linomenesesJr.

Geographer🌎 #mapping
GIS enthusiast 💻 (Qgis)
#R c ( Spatial Analysis, Spatial Statistics) 📊

[Foro Nacional Ambiental @FNAmbiental]

📖 Manual de restauración ecológica en la Orinoquia colombiana.
- Una herramienta para orientar a quienes trabajan, habitan y toman decisiones en ese territorio.

"La restauración ecológica es más que una acción ambiental, es un acto de compromiso con el presente y el futuro"

-> Fuente: @fundacionnatura @MinAmbienteCo
https://t.co/8q1LqN5uJS

[Physiologia Plantarum @PPLplantarum]

Check out our newly compiled Special Issue compiling reviews on "Methodologies to Assess Crop Stress Resilience"
https://t.co/HBHe1iHKBg

You can read the editorial here https://t.co/r1wjiB8DL4

All articles are free to read for 2 months.

[Qiusheng Wu @giswqs]

Working with Planet Tanager Hyperspectral Data (426 Bands) in QGIS with HyperCoast

Learn how to access, visualize, and analyze Planet’s Tanager hyperspectral imagery in QGIS using the HyperCoast plugin. In this tutorial, you will explore freely available Planet Tanager open data with over 426 spectral bands and learn how to stream imagery, download HDF5 datasets, inspect spectral signatures, and perform hyperspectral visualization directly inside QGIS.

Video Tutorial: https://t.co/QfPIevlsNR
HyperCoast: https://t.co/1GPh3h73Rv
QGIS Plugin: https://t.co/6CB4ApMkhd
Tanager STAC Repo: https://t.co/8Bn8I9T5bR
Tanager STAC Browser: https://t.co/ll9Hq1VEQS

#opendata #geospatial #hypercoast #QGIS #Tanager #hyperspectral

[Luis Berbiela @InForestMed]

Muy interesante mapa nacional de alta resolución sobre la vulnerabilidad de la biomasa forestal frente a incendios forestales en #Francia.
https://t.co/4ZIzWxDvVs

[Valeriy M., PhD, MBA, CQF @predict_addict]

The best probability books are always short & sweet.

#math

[Yohan @yohaniddawela]

A crop-yield model looked useful at R² = 0.30, then collapsed below zero the moment it had to predict yields across an African border.

That’s the surprising result from a new paper on maize-yield prediction in sub-Saharan Africa.

The researchers tested 6,404 smallholder maize observations from Kenya, Malawi, Nigeria, Rwanda, and Tanzania, covering the years 2017 to 2022.

The basic question was: can modern geospatial AI models trained in some countries predict crop yields in a country they’ve never seen before?

That question matters because food-security systems often need exactly that. The countries with the greatest need for early yield forecasts are often the countries with the weakest ground data. If a satellite model only works where you already have good field surveys, its usefulness is limited.

So the paper set up a harder test.

The researchers compared three ways of describing each maize field from satellite data.

First, a standard Sentinel-2 baseline using satellite bands, vegetation indices, and rainfall. This is the kind of engineered feature set used in many practical remote-sensing workflows.

Second, Prithvi-EO embeddings. Prithvi-EO is the IBM and NASA geospatial foundation model trained on Earth observation data. In simple terms, it has already looked at a large amount of satellite imagery and learned general patterns about land, crops, water, vegetation, and surface change.

Third, ViT-Base embeddings. That’s a general vision model trained on ordinary images, included as a comparison against a model built specifically for Earth observation.

The researchers then tested these feature sets with three common machine-learning models: Ridge regression, Random Forest, and XGBoost.

Under the standard benchmark, the results looked respectable.

With random five-fold cross-validation, all models achieved positive R² values between 0.17 and 0.30. Prithvi-EO with XGBoost came out on top at R² = 0.300.

That sounds like a decent result until you look closer.

The ordinary Sentinel-2 baseline with XGBoost scored R² = 0.291. So a 100-million parameter geospatial foundation model beat a simple hand-engineered satellite baseline by only 0.009 R² points under the easier test.

Then the researchers ran the test that actually matters for deployment.

They used leave-one-country-out validation.

This is a much more realistic test. It essentially asks whether the model can travel to a new national context instead of interpolating inside a familiar pooled dataset.

That’s where everything broke.

Every model produced negative R²....

Spectral features failed, Prithvi-EO failed, ViT-Base failed, etc.

The best result was Prithvi-EO with Ridge at R² = -0.027. The worst was spectral features with Ridge at R² = -0.093.

A negative R² is pretty nuts... It means the model is doing worse than a basic reference point based on the test data average. Or in simple terms, the model basically learned patterns that looked useful inside familiar countries, but those patterns didn’t survive contact with a new one.

The reason is visible in the yield numbers. Rwanda’s average yield was more than three times Nigeria’s. A model trained across these countries isn’t simply learning the relationship between satellite reflectance and maize productivity. It’s also absorbing country-level yield regimes.

That’s where random cross-validation can mislead.

When the same countries appear in both training and test folds, the model gets help from familiar national patterns. The benchmark looks like crop-yield prediction, but part of the performance comes from recognising the statistical environment it has already seen.

Leave-one-country-out removes that hidden support.

The paper finds a generalisation gap of roughly 0.22 to 0.38 R² points when moving from random cross-validation to cross-country testing. That’s the gap between a model that looks promising in a paper and a model that struggles in the setting where it would actually be used.

The foundation model result is the part people should sit with.

Prithvi-EO didn’t meaningfully outperform simple Sentinel-2 band medians for cross-country yield prediction. A specialised geospatial foundation model, used as a frozen feature extractor, couldn’t solve the transfer problem.

There’s an important caveat. Prithvi-EO was designed for multi-temporal satellite input, meaning it’s meant to see how land changes across time. This paper used a single growing-season composite, which removes much of the seasonal signal the model was built to use.

So the result should be read carefully.

It doesn’t prove Prithvi-EO has no value for agriculture. It shows that frozen foundation-model embeddings from one annual snapshot don’t automatically contain the country-invariant signal needed for crop-yield transfer.

That’s a useful finding because crop yield is a difficult target.

Satellites can see greenness, water stress, canopy structure, vegetation indices, and broad crop condition. They can’t fully see seed variety, fertiliser use, labour, pest pressure, soil nutrients, farm management, local institutions, or the way yield was measured in a survey.

Two maize fields can look similar from space and belong to very different agricultural systems.

That’s the broader lesson.

Anyway, the path forward is clear enough. Multi-temporal satellite inputs, fine-tuning on African agricultural data, soil and management covariates, country-level normalisation, domain adaptation, and meta-learning.

But the eval has gotta stay honest.

A model meant for food-security decisions shouldn’t be judged only on whether it can interpolate inside a familiar dataset. It should be judged on whether it can cross borders without falling apart.

Link to paper: https://t.co/mdMFBmmoBZ

[World Meteorological Organization @WMO]

This new app can be your time machine to revisit past weather.

Weather Replay from @copernicusecmwf enables anyone to revisit the weather anywhere on the globe, hour by hour, from January 1940 up to a few days before present.

Try it out 👉 https://t.co/ieoaI4fVs0

[Mashford Mahute @MashfordMahute]

𝗛𝗼𝘄 𝗜 𝗖𝗿𝗲𝗮𝘁𝗲𝗱 𝗧𝗵𝗶𝘀 “𝗠𝗶𝗻𝗶𝗺𝘂𝗺 𝗥𝗶𝘃𝗲𝗿 𝗧𝗲𝗺𝗽𝗲𝗿𝗮𝘁𝘂𝗿𝗲” 𝗠𝗮𝗽 𝗶𝗻 𝗤𝗚𝗜𝗦

🪡🧵

[Antonio Lupetti @antoniolupetti]

"Algorithms for Decision Making" is a free book about the mathematical foundations of artificial intelligence, autonomous decision systems and modern machine learning.

Published by MIT Press, the book connects probability, optimisation, planning, search, reinforcement learning, Markov decision processes, utility theory, and sequential decision-making in a rigorous yet modern way.

With more than 700 pages, it provides a remarkably broad view of how intelligent systems reason, evaluate uncertainty, and make decisions under constraints.

One of the most interesting aspects of the web is the enormous amount of high-quality free knowledge available today. Complex subjects that once required access to expensive institutions or specialised libraries are now accessible to anyone willing to study!

https://t.co/I9cHSCvvlm

[CONAE @CONAE_Oficial]

📢 Anunciamos el próximo llamado a inscripción al curso virtual de capacitación sobre uso de imágenes satelitales, destinado a personas interesadas en iniciarse en esta temática, sin necesidad de contar con experiencia previa.

✅ Curso de Teledetección Óptica 🛰️

📲 La inscripción en línea se abre el miércoles 15/04/2026 a las 10:00hs y permanece abierta hasta el 26/04/2026 o hasta agotar las 1.000 vacantes.

🖥 Cursado en modalidad virtual, del 27/04/2026 al 14/06/2026, nivel introductorio, no arancelado.

🔗 Encontrá la información completa aquí 👉 https://t.co/q4vR1oe3Pc

📧 Consultas: [email protected]

[Hydrology MDPI @Hydrology_mdpi]

#HiglyCited
The Implementation of Multimodal Large Language Models for Hydrological Applications: A Comparative Study of GPT-4 Vision, Gemini, LLaVa, and Multimodal-GPT

👉https://t.co/bUQemu50Xo
#GPT_4_Vision #hydrology #intelligent_assistants

[ESA Climate Office @esaclimate]

Coastal cities, island nations, millions of people: all affected by a number measured in millimetres.

On 18 March, scientist Anny Cazenave will show how satellites track sea level rise on a global, regional and coastal scale, and what it tells us about our changing climate🌊

Register here: https://t.co/9YthvJybvW

[Isaac Corley @isaaccorley_]

Caleb (@calebrob6) and I are trying something new: we started a blog where we'll share our experiments, paper highlights, and explorations in geospatial machine learning including TorchGeo use-cases.

First posts are live at https://t.co/oHaixiRr28 and you can subscribe on Substack at https://t.co/YaQ3eaiwGd

[Mario Picazo @picazomario]

CUANDO LA CORRIENTE EN CHORRO SE DEFORMA

La corriente en chorro solía mantener el aire frío del Ártico separado del aire más cálido en latitudes más bajas. Una corriente en chorro distorsionada provoca un fuerte calentamiento del Ártico, mientras que las latitudes más bajas se enfrían, como lo ilustra la imagen de abajo, que muestra la anomalía de temperatura del 24 de enero de 2026 a las 18:00 h. Esto se ha denominado retroalimentación de "puertas abiertas", es como si se dejara abierta la puerta del congelador.

El aumento de la temperatura global conlleva numerosas retroalimentaciones, como un mayor vapor de agua en la atmósfera, la amplificación polar del aumento de temperatura y la distorsión de la corriente en chorro, que en ocasiones puede provocar temperaturas inusualmente bajas en los continentes del hemisferio norte.

Es importante destacar que la distorsión de la corriente en chorro puede a veces provocar que grandes cantidades de calor del océano sean transportadas al océano Ártico, calentando abruptamente el agua de este océano y amenazando con desestabilizar los hidratos de metano contenidos en los sedimentos del fondo marino, lo que da lugar a enormes erupciones de metano.

mapa. Climate Reanalizer - Thomas Reis

[WEkEO_dias @WEkEO_dias]

New to WEkEO or looking to deepen your skills?

WEkEO offers regular online trainings and workshops to help you:

📈 Discover Copernicus datasets
📈 Use cloud computing tools
📈 Build workflows for real world applications

Training is free and open to all users.

Check upcoming sessions and register here: https://t.co/mvft73IyR4

#CopernicusData #OpenScience

[Land MDPI @Land_MDPI]

🚀 New Special Issue Reprint in journal #Land Alert!

📚 Feature Papers for Land: Innovations, Data, and Machine Learning
👥 Edited by Prof. Dr. Chuanrong Zhang
👉 Explore the volume here:
https://t.co/Q5GI0lFTsC

#Land #MachineLearning #Geospatial #DataScience #RemoteSensing

[Colegio de Geografos @colegeografos]

FGUMA - Oferta formativa on-line QGIS @fguma @andalgeos https://t.co/6OzIxIboMG

#ventajascolegiales #qgis

[Rosana Ferrero 📈📊🙌 @RosanaFerrero]

🎉 Top Libros de Análisis Espacial con R (con acceso online gratuito) 🗺️
Si trabajas con SIG y quieres potenciar tus habilidades de análisis de datos geoespaciales, R es una excelente opción.

#rstats #datascience #rspatial #analytics #AnalisisEspacial #GIS #SIG #Geoespacial

[ESA Earth Observation @ESA_EO]

Three decades of satellite data reveal a surprising culprit behind tropical carbon loss.

Using data from @esaclimate’s RECCAP-2 and Biomass projects, scientists found that small deforestation clearings – often under two hectare – are behind more than half of total losses since 1990.

https://t.co/7c0w9FIKLg

📸LSCE–Y. Xu/ESA

[datos.gob.es – Dirección General del Dato @datosgob]

☀️10 repositorios de datos públicos sobre el clima
Descubre fuentes abiertas clave para acceder a información climática fiable y apoyar la acción frente al cambio climático. Lee aquí 👉 https://t.co/yRa5QssA40