AS2.3

Global eddy-covariance (EC) flux measurement networks have provided invaluable insights into ecosystem-atmosphere exchanges of gases, energy and momentum. However, EC technique underestimates surface fluxes during periods when the turbulent flow is decoupled from the surface and this deficiency casts a shade on the validity of EC flux networks. The decoupling can happen for instance when strongly stably stratified air layers or thick forest canopies inhibit vertical mixing. These so-called decoupling periods are typically identified using friction-velocity (u*) and periods when u* is below a site-specific threshold are removed from EC flux time series. This approach has at least two weaknesses: 1) it relies on uncertain site-specific threshold values and 2) it does not consider changes in processes hindering the flow coupling to the surface. Furthermore, it can be questioned whether u* is a correct metric for the strength of turbulent mixing. In this study we utilize recently proposed method which overcomes the above-mentioned weaknesses of u* filtering. The method is based on a comparison between vertical wind speed standard deviation and work done against forces (buoyancy and canopy drag) hindering the movement of a downward propagating air parcel. Via this comparison the need for site-specific thresholds is in theory alleviated. We utilize data from various contrasting EC sites to 1) evaluate whether the new method is free from site-specific thresholds also in practice, 2) compare the flux filtering methods in different conditions and 3) assess the effect of these methods on ecosystem respiration, gross primary productivity and carbon (C) balance estimates. These results will help to assess the robustness of ecosystem C flux estimates made in the past with EC and give clues on how to move forward with EC measurements during the decoupled periods.

How to cite: Peltola, O., Aurela, M., Kolari, P., Mammarella, I., Papale, D., Vesala, T., and Laurila, T.: Beyond friction-velocity thresholds: a new look on eddy-covariance flux filtering and impact on ecosystem C flux estimates, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-3994, https://doi.org/10.5194/egusphere-egu21-3994, 2021.

Previous studies showed at a forest site, that small air pressure fluctuations that are generated during periods of high wind speed significantly enhance topsoil gas transport, which is called pressure-pumping. The strength of these air pressure fluctuations can be described by the pressure pumping coefficient (PPC) which is defined as the mean absolute slope between two measurements (0.5 s) per 30 min interval. It was shown that at this site a quadratic relationship exists between the PPC and above canopy wind speed.

To investigate the variability of small air pressure fluctuations, high-frequency airflow and air pressure measurements were carried out at ten European and American sites with different land use (grassland, crop, forest, urban). The air pressure fluctuations were generally measured above the soil surface and airflow above the site-specific canopy (above trees in forests, on the top of a high building in the city). The measurements took place between 2016 and 2020 and commonly lasted at least one month per site.

Results show that the site-specific PPC increases in a quadratic relationship with above-canopy wind speed at all sites. The data was very close to a quadratic relationship at sites with rather uniform forests and level topography (R² > 0.92), while more complex sites revealed a larger scattering of this correlation (R² > 0.65).

At some sites, the PPC is also highly dependent on the prevailing wind direction. It is shown that the local surface roughness of the plant canopy can be excluded as a main driver of the PPC. Moreover, analysis of surface roughness parameters suggests that the topographic exposure around the measurement sites is responsible for the variability in the PPC.

However, due to the limited data availability and the complexity of the sites (topography, canopy, buildings), it cannot yet be ruled out that other effects have an influence on the PPC. In any case, from the results it can be inferred that wind-induced air pressure fluctuations responsible for pressure-pumping are detectable over a variety of natural and artificial surfaces. It must, therefore, be assumed that they have the potential to increase the diffusion-limited transport rate of trace gases in the soil as well as the soil-atmosphere exchange of trace gases over a large number of surfaces during periods of high wind speed.

How to cite: Mohr, M., Laemmel, T., Maier, M., Kolbe, S., Jung, C., Zeeman, M., Longdoz, B., Knohl, A., Thomas, C., and Schindler, D.: Comparison of wind-induced air pressure fluctuations at sites with different land use, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-3341, https://doi.org/10.5194/egusphere-egu21-3341, 2021.

The reduced availability of evaporative cooling resulting from a hotter and drier climate can lead to high leaf temperatures resulting in overheating. This can affect a variety of biophysical and biochemical processes that could enhance mortality. Plant resilience to these increasingly stressful conditions could rely on non-evaporative cooling. However, to what extent this plays a role is poorly known at present.

In order to assess heat dissipation under the long summer drought conditions, we measured leaf-to-air temperature differences ΔT_leaf-air of pine needles in semi-arid conditions in a drought-exposed and in an experimentally irrigated plot. For this purpose, we developed a novel, high accuracy system based on an infrared camera capable of continuous measurements of leaf temperature under field conditions. Both drought-exposed and irrigated trees, which had a 10x higher transpiration rate, exhibited a similar ΔT_leaf-air that remained mostly below 3.5°C. Variations in mean wind speed did not strongly affect ΔT_leaf-air, but it depended highly on within-canopy turbulence. This suggests a non-evaporative cooling mechanism that relies on the low leaf resistance to heat transfer, thus generating a large sensible heat flux. The ~30% reduction in resistance between leaves of drought-exposed and irrigated trees in the same species must be a result of changes in leaf characteristics and differences in canopy structure influencing wind penetration into the canopy. This reduction in resistance is required to generate the sufficiently larger sensible heat flux of nearly 100 W m^-2 observed between both treatments under high radiation.

Non-evaporative cooling was demonstrated to be an effective leaf- and leaf-branch-scale cooling mechanism in trees with small leaves, which can be a critical factor in forest resistance to drying climates. The generation of a leaf-scale sensible heat flux is considered as a possible mechanism leading to the development of the previously identified canopy-scale ‘convector effect’.

How to cite: Muller, J. D., Rotenberg, E., Tatarinov, F., Oz, I., and Yakir, D.: Evidence for efficient non-evaporative leaf cooling mechanism in a pine forest under drought conditions, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-8527, https://doi.org/10.5194/egusphere-egu21-8527, 2021.

An increased occurrence of persistent heatwaves, as one possible future scenario, generates favorable conditions for the formation of ambient air ozone. Vegetation highly specialized to sub-arctic climate is vulnerable to rapid environmental changes inflicted by global warming and might become more susceptible to ozone in the future. Over large parts of Europe the summer 2018 had been extraordinarily hot and dry and caused large wildfires in northern Sweden in particular. This can be regarded as a test case for such a future scenario. In both 2018 and 2019, we have monitored ambient air ozone concentrations at the Norwegian Institute of Bioeconomy Research (NIBIO) Environment Centre Svanhovd in Northern Norway. Due to
data acquisition problems, ozone concentrations for two weeks in July 2018 were missing from our record. We present a reconstruction based on probability density function with respect to the Swedish and Finnish atmospheric monitoring sites in the region. Over all, ozone concentrations did not differ significantly between the two years. While temperatures and global irradiance diverged significantly from multi annual mean, precipitation varied only to some extend. Coincidentally, we have observed ozone-induced visible injuries on clovers in the ozone garden at Svanhovd in 2018, but not in 2019. We investigate the difference in uptake of ozone using the DO3SE model, with respect to the typical vegetation (e.g., birch and conifers) at
the location. We assess whether critical levels on POD1 for these species were breached. We find that an unadjusted transfer of currently used standard parameters and methodes on ozone damage assessment (IPC Mapping Manual) to vegetation in the subarctics will result in an missinterpretation of POD1 values.

How to cite: Falk, S., Vollsnes, A. V., Emberson, L., O'Neill, C., Berglen Eriksen, A. E., Stordal, F., and Koren Berntsen, T.: The 2018 heatwave and its implications on ozone induced damage on vegetation in a subarctic climate, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-15000, https://doi.org/10.5194/egusphere-egu21-15000, 2021.

Our understanding of the carbon and water cycle was greatly improved through application of eddy covariance measurements in recent decades. Though powerful, this micrometeorological approach relies on a number of assumptions that can be affected by a selection of station location. Most importantly, terrain of the target area should be flat, target area should be homogeneous and adequate air mixing should be achieved. Although possible shortcomings can be reduced by careful site inspection before tower installation (flat terrain) or can be corrected for during data post-processing (filtering of periods with low mixing), preliminary assessment of target area homogeneity is difficult as well as correction of its impacts afterwards. The influence of such inhomogeneities can lead to a bias in the flux annual sums but also a bias in their relationships with environmental variables. Certain solutions were already proposed, but target area homogeneity was so far assessed only at a few selected sites. Here we aim to provide a suit of software tools that build on the existing software packages (REddyProc, Flux Footprint Prediction, openair, openeddy) and allow easy diagnosis of the situation at the given ecosystem station. We plan to provide directional analyses of variables of interest. This will allow to identify the wind sectors that show large deviations from the mean value of the whole target area. In a further step, we plan to combine footprint modeling with CO₂ and energy flux measurements and thus provide attribution of mean (weighted) fluxes to their source area. Based on the differences with the directional analyses we will assess whether the higher computational expenses of footprint modeling are justified and bring additional information. Finally, we plan to separate the target area to a limited amount of wind sectors and attempt separate gap-filling and flux partitioning for areas identified by preceding homogeneity evaluation. The limitations and feasibility of this approach will be assessed.

This work was supported by the Ministry of Education, Youth and Sports of CR within Mobility CzechGlobe2 (CZ.02.2.69/0.0/0.0/18_053/0016924).

How to cite: Šigut, L., Wutzler, T., El-Madany, T., Fischer, M., and Migliavacca, M.: A tool for evaluation of target area homogeneity at ecosystem stations employing eddy covariance method, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-8919, https://doi.org/10.5194/egusphere-egu21-8919, 2021.

Trees have a large role in improving urban air quality, among other mechanisms, through dry deposition of scalars and aerosols on leaf surfaces. We tested the role of leaf density and canopy structure in modulating the rate of dry deposition. We simulated the interactions between a virtual forest patch and deposition rate of an arbitrary scalar using the Parallelized Large Eddy Simulation Model (PALM). Two canopy structures were considered: a homogenous canopy; and canopy stripes perpendicular to the wind direction. For each canopy stripe scenario, we considered thin, intermediate, and wide stripes, while the space between stripes equals the stripes’ width. Four leaf area densities were considered for each case. The results showed that stripes perpendicular to the wind direction had a larger deposition per leaf area than homogeneous canopies, and denser canopies had more total deposition, but lower per-leaf area rate. Our results can be explained by canopy-induced turbulence structures that couple the air within and above the canopy and lead to more effective leaf area where this coupling is stronger. We aggregate our results to the whole-patch scale and suggest a canopy-structure and leaf-area dependent correction to the canopy resistance parameter so to be used in coarse models that resolve dry deposition. 

How to cite: Yazbeck, T., Bohrer, G., Vines, C., De Roo, F., Mauder, M., and Bakshi, B.: Effects of spatial heterogeneity of leaf density and crown spacing of canopy patches on dry deposition rates, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-414, https://doi.org/10.5194/egusphere-egu21-414, 2021.

Trees play an important role in the urban heat island effect and urban air quality due to their impact on the transfer of radiation, momentum, heat, moisture and pollution. However, the effects of trees are hard to quantify due to their complex interactions with urban surfaces and the turbulent atmosphere overhead.

We present a complete tree model for large-eddy simulations (LES) that represents the effects of trees on drag, transpiration, shading and deposition at resolutions of O(1 m, 0.1 s) whilst minimising the number of model parameters. The tree model avoids the necessity to resolve the leaf temperature via a derivation of the Penman-Monteith equation and distinguishes between cooling via transpiration and shading. The latent heat flux is further broken down into radiative and advective components in order to better understand the mechanism behind transpirational cooling (e.g. the ‘oasis’ effect).

The new tree model is investigated analytically to provide insight into tree cooling regimes, and is applied to field studies to contextualise the analysis. The combined cooling effect of trees due to transpiration and shading processes can be reduced to a four-dimensional parameter space. The net tree cooling (NTC) and tree cooling ratio (TCR) parameters are defined to enable a systematic categorisation of the thermal effect of a tree into five regimes: net heating, net reduction (shading dominated), net reduction (transpiration dominated), net cooling (shading dominated) and net cooling (transpiration dominated). Existing parameterisations for tree cooling are reviewed, illustrating their limitations and highlighting the need for complete models to determine tree cooling.

The tree model is implemented into the LES model uDALES. The drag and canopy energy balance models are validated, and results are presented for domains that are 1) fully covered by trees; 2) partially covered by trees; and 3) have a single line of trees. These simulations provide physical insight into the effect of trees on the microclimate and provide evaluation data for future studies.

How to cite: van Reeuwijk, M. and Grylls, T.: Tree model with drag, transpiration, shading and deposition: Identification of cooling regimes and large-eddy simulation, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-15722, https://doi.org/10.5194/egusphere-egu21-15722, 2021.

Eddy covariance (EC) scalar flux loss at high frequency is due to the incapability of the measurement system to detect small-scale variation of atmospheric turbulent signals. This systematic bias is particularly important for closed-path systems, and it is mainly related to inadequate sensor frequency response, sensor separation and the air sampling trough tubes and filters. Here, we investigate the limitations of current approaches, based on measured power spectra (PSA) or cospectra (CSA), to empirically estimated the spectral transfer function of the EC system needed for the frequency response correction of measured fluxes. We performed a systematic analysis by using EC data from a wetland and forest site for a wide range of attenuation levels and signal-to-noise ratio. We proposed a novel approach for PSA that uses simultaneously the noise and the turbulent signals present in the power spectrum, providing robust estimates of spectral transfer function for all conditions. We further theoretically derived a new transfer function to be used in the CSA approach which specifically accounts for the interaction between the low-pass filtering induced phase shift and the high frequency attenuation. We show that current CSA approaches neglect such effect, giving a non-negligible systematic bias to the estimated scalar fluxes from the studied sites. Based on these findings, we recommend that spectral correction methods, implemented in EC data processing algorithms, are revised accordingly.

How to cite: Mammarella, I., Peltola, O., Aslan, T., Ibrom, A., Nemitz, E., and Rannik, Ü.: Revising the high frequency response correction of scalar fluxes measured by closed-path eddy covariance systems, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-14464, https://doi.org/10.5194/egusphere-egu21-14464, 2021.

Agroforestry is an integration of trees in cropland or grassland and is discussed, within Germany and the EU, as a potential “Green Solution” for agriculture. Agroforestry alters the microclimate, productivity, biodiversity, and nutrient and water usage – as compared to standard agricultural practise. A potentially key benefit is the higher carbon sequestration of agroforestry, relative to monoculture systems, which could provide an interesting option for mitigating climate change, while still providing valuable arable land. Net ecosystem exchange studies of CO₂ (NEE) of agroforestry systems are rare, in comparison to the extensive studies of NEE of agricultural systems (croplands and grasslands). Therefore, the current study, as part of the SIGNAL (sustainable intensification of agriculture through agroforestry) project, investigates the NEE of agroforestry compared to that of monoculture agriculture.

At five locations across Germany, paired flux measurements above agroforestry and monoculture agronomy are performed using innovative low-cost CO₂ eddy covariance sensors (slow response Vaisala GMP343 IRGA, with custom made housing). During the summer of 2020 simultaneous measurements of the low-cost setup and a LI-COR 7200 are performed, above grassland at 3.5 m and adjacent agroforestry grassland at 10 m measurements height.

The low-cost eddy covariance system is able to capture the turbulent (diurnal) CO₂ flux dynamics and the response to management activities. After spectral corrections and applying quality control, the low-cost system at the agroforestry site (slope = 0.92, R² = 0.88) performs better than the low-cost system at the grassland site (slope = 0.67, R² = 0.80), when compared to the LI-COR measurements. This is probably due to the difference in turbulence caused by different surface roughness and measurement height. The preliminary cumulative carbon flux during the four-month measurement campaign shows a significant difference between the grassland (source of (+) 16-38 gC/m²) and agroforestry grassland (sink of (-) 148-164 gC/m²), in favour of agroforestry. By applying post processing software, we aim to further optimize the frequency corrections for the low-cost system. In the future the obtained post processing scheme will be applied to the other low-cost eddy covariance systems within the project.

How to cite: van Ramshorst, J., Markwitz, C., Hill, T., Clement, R., Knohl, A., and Siebicke, L.: Evaluation of low-cost eddy covariance for CO2 fluxes over agroforestry and grassland, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-2181, https://doi.org/10.5194/egusphere-egu21-2181, 2021.

The implementation of higher-order turbulence closure schemes in Earth system models (e.g., the Cloud Layers Unified by Binormals; CLUBB) aims to improve the modeling of convection and radiative transfer in numerical weather prediction and climate models. However, the added value of these schemes is constrained by the specification of boundary conditions on higher-order statistics. At the land surface, many of the higher order turbulence statistics that are required as boundary conditions are parameterized using formulations more appropriate for stationary and planar-homogeneous flow in the absence of subsidence. A case in point is the variance of the potential temperature fluctuations.  Because of the additive nature of variances arising from non-uniformity in surface heating, current parameterizations are not readily generalizable. The current scheme used in CLUBB, as well as other models, relies on limited studies over uniform terrain, with the variance entirely determined by local sensible heat flux, friction velocity, and the Obukhov stability parameter without regard to local site characteristics. This presentation aims to address this weakness by leveraging the National Ecological Observation Network (NEON) network of eddy covariance towers to validate the current parameterization scheme for potential temperature variance, as well as propose improvements for more heterogeneous terrain.

The turbulence fluctuations of temperature at 39 NEON sites are processed and quality controlled, removing points occurring at night, while precipitation is falling, and with sub-zero temperatures. Results overall indicate the current scheme performs well, especially over flat homogeneous terrain where local flux relationships dominate. When there is sufficiently heterogeneous, rough terrain or non-closure of the local energy balance, however, existing schemes fail to accurately estimate the variances in temperature. In these cases, the parameterization needs to be modified, and initial results suggest simple adjustments can yield improvements and reduce error close to that of the uniform sites with local energy balance closure. The successful improvement of the temperature variance parameterization scheme implies high potential for similar, new, empirically derived parameterizations for the surface boundaries for other higher order turbulent statistics (e.g. temperature skewness) in atmospheric turbulence models.

How to cite: Waterman, T., Katul, G., Bragg, A., and Chaney, N.: Evaluating and Improving Parameterizations of the Variance of Temperature Fluctuations Over Heterogeneous Landscapes for Surface Boundary Conditions in Atmospheric Models, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-13548, https://doi.org/10.5194/egusphere-egu21-13548, 2021.

Land-surface heterogeneity is known to play an important role in land-surface hydrology, which drives the bottom boundary condition for atmospheric models in numerical weather prediction (NWP) applications. However, the ultimate impact of land-surface heterogeneity on atmospheric boundary layer (ABL) development is still an open problem with implications for sub-grid scale (SGS) parameterizations for both NWP and climate modeling. Large-eddy simulation (LES) is often used to study the effects of land-surface heterogeneity on ABL development, most typically via specified surface fields which are not influenced by the atmosphere (i.e. semi-coupled). Heterogeneous land surfaces have been seen in previous studies to have a significant influence on ABL dynamics, particularly cloud production, in certain cases when semi-coupled to the atmosphere.

Here we use the Weather Research and Forecasting (WRF) model as an LES with both semi-coupled and fully-coupled land surfaces to investigate the impact of two-way coupling on the interaction between heterogeneous land surfaces and daytime ABLs. For semi-coupled simulations, the HydroBlocks land-surface model is run offline, driven by 4-km NLDAS-2 meteorology with Stage-IV radar rainfall data, and then used to specify the bottom boundary in WRF. The WRF-Hydro model is used for cases where the land surface is fully coupled to the WRF model. Both land-surface models use the Noah-MP model as their underlying physics package and add both subsurface and overland flow routing. The WRF model uses a 100-m horizontal resolution, and the land-surface models use high resolution (30 m) datasets that were upscaled to match the LES resolution for elevation, landcover, and soil type using NED, NLCD, and POLARIS respectively. These LES experiments are performed over the ARM Southern Great Plains Site atmospheric observatory in Oklahoma during the Summer of 2017 with a grid size of 100 km x 100 km to imitate a single cell in a modern climate model. The impact of land-surface heterogeneity on the atmosphere is evaluated by comparing simulations using the fully heterogeneous land surfaces with simulations where the land surface is homogenized at each timestep, taking a domain-wide spatial mean value at every grid cell. Results are evaluated primarily by the differences in the development of clouds and evolution of turbulent kinetic energy in the ABL.

How to cite: Simon, J., Waterman, T., Hay-Chapman, F., Dirmeyer, P., Bragg, A., and Chaney, N.: An Analysis of the Importance of a Fully-Coupled Atmosphere and Land-Surface When Considering the Impact of Multi-Scale Land Spatial Heterogeneity on Cloud Development, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-13379, https://doi.org/10.5194/egusphere-egu21-13379, 2021.

The surface energy balance (SEB) is defined as the balance between incoming energy from the sun and outgoing energy from the Earth’s surface. All components of the SEB depend on land surface temperature (LST). Therefore, LST is an important state variable that controls the energy and water exchange between the Earth’s surface and the atmosphere. LST can be estimated radiometrically, based on the infrared radiance emanating from the surface. At the landscape scale, LST is derived from thermal radiation measured using  satellites.  At the plot scale, eddy covariance flux towers commonly record downwelling and upwelling longwave radiation, which can be inverted to retrieve LST  using the grey body equation :
             R_lup = εσ T_s⁴ + (1 − ε) R _ldw         (1)
where R_lup is the upwelling longwave radiation, R_ldw is the downwelling longwave radiation, ε is the surface emissivity, T_s  is the surface temperature and σ  is the Stefan-Boltzmann constant. The first term is the temperature-dependent part, while the second represents reflected longwave radiation. Since in the past downwelling longwave radiation was not measured routinely using flux towers, it is an established practice to only use upwelling longwave radiation for the retrieval of plot-scale LST, essentially neglecting the reflected part and shortening Eq. 1 to:
               R_lup = εσ T_s⁴                        (2)
Despite  widespread availability of downwelling longwave radiation measurements, it is still common to use the short equation (Eq. 2) for in-situ LST retrieval. This prompts the question if ignoring the downwelling longwave radiation introduces a bias in LST estimations from tower measurements. Another associated question is how to obtain the correct ε needed for in-situ LST retrievals using tower-based measurements.
The current work addresses these two important science questions using observed fluxes at eddy covariance towers for different land cover types. Additionally, uncertainty in retrieved LST and emissivity due to uncertainty in input fluxes was quantified using SOBOL-based uncertainty analysis (SALib). Using landscape-scale emissivity obtained from satellite data (MODIS), we found that the LST  obtained using the complete equation (Eq. 1) is 0.5 to 1.5 K lower than the short equation (Eq. 2). Also, plot-scale emissivity was estimated using observed sensible heat flux and surface-air temperature differences. Plot-scale emissivity obtained using the complete equation was generally between 0.8 to 0.98 while the short equation gave values between 0.9 to 0.98, for all land cover types. Despite additional input data for the complete equation, the uncertainty in plot-scale LST was not greater than if the short equation was used. Landscape-scale daytime LST obtained from satellite data (MODIS TERRA) were strongly correlated with our plot-scale estimates, but on average higher by 0.5 to 9 K, regardless of the equation used. However, for most sites, the correspondence between MODIS TERRA LST and retrieved plot-scale LST estimates increased significantly if plot-scale emissivity was used instead of the landscape-scale emissivity obtained from satellite data.

How to cite: Thakur, G., Schymanski, S., Mallick, K., and Trebs, I.: Plot-scale retrieval of land surface temperature and emissivity estimation, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-3962, https://doi.org/10.5194/egusphere-egu21-3962, 2021.

Land surface temperature (LST) is a key factor in earth–atmosphere interactions and an important indicator for monitoring environmental changes and energy balance on Earth's surface. Thermal infrared (TIR) remote sensing can only obtain valid observations under clear-sky conditions, which results in the discontinuities of the LST time series. In contrast, passive microwave (PMW) remote sensing can help estimate the LST under cloudy conditions and the LST generated by PMW observations is an important input parameter for generating medium-resolution (e.g., 1km) all-weather LST. Neural networks, especially the latest deep learning, have exhibited good ability in estimating surface parameters from satellite remote sensing. However, thorough examinations of neural networks in the estimation of LST from satellite PMW observations are still lacking. In this study, we examined the performances of the traditional neural network (NN), deep belief network (DBN), and convolutional neural network (CNN) in estimating LST from the Advanced Microwave Scanning Radiometer for EOS (AMSR-E) and Advanced Microwave Scanning Radiometer 2 (AMSR2) data over the Chinese landmass. The examination results show that CNN is better than NN and DBN by 0.1–0.4 K. Different combinations of input parameters were examined to get the best combinations for the daytime and nighttime conditions. The best combinations are the brightness temperatures (BTs), NDVI, air temperature, and day of the year (DOY) for the daytime and BTs and air temperature for the nighttime. Compared with the MODIS LST, the CNN LST estimates yielded root-mean-square differences (RMSDs) of 2.19–3.58 K for the daytime and 1.43–2.14 K for the nighttime for diverse land cover types for AMSR-E. Validation based on the in-situ LST demonstrates that the CNN LST yielded root-mean-square errors of 2.10–5.34 K and the error analysis confirms that the main reason for the errors is the scale mismatching between the ground stations and the MW pixels. This study helps better the understanding of the use of neural networks for estimating LST from satellite MW observations.

How to cite: Wang, S., Zhou, J., Zhang, X., and Jin, Z.: Estimating Land Surface Temperature from AMSR-E and AMSR2 Data with Convolutional Neural Network, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-14126, https://doi.org/10.5194/egusphere-egu21-14126, 2021.

As an important indicator of land-atmosphere energy interaction, land surface temperature (LST) plays an important role in the research of climate change, hydrology, and various land surface processes. Compared with traditional ground-based observation, satellite remote sensing provides the possibility to retrieve LST more efficiently over a global scale. Since the lack of global LST before, Ma et al., (2020) released a global 0.05 ×0.05  long-term (1981-2000) LST based on NOAA-7/9/11/14 AVHRR. The dataset includes three layers: (1) instantaneous LST, a product generated based on an ensemble of several split-window algorithms with a random forest (RF-SWA); (2) orbital-drift-corrected (ODC) LST, a drift-corrected version of RF-SWA LST at 14:30 solar time; and (3) monthly averages of ODC LST. To meet the requirement of the long-term application, e.g. climate change, the period of the LST is extended from 1981-2000 to 1981-2020 in this study. The LST from 2001 to 2020 are retrieved from NOAA-16/18/19 AVHRR with the same algorithm for NOAA-7/8/11/14 AVHRR. The train and test results based on the simulation data from SeeBor and TIGR atmospheric profiles show that the accuracy of the RF-SWA method for the three sensors is consistent with the previous four sensors, i.e. the mean bias error and standard deviation less than 0.10 K and 1.10 K, respectively, under the assumption that the maximum emissivity and water vapor content uncertainties are 0.04 and 1.0 g/cm², respectively. The preliminary validation against in-situ LST also shows a similar accuracy, indicating that the accuracy of LST from 1981 to 2020 are consistent with each other. In the generation code, the new LST has been improved in terms of land surface emissivity estimation, identification of cloud pixel, and the ODC method in order to generate a more reliable LST dataset. Up to now, the new version LST product (1981-2020) is under generating and will be released soon in support of the scientific research community.

How to cite: Ma, J. and Zhou, J.: Global long-term land surface temperature for NOAA AVHRR: extension from 1981-2000 to 1981-2020, EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-14373, https://doi.org/10.5194/egusphere-egu21-14373, 2021.