Skip to content

GeoChrono: Benchmarking and Rethinking Long-Term Temporal Understanding in Remote Sensing

Jul 2026 · arXiv.org · Vol abs/2607.15768 · 0 citations · 50 references
Computer Science

TL;DR

GeoChrono is proposed, an MLLM with enhanced capabilities for tracing, memorizing, and reasoning about long-term geographic evolution, surpassing the leading commercial MLLMs by over 20%, while C2FComp reduces visual tokens by over 56% while retaining GeoChrono's 94.6% performance on ChronoBench.

Abstract

Remote sensing offers an unparalleled vantage point for observing the Earth's long-term surface evolution, yet it demands that a model not only perceive land cover at isolated moments, but also track changes, memorize evolution histories, and reason across time and space. However, existing studies lack a systematic evaluation that dissects these distinct competencies. To fill this gap, we introduce ChronoBench, a multidimensional benchmark that decomposes this task into four progressive cognitive levels (i.e., Land Cover Perception, Temporal Recognition, Long-Term Memory, and Spatio-Temporal Reasoning). The ChronoBench comprises 12 sub-tasks and 17,689 rigorously validated QA (Question-Answer) pairs. Extensive evaluations reveal that mainstream MLLMs fall drastically behind human experts, with Long-Term Memory emerging as the most critical bottleneck. Motivated by this finding, we further propose GeoChrono, an MLLM with enhanced capabilities for tracing, memorizing, and reasoning about long-term geographic evolution. Leveraging the physical prior that geographic parcels remain spatially fixed while their semantics evolve, we design a Temporal Trajectory Encoder~(TempEnc) that constructs per-location temporal trajectories for dedicated land cover evolution modeling, and we introduce a Coarse-to-Fine Token Compressor~(C2FComp) that adaptively preserves dynamic regions while compressing the static background. To support training, we also construct ChronoInstruct, a 104K-sample instruction-tuning dataset spanning all competency levels for training. GeoChrono achieves state-of-the-art performance on ChronoBench, surpassing the leading commercial MLLMs by over 20%, while C2FComp reduces visual tokens by over 56% while retaining GeoChrono's 94.6% performance. The code and data will be available at https://github.com/IntelliSensing/GeoChrono

View source

Similar papers

Preprint Aug 2026

GeoPhysAdapter: Scale-Matched Geophysical Adaptation for Cross-Domain Landslide Mapping with Vision Foundation Models

Newly triggered landslides rarely carry immediate annotations, so cross-domain transferability determines the value of landslide mapping for emergency response and regional risk assessment. Vision foundation models have strengthened representational transfer, yet on unseen regions, events, and data sources they still generate high-confidence false alarms. Terrain, material, and rainfall triggering can constrain such errors, but their supports are local, regional, and event-scale, so that resampling onto a 10~m grid misaligns them with the segmentation decision unit and compounds the uncertain geographic context problem (UGCoP). We propose GeoPhysAdapter, which anchors on a frozen vision foundation model, restricts terrain, material, and triggering to dense spatial guidance, regional modulation, and event-timing forcing, and applies bounded adaptation at two decision units, the pixel and the candidate landslide body, reverting exactly to the visual prediction where support is insufficient. On an event-isolated PILD dataset of four public sources, 55 global landslide events, and 7,890 test samples, 70.3% of cross-domain false-positive mass lies in near-pure spurious bodies of median equivalent diameter 207m, matching coarse-prior support rather than the pixel. Pixel-level adaptation removes a net 507,817 erroneous pixels and reduces error by 7.76%, whereas raising the decision unit to the candidate body, under identical samples, anchor, and baseline, increases error reduction to 23.99%, approximately 3.1 times the pixel-level effect, improves IoU by 0.031 (14.2% relative), and corrects 9.92 pixels per pixel harmed. The data and code are publicly available at: https://github.com/Liu-Zhihang/geophysadapter.

Zhihang Liu, Mei-Po Kwan, Jinlin Wu et al. · 0 citations
Open access Sep 2026

Remote Sensing Hydrology Revisited: Advances, Persistent Challenges, and a Science‐Driven Agenda for the Next Decade

In the WRR 50th Anniversary Special Collection, Lettenmaier et al. (2015, https://doi.org/10.1002/2015WR017616 ) provided a timely account of how satellite remote sensing evolved from a niche capability in hydrology into a mainstream driver of hydrologic discovery, as the community shifted from imagery toward validated Earth Observation (EO) products and open‐access data systems. In this Commentary for the WRR 60th Anniversary Special Collection, we revisit and extend that synthesis and argue that the next step is not just more EO data, but a more accountable remote sensing hydrology. The challenge is to integrate remote sensing more directly into hydrologic science questions in ways that ground networks alone cannot provide consistently across scales. We center this Commentary on a core question: How can EO be used, credibly and consistently, to quantify where water is stored, how it moves, and how the hydrologic cycle is changing, in ways that improve hydrologic understanding and decision value beyond what ground networks alone can provide? We summarize key advances since 2015 that sharpen water‐cycle observability, expand analysis‐ready data systems and computation, and accelerate hybrid inference, while highlighting persistent challenges that continue to limit credible hydrologic inference near thresholds and extremes. We conclude with a science‐driven agenda focused on closing observability gaps, improving record coherence, advancing uncertainty‐aware inference, strengthening end‐to‐end traceability from measurements to societal impacts, and sustaining climate‐quality EO records for hydrologic science in a changing climate, while recognizing that these advances will depend on continuity of observing capability and credible follow‐on mission strategies.

Yang Hong, D. Lettenmaier, E. Foufoula‐Georgiou et al. · 0 citations
#machine learning Preprint Sep 2026

Chronosphere: Space-Time Tessellation of Local Climate Experts

We introduce Chronosphere, a spatio-temporal neural field that learns representations of climate. A central challenge in geographic representation learning is modeling environmental processes whose spatial and temporal complexity varies widely. Yet existing location encoders typically fix a single level of detail everywhere. Global bases such as spherical harmonics spread capacity uniformly across space and time. Localized bases resolve only predefined regions. Learned tessellations adapt, but are inefficient at representing higher frequencies. Chronosphere unifies these approaches, pairing an adaptive tessellation of learnable sites on the spacetime torus $S^2\times S^1$ with a shared bank of local basis functions. Both where capacity is placed and how much detail each region carries adapt to the data, across space and time. Trained to reconstruct climatology, Chronosphere matches or leads state-of-the-art location encoders across spatial and temporal tasks, with the largest gains under spatial and temporal transfer.

Dan Cher, Eric P. Xing, Ke-Xing Li et al. · 0 citations
Open access Aug 2026

Resolving circumarctic zero-curtain phenomena with AI-integrated earth observations

Across the circumarctic, permafrost landscapes store approximately 1700 billion metric tons of organic carbon—nearly twice atmospheric levels—yet reservoir stability depends on the zero-curtain, a subsurface thermal plateau that emerges when latent heat maintains soil temperatures near 0 °C during freeze–thaw phase transitions. The zero-curtain sustains liquid water through cryosuction within the active layer, enabling microbial activity to persist well into the cold season and regulating permafrost thermal stability. However, zero-curtain intensity, duration, and spatial extent remain inadequately quantified during transitional seasons when isothermal buffering exhibits maximum variability, limiting our understanding of permafrost-climate feedbacks. Using GeoCryoAI—a hybridized physics-informed (PI) transfer learning framework that integrates 62.71 million in situ measurements and 3.3 billion remote sensing observations—we show that zero-curtain candidate phenomena exhibit pronounced seasonal asymmetry, with extended vernal intensification (1000–4000 h) relative to compressed winter occurrence (100–500 h), and significant longitudinal variation: moderate intensity patterns across the North American Arctic, enhanced vernal amplification in Siberia, reduced winter suppression in Fennoscandia, and a delayed vernal response in the Canadian Archipelago. The framework quantifies these dynamics at statistically downscaled 30 m gridded resolution, achieving 96.4% candidate detection accuracy. Ablation studies confirm that the full GeoCryoAI architecture achieves 11.8% improvement over baseline multilayer perceptron architectures (93.4% vs. 81.6% in component validation experiments), with PI constraints providing essential regularization for thermodynamic consistency. Mechanistic analysis reveals soil moisture-latent heat coupling is the dominant framework-identified predictor of 60–90% duration variability; within the sampled range, this dependence is 20–40% stronger under warmer, high moisture conditions. This framework establishes a NISAR-ready circumarctic monitoring protocol, enabling 3–6-month forecasts and spatially explicit boundary inputs for Earth system models simulating carbon-climate feedbacks in a warming Arctic.

B. Gay, K. Miner, N. Rietze et al. · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.