Using assembled bio-geo-environmental data for a common study area – the Vaud Alps - the SNF EcoGeoIntegralp project aimed at further understanding and predicting the geographic distribution of four main components – vegetation, soils, geomorphology and hydrology – and their interelations as inputs to the spatial assessment of two ecosystem services: water provision and scenic value of the landscape (Fig. 1).
Figure 1: The initial organisation of the EcoGeoIntegralp project, with the six modules, and their interactions.
Successful research was conducted and published on all project dimensions, most planned outputs were produced, and the two project-funded PhD students defended their PhD successfully within the time frame of the project (E. Giaccone, J. Thornton). The third, UNIL-funded, PhD student in the soil module also completed her PhD successfully in July 2019 (A. Buri). However, the main postdoc and coordinator of the project (C. Cianfrani) had a 5-months maternity leave during the project, and then left earlier for a permanent position, which limited some developments of the project, such as the 3D simulations of landscape scenic value, which also proved more difficult than expected, but related results could still be obtained with other aspects of landscape’s cultural values as ecosystem services in module 6. Two UNIL-funded external postdocs from the Guisan group performed some specific tasks such as forest tree modelling (D. Scherrer) and took over two unfinished studies at the end of the project (temporal soil-vegetation survey and snow/ndvi trends in the Alps; S. Rumpf). Another UNIL-funded postdoc, Daniel Scherrer, conducted all forest modelling in the project. Overall, all project collaborators did an excellent work. Also, the second workshop with stakeholders could not be organized as planned by the end of the project (May 2020) due to the covid situation, and was aimed instead for Fall 2020, but was again postponed due to covid. It should take place as soon as the situation will allow a presential meeting, in Spring or summer 2021, again in Château d’Oex, and the final outputs of EcoGeoIntegralp will be presented. Hereafter, we use the project’s module structure to report on the main results and outputs obtained during the three to four years of research (depending on the collaborators involved), from M1 to M6.
M1- GeoDataHub
The GeoDataHub module gathered the geo- and remote sensing data necessary to the project and stored them on a common NAS server accessible by all project members. Worldview 3 satellite image at 1 m resolution were acquired for the whole Vaud Alps study area (Boserup 2018) and some more specifically for the focal area Vallon de Nant. For higher-resolution data, a series of optical and thermal unmanned aerial vehicle (UAV) surveys were carried out in the Vallon de Nant, aimed at identifying interactions between groundwater and surface waters (Vallat 2017). Additionally, codes were developed to access spatially- and temporally-integrated Landsat satellite images using innovative approaches within Google Earth Engine, which allowed generating several new environmental maps for the study area, such as of snow and vegetation indices (Rumpf et al. 2022., Panchard et al. In review) for use in further analyses (task 5.1). The previously sampled soil data were also generalized in space for use in plant models in M2 (Buri et al. 2017, Cianfrani et al. 2018, Cianfrani et al. 2019, Buri et al. 2020)(tasks 5.2 and 5.3). The same spatialization was performed for geomorphological data (Giaccone et al. In prep.-a, Giaccone et al. In prep.-b) and new hydrological maps can be produced for the Nant Valley by the hydrological model in M5.
M2- GeoVegetation
The GeoVeg module produced an integrated review of trends and factors affecting biodiversity and ecosystems in mountain areas under climate and landuse changes (task 2.1), which also contributed to the 1st assessment of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES) for Europe and Central Asia (IPBES 2018, Guisan et al. 2019a). A contribution was also made to biodiversity modelling standards (Araujo et al. 2019) and implemented in a spatial modelling tool (Di Cola et al. 2017). Next, the improvement brought by adding new soil predictors (Buri et al. 2017, Cianfrani et al. 2019, Buri et al. 2020) and new snow predictors (Boserup 2018, Panchard et al. In review) in plant species distribution models (SDMs) was successfully evidenced (task 2.2a), and the influence of geomorphological variables on plant richness and vegetation cover was additionally shown (Giaccone et al. 2019). The role of climate, and the effects of climate scenarios, to predict plants and other above- and below-ground organisms was also assessed, and showed the lower importance of climate to predict below-ground organisms (Mod et al. 2020), except for protist richness (Seppey et al. 2019). A spatial model for the most impactful invasive plant in the study area – Heracleum mantegazzianum S&L – was also finalized and used to discuss socio-economic implications (Shackleton et al. 2020). Some aspects of the plant community modelling approaches developed in previous projects were also finalized in this project, especially regarding the role of top-down macroecological constraints (Mateo et al. 2017), evaluation of community predictions (Scherrer et al. 2018, Scherrer et al. 2020a) and the spatial mismatch in trait and niche characteristics used to assemble species into communities (Guisan et al. 2019b). To assess our capacity to predict in time, two temporal surveys were conducted in the study area by revisiting old vegetation plots, one in forests (Scherrer et al. 2017) and one in grasslands (Rumpf et al. Submitted), providing baselines for future changes. As an input to the geo-hydrology model in M5, tree species distribution models were more specifically built using both correlative (SDM) and mechanistic (TREEMIG) approaches, their capacity to predict future distributions was compared, especially at the upper tree limit (Scherrer et al. 2020b), and the spatial predictions transferred to M5. Finally, plant SDMs were used as input together with Ecosystem Service (ES) maps for spatial conservation planning prioritization in M6 (Vincent et al. 2019, Ramel et al. 2020). The esthetic value of the landscape (task 2.3) proved difficult to realize due to the loss of some resources (maternity leave and earlier departure of the postdoc) and failure, without dedicated budget, to find external partners on the needed 3D simulations (typically working with game development companies), but other approaches are currently under consideration or development, e.g. based on hikers’ walking path utilization rates using connected watches (Rey et al. in prep.) or through social surveys (ongoing in the ValPar.ch project).
M3- GeoSoil
The GeoSoil module synthesized the available knowledge on spatializing soil data in a review paper (task 3.1), which showed that predictive and hybrid (predictive+geostatistical) approaches proved better than purely geostatistical ones, but also that maps including a geostatistical component could not be reliably projected in the future (Cianfrani et al. 2018). Accordingly, data and a predictive modelling pipeline were developed to spatialize soil properties (task 3.2), and the resulting maps were then used to feed plant SDMs in M2 (Buri et al. 2017, Cianfrani et al. 2019, Buri et al. 2020). It showed that pH is both the most important soil predictor for plants but also the soil characteristic (among >40 tested) mapped with the greatest accuracy (Buri et al. 2017, Buri et al. 2020). As pH is also the most important predictor for soil bacteria, the map developed here could also be used in a modelling study of bacteria distribution (Mod et al. In review). Interestingly, pH was also the variable to change most in the temporal study comparing soil and vegetation in resurveyed plots in a > 40-years time interval (Rumpf et al. Submitted), which could serve as a baseline (together with changes in total organic carbon) to define simple soil change future scenarios as input for the prediction of future bacteria distribution under climate changes (Mod et al. In review). Fine-scale modelling of soil properties (task 3.3) was also conducted, with some success to model soil water holding capacity based on the set of vertical soil samples available for the focal part of the study area (Vallon de Nant)(Cianfrani et al. 2019), generalized to the whole Vaud Alps and introduced in improved plant SDMs in M2.
M4- GeoMorphology
The GeoMorpho module accomplished the two main objectives presented in the initial project: 1) it investigated the link between vegetation and geomorphic parameters in three focus sites in the Vaud Alps (task 4.1) and 2) it provided spatially-distributed geomorphological data for improving vegetation models, in particular producing grain size maps and geomorphological maps for significant parts of the study area and a high-resolution permafrost map for the entire study area (tasks 4.2 and 4.3), following the methodology of Deluigi et al. (2017) in the same research group. The investigation of the influence of microclimate and geomorphological factors on vegetation development was based on fieldwork carried out between 2016 and 2019 in the Vallon de Nant. Data about vegetation, ground surface temperature, permafrost occurrence and earth surface processes were collected at around 80 plots. The results show that landform morphodynamics is a key factor, together with growing degree days, to explain alpine plant distribution and community composition (Giaccone et al. 2019). In parallel, a method for the measurement of grain size from UAV-based images was developed. Different algorithms were tested and finally the Basegrain approach was retained (Giaccone et al. In prep.-a). Next, two approaches were used to develop semi-automated geomorphological mapping (SAGM). The first one is the Direct Sampling method, from the multiple point geostatistics family, whereas the second one is the Random Forest, a machine learning technique. Both methods provided encouraging results with slight differences (Giaccone et al. In prep.-b). From this, a geomorphological map of the Vaud Alps was recently produced, but its power to predict plant distribution remains to be tested.
M5 - GeoHydrology
The GeoHydro module produced the model initially planned (tasks 5.1 to 5.3). This model sought to evaluate the utility of one of the most advanced fully-integrated surface subsurface flow codes for simulating hydrological dynamics in steep, snow-dominated, and geologically complex Alpine headwaters, under both present and plausible future climate, and accounting for forest and permafrost conditions. The model was built in three phases. In the first, initial spatial data was gathered to build the model (task 5.1) and additional data was included when available from the other modules (e.g. snow, permafrost, forest scenarios). Given the geological complexity of the study area, the research plan was revised to include an additional task: the development of a 3D model of bedrock geology. This work demonstrates that 3D geological models with appropriate characteristics for hydrogeological applications can be developed in even the most complex settings, and that the lack of such data (at least) should not form an impediment to progressing beyond simple conceptual hydrological models (Thornton et al. 2018). In the second phase, given the importance of complex snow processes to the hydrological functioning of such regions, a novel, code independent, and high resolution (hourly, 25 m) snow simulation, optimisation, and uncertainty framework was proposed (Thornton et al. In revision). Being energy balance-based and additionally accounting for gravitational redistribution, the snow modelling approach extends well beyond that taken in many hydrological models – including otherwise advanced fully-integrated ones – which still mostly rely on index-based snowmelt modelling approaches whose ability to realistically reproduce snow dynamics in complex Alpine terrain is questionable. Two complementary types of snow observations – namely snow extent maps and snow water equivalent time-series – contributed to the estimation of several important but uncertain parameters (Thornton et al. In revision). The results of that model then informed the third phase: the development and calibration of fully-integrated surface-subsurface model was developed using the code HydroGeoSphere (HGS)(Thornton et al. In prep.-b). Streamflow was reproduced at the main gauging station over an independent 11-month evaluation period with a Nash-Sutcliffe Efficiency coefficient of 0.75. The main seasonal signal of the observed groundwater levels could also be broadly replicated, although capturing the observed differences between sites remained elusive, probably due to local scale variability in hydraulic properties. Simulated spatio-temporal patterns of several other important hydrological variables were also visualised to illustrate the model’s coherence and the capabilities of such an approach (Thornton et al. In prep.-b). Finally, in an attempt to assess the potential magnitude of future hydrological change in such regions and unravel its dominant drivers, the model chain was forced with climate, vegetation, and permafrost scenarios that could be expected under “moderate” warming by approximately the year 2075 (Thornton et al. In prep.-a). Direct climatic changes were found to dominate, but increased evapotranspiration due to more extensive forests were predicted to reinforce declining annual streamflows.
M6 – Ecosystem Services assessment
The first ecosystem service (ES) on water provision was assessed using the model developed in M5 (Thornton et al. In prep.-b) and accounting for the geomorphological (e.g. permafrost, snow) and forest tree distribution maps developed in M2 and M4, and as noted above, showed that whilst climate changes are expected to dominate changes in water provision in the medium term, the hitherto rarely assessed impact of contemporaneous forest change are not negligible (Thornton et al. In prep.-a). In contrast, due to the extremely limited present-day permafrost distribution, the hydrological impacts of simulated complete thaw were barely discernable, although this would not be the case at a higher elevation site.
For the second ES, using the Zonation spatial prioritization tool, we combined the plant distribution models with predictions for other taxonomic groups (insects, amphibians and reptiles) to assess present and future spatial conservation priorities in the study area compared to existing ones (Vincent et al. 2019), and expanded this study to include the mapping of 10 ecosystem services (4 provisioning, 4 regulating and 2 cultural ESs) in the prioritization process and showed that putting too high weights on ES could be at the cost of lowering the protection of biodiversity (Ramel et al. 2020). As reported in M2, the landscape scenic value could not be developed as expected, but research is still going on in one of our groups concerning the cultural value of the landscape, notably in the recently started national confederation-funded ValPar project (http://www.valpar.ch) which will be able to use and acknowledge the early developments made in the EcoGeoIntegralp project.
