Tag
Allen AI 发布 OlmoEarth v1.3 卫星嵌入模型，可为全球每个 10 米像素生成 128 维向量，并同时公开全球预计算嵌入数据与浏览器端嵌入探索工具，在 27 项评测中 20 项取得最佳成绩。
Google DeepMind introduces ForTy, a system that classifies every 10x10m patch of land on Earth into distinct forest and tree-crop types, highlighting how untouched natural forests store far more carbon and support more biodiversity than single-species timber plantations.
Google DeepMind, Google Research, WRI and IIASA released Forest Typology (ForTy) 2020, a 10-meter global map classifying every land patch into six forest and land-use types, plus a 2017–2025 time series showing what replaced natural forests, supporting the EU Deforestation Regulation.
本文评估了五个天气基础模型作为冻结编码器在增强岩石风化（ERW）碳去除量化中的迁移能力，其中基于 Aurora 1.5 构建的 TerraNova-ERW 在未见农田上表现最佳，支持少样本预测并可显著减少土壤采样需求。
NSIDC 深度科普文章解释“喀喇昆仑异常”现象：在全球多数山地冰川消融的背景下，喀喇昆仑地区的冰川自1990年代起出现小幅增厚，科学家提出了包括碎屑覆盖、天气模式等在内的多种解释，并讨论该异常在持续变暖下还能维持多久。
提出了Spectral-Spatial Interpretation (SSI)，一种在位置-通道玩家上进行全网格归因的方法，利用局部Shapley值生成光谱-空间归因矩阵，在HLS Burn Scar、CUB-200和EuroSAT数据集上优于现有基线方法。
该论文评估了基于循环神经网络(RNN)的Sentinel-2时间序列树种分类方法，并使用积分梯度(Integrated Gradients)进行可解释性分析，重点研究了纯林与混交林挪威云杉像元的区分，最佳RNN模型Macro-F1达0.851，优于XGBoost和随机森林。
论文提出 FireGraphNet，将12小时野火蔓延预测建模为有向 H3 图上的节点分类任务，在加州测试火灾上平均精度达到最强距离基线的1.7–1.9倍，并分析了图消息传递对预测能力的贡献。
YerevaNN researchers show that for image self-distillation, sample repetition effects are dataset-dependent and can violate standard scaling law assumptions — a 1.5% random subset of Maxar even outperforms the full 50M-image pool under equal compute — and propose a modified Chinchilla-style law accounting for unique samples and repetition.
该论文综述了人工智能在人道主义任务中的应用，涵盖气候变化、灾害响应、冲突监测、损失评估与流离失所追踪等场景，展示了AI提供可扩展数据驱动解决方案的潜力。
PyTorch Conference North America在圣何塞举办，Steve Pousty介绍今年新增的入门分论坛、海报展示、展位及协同活动，并将分享基于PyTorch的卫星影像遥感应用。
The paper introduces FairRSFM, a biome-aware benchmark and debiasing framework that reveals ecological performance disparities in remote-sensing foundation models, evaluating Prithvi-EO-2.0, SatMAE, and DOFA across four downstream datasets alongside mitigation baselines like BOLP, Dynamic Biome Reweighting, and GroupDRO.
GeoSET introduces the first generalist foundation model for SAR-to-EO image translation, pretrained on over 3 million SAR–EO pairs and adaptable to downstream datasets via LoRA (0.60% parameter updates), achieving state-of-the-art FID and DISTS across six benchmarks.
GeoCR introduces a generalist cloud removal model that unifies RGB-only and multispectral cloud removal (with optional SAR guidance) in a single flow transformer, trained on ten datasets with 883,331 cloud-free images, achieving state-of-the-art FID/DISTS without dataset-specific fine-tuning.
This paper introduces a modified Stable Diffusion 3 architecture with multimodal conditioning to refine photogrammetric digital surface models, improving elevation accuracy in French cities by reducing noise and errors.
This paper introduces Modl, a technique for creating cross-lingual remote-sensing multimodal large language models by composing domain and language LoRAs with mutual orthogonality, achieving superior performance without paired multilingual data.
This paper introduces an LLM-driven, training-free closed-loop paradigm for fusing dual-source encrypted POIs, improving location alignment and attribute matching for land-use/land-cover mapping, outperforming existing methods.
This paper proposes an alternative signal-centric method for processing sonar data in remote sensing applications, using CSV format and acoustic processing to reduce processing time by 91.18% and improve machine learning-driven object detection.
A systematic evaluation of the COTQ provincial land cover product for Quebec, comparing it with ESA WorldCover, ESRI LandCover, and Google DynamicWorld through structural consistency, spectral separability, and relative positioning analyses.
The paper introduces a weakly supervised multi-label proportion learning framework for sea-ice type prediction using SAR imagery and multimodal data, achieving significant accuracy improvements over supervised baselines.