecosystems with unpredictable and potentially cascading effects on complex physiological processes and interactions within ecosystems (Anderegg et al. 2018). For example, the 2018 heat wave over Europe induced canopy die-back and hydraulic failure in many forests (Schuldt et al. 2020, Walthert et al. 2021). In contrast to managed agricultural systems, response and regeneration times of forest ecosystems are much longer (Williams et al. 2014). Legacy effects, for example, altered response after previous stress (McDowell et al. 2020), are still poorly understood. Identifying tipping points and thresholds of forest functioning will be crucial to ensure effective conservation and mitigation actions. Thus, there is an urgent need to better quantify and predict ecosystem processes in response to climate extremes and characterise the key processes determining forest resilience.

Multiple interactions of biotic and abiotic drivers and complex feedback loops foster spatio-temporal dynamics and heterogeneity in forests from leaf to ecosystem. While it is known that climatic extreme events, such as heat waves, droughts, frosts or beetle attacks, induce stress effects, the observed strong local variability, for example, single tree die-back (Schuldt et al. 2020), is not well understood. Sites/events with a disproportional impact on net carbon and water fluxes, i.e. “hot spots” and “hot moments” (McClain et al. 2003), can have long-lasting impacts on ecosystem functioning, in both a positive or negative manner. For example, over 10% of the annual carbon uptake of diverse ecosystems was observed to take place in the 2% most productive hours (Kannenberg et al. 2020). Inversely, a significant amount of carbon can be lost during heavy rainfalls after dry periods, so-called Birch effects (Unger et al. 2010, Waring and Powers 2016). Moreover, the role of structural and functional heterogeneity of the forest is yet to be explored. Currently, only a few highly equipped sites exist in Germany (e.g. TERENO sites), which are mostly in flat terrain and do not well represent the complexity of our forest ecosystems.

Consequently, ecosystem research faces great challenges to capture relevant processes and their interactions. ECOSENSE will tackle these challenges. We will target the spatio-temporal dynamics of carbon and water pools and fluxes and relate them to the heterogeneous structure of a forest, from the soil to the canopy and the atmosphere. We will use, on the one hand, chlorophyll fluorescence as a sensitive stress indicator (Werner et al. 2002) for hot spots and hot moments, for example, during extreme heat and drought. On the other hand, photosynthetic isotope discrimination and analysis of Volatile Organic Compounds (VOC) will provide insights into plant regulatory responses (Haberstroh et al. 2019, Kreuzwieser et al. 2021), such as water-use efficiency (Werner et al. 2012) and induced stress responses (e.g. after bark beetle attacks).

A thorough quantification is labour-intensive and requires extensive equipment, logistics and computing capacity, which currently presents a major constraint to understand highly dynamic ecosystem processes. This urgent need can be met by novel smart sensor systems tailored to the requirements of remote forest areas (Gabrys 2020). In recent years, smart sensor systems have undergone tremendous development and now provide a key technology for all aspects of automation and Internet of Things (IoT). Miniaturised sensors, communication protocols (Mekki et al. 2019), energy harvesting (Cepnik et al. 2013) and low-power electronics (Wang et al. 2017) have been explored for years for a huge variety of application scenarios in industry (Talari et al. 2017), healthcare (Neves et al. 2008) and smart homes (Stoppa and Chiolerio 2014). Recently, also the integration of deep-learning (DL) for efficient sensor application has come into focus (Guo et al. 2019, Sepasgozar et al. 2020).

However, especially in the harsh and remote environment of a forest, the sensors must be robust and functional over long time spans at least for a vegetation period and preferably for many years. Many opportunities for applying environmental sensing exist and equipment is partly commercially available; however, these devices are typically bulky, heavy, maintenance-intensive, invasive and expensive. Consequently, due to limited resources, the number of sensors deployed in forest ecosystems is low, resulting in sparse spatio-temporal data coverage. This, in turn, leads to huge data gaps for detailed understanding and modelling of complex, multi-scale ecosystem processes. Furthermore, existing solutions lack mobility and are not easily transferable amongst sites or ecosystems. Finally, we still lack miniaturised energy-autonomous sensors suitable for the also energetically harsh conditions present in a forest and equipped with “smartness” to cope with it.

Future sensors for forest ecosystem research must be affordable and easily deployable in large numbers and provide the capability of being widely distributed. They need to be small, minimally invasive, intelligent, energy-autonomous and wirelessly connected in a smart sensor network. Furthermore, they must be robust in terms of temperature, humidity, dirt or biofouling and, finally, if small sensors are lost, they need to be located to retrieve them from the environment. A multitude of such distributed sensors will provide a plethora of data for which deep-learning-based methods enable the deconvolution of the underlying processes driving spatio-temporal heterogeneity. This will then allow the implementation of these processes into existing ecosystem models, which currently address structural and functional heterogeneity based on widely simplified assumptions. The new information about such spatial heterogeneity will, on the one hand, enable the refinement of the model structure and, on the other hand, the optimisation of sensor distribution and activity.

To overcome our: (i) restricted view on processes driving ecosystem functioning, (ii) limitation in suitable sensors and smart network technology and (iii) limitation in models representing spatio-temporal dynamics, ECOSENSE proposes a fully interdisciplinary approach, bridging science and technology across all relevant scales in a next-generation ecosystem research assessment. We will integrate measurements of the hydrosphere, pedosphere, biosphere and atmosphere within one smart, distributed, autonomous and intelligent ecosystem assessment system. The vast amount of data will be integrated in real-time into deep-learning and process-based modelling approaches that will be jointly used for now- and forecasting of carbon and water fluxes at leaf to ecosystem scale.

We hypothesise that only a novel concept of ecosystem analysis, seamlessly combining continuous observations and simulations of each relevant component across scales, will enable the identification of hierarchies and scale coherence of the driving abiotic and physiological processes and environmental triggers. Such a novel concept of integrated sensors, remote sensing, real-time data assimilation and model-based ecosystem analysis will enable us to identify thresholds of ecosystem functioning in response to global climate change and predict future developments.

Forests provide important ecosystem services due to their high ecological and socio-economic value, but climate extremes have enhanced tree mortality worldwide (Anderegg et al. 2019). Therefore, increasing the resilience and resistance of our forests to climate change is a major challenge. However, strong local differences of the stress impacts on our forest systems during extreme climatic events, such as local tree die-back during the 2018 drought in Germany (Schuldt et al. 2020), are not yet understood, but point towards the complex interplay between pedosphere (soils), hydrology, atmosphere and biosphere. These processes can amplify or buffer stress impacts on ecosystems through a cascade of biotic (e.g. competition and facilitation, Haberstroh et al. (2021)) and abiotic factors (e.g. soil moisture, air temperature and humidity). However, knowledge on the underlying mechanisms enabling informed predictions on a spatially resolved scale is lacking.

It has been recognised that structural heterogeneity in forests can be an important factor for ecosystem stability and resilience against disturbances, such as storms (Bauhus et al. 2017, Ratcliffe et al. 2017). Generally, positive relationships have been found amongst species richness, ecosystem functioning and resilience to stress (Dı́az and Cabido 2001, Grossiord 2020). More recently, there is also increasing evidence that spatial heterogeneity in communities may confer important aspects of functional stability, promoting facilitative interactions amongst species. For example, it has been shown that the sensitivity to changes in climate is dependent not only on the intensity of stress and soil type, but also, importantly, on the dynamics of individual-level competition within plant canopies (Aguirre‐Gutiérrez et al. 2019), microclimatic conditions within the forest (Zellweger et al. 2020) and their interplay (Werner et al. 2021). Thus, these interactions can modulate ecosystem sensitivity, which is generally not incorporated in traditional models (Levine et al. 2016). Moreover, spatial heterogeneity is increasingly recognised as an important driver at the macro- and micro-scale. In soils, probably the most heterogeneous parts of the biosphere, it has been shown that hot spots and hot moments are important drivers of soil processes (Kuzyakov and Gavrichkova 2010, Miguez-Macho and Fan 2021). Even though the importance of these processes has been identified, we lack affordable and operational non-invasive measurement techniques to cover the relevant scales in complex forest ecosystems.

Essential processes in forests driving evapotranspiration and the carbon sink/source capacity are soil-plant-atmosphere CO₂ and H₂O exchange, which are thereby relevant to buffering or accelerating climate change (Bonan 2008). Moreover, forest ecosystems influence global climate and atmospheric chemistry by the release of volatile organic compounds (VOCs) into the atmosphere (Sindelarova et al. 2014), by impacting atmospheric chemistry and secondary organic aerosol formation (Fuentes et al. 2000). Typically, the scale of process function differs from the scale of process impact. Total ecosystem fluxes are not simply the sum of local processes that can substantially vary at small scale (e.g. soil microbe, fine root or leaf scale). However, process description is highly simplified in models: in extreme cases, single-leaf measurements are scaled up through big-leaf models to the ecosystem scale (assuming that the entire forest responds similarly to a single leaf (e.g. Farquhar et al. (1980)). In more sophisticated models, variability within the canopy species diversity is considered, but not the variability due to different tree sizes or the relationship between single tree canopies. In addition, also processes of plant-soil feedbacks are either neglected or not dynamically represented, specifically the interactions amongst root systems, soil organism activity and abiotic soil conditions, which strongly influence plant performance, ultimately driving ecosystem processes (Pugnaire et al. 2019). How to include processes for which we have limited process understanding is an emerging research field (Reichstein et al. 2019): deep learning can be used both as part of ecosystem models or as a bias-correcting step after the model (Zhao et al. 2019).

Obviously, to date, little is known of how these interactions quantitatively scale to ecosystem fluxes and even less about their potential changes under environmental or climate change. Hence, in spite of the increasing awareness of the relevance of small-scale processes, feedbacks and their interactions, we are currently unable to sufficiently quantify and evaluate their contribution to total ecosystem fluxes. This shortcoming is largely due to a lack of suitably-sized sensors, explicitly tailored to the measurement tasks in forests, which should be connected to a widely distributed sensor network capturing state variables and fluxes rapidly, locally and non-invasively. Moreover, we even lack knowledge on efficient sampling strategies: how often do we have to measure? How many sensors are needed at which spatial distribution? The answers will not only depend on the processes, but also on their spatio-temporal dynamics.

This is particularly relevant as hot spots and hot moments develop dynamically in space in time. A drought event or heat waves can be such a hot moment, with a disproportionate effect on net water and carbon balance (Ciais et al. 2005). Similar effects can be caused by locally restricted events, i.e. hot spots, such as local late frost injuries or beetle attacks of susceptible trees. Little is known on how hot spots and hot moments influence ecosystem fluxes as their impact can persist through time by locally altering plants' performance and resilience. These legacy effects are not well understood either, but may gain importance with increasing frequency and intensity of climatic extreme events. Neither these legacy effects nor the structural and functional heterogeneity within forests is currently represented in existing models due to missing data and incomplete process knowledge.

Therefore, large networks of distributed micro-sensors are mandatory to capture these spatio-temporal stress indicators and flux dynamics from the micro- (mm/s) to macro- (km/years) scales. They need to be combined with intelligent data acquisition and evaluation to feed, in a timely manner, deep-learning tools and process models for immediate assessment, as well as scenario analysis.

Sensor networks should be flexible in up- or down regulating measurement frequencies and require efficient handling of large data packets. Mature forest sites present a particular challenge, as canopy access is limited, energy supply and communication of sensors is out of reach of centralised grid solutions and seasons cause tremendous physical stress on in-situ technical systems. This is why tall forest stands are often restricted to point measurements at canopy towers (Pepin and Körner 2002), which are, however, unable to cover representative areas of the ecosystem.

Despite the manifold of existing IoT (Internet of Things) sensors and sensor networks, significant knowledge gaps remain for the application of wireless sensor networks in forests representing a specific kind of harsh environment. A recent review from Ullo and Sinha (Ullo and Sinha 2021) argues that, in the field of remote sensing and agriculture, there is still a research need for new robust sensing methods and sensor systems. Particularly, this argument holds for forests, which stand for other timescales and hardly accessible areas.

A robust wireless network has been already installed for animal tracking in the rain forests of Borneo (Sethi et al. 2018), however, without environmental sensors. Several forest sensor networks have been explored for the specific task of fire detection (Lloret et al. 2009) and research in this direction is ongoing, for example, considering energy efficiency (Vikram et al. 2020) or data fusion (Varela et al. 2020). However, the need for fire detection arises typically in dry and hot regions. For a comprehensive assessment of Central European ecosystems as planned in ECOSENSE, other research tasks emerge, such as changes in ecosystem competition and facilitation under cold or heat spells, droughts or floods and other disturbing events. Given that large numbers of miniaturised sensor nodes need to be deployed, they should not interfere with the ecosystem, i.e. with leaves, stems, roots and water and gas fluxes. For example, leaf sensors should avoid shading, be lightweight, enable natural gas exchange and be energy-autonomous, as cables and batteries are inefficient, expensive and maintenance-intensive. It is still an open question which existing sensing principle can be miniaturised with sufficient sensitivity and selectivity. Whenever miniaturisation is not possible, new sensing mechanisms are mandatory and need to be explored. Furthermore, it is not yet known how the sensor size, number and way of deployment will guarantee an undisturbed metabolic function of ecosystem components. Additionally, the robustness of sensors is essential in terms of temperature cycles, UV exposure and fouling processes. How can we guarantee to maintain sensitivity over long periods, in soil or air? Furthermore, the overall energy demand is completely uncertain. We do not know whether distributed solar energy harvesting in the shaded forest or thermoelectric energy harvesting in the relatively stable temperature conditions in soils, will be sufficient. We even do not know the meaning of “sufficient” for our network regarding the frequency of sensing with regard to the actual energy supply condition. This is superposed by the variable need for data under rapidly changing environmental conditions and the avoidance of redundant data accumulation under stable conditions.

Finally, a data- and energy-efficient operation of the network can only be achieved by a continuous interplay and feedback between technical advances and scientific process descriptions and modelling of ecosystem functions.

So far, ecosystem sensing and modelling are not well connected since data availability has time lags of months to years, particularly for classical monitoring programmes, such as Fluxnet and TERENO. Thus, ECOSENSE is breaking new ground in: i) sensing, ii) real-time data assimilation and modelling and iii) combining both for now- and forecasting.

The central aim of ECOSENSE is to enable a novel dimension of ecosystem process sensing, understanding and modelling. We consider this fundamental to predict sustainable forest functioning, this being the most important ecosystem service that buffers climate change with high socio-economic value in Germany and elsewhere.

We argue that a holistic understanding of ecosystem functioning can only be achieved with in-depth understanding of spatial and temporal dynamics of relevant ecosystem processes across scales.

Heterogeneity and functional diversity of complex forest systems and their interactions and feedbacks are key factors able to buffer or accelerate stress impacts after extreme events.

We identify four topics, which are currently not appropriately addressed, but are of utmost importance to quantify and predict complex ecosystem processes:

● scaling of key processes considering their spatio-temporal heterogeneity and dynamics, hot spots and hot moments and their contributions to ecosystem fluxes;

● developing sensing technology and efficient sensor network communication for providing sufficient spatio-temporal resolution of relevant processes;

● designing size and gridding of sensor networks to capture important dynamics of ecosystem processes from leaf to ecosystem scale;

● big data processing, analysis and assimilation and multi-scale ecosystem modelling considering non-linear scaling laws for improved ecosystem analysis and prediction.

Accordingly, ECOSENSE will address four main research aims as illustrated in Fig. 2.

Finally, the overarching long-term outcome of ECOSENSE will be a scientific framework of hardware, methods and workflows to transfer ecosystem sensing and modelling into other (forested) ecosystems. This Forest ECOSENSE Toolkit (Fig. 3) builds on:

● a transferable, verified multi-scale ecosystem model extended for spatio-temporal dynamics and scaling;

● an in-depth scientific understanding of the ecosystems studied in ECOSENSE;

● a modular and energy-autonomous sensing system, which will allow rapid deployment in remote areas to evaluate ecosystem processes affected by climate extremes.

These aims will be approached step-wise over 12 years throughout the three funding phases of the CRC.

To achieve our aims, expert scientists of this interdisciplinary CRC will implement a comprehensive research programme, which approaches the challenges in close, mutual cooperation and from two central aspects: from a technical and from an ecological point of view.

To this end, we will explore, develop and implement an intelligent distributed micro-sensor network autonomously measuring fluxes from soil to vegetation and atmosphere. Consequently, we will address the miniaturisation of established sensing principles for enhanced spatio-temporal resolution and drive the research and implementation of new sensing principles in ecosystem monitoring.

We will assess relevant ecosystem processes covering scales from root/leaf to forest stands and, temporally, from seconds to years, capturing fluxes from the hydrosphere, pedosphere, biosphere and atmosphere. In the first phase, ECOSENSE will focus on water and carbon fluxes, namely H₂O, CO₂ and VOCs and their carbon isotopes, as the most prominent drivers and chlorophyll fluorescence as a well-established stress marker. These data will be evaluated in the light of detailed information of forest structure and microclimate, tree and soil nutrition and soil physical properties, which will be assessed by a thorough characterisation of the ECOSENSE Forest.

We will start with a limited number of commercially available sensors, to initiate ECOSENSE’s systems research right from the start. They will be substituted step-by-step with the new distributed sensors and sensor systems once they are ready for field deployment. Sensing and measuring principal and key tree responses to extreme environments will be first tested in the newly-founded “Ecosphere Experimental Platform (EcoExPo)” of the UNR on the campus of the Faculty of Engineering. Subsequently, they will be transferred to our joint field site, the ECOSENSE Forest, in the state forest of Ettenheim near Freiburg. The ECOSENSE Forest represents a well-grown mixed needle and broadleaf forest at the foothills of the Black Forest. Given the complexity and challenges, both in developing the intelligent sensor network and the deconvolution of ecological processes across all relevant scales, we will start this CRC, as stated, focusing on carbon and water fluxes, while taking forest structure, soil characteristics and nutrient contents into account. Moreover, our vision is that the system can be expanded to include further factors/elements such as nutrients fluxes, other trace gases and biotic interactions in the future.

These research activities are jointly implemented into the ECOSENSE research network comprising three Research Areas as shown in Fig. 4:

Area A: the assessment of H₂O, CO₂ and VOC fluxes and their isotopes from soil, trees, to atmosphere;

Area B: the assessment of stress indicators through chlorophyll fluorescence from individual leaves to forest stands;

Area C: sensor network implementation, materials for environmental robustness, efficient data management, deep-learning and modelling.

The outstanding characteristic of ECOSENSE is the consequent entanglement of both disciplines in almost every project. Therefore, the CRC has a unique structure: The three Research Areas are divided into ten projects (Fig. 4), of which the projects in Area A and Area B each comprise subprojects from the technological and from the ecological side, while Area C addresses overarching research tasks. This results in a total amount of 18 scientific subprojects and includes 23 doctoral and postdoctoral funded researchers, as outlined below. The structure reflects the close cooperation, which ensures a highly dynamic knowledge exchange, joint development of novel approaches and strongly fosters cross-disciplinary team building. It will be further enforced by the joint experimental platforms (EcoExPo and ECOSENSE Forest, Z1), a common ECOSENSE research training group (RTG) and cross-disciplinary workshops and communication (Z2).

In the following, we will provide an overview of the Areas and projects in a nutshell. Fig. 5 shows the circle of interacting and data exchanging projects and illustrates the interlinkage of the technical (blue) and ecological (green) subprojects spanning from pedosphere through biosphere to atmosphere. Thereby, Area A focuses on water and carbon pools and fluxes, Area B on stress indicators (Chlorophyll Fluorescence, ChlF). The fluxes of VOC measured in A3 and A4 are also stress indicators and thus reach into Area B. Area C covers cross-spanning projects, as all sensors will require low power electronics and communication from C1 and environmental robustness from C2. Finally, C3-INF collects all data from Areas A and B and the interdependence of Area A and Area B data is analysed and modelled in C4.

Research Area A - H₂O, CO₂ and VOC fluxes along the different ecosystem compartments and scales

Underlying plant and soil processes such as photosynthesis, respiration, ecohydrological H₂O fluxes, as well as soil-plant and plant-atmosphere interactions, will be emphasised (Fig. 5). Process-based understanding of pools and H₂O, CO₂ and VOC fluxes will allow CO₂-flux to be partitioned into: (i) ecosystem respiration (Re) and photosynthesis (gross primary production, GPP), (ii) the contribution of different functional units within a plant (leaves, stem, roots) and soils (heterotrophic and autotrophic respiration); and (iii) above-ground and below-ground processes and evaluate hot spots and hot moments within the mixed forest. Furthermore, Research Area A aims at the partitioning of VOC fluxes within the canopy, between species and its significance for the ecosystem VOC flux.

A1 will assess the spatio-temporal heterogeneity in soil fluxes and microclimatic/edaphic conditions to evaluate the importance of rhizosphere and microbial sources and identify hot spots and hot moments in microbial and tree root functioning and their interactions (A1.1). To this end, we will develop novel soil probes measuring CO₂ and temperature, which can be installed with high spatial coverage in a measuring grid. They will be energy-autonomous through thermo-electric and solar energy harvesting and, thus, perfectly fulfil a “deploy-and forget” strategy (A1.2).

A2 addresses ecohydrological fluxes and processes in heterogeneous forest patches investigating dynamics and spatial patterns in root water uptake, tree sap flow and phloem carbon isotopes, its feedbacks on spatio-temporal soil moisture variability and heterogeneity and how this, in turn, affects tree water use efficiency and phloem sugar transport. We aim to trace water and carbon fluxes in trees and sharpen our picture of biotic controls (A2.1). Here, we will develop a fully novel methodology through compact Magnetic Resonance Imaging (MRI)/Nuclear Magnetic Resonance (NMR) sensors, based on permanent magnets attached to small-sized branches. They will allow for in situ imaging of the H₂O fluxes without interfering with the branch and NMR analysis of xylem and phloem sap flows (A2.2). Continuous in situ phloem sap NMR measurements will allow a new dimension of quantifying integrated carbon transport in trees.

The goal of A3 is to capture spatial and temporal dynamics in photosynthetic carbon isotope discrimination, stomatal conductance and leaf H₂O and VOC fluxes within and amongst tree crowns in a mixed forest ecosystem. Environmental impact on photosynthetic CO₂ fixation can be decoded by ¹³C-isotope discrimination, determining hot spots and hot moments of photosynthetic C fixation, C respiration and VOC emissions (A3.1). Micro gas cuvettes will be developed and deployed in large numbers to monitor the 3D-variability within the canopy. Equipped with an integrated open-close mechanism, they will be connected to multiple small, low-cost carbon isotope laser spectrometers, which will also measure leaf H₂O fluxes. As the laser spectrometers cannot be miniaturised to a similar extent as the leaf cuvettes, they will be placed at a central position and connected by tubing (A3.2).

Finally, to bridge the gap between the relatively small scale of an individual tree and a forest stand, A4 will assess ecosystem-atmosphere exchange by eddy covariance measurements of H₂O, CO₂ and its isoflux (δ¹³CO₂), to separate fluxes into components (ecosystem respiration and photosynthesis (gross primary production, GPP) at an integrated scale. Further VOC uptake and release by our forests will be measured, linking to important ecosystem functions, which are strongly sensitive to environmental changes (A4.1). Here, we will develop, for the first time, an optical spectroscopic sensing technology, to measure VOCs using tunable laser absorption spectroscopy (TLAS) along a concentration gradient at the tower (A4.1b) and connected to single-leaf cuvettes, as developed in A3.2.

Research Area B - Active chlorophyll fluorescence measurements as a sensitive stress parameter addressing the relevance of microclimatic heterogeneity

In Research Area B, we aim for a full picture of chlorophyll fluorescence (ChlF), as a sensitive parameter to trace effects on environmental changes and biotic and abiotic stresses on the photosynthetic efficiency, ranging from leaf-based networks up to canopy and forest stand. This Research Area will identify hot spots and hot moments of the heterogeneous 3D structure of the forest canopy and enable us to relate the underlying process to small-scale microclimatic changes, for example, incident light, temperature and humidity at the leaf surface, as well as to macro–scale stress events.

B1 will capture heterogeneity in leaf-level ChlF and associated microclimate in a novel dimension of a spatial network within and amongst tree canopies, to efficiently identify hot spots and hot moments of stress-induced changes of photosynthetic efficiency (B1.1). To this end, we will develop novel wireless, energy-autonomous ChlF sensors, using flexible, multifunctional (microclimate) and highly integrated microsensor probes (B1.2). A new kind of leaf sensor (< 1 cm²) to be deployed, covering multiple positions within the canopies, will strongly profit from the advantages of microsystem technology: they will have the highest degree of miniaturisation, to ensure the least interference with leaves and they will need to be mass produced. These sensors will act as independent sensor nodes, due to solar energy harvesting enabling them to communicate their data wirelessly and can be installed at the beginning of the season on the leaves in a deploy-and-forget strategy.

To bridge the gap from single leaves and trees to even larger forest stands, B2 will use remote assessments of active ChlF via an unmanned aerial vehicle (UAV) to generate detailed repeated spatial maps to visualise and assess tree vitality, identify hot spots within the forests and transfer the sample based findings into area wide monitoring (B2.1). In doing so, we will further develop the models for chlorophyll fluorescence assessment, based on UAV-carried remote sensing devices, calibrated against the tree leaf measurements. We will develop a strongly miniaturised, highly integrated lightweight UAV-based active ChlF sensing instrument using LiDAR techniques with an efficient new scanning paradigm. It will allow for largely extended flight campaigns with high spatial resolution (B2.2). Thereby, the calibration of the UAV-borne data will directly profit from the leaf sensor network of B1.

Research Area C - Intelligent sensor network, robustness, data management, ecosystem model and deep-learning

This Research Area builds up the full chain from data acquisition, through an intelligent, energy-autonomous and robust sensor network, over comprehensive and quality-assured database to process-based and deep-learning model assessment and forecasting. The key challenges addressed are: (i) to develop sensors operating and communicating reliably under field conditions, with fluctuating input of ambient energy and exposure to weathering and biofouling; (ii) a real-time collation of sensor data into a central database in order to make it accessible to our new process-based and deep-learning models.

C1 will perform research on specific needs for all distributed sensors (from Areas A and B) creating a number of innovations for embedded autonomous sensor systems. All sensors will be equipped with ultra-low power electronics and communication capabilities (C1.1) to go beyond their pure sensing function and to enable them to be used as active and intelligent sensor nodes in our ECOSENSE network. Self-test features and energy awareness will be incorporated to improve a sensor node’s performance over time. Wireless communication between individual sensor nodes and central sensor hubs will be provided in multiple ways (C1.2), by using newly-established communication networks (e.g. narrow-band Internet-of-Things or NB-IoT), as well as for specially tailored low-power local radio (e.g. LoRa). The energy supply of such widely distributed sensor nodes has to happen either via energy harvesting from thermal, solar or mechanical ambient energy or via passive remote sensing approaches similar to today’s RFID technology. Our long-term goal is to achieve a maintenance free, reliably working network.

C2 will investigate how sensor nodes, being in direct contact to the environment as soil probes and leaf sensors, can be designed to be long-lasting and stable by preventing fouling processes, such as the accumulation of dirt or bacterial growth. This requires tailored surface structures and anti-fouling coatings, whose specification strongly depends on sensor location. In soil, they are exposed to long wet phases, while in air, they are exposed to large temperature and humidity variations or simply to the accumulation of dirt. Furthermore, trade-offs have to be found between, for example, being insect and bacteria repellent, while at the same time, not out-gassing unwanted VOCs.

C3-INF will generate, manage, adapt and extend a joint database for all project data, securing homogenised formats, storage, selection and accessibility by all groups. As ECOSENSE relies on data interoperability within and between its projects, as well as with the scientific research community, the challenge lies in managing easy data access using existing data formats where appropriate and creating new research data formats where necessary. To achieve this, C3-INF will overcome the burdens of data management policies by providing a central platform based on AQUARIUS (Aquatic Informatics, Vancouver) and additional tools for joint data storage, management and backup of all ECOSENSE projects. C3-INF will further provide services for data interoperability and exchange, for shared data analysis tasks (e.g. data quality control) and for managing project workflows.

C4 will develop process-based model simulations and deep-learning tools for data analysis to interact with and optimise the sensor-based monitoring, as well as to deepen our understanding of impacts of spatio-temporal heterogeneity and dynamics for total ecosystem water and carbon exchange. An existing 2D process-based model will be extended, calibrated and run in a now- and forecasting 3D mode, covering spatio-temporal heterogeneity of small-scale processes and integrating new scaling laws for non-linear interactions (C4.1). Using deep-learning algorithms, the plethora of data will be efficiently evaluated to distinguish between important and redundant data. The aim is to, on the one hand, provide sufficient spatio-temporal resolution, thus allowing for a reliable ecosystem modelling and, on the other hand, save sensor node energy and reduce redundant data accumulation (C4.2). Thereby, deep-learning and process simulations interact with the sensor network in two directions; i.e. i) data assimilation from the sensor network into the model systems and ii) adjustment and optimisation of the measuring design based on simulated outputs and predictions.

Z-projects and RTG – Common infrastructure, administration and Research Training Group (RTG)

The research of ECOSENSE is also accompanied by two Z-projects. They include our infrastructure, i.e. joint research platforms (Z1) and an administrative coordination project (Z2). Furthermore, we consider it important to offer a dedicated Research Training Group (RTG) for the doctoral researchers of our highly interdisciplinary CRC programme.

In our joint research platforms, Z1 we will take the ECOSENSE sensor network from the laboratory into a test-bed for experimental climate simulations in our new campus research facility and to the real world, our ECOSENSE Forest, a well grown mixed forest near Freiburg.

EcoExPo (Ecosphere Experimental Platform)

To test the newly-developed sensors and equipment, as well as novel ecological hypotheses, we will conduct pilot studies at the newly-founded Ecosphere Experimental Platform (EcoExPo, Eva Mayr-Stihl Umwelttechnikum) of the UNR, located directly on the Technical University Campus of the University of Freiburg. It comprises a greenhouse, outdoor experimental fields with large-scale weighing lysimeter fields, rainout shelter with different vegetation and soils, as well as an indoor experimental facility with four Ecotron units for controlled climate change experiments and simulation of environmental stresses.

The outdoor experimental platform with several large-scale weighing lysimeters and different rainout shelters will allow us to test sensors and hypotheses in different soil types and with rain exclusion or flooding experiments under natural fluctuations of environmental conditions. In order to support ECOSENSE directly and more realistically, we will extract large-scale soil monoliths for the lysimeters (2 m diameter and 1.5 m depth) directly in the ECOSENSE Forest from the three sites including young beech and Douglas fir trees (up to 8 m). These monoliths including the trees in the centre allow direct observations in the soil and trees. We will have the possibility to cover each lysimeter with a mobile Ecotron (Fig. 6) to allow specific environmental and climate conditions including different air temperatures, air humidity, atmospheric CO₂ concentrations and rainfall amounts. An Ecotron allows comprehensive investigations of several important plant and ecosystem functions, fluxes and processes under controlled environmental conditions. The four Ecotron units (with four lysimeter subunits each) will be equipped with environmental sensors (such as above- and belowground moisture, temperature), incident light (PAR), soil temperature and air temperatures and humidity as well as watering regimes can be fully controlled. Above and belowground compartments will allow continuous measurement of carbon (CO₂ and VOC) and water fluxes and their respective isotopic composition through an automatic sampling system connected to carbon and water isotope laser spectrometers (Picarro). Additionally, automatic VOC measurement will be possible by linking the PTR-TOF-MS of the Werner group during measurement campaigns to the sampling streams. Belowground compartments include smaller-scale portable lysimeters (surface area up to 1 m²), which can be moved from the greenhouse, outdoor lysimeter fields to Ecotrons or other test environments. The mobile lysimeter allows planting various trees or other vegetation to the typical soil of the ECOSENSE Forest or other soil types of interest.

ECOSENSE Forest

A well-suited forested area (Fig. 7) has been selected and permission granted by local authorities. The area is located in the municipal forest of the city of Ettenheim, approximately 45 minutes’ drive north of Freiburg and easily accessible by forest roads (Fig. 8). The forest is nested in the hilly transition between the upper Rhine Valley and the Black Forest providing a mixed forest system, dominated by beech with patches of spruce, oak, fir, Douglas fir and pine trees of different ages in mixed and pure patches. The Ettenheim forest site is a managed forest, with a low-impact sustainable management strategy. Three stands were selected for the first phase of ECOSENSE with pure beech, mixed beech-Douglas fir and pure Douglas fir trees, suitable for investigating the hypotheses on spatio-temporal heterogeneity (Fig. 9). The orographic location of the forest on an extensive plateau was selected to fulfil the requirements for an eddy covariance (EC) tower and two canopy access towers. The requirements for EC measurements are met with relatively flat or uniformly sloped terrain (< 5°). The forest provides similar roughness in the fetch of the prevailing wind directions up to 1-1.5 km. In addition, there are many forest patches on the same plateau with different composition and ages that can be used in the next phases of the project. The Mayor of Ettenheim and the local forester support ECOSENSE and will provide long-term access (12 years) and adjust management of the forest surrounding our sites in order to provide the best environment possible.

The field site is located on Triassic sedimentary rock. The plateau area, which comprises the selected stands, has relative homogeneous geological and pedological conditions of “Plattensandstein-Formation”, a sandstone with layers of clay and some smaller outcrops of the “Rotton Formation”. The soils are Cambisols with a silty loam to loamy clay texture, carbonate free and well developed to a depth between 60 to 120 cm. The field capacity is between 230 and 350 mm with average permeability and low stone content. The plateau is drained by four creeks in steep valleys in different directions. The elevation ranges between 450 to 520 m a.s.l. A recently funded DFG project (Christiane Werner, Natalie Orlowski, Markus Weiler) already established the first sensor network near the selected field site examining ecohydrological processes and tree interaction in the mixed spruce-beech forest.

Finally, Z2 ensures internal and external communication, fosters the collaboration within and between the projects, monitors and controls the timeline of the work in progress and the quality of the scientific content, supports publication processes and controls for good scientific practice.

ECOSENSE is strongly interdisciplinary and we want to encourage our young scientist members, namely doctoral and postdoctoral researchers, to cross the borders of their field and to develop a broader perspective. We are planning for an integrated research training group (RTG) which, on the one hand, uses existing doctoral researcher programmes of the University, but, on the other hand, is complemented by ECOSENSE specific measures. Thereby, we will train them in all important aspects and skills of the various disciplines and Research Areas involved in ECOSENSE and support them towards scientific independence already at an early career state, to develop their own scientific network and visibility in the international research community and to provide them with excellent career perspectives.

To wrap up the research programme, we summarise in Table 1 our research goals on a shorter, a medium and a longer perspective, i.e. on a 4 year, 8 year and 12 year timescale, representing the three proposed phases of ECOSENSE.

Starting from the second phase, we will evaluate the possibility to derive side projects for a transfer of the developed concepts and solutions into commercially available sensor products, either via the foundation of start-up companies or in cooperation with industrial partners (e.g. within the transfer opportunities of the DFG). We aim to provide the scientific community and governmental institutions with robust, long-lasting and low-cost sensors to be deployed and operated with minimal effort (“deploy and forget”) and to be designed and fabricated in an environmentally compatible manner.

The ECOSENSE project composition resulting from the research programme described above is summarised and illustrated in Fig. 10, which lists all project and subproject titles, as well as the associated PIs (in green: partners from the environmental sciences, in blue from the technological sciences). Research Areas A and B are the building blocks to assess ecosystem pools and fluxes and stress effects, while in Area C, the ecosystem model, data assimilation and deep-learning span the roof across all projects, whereas the integrated intelligent sensor network will interlink all sensor projects at all levels. The basis of ECOSENSE is the common infrastructure comprising the joint experimental platforms (EcoExPo and ECOSENSE Forest), joint data management and modelling and all Z-activities.

Most of our PIs are members of the Faculty of Engineering or Faculty of Environment and Natural Resources (University of Freiburg) located within close vicinity (< 5 km). Karlsruhe Institute of Technology (KIT) provides complementary expertise, with established long-term collaborations and associations to the University of Freiburg. ECOSENSE will further profit from the involvement of some colleagues of the Fraunhofer Institute for Physical Measurement Techniques in Freiburg, also located on the campus of the Faculty of Engineering.

Research data management is regarded as critically important to the ECOSENSE project itself, as well as the University of Freiburg. In ECOSENSE, we devote a project (C3-INF) to this topic and refer to the specific data management strategy details there. Here, we only briefly outline the underlying principles.

● Our data management follows the FAIR principles;

● Our data publication strategy is aligned with the DFG-funded NFDIs, specifically NFDI4Biodiversity (overlap-ping with GFBio). Harmonisation between our database (project C3-INF) and the NFDI4Biodiversity will be part of the work in the first funding phase;

● Data will be made openly and publicly available within 3 years of collection;

● For sensor construction, we encourage all PIs to follow Open-Source Hardware idea, i.e. releasing detailed construction descriptions for other researchers to copy under a non-commercial licence;

● At the university, research data management activities are being supported by a working group at the computing centre (https://www.rz.uni-freiburg.de/rz/aktuell/rdmg-start/?searchterm=Forschungsdatenmanagement).

ECOSENSE will be jointly managed by two spokespersons having equal rights, with one expert representing the environmental sciences (Christiane Werner) and one the engineering sciences (Ulrike Wallrabe). Together with the spokespersons of our three research areas, the embedded Research Training Group RTG and two representatives from the group of doctoral and postdoctoral researchers, they will form the Board of ECOSENSE which controls and monitors the timely progress and content flow of the project. Furthermore, the Board will be responsible to guide and enable a gender- and diversity-balanced working attitude.

All staff employed in the CRC will be ECOSENSE Members. They will elect all spokespersons of ECOSENSE and an equal opportunity officer. The ECOSENSE Members will furthermore suggest and accept new members in conjunction with the Board. Eventually, they will have the power to accept the finance plan for each year as suggested by the Board.

The doctoral and postdoctoral researchers will implement their own network, the Network of Young Researchers delegate two elected peers to the Board.

The Board will be supported by a coordinator (Nina Stobbe) responsible for finances and the scientific coordination of workshops, retreats and the international ECOSENSE conference to take place in Freiburg, Germany, in summer 2025. The Research Training Group, as well as our joint research infrastructure, will be coordinated by Julian Müller.

We are aware that an intense and open communication within the CRC, but also to the community, is essential for a high quality research output and a strong positive national and international visibility. An intense interdisciplinary cooperation and information flow is already embedded in the unique structure of ECOSENSE, as each project in Research Areas A and B will join partners from different disciplines and the overarching Research Area C will assimilate data from all projects into a common model. One key focal activity will be our joint experimental platform, most importantly our ECOSENSE Forest.

The synergy of the observations of data collected at ECOSENSE Forest will be assured not only by our coordinated data management, but also through various joint workshops, fostering extensive interdisciplinary collaboration. The Board of ECOSENSE will provide the mandatory infrastructure for knowledge exchange (internet platform, ECOSENSE homepage, common data platform) and foster the information flow within the CRC amongst participants.