Upscaling hotspots – Understanding the variability of critical land-atmosphere fluxes to strengthen climate models (Spot-On)

• Pirk, Norbert; Aalstad, Kristoffer; Holmlund, Erik Schytt; Clayer, Francois; de Wit, Heleen & Christiansen, Casper Tai [Show all 9 contributors for this article] (2024). Disaggregating the Carbon Exchange of Degrading Permafrost Peatlands Using Bayesian Deep Learning. Geophysical Research Letters. ISSN 0094-8276. 51(10). doi: 10.1029/2024GL109283.
• Pirk, Norbert; Aalstad, Kristoffer; Yilmaz, Yeliz A.; Vatne, Astrid; Popp, Andrea & Horvath, Peter [Show all 12 contributors for this article] (2023). Snow-vegetation-atmosphere interactions in alpine tundra. Biogeosciences. ISSN 1726-4170. 20(11), p. 2031–2047. doi: 10.5194/bg-20-2031-2023. Full text in Research Archive Show summary

  The interannual variability of snow cover in alpine areas is increasing, which may affect the tightly coupled cycles of carbon and water through snow–vegetation–atmosphere interactions across a range of spatio-temporal scales. To explore the role of snow cover for the land–atmosphere exchange of CO2 and water vapor in alpine tundra ecosystems, we combined 3 years (2019–2021) of continuous eddy covariance flux measurements of the net ecosystem exchange of CO2 (NEE) and evapotranspiration (ET) from the Finse site in alpine Norway (1210 m a.s.l.) with a ground-based ecosystem-type classification and satellite imagery from Sentinel-2, Landsat 8, and MODIS. While the snow conditions in 2019 and 2021 can be described as site typical, 2020 features an extreme snow accumulation associated with a strong negative phase of the Scandinavian pattern of the synoptic atmospheric circulation during spring. This extreme snow accumulation caused a 1-month delay in melt-out date, which falls in the 92nd percentile in the distribution of yearly melt-out dates in the period 2001–2021. The melt-out dates follow a consistent fine-scale spatial relationship with ecosystem types across years. Mountain and lichen heathlands melt out more heterogeneously than fens and flood plains, while late snowbeds melt out up to 1 month later than the other ecosystem types. While the summertime average normalized difference vegetation index (NDVI) was reduced considerably during the extreme-snow year 2020, it reached the same maximum as in the other years for all but one of the ecosystem types (late snowbeds), indicating that the delayed onset of vegetation growth is compensated to the same maximum productivity. Eddy covariance estimates of NEE and ET are gap-filled separately for two wind sectors using a random forest regression model to account for complex and nonlinear ecohydrological interactions. While the two wind sectors differ markedly in vegetation composition and flux magnitudes, their flux response is controlled by the same drivers as estimated by the predictor importance of the random forest model, as well as by the high correlation of flux magnitudes (correlation coefficient r=0.92 for NEE and r=0.89 for ET) between both areas. The 1-month delay of the start of the snow-free season in 2020 reduced the total annual ET by 50 % compared to 2019 and 2021 and reduced the growing-season carbon assimilation to turn the ecosystem from a moderate annual carbon sink (−31 to −6 gC m−2 yr−1) to a source (34 to 20 gC m−2 yr−1). These results underpin the strong dependence of ecosystem structure and functioning on snow dynamics, whose anomalies can result in important ecological extreme events for alpine ecosystems.

• Alonso-Gonzalez, Esteban; Aalstad, Kristoffer; Pirk, Norbert; Mazzolini, Marco; Treichler, Désirée & Leclercq, Paul [Show all 9 contributors for this article] (2023). Spatio-temporal information propagation using sparse observations in hyper-resolution ensemble-based snow data assimilation. Hydrology and Earth System Sciences (HESS). ISSN 1027-5606. 27(24), p. 4637–4659. doi: 10.5194/hess-27-4637-2023. Full text in Research Archive Show summary

  Data assimilation techniques that integrate available observations with snow models have been proposed as a viable option to simultaneously help constrain model uncertainty and add value to observations by improving estimates of the snowpack state. However, the propagation of information from spatially sparse observations in high-resolution simulations remains an under-explored topic. To remedy this, the development of data assimilation techniques that can spread information in space is a crucial step. Herein, we examine the potential of spatio-temporal data assimilation for integrating sparse snow depth observations with hyper-resolution (5 m) snow simulations in the Izas central Pyrenean experimental catchment (Spain). Our experiments were developed using the Multiple Snow Data Assimilation System (MuSA) with new improvements to tackle the spatio-temporal data assimilation. Therein, we used a deterministic ensemble smoother with multiple data assimilation (DES-MDA) with domain localization. Three different experiments were performed to showcase the capabilities of spatio-temporal information transfer in hyper-resolution snow simulations. Experiment I employed the conventional geographical Euclidean distance to map the similarity between cells. Experiment II utilized the Mahalanobis distance in a multi-dimensional topographic space using terrain parameters extracted from a digital elevation model. Experiment III utilized a more direct mapping of snowpack similarity from a single complete snow depth map together with the easting and northing coordinates. Although all experiments showed a noticeable improvement in the snow patterns in the catchment compared with the deterministic open loop in terms of correlation (r=0.13) and root mean square error (RMSE = 1.11 m), the use of topographical dimensions (Experiment II, r=0.63 and RMSE =  0.89 m) and observations (Experiments III, r=0.92 and RMSE =  0.44 m) largely outperform the simulated patterns in Experiment I (r=0.38 and RMSE = 1.16 m). At the same time, Experiments II and III are considerably more challenging to set up. The results of these experiments can help pave the way for the creation of snow reanalysis and forecasting tools that can seamlessly integrate sparse information from national monitoring networks and high-resolution satellite information.

• van Hove, Alouette; Aalstad, Kristoffer & Pirk, Norbert (2023). Using reinforcement learning to improve drone-based inference of greenhouse gas fluxes. Nordic Machine Intelligence (NMI). ISSN 2703-9196. 3, p. 1–6. doi: 10.5617/nmi.9897. Full text in Research Archive Show summary

  Accurate mapping of greenhouse gas fluxes at the Earth's surface is essential for the validation and calibration of climate models. In this study, we present a framework for surface flux estimation with drones. Our approach uses data assimilation (DA) to infer fluxes from drone-based observations, and reinforcement learning (RL) to optimize the drone's sampling strategy. Herein, we demonstrate that a RL-trained drone can quantify a CO_2 hotspot more accurately than a drone sampling along a predefined flight path that traverses the emission plume. We find that information-based reward functions can match the performance of an error-based reward function that quantifies the difference between the estimated surface flux and the true value. Reward functions based on information gain and information entropy can motivate actions that increase the drone's confidence in its updated belief, without requiring knowledge of the true surface flux. These findings provide valuable insights for further development of the framework for the mapping of more complex surface flux fields.

• Filhol, Simon; Fiddes, Joel & Aalstad, Kristoffer (2023). TopoPyScale: A Python Package for Hillslope Climate Downscaling. Journal of Open Source Software (JOSS). ISSN 2475-9066. 8(86), p. 1–6. doi: 10.21105/joss.05059. Full text in Research Archive Show summary

  Global climate reanalyses and projections are available worldwide for the past and coming century. However, model grids remain too coarse to be directly relevant at the hillslope-scale (Fan et al., 2019), requiring adapted downscaling tools to account for the effects of local topography. Mountain regions are experiencing accelerated warming with the cryosphere rapidly responding to climate change. To understand and study geomorphological, hydrological, and glaciological changes, we need tools to downscale meteorological timeseries for the basic atmospheric variables used to solve surface energy and mass balance in process-based models. Advanced dynamical downscaling methods exist, though they come with a high computational cost and complex technical setup (Kruyt et al., 2022). TopoPyScale uses a pragmatic approach to downscaling by minimizing complexity, reducing computational cost, simplifying interoperability with land surface models, while retaining physical coherence and allowing the primary drivers of land surface-atmosphere interaction to be considered. The toolbox is designed to be flexible in its usage and development.

• Alonso-Gonzalez, Esteban; Aalstad, Kristoffer; Baba, Mohamed Wassim; Revuelto, Jesus; Lopez-Moreno, Juan I & Fiddes, Joel [Show all 8 contributors for this article] (2022). The Multiple Snow Data Assimilation System (MuSA v1.0). Geoscientific Model Development. ISSN 1991-959X. 15(24), p. 9127–9155. doi: 10.5194/gmd-15-9127-2022. Full text in Research Archive Show summary

  Accurate knowledge of the seasonal snow distribution is vital in several domains including ecology, water resources management, and tourism. Current spaceborne sensors provide a useful but incomplete description of the snowpack. Many studies suggest that the assimilation of remotely sensed products in physically based snowpack models is a promising path forward to estimate the spatial distribution of snow water equivalent (SWE). However, to date there is no standalone, open-source, community-driven project dedicated to snow data assimilation, which makes it difficult to compare existing algorithms and fragments development efforts. Here we introduce a new data assimilation toolbox, the Multiple Snow Data Assimilation System (MuSA), to help fill this gap. MuSA was developed to fuse remotely sensed information that is available at different timescales with the energy and mass balance Flexible Snow Model (FSM2). MuSA was designed to be user-friendly and scalable. It enables assimilation of different state variables such as the snow depth, SWE, snow surface temperature, binary or fractional snow-covered area, and snow albedo and could be easily upgraded to assimilate other variables such as liquid water content or snow density in the future. MuSA allows the joint assimilation of an arbitrary number of these variables, through the generation of an ensemble of FSM2 simulations. The characteristics of the ensemble (i.e., the number of particles and their prior covariance) may be controlled by the user, and it is generated by perturbing the meteorological forcing of FSM2. The observational variables may be assimilated using different algorithms including particle filters and smoothers as well as ensemble Kalman filters and smoothers along with their iterative variants. We demonstrate the wide capabilities of MuSA through two snow data assimilation experiments. First, 5 m resolution snow depth maps derived from drone surveys are assimilated in a distributed fashion in the Izas catchment (central Pyrenees). Furthermore, we conducted a joint-assimilation experiment, fusing MODIS land surface temperature and fractional snow-covered area with FSM2 in a single-cell experiment. In light of these experiments, we discuss the pros and cons of the assimilation algorithms, including their computational cost.

• Pirk, Norbert; Aalstad, Kristoffer; Westermann, Sebastian; Vatne, Astrid; van Hove, Alouette & Tallaksen, Lena Merete [Show all 8 contributors for this article] (2022). Inferring surface energy fluxes using drone data assimilation in large eddy simulations. Atmospheric Measurement Techniques. ISSN 1867-1381. 15(24), p. 7293–7314. doi: 10.5194/amt-15-7293-2022. Full text in Research Archive Show summary

  Spatially representative estimates of surface energy exchange from field measurements are required for improving and validating Earth system models and satellite remote sensing algorithms. The scarcity of flux measurements can limit understanding of ecohydrological responses to climate warming, especially in remote regions with limited infrastructure. Direct field measurements often apply the eddy covariance method on stationary towers, but recently, drone-based measurements of temperature, humidity, and wind speed have been suggested as a viable alternative to quantify the turbulent fluxes of sensible (H) and latent heat (LE). A data assimilation framework to infer uncertainty-aware surface flux estimates from sparse and noisy drone-based observations is developed and tested using a turbulence-resolving large eddy simulation (LES) as a forward model to connect surface fluxes to drone observations. The proposed framework explicitly represents the sequential collection of drone data, accounts for sensor noise, includes uncertainty in boundary and initial conditions, and jointly estimates the posterior distribution of a multivariate parameter space. Assuming typical flight times and observational errors of light-weight, multi-rotor drone systems, we first evaluate the information gain and performance of different ensemble-based data assimilation schemes in experiments with synthetically generated observations. It is shown that an iterative ensemble smoother outperforms both the non-iterative ensemble smoother and the particle batch smoother in the given problem, yielding well-calibrated posterior uncertainty with continuous ranked probability scores of 12 W m−2 for both H and LE, with standard deviations of 37 W m−2 (H) and 46 W m−2 (LE) for a 12 min vertical step profile by a single drone. Increasing flight times, using observations from multiple drones, and further narrowing the prior distributions of the initial conditions are viable for reducing the posterior spread. Sampling strategies prioritizing space–time exploration without temporal averaging, instead of hovering at fixed locations while averaging, enhance the non-linearities in the forward model and can lead to biased flux results with ensemble-based assimilation schemes. In a set of 18 real-world field experiments at two wetland sites in Norway, drone data assimilation estimates agree with independent eddy covariance estimates, with root mean square error values of 37 W m−2 (H), 52 W m−2 (LE), and 58 W m−2 (H+LE) and correlation coefficients of 0.90 (H), 0.40 (LE), and 0.83 (H+LE). While this comparison uses the simplifying assumptions of flux homogeneity, stationarity, and flat terrain, it is emphasized that the drone data assimilation framework is not confined to these assumptions and can thus readily be extended to more complex cases and other scalar fluxes, such as for trace gases in future studies.

• Oehri, Jacqueline; Schaepman-Strub, Gabriela; Kim, Jin-Soo; Grysko, Raleigh; Kropp, Heather & Grünberg, Inge [Show all 74 contributors for this article] (2022). Vegetation type is an important predictor of the arctic summer land surface energy budget. Nature Communications. ISSN 2041-1723. 13. doi: 10.1038/s41467-022-34049-3. Full text in Research Archive

View all works in Cristin

• Aalstad, Kristoffer; Alonso-Gonzalez, Esteban; Bazilova, Varvara; Bertino, Laurent; Fiddes, Joel & Pirk, Norbert [Show all 7 contributors for this article] (2021). Leveraging emerging Earth observations using data assimilation. Show summary

  The task of mapping from noisy observations to the states that (given a forward model) may have generated them falls under the umbrella of inverse problems. These problems are abundant in Earth system science since our uncertain mechanistic models need to be fully specified while the system is only partially and imperfectly observed. This abundance has, combined with a steadily growing observing system, fueled the development of probabilistic Data Assimilation (DA) schemes that use Bayesian inference to fuse uncertain information from models and observations. Notable operational applications of DA include the production of global atmospheric reanalyses and initializing numerical weather predictions. Perhaps less appreciated is the added value that DA can bring to Earth Observations (EO) as a generalized framework for building retrieval algorithms. In our work we present two completely different inverse problems that show how DA can help us to make the most out of emerging EO. The first problem is snowpack reconstruction where we constrain snow models using highly informative observations of the dynamics of snow cover depletion retrieved using satellite remote sensing. We provide examples assimilating retrievals from moderate (MODIS) and higher resolution (Landsat, Sentinel-2, PlanetScope cubesats) optical sensors as well as from all-weather radar data (Sentinel-1) in the Californian Sierra Nevada, Svalbard, Finse, the Pyrenees, the Swiss Alps, and Lebanon. This method is being developed to build tailored snow and permafrost reanalysis frameworks that lead to improved global cryospheric mapping capabilities and provide new benchmarks for Earth system models. The second problem is flux inversion where we seek to infer surface fluxes of carbon, water, and heat using a drone swarm that provides distributed measurements of temperature, gas concentrations, and wind in the atmospheric boundary layer. To achieve this, we assimilate drone data into various boundary layer models, building up complexity from analytical flux-profile relationships based on the widely used Monin-Obukhov Similarity Theory to turbulence resolving large eddy simulations. Through flux inversion with the latter the hope is that we will be able to map fluxes in highly heterogeneous landscapes at the scale of Earth system models (10 km). This is not possible with existing methods like eddy covariance and can thereby shed new light on the role of flux heterogeneities in land-atmosphere coupling. When solving these problems we test various probabilistic DA schemes including variants of the ensemble Kalman filter, the particle filter, and Markov chain Monte Carlo. These schemes have been adapted to our problems by casting them as smoothers that condition the model on future observations, rather than as sequential filters, which crucially allows information to propagate backwards in time. We also emphasize how we can use DA to move beyond the usual first level of inference where we “fit” our model to data, up to the second level of inference where we can compare different competing model structures and parametrizations using the model evidence framework.

• Aalstad, Kristoffer; Fiddes, Joel; Martin, Leo Celestin Paul; Alonso-Gonzalez, Esteban; Yilmaz, Yeliz A. & Pirk, Norbert [Show all 7 contributors for this article] (2021). Workshop on downscaling with TopoSCALE for cryospheric applications. Show summary

  Meteorological forcing data is a crucial ingredient when modeling the past, present, and future state of the cryosphere. At the same time, cold regions (i.e. the high elevations and/or latitudes of our planet) where most of the cryosphere is found are typically remote with a very low density of in-situ observations. The few meteorological stations that do exist are typically quite unrepresentative due to the often extreme surface heterogeneity and complex terrain in these regions. This makes empirical-statistical and geostatistical downscaling techniques that rely heavily on station data somewhat impractical for cryospheric applications. On the other side of the scale, we have dynamical downscaling techniques that rely on using sophisticated regional climate models to downscale historical global reanalysis data or projections from global climate models. Although these are not as dependent on local observations in that they are able to mechanistically model the state of the atmosphere, they are prohibitively expensive to run at the decadal timescales and hillslope (1 km - 100 m) spatial scale that is often sought in cryospheric applications. In this workshop, we will provide an overview of various downscaling techniques and introduce the topography-based downscaling routine TopoSCALE as well as its many applications and downstream methods listed below. TopoSCALE was developed to provide a computationally feasible technique for generating hourly hillslope scale atmospheric forcing for cryospheric modeling from global reanalysis data or other atmospheric model outputs, without the need for in-situ data. It relies heavily on topographic parameters derived from digital elevation models to be able to scale the input atmospheric forcing to the local topography. It has been tested quite extensively in a variety of environments, including: the Swiss Alps, Svalbard, the California Sierra Nevada, and High Mountain Asia. TopoSCALE is also being used as part of ongoing permafrost reanalysis efforts in the ESA Permafrost_CCI project. The scheme can be coupled to a clustering framework (TopoSUB) which can speed up distributed simulations by orders of magnitude compared to the more traditional gridded approach to land surface modeling. TopoSCALE can also be used together with bias-correction techniques to help downscale future regional climate projections to the hillslope scale with applications to cryospheric climate impact studies. Recent and ongoing applications of TopoSCALE together with snow data assimilation have been particularly fruitful in being able to handle biases in solid precipitation, which has been one of the most challenging problems for downscaling in cryospheric applications. Importantly, TopoSCALE lets modellers shift limited computational resources away from the downscaling exercise to probing uncertainties through ensemble simulations and constraining these with Earth observations.

• Aalstad, Kristoffer; Westermann, Sebastian; Pirk, Norbert; Fiddes, Joel & Bertino, Laurent (2021). Retrieving fractional snow-covered area from optical satellites using data assimilation. Show summary

  Mapping from noisy observations to the latent states that may have generated them falls under the umbrella of inverse problems. These problems are abundant in Earth system science since our uncertain mechanistic models need to be fully specified while the system is only partially and imperfectly observed. Combined with a steadily growing observing system, this abundance has fueled the development of probabilistic Data Assimilation (DA) schemes that use Bayesian inference to fuse uncertain information from models and observations. Widely used applications of DA include the production of global atmospheric reanalyses and initializing numerical weather predictions. At the same time, the added value that DA can bring to remote sensing as a generalized framework for building retrieval algorithms remains largely untapped. In our contribution, we demonstrate the potential of data assimilation in the task of retrieving fractional snow-covered area (fSCA) from multispectral satellite imagery from moderate (MODIS) and higher (Sentinel-2 MSI, Landsat 8 OLI) optical sensors. In this analysis, we build on our previous work by focusing on the Bayelva catchment near Ny-Ålesund we have access to independent high-quality validation data obtained from terrestrial photography. We show how the general problem of linear spectral unmixing that is widely used for land cover classification can be recast as a Bayesian inverse problem. This can then be readily solved using ensemble-based data assimilation schemes, where we test both vanilla and sophisticated flavors of the particle filter and the ensemble Kalman filter, as well as Markov chain Monte Carlo benchmarks. By solving the problem in a transformed parameter space, the physical constraints of spectral unmixing are satisfied while reducing the need for ad hocery. The Bayesian data assimilation fSCA retrieval approach lets us deal with ill-posedness, incorporate physical knowledge, and account for uncertainty in the observed reflectances. It performs favorably compared to widely used techniques for fSCA retrieval such as thresholding of the NDSI, regression on the NDSI, and classical (non-negative least squares) spectral unmixing. This method is also much more scalable than classical unmixing since iterations are pre-determined and can fully exploit vectorization. Furthermore, it does not require any tuning on in-situ observations and it can also be used to solve the endmember selection problem using the concept of model evidence. Crucially, the retrieved fSCA includes dynamic uncertainty estimates that are required for satellite retrievals to be of any use in dynamic data assimilation frameworks. We envisage further validation by leveraging the network of terrestrial cameras operated by our partners in the PASSES consortium (Salzano et al., 2021; SESS Report 2021, Ch. 10). Our aim is to exploit these satellite retrievals in our ongoing efforts to produce tailored high resolution permafrost and snow reanalyses in cold regions, including Svalbard. At the same time, the approach outlined here could also be modified to retrieve surface albedo and (sub-pixel) land cover globally with even broader implications to Earth system science.

• Nickelsen, Trine (2021). Verdens største karbonlager lekker – han måler utslippene. Apollon : Forskningsmagasin for Universitetet i Oslo. ISSN 0803-6926.
• Nickelsen, Trine (2021). Verdens største karbonlager lekker – han måler utslippene. Apollon : Forskningsmagasin for Universitetet i Oslo. ISSN 0803-6926.
• Aalstad, Kristoffer; Alonso-Gonzalez, Esteban; Bazilova, Varvara; Bertino, Laurent; Fiddes, Joel & Filhol, Simon [Show all 9 contributors for this article] (2021). Unmixing satellite imagery using data assimilation to map snow and albedo.
• Aalstad, Kristoffer; Alonso-Gonzalez, Esteban; Bertino, Laurent; Fiddes, Joel; van Hove, Alouette & Pirk, Norbert [Show all 8 contributors for this article] (2021). Learning from Earth observations using data assimilation.

View all works in Cristin