Earth Observation Lab

@humboldteolab

Humboldt´s Earth Observation Lab focuses on a better understanding of coupled human-environment systems based on remote sensing data and geoinformation.

391 Siguiendo475 Seguidores87 Publicaciones

Bluesky

¿Es tu perfil?Reclámalo y humboldteolab.dirty.social es tuyo.Reclamar perfil

Earth Observation Lab@humboldteolab· 8d

Simone Stuenzi, @smstuenzi.bsky.social, postdoc at the EOLab, and colleagues from @awi.de, @vuamsterdam.bsky.social and @uio.no, use process-based permafrost modeling, satellite time-series, and field data to study canopy-ground thermal dynamics, addressing a blind spot in Earth system models.

A photograph of Simone Stuenzi standing in an open boreal forest. The ground is covered in green grass and scattered fallen logs, with sparse, tall, thin trees in the background under a cloudy sky. Field gear sits on the ground nearby.

A flowchart diagram titled "Fig. 8. Study workflow." It illustrates how bioclimatic classification, climate data extraction, field observations, and canopy density assessments feed into process-based modelling (using the CryoGrid model) to calculate soil organic carbon exposure and assess permafrost carbon vulnerability.

Earth Observation Lab@humboldteolab· 14d

To learn more about this complex canopy-ground climate dynamic, have a look through our new StoryMap: arcg.is/18Wa810

Earth Observation Lab@humboldteolab· 14d

When these forests are lost to wildfire, logging, or pests, the ground warms rapidly. Modeling shows that within five years of complete forest loss, ground warming reaches the same scale as the regional warming projected out to 2100.

A photograph of a boreal forest area following a wildfire, showing charred, black, leafless tree trunks standing closely together against a bright sky. The forest floor is littered with fallen, blackened branches and dark, burned soil, with new, bright-green vegetation starting to sprout through the ash.

An aerial landscape view of rolling, forested hills under a cloudy sky. The slopes show a mosaic of forest conditions, with some areas dark and recovering from past wildfires, and others covered in healthy green and vibrant autumn-yellow trees.

Earth Observation Lab@humboldteolab· 14d

In summer, boreal canopies intercept up to 90 percent of incoming solar radiation, limiting summer thaw depths. In winter, denser evergreen crowns trap snow early to reduce ground insulation, while open larch forests allow winter heat to escape readily, preserving lower ground temperatures.

A photograph of a dense, open-canopy boreal forest with thin, light-green foliage. The entire forest floor is a continuous, soft green carpet of low-growing plants and moss, with patches of pale white lichen interspersed, creating a uniform understory. Small fallen branches lie scattered. The background shows more forest fading into natural light.

Earth Observation Lab@humboldteolab· 14d

How do trees keep permafrost frozen? Research by @smstuenzi.bsky.social shows forest cover cools the ground surface and can reduce active layer thickness by up to 60 percent compared to open ground. The forest canopy regulates ground temperature through complex, multi-directional processes.

A diagram titled "Fig. 5. Forest canopy regulation," showing a wheel of interacting pathways through which boreal forest canopies influence permafrost. Key factors highlighted include canopy shading, reflectivity, evapotranspiration, photosynthetic activity, snow interactions, hydrology, active layer depth, regional climate, and canopy height.

Earth Observation Lab@humboldteolab· 16d

Beneath the tundra and forests of the Arctic lies permafrost: ground frozen year-round. @smstuenzi.bsky.social's research focuses on how boreal forests and permafrost interact, and what happens when that relationship breaks down. Read all about it in this new StoryMap! arcg.is/18Wa810

Earth Observation Lab@humboldteolab· 27d

Intra-annual time series capture gradual dieback and can give important insights into dynamics following different disturbance agents. While natural crown dieback mostly increased NPV fractions, clear-cuts and salvage logging mainly introduced high soil exposure.

$Detailed time-series analysis (Figure 6) of six selected forest pixels (a to f) in Germany from 2017 to 2023, showcasing different disturbance agents. Each of the six rows (a-f) features a line graph on the left plotting the temporal fractional cover of green canopy (PV, green line), deadwood/bark (NPV, red line), and bare soil (blue line). To the right of each plot are two close-up color-composite satellite images before and after the event, alongside a high-resolution aerial photograph for validation. A small map of Germany in the bottom right marks the geographic location of each case study.$

Earth Observation Lab@humboldteolab· 27d

From 2019–2022, dieback affected ~487,000 ha (5.4%) of Germany’s forests. Species showed distinct patterns: spruce had the highest annual dieback rate at 2.7%, followed by pine at 1.5%, while oak (0.7%) and beech (0.5%) remained low.

National-scale maps and line graphs (Figure 8) of species-specific forest dieback rates in Germany between 2019 and 2022.

Part a) displays four maps of Germany showing the spatial distribution of annual dieback rates in 10x10 km grid cells. Spruce (red, up to 15%) shows severe impact in central German low mountain ranges like the Harz. Pine (orange, up to 8%) is primarily affected in eastern Germany. Beech (green, up to 2%) and Oak (blue, up to 2%) show much lower, localized dieback rates.

Part b) displays line graphs of annual national-scale dieback rates over time. Spruce dieback rises steeply and steadily (from about 2% to over 3.5% annually), while other species fluctuate at much lower levels (below 2%).

Earth Observation Lab@humboldteolab· 27d

Validation against high-res imagery confirms that the models generalize well across species and environmental gradients. During the growing season, NPV fractions were mapped with a low Mean Absolute Error of 11.7%, demonstrating strong spatial and temporal stability.

$Model validation scatter and error plots (Figure 4) for predicted forest dieback. Part a) displays a scatter plot comparing observed fractional cover (%) on the x-axis with predicted NPV fractional cover (%) on the y-axis. The red point cloud closely aligns with the dashed 1:1 line, supported by a solid regression line and key statistics: Mean Absolute Error (MAE) = 13.35, RMSE = 17.7, and a correlation coefficient r = 0.86. Part b) shows a point plot of the Mean Absolute Error (MAE in %) for different tree species (Beech, Oak, others, Pine, Spruce). Three points per species represent the error for all months, the growing season, and outside the growing season, complete with error bars representing standard deviations. Errors remain consistently low, mostly between 10% and 20%.$

Earth Observation Lab@humboldteolab· 27d

The method uses regression-based spectral unmixing to estimate fractional cover of non-photosynthetic vegetation (NPV), green vegetation, and soil. Using synthetic training data from a spectral library, it works without extensive field data collection.

Scientific figure (Figure 3) illustrating the spectral unmixing of tree species.

Part a) shows three line plots of reflectance values across wavelengths from 500 to 2000 nanometers for three land cover classes: non-photosynthetic vegetation (NPV, red), photosynthetic vegetation (PV, green), and bare soil (blue).

Part b) displays six density scatter plots for different tree species (Beech, Oak, Pine, Spruce, other deciduous, other coniferous) in a 2D feature space of NDVI (x-axis) and SWIR ratio (y-axis). Pixel density is color-coded from purple (low count) to yellow (high count), framed by dashed lines connecting the mean endmember points for PV, NPV, and bare soil.

Earth Observation Lab@humboldteolab· 27d

A new study by @jalsleben.bsky.social et al. introduces a national approach to map tree dieback across Germany using Sentinel-2 time series. Instead of just measuring greenness, it directly quantifies non-photosynthetic vegetation at a sub-pixel scale. Let's have a closer look ⬇️

$Infographic announcing a new scientific publication titled "Mapping tree dieback using Sentinel-2 time series and generalized regression-based unmixing models reveals species-specific patterns across Germany" by Jonas Alsleben et al., published in the journal "Remote Sensing of Environment". The top portion features a multi-panel graphic visualizing the progression of spruce dieback from 2019 to 2022 using Sentinel-2 satellite data. Color-coded from light beige (0% dieback fraction) to dark red (100% dieback fraction), it shows the spatial expansion of forest mortality in a selected area over time, paired with 2024 aerial imagery for validation. Earth Observation Lab Logo is shown in the bottom right corner.$

Earth Observation Lab@humboldteolab· Jun 11

Ultimately, this multidecadal remote sensing approach provides a reliable, spatially explicit data source for environmental monitoring. The generated dataset can be directly integrated into biodiversity assessments, ecosystem service modeling, and soil carbon reporting.

Two comparative maps of Germany titled "Fig. 7" broken down by Soil-Climate Region, marked with lowercase letters a through f to pinpoint specific geographic areas.
The left map shows the "Share of grassland on agricultural area (2017)" using a purple-to-yellow gradient. The highest shares (green/yellow) are visible in the Alps (a), Black Forest (b), Sauerland (c), and North Sea coast (d).
The right map shows the "Share of grassland established prior to 1990" using a purple-to-orange gradient. It reveals that the vast majority of Germany's grasslands are old/permanent (yellow/orange), except for a distinct dark purple patch in the Central German black soil region (e), indicating much younger grasslands there.

Earth Observation Lab@humboldteolab· Jun 11

Spatiotemporal patterns align with historical policy shifts. High rates of grassland establishment in the early 1990s correspond to land abandonment following the GDR collapse. Subsequent trends reflect adaptations to European Common Agricultural Policy reforms.

A grid of 15 maps of Germany titled "Fig. 10. Share of grassland established within respective bi-annual intervals per Soil-Climate Region" from 1990–1991 to 2016–2017.
The maps use a color gradient where dark purple indicates 0% new grassland and bright red indicates over 10%. Most maps across the decades are dark blue/purple, indicating low rates of new grassland establishment, with a major exception in the 1992–1993 period, which shows a dramatic surge (dark red and orange) concentrated across eastern Germany.

Earth Observation Lab@humboldteolab· Jun 11

Grassland establishment after 1990 was detected with an F-score of 78.92, and differentiated from persist grassland with 99.21% overall accuracy.  For correctly identified recent grasslands, the framework derived the exact year of establishment with a Mean Absolute Error of 1.3 years.

A multi-panel scatter plot titled "Fig. 6" showing the accuracy of predicted grassland age compared to reference years (2006 to 2022) across seven German federal states: North Rhine-Westphalia, Lower Saxony, Brandenburg, Rhineland-Palatinate, Hesse, Baden-Württemberg, and Bavaria.
Each plot maps reference year vs. predicted year, showing tight alignments along the 1:1 diagonal line. Orange dots represent a single land-use change (LUC), while blue dots represent multiple changes. The Mean Absolute Error (MAE) ranges remarkably low, roughly between 0.84 and 1.46 years for single land-use changes, demonstrating high prediction accuracy.

Earth Observation Lab@humboldteolab· Jun 11

Utilizing Landsat and Sentinel-2 archives (1986–2023), every clear-sky observation was classified to map seasonal bare soil frequencies.  Since permanent grasslands rarely experience tillage, a historical increase in bare soil indicates preceding cropland use.

A technical flowchart titled "Fig. 2. Overview of the proposed approach" mapping out how grassland age is calculated.
It starts at the top with "LUCAS data" passing through "Photo interpretation" to create a "Bare soil reference". This reference, combined with a multidecadal stack of "Landsat and Sentinel-2 imagery (1986–2023)", is fed into a "Random forest classifier" to map "Bare soil occurrence (1986–2023)".
This passes through "Seasonal binning" to create a "Seasonal bare soil frequency" stack. Finally, a "Rule set & thresholds" step incorporates "IACS data" to output "Grassland age" and its corresponding accuracy assessments.

Earth Observation Lab@humboldteolab· Jun 11

Grassland age is a critical parameter influencing carbon sequestration potential, biodiversity support, and ecosystem resilience.  A new study published in Remote Sensing of Environment presents a national-scale framework to estimate grassland age across Germany using satellite time series:

An announcement graphic titled "NEW PUBLICATION!" featuring a scientific paper by Lukas Blickensdörfer et al., published in "Remote Sensing of Environment". The title is "Leveraging multidecadal satellite time series to estimate grassland age on national scale".
The background is blue and shows 15 small maps of Germany tracking the "Share of grassland established since 1990" in bi-annual steps from 1990 to 2017. A color scale ranges from 0% (dark purple) to over 10% (red). Notably, the 1992–1993 map shows a massive red spike in eastern Germany. Logos for THÜNEN and Earth Observation Lab are at the bottom.

Earth Observation Lab@humboldteolab· Jun 10

"Detectives from Space - Discovering Geography from Above": We were delighted to present the fundamentals and possibilities of remote sensing at this year's Long Night of Science at HU Berlin's Campus Adlershof. A great opportunity to engage with and inspire potential future students! #LNDW2026

EOL's Dirk Pflugmacher and two guests, a mother and a child, looking at various "before and after" satellite images that show different kinds of land cover change.

An indoor wide shot of the event at our geography institute, featuring people standing around information tables and interactive displays.

A tabletop scientific setup featuring a spectrometer on a tripod, a spotlight shining on a pile of green leaves and a nearby laptop displaying the a graph of the reflectance data.

EOL's Lennart Eichfuss and of our guests, a sixth grader, discussing one of the displayed croptype maps.

Earth Observation Lab@humboldteolab· Jun 5

The Geography Department of the HU Berlin is part of The Long Night of Science 2026! Tomorrow, June 6th, 5-10pm, Alfred-Rühl-Haus, Rudower Chaussee 16. The Earth Observation Lab, @biogeoberlin.bsky.social and many more will await you with exhibitions, games, lectures and hands-on experiments!

Lange Nacht der Wissenschaften auf dem Campus Adlershof:
grafische Darstellung des Programms, als Datei hier zu finden:
https://www.hu-berlin.de/fileadmin/Mediathek/Zentrale_Seiten/Veranstaltungen/Dokumente/Lange_Nacht_der_Wissenschaften/2026_LNDW_HU_Programm_Campus_Adlershof.pdf

Earth Observation Lab@humboldteolab· Jun 4

EARSeL VP Jean-Christophe Schyns congratulates @lasseharkort.bsky.social for winning the Young Scientist Award at the Imaging Spectroscopy SIG. Lasse's presentation "Capturing the Pulse of the Dry Season" showed how EnMAP time series can disentangle different NPV types from space. Congrats, Lasse!

Picture with headline: "CONGRATULATIONS LASSE!"
Picture shows Jean-Christophe Schyns, EARSeL Vice President, he congratulates Lasse Harkort on Stage for winning the Young Scientist Award at the EARSeL Special Interest Group on Imaging Spectroscopy. Lasse presented on "Capturing the Pulse of the Dry Season", where he showed how EnMAP time series allow disentangling different NPV types from space.

Earth Observation Lab@humboldteolab· Jun 4

EOLab’s @jakimowb.bsky.social , @lasseharkort.bsky.social and Shawn Schneidereit participated in the 14th EARSeL Workshop on Imaging Spectroscopy, hosted by Aalto University (@aalto.fi) in Helsinki, presenting their recent research utilizing EnMAP hyperspectral time series and the EnMAP-Box.

Pictures of presentation:
Speaker: Lasse Harkort
Title: Capturing the Pulse of the Dry Season: Decoupling Woody and Herbaceous NPV with EnMAP Time Series

Pictures of presentation:
Speaker: Shawn Schneidereit
Title: Before The Burn, After The Ash: Tracking Fire Severity And Vegetation Recovery With EnMAP Hyperspectral Time Series

Earth Observation Lab@humboldteolab· May 4

Larger fields, more deforestation. Linking field size to forest cover change indicates a pattern where areas with larger mean field sizes lost more forest. This challenges the narrative that smallholders are responsible for most agricultural deforestation; larger farms matter more.

$A map and scatterplot series illustrating the correlation between farm size and deforestation in Mozambique. The data uses orange and red color coding to highlight regions where larger mean field sizes (above 1 and 2.5 hectares) are linked to significant forest cover loss, challenging the narrative that smallholder farms are the primary drivers of agricultural deforestation. „Fig 5. Relationship between field size and net forest cover change. Bivariate map of tree cover change categories (loss and gain refer to at least 25% of tree cover change relative to 2010 tree cover) for the period 2010–2020 calculated at 0.05° resolution from Potapov et al (2022) and mean field size categories. Selected zoom-ins show diverse configurations of field size and deforestation (A)–(D). Scatterplots show the relationship between cropland fraction and net forest cover loss (E) and mean field size and forest cover loss (F). Point colors relate to baseline forest cover (2010) and point size reflects population density. Trend lines smoothed using locally weighted regressions (loess) weighted by 2010 forest cover across 0.05° grid cells (N = 14 144). Trends over full forest cover change value range are depicted in figure S9.“$

Earth Observation Lab@humboldteolab· May 4

Most fields are tiny, but large fields take up a lot of land. 78% of fields are under 0.5 ha. Nevertheless, fields above 1 ha account for 37% of total cropland area. The far end of the field size distribution matters more for land use than suggested by plain field counts.

$Two stacked bar charts illustrating the difference between field counts and land area. The data shows that although nearly 80% of fields are smallholder plots under 0.5 hectares, larger fields over 1 hectare account for over a third of the total cropland area. „Fig 2. Fractions of fields and cropland area by field size category at national scale. Fraction of fields is the number of fields within a size category divided by the total number of fields, fraction of cropland area is cropland area per size category relative to the total mapped cropland area.“$

Earth Observation Lab@humboldteolab· May 4

There’s more cropland than global maps suggest. Our error-adjusted estimate puts active cropland at around 76,300 km², well above ESA WorldCover (50,900 km²) or GLAD (40,900 km²). We found agricultural activity in frontier regions that are home to an estimated 1.5-2.8 million people.

$A data graphic comparing Mozambique cropland mapping datasets. A large heat map shows national cropland density, while a side-by-side comparison grid demonstrates that the new MozFields 2023 model captures many more smallholder farm plots than existing global maps (WorldCover and GLAD), revealing much higher levels of agricultural activity than previously recorded. „Fig 1. Mapped cropland fraction in MozFields 2023 and zoom-ins of diverse agricultural landscapes with MozFields 2023, WorldCover 2021 and GLAD 2019 cropland extent overlaid (right). Reproduced with permission from SPOT6/7 data © AIRBUS DS (2023) accessed via Descartes Labs ©.“$

Earth Observation Lab@humboldteolab· May 4

Smallholder farming systems are notoriously difficult to map from space. Landscapes are fragmented and global maps often disagree, yet data on where farming actually happens is fundamental to designing effective land use and sustainability policies.  A few things stood out in the data! 📊 ⬇️

Earth Observation Lab@humboldteolab· Apr 29

Don't miss tomorrow's EOLab contribution to the "National Forum for Remote Sensing and Copernicus"! ➡️ archiv.d-copernicus.de/infothek/ver...

Graphic announcement for a session titled "Hyperspektrale Anwendungen" (Hyperspectral Applications) at the National Forum for Remote Sensing and Copernicus.
Details listed:
Moderation: Prof. Dr. Patrick Hostert (Humboldt-Universität zu Berlin).
Presentations: 1. "EnMAP: Wissenschaftliche Highlights vier Jahre nach dem Start" by Dr. Akpona Okujeni (GFZ Helmholtz Center for Geosciences). 2. "EnMAP-Box: hyperspektrale Fernerkundungsdaten mit QGIS in Wert setzen" by Dr. Benjamin Jakimow (Humboldt-Universität zu Berlin).
Time & Location: Thursday, April 30, 2026, at 13:30. Fachsessions Block H, Room: EnMap.
Visuals: The design features a dark blue background with a white central card, Earth Observation Lab pictogram in the bottom right corner.

Earth Observation Lab@humboldteolab· Apr 23

New publication out now! @philrufin.bsky.social, Pauline Hammer et al. mapped 17M fields in Mozambique using deep learning and satellite imagery. With 78% of fields being <0.5 ha, this provides critical data for sustainable land use policies! Read more here: iopscience.iop.org/article/10.1...

$Announcement graphic for a scientific publication in Environmental Research Letters titled "National-scale field delineation in Mozambique refines our understanding of cropland distribution, field size, and deforestation actors" by Rufin, Hammer et al. The central feature is a map of Mozambique color-coded by Mean Field Size, ranging from 0.01 hectares (tan) to 2.50 hectares (teal). Additionally, various maps and statistical graphs regarding mean field size show close-up field boundaries across the country and correlate field size with variables such as cropland fraction, travel time, and population density. The background is dark navy blue, right bottom corner shows the Earth Observation Lab pictogram, a satellite orbiting a globe.$

Earth Observation Lab@humboldteolab· Feb 16

Advancing optical #EarthObservation for #EU policies! We systematically link EU land-related agricultural & environmental legislation to EO-derived variables from next-generation optical missions (CHIME, Sen-2 NG) and assess their TRLs for operational uptake. ➡️ link.springer.com/article/10.1...

Announcement of a new paper by Berger et al. titled "Improving spectral Earth observations for EU policies: needs, opportunities, recommendations," published in Environmental Sciences Europe. The image features an artist's impression of the CHIME satellite in orbit, projecting a green scan beam onto the Earth’s surface. The image is from the website https://sentiwiki.copernicus.eu/web/chime.

Earth Observation Lab@humboldteolab· Feb 4

EOLab’s Florian Poetzschner is presenting current progress of Field delineation and water use assessment for EOAgriTwin - ‘A Digital Twin for Agriculture under multiple stressors’ at ESA's Open Science Meeting in Frascati Find out more about the project: www.eoagritwin.eu

Earth Observation Lab@humboldteolab· Jan 25

The #GreenGrass 2.0 Project aims to enhance the resilience and long-term protection of grassland pasture systems in Germany with smart farming systems. Check out our new StoryMap to see how #remotesensing helps to develop these sustainable pasture management strategies ➡️ arcg.is/1LOKDC1

A scientific infographic titled "Leverage multi-decadal Landsat archive" illustrating how satellite data tracks grassland changes.
Credits: Image: Jonas Alsleben. Graphic: Alsleben, J., Jauker, F. Hostert, P. (November 07, 2025). Satellite-based grassland diversity indicators on a landscape scale (unpublished poster). AdZ Meeting Berlin.

Earth Observation Lab@humboldteolab· Jan 21

From local fields to national scales: EOLab’s @jalsleben.bsky.social is contributing remote sensing expertise to the interdisciplinary GreenGrass 2.0 Project! 🌱🐄🛰️ Find out more about the project, research challenges, and methodological approaches in this StoryMap: arcg.is/1LOKDC1 #EarthObservation