Pith. sign in

REVIEW 3 major objections 8 minor 63 references

CarbonCLIP: Enhance Carbon Prediction from Satellite Imagery via Integrated Street-View Semantics and Temporal Context Training

T0 review · 3 major / 8 minor · reviewed 2026-07-09 · glm-5.2

Pith's one-line read Street-view text distillation lifts satellite carbon prediction to R²=0.73

desk verdict Solid multimodal distillation idea for satellite-based carbon prediction, but the train/test split protocol is unspecified — a foundational gap for a spatiotemporal prediction paper. read the letter →

arxiv 2607.07292 v1 pith:EKN447MZ submitted 2026-07-08 cs.CV cs.AI

classification cs.CVcs.AI
keywords carbonclipsatellitecarbonmultimodalurbandatatemporalactivities
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proposes CarbonCLIP, a framework that improves monthly urban carbon emission prediction from satellite imagery by distilling two kinds of knowledge into satellite features during pretraining: ground-level urban semantics and month-level temporal context. The spatial branch aligns satellite image embeddings with textual descriptions that a Large Multimodal Model automatically generates from street-view panoramas, capturing building functions, infrastructure, and street-level activity that top-down satellite views cannot see directly. The temporal branch aligns satellite features with learnable month embeddings that encode cyclic seasonal patterns in emissions. After this dual-branch contrastive pretraining, the street-view text and month context are discarded: the downstream carbon predictor uses only satellite imagery as input. The central design claim is that using LMM-generated text descriptions as the alignment medium—rather than raw street-view image embeddings—produces a stronger satellite representation, because textual abstraction filters out view-specific noise such as differing panorama directions, occlusions, and uneven sampling within a 1 km² tile. Experiments on Beijing and Singapore show the method reaches R² of 0.728 and 0.704 respectively, outperforming ResNet, ViT, and UrbanCLIP baselines, and a targeted ablation confirms that text-based supervision outperforms direct street-view image alignment in both cities.

What carries the argument

Dual-branch contrastive learning: (1) an image-text branch that aligns satellite spatial embeddings with CLIP text embeddings of LMM-generated street-view descriptions via symmetric contrastive loss, and (2) a temporal branch that aligns satellite temporal embeddings with month embeddings via a soft contrastive loss using circular month distance and Gaussian kernel weighting. The month encoder combines multi-frequency sinusoidal encoding of month position with learnable coarse calendar-group embeddings, blended by a cosine transition weight. At inference, the frozen satellite encoder produces both spatial and temporal embeddings through separate projection layers, which are concatenated and喂

What would settle it

If a future study finds that direct street-view image alignment matches or exceeds text-based alignment in cities with different morphology or denser, more uniform street-view coverage, the central design claim—that textual abstraction is the superior alignment medium—would be weakened.

Watch

Extended reading notes

Core claim

The paper's central finding is that ground-level semantic knowledge and temporal context can be transferred into a satellite-only representation through contrastive pretraining, and that LMM-generated textual descriptions of street-view imagery serve as a more effective alignment target than raw street-view image embeddings. The text abstraction collapses heterogeneous panorama-level visual cues within a spatial tile into a coherent functional description, reducing view-specific noise before alignment with satellite features. This distilled representation, combined with a cyclic month encoder that captures recurrent temporal patterns, allows a model that sees only satellite imagery at test时间

Load-bearing premise

The framework assumes that LMM-generated textual descriptions of street-view panoramas provide a more effective alignment target for satellite features than direct street-view image embeddings. This is supported by a targeted ablation in two cities, but the result depends on the specific text generator, prompt design, and street-view coverage used. If text quality degrades for cities with substantially different urban morphology or sparse street-view data, the semantic disti

Editorial extensions

If this is right

  • If the text-as-alignment-medium claim generalizes, cities without street-view coverage at inference could still benefit from street-level semantic supervision gathered during a one-time pretraining phase, lowering the data barrier for satellite-based carbon monitoring in data-scarce regions.
  • The finding that abstracted text outperforms raw image embeddings as a contrastive target suggests that LMM-generated descriptions may serve as a general-purpose semantic bridge for other cross-view or cross-modality geospatial alignment tasks beyond carbon prediction.
  • The cyclic month encoder design could be extended to finer temporal granularity (weekly, daily) if higher-resolution emission inventories become available, potentially capturing short-term emission drivers the current discrete-month formulation misses.
  • The asymmetric training paradigm—multimodal supervision during pretraining, single-modality inference—offers a template for other environmental prediction tasks where auxiliary data is available for training cities but not for deployment cities.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 8 minor

Summary. The manuscript proposes CarbonCLIP, a contrastive learning framework that distills street-view-derived textual semantics and month-level temporal context into satellite image representations for monthly urban carbon emission prediction. The key design choice is asymmetric: multimodal data (street-view text, month embeddings) are used only during pretraining, while inference relies solely on satellite imagery. The spatial branch aligns satellite features with LMM-generated street-view descriptions via CLIP-style contrastive learning; the temporal branch aligns satellite features with cyclic month embeddings using a soft Gaussian-kernel contrastive loss. A lightweight MLP predictor is then trained on frozen satellite embeddings for monthly emission regression. Experiments on Beijing and Singapore using Planet satellite imagery and ODIAC emission data show improvements over ResNet, ViT, and UrbanCLIP baselines, with ablations supporting the contribution of both spatial and temporal branches.

Significance. The paper addresses a practically important problem: scalable, satellite-only carbon emission prediction that benefits from ground-level semantic knowledge without requiring auxiliary data at inference. The asymmetric training-inference design is well-motivated and practically valuable for deployment in data-scarce regions. The dual-branch contrastive framework with LMM-generated text as semantic anchors is a reasonable and novel combination for this task. The controlled evaluation protocol (same splits, labels, and regressor across all baselines) and the targeted ablation comparing text-based vs. direct image-based street-view alignment (Table 8) are commendable. The dataset construction pipeline and plan for public release are additional strengths. However, the empirical foundation has a critical gap regarding the train/test split protocol, which must be addressed before the central claims can be fully evaluated.

major comments (3)
  1. §4.1.3: The train/validation/test split strategy is never described. The paper states that all methods use 'the same train, validation, and test partitions' but does not specify whether the split is performed at the tile-month level, the tile level, or with spatial/temporal blocking. For a spatiotemporal prediction task with 1 km² tiles and monthly observations, this is foundational. If the same tiles appear in both train and test sets (for different months), the model can learn tile-specific emission baselines rather than generalizable satellite-to-emission relationships. If tiles are randomly assigned without spatial blocking, spatial autocorrelation between neighboring tiles could inflate all reported R² values. This concern is load-bearing because the central claim — that CarbonCLIP outperforms baselines — cannot be properly interpreted without knowing whether the evaluation protocol
  2. §4.2, Table 7: Results are reported for only two cities (Beijing and Singapore). While these represent different climatic regimes, the generalizability claim ('scalable deployment when ground-level data are unavailable at inference') is not substantiated by testing in cities where street-view coverage is sparse during pretraining. The paper acknowledges in §6.2 that pretraining still depends on street-view data, but the experimental validation does not test the transfer scenario. At minimum, the authors should qualify their claims or provide cross-city transfer experiments (e.g., pretrain on one city, test on another) to support the scalability argument.
  3. §3.2, Eqs. (1)–(6): The month encoder introduces several hyperparameters (n, γ, T, φ, σ, τ, calendar-group definitions) whose values are not reported. The calendar groups differ between cities (four for Beijing, two for Singapore), which means the temporal model is city-specific. This raises a question about whether the temporal branch's contribution comes from the encoding design or from city-specific tuning. The authors should report all hyperparameter values and ideally test sensitivity to the calendar-group definition.
minor comments (8)
  1. §4.1.1: The ODIAC dataset is model-derived (not directly measured emissions), which means the prediction targets themselves contain modeling assumptions. This should be acknowledged as a limitation on the interpretability of all reported metrics.
  2. Table 4: The qualitative comparison of Qwen2.5-VL model sizes is informative but the manual scoring in Table 5 is based on only 20 samples. The authors should note this is a preliminary check rather than a rigorous evaluation.
  3. §3.1: The prompt instruction includes 'ignore the weather and other details that change over a short period of time,' but the temporal branch relies on monthly variation. There is a potential tension here: if weather-related visual cues are suppressed in the text descriptions, the spatial branch may lose information relevant to seasonal emission drivers. This should be discussed.
  4. Figure 3: The x-axis labels are rendered as glyph codes (e.g., /uni0000002d) and are unreadable. This must be fixed for the camera-ready version.
  5. Figure 4: Same rendering issue as Figure 3 — axis labels are unreadable.
  6. §4.1.3: The downstream regressor architecture (3-layer MLP with 512, 256, 1 neurons) is described, but the learning rate, optimizer, batch size, and total training epochs are not reported. These should be included for reproducibility.
  7. Table 6: UrbanCLIP uses a CoCa-ViT-L/14 backbone while CarbonCLIP uses ViT-B/32. This backbone difference is not discussed as a potential confound. A brief note acknowledging this would strengthen the analysis.
  8. §5.2, Figure 5: The UMAP visualization is qualitative and city-specific. The claim that 'part of the street-level semantic information introduced during pretraining is reflected in the satellite representation' would be strengthened by a quantitative nearest-neighbor or retrieval evaluation rather than visual inspection alone.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: pretraining supervision and prediction targets are independent; all comparisons use external benchmarks

full rationale

The paper's derivation chain is self-contained and does not exhibit circular reasoning. The contrastive pretraining objectives (Eqs. 7 and 9) use street-view-generated text embeddings and month indices as supervision signals, both of which are independent of the downstream prediction target (ODIAC carbon emissions). The carbon emission predictor (Eq. 12) minimizes MSE against ODIAC labels, which are a separate external dataset. The central empirical claim—that CarbonCLIP outperforms ResNet, ViT, and UrbanCLIP—is validated against external baselines under a controlled evaluation protocol (Table 7). The ablation in Table 8 comparing text-based vs. direct image-based street-view alignment provides independent evidence for the design choice, as the two variants use different alignment targets while keeping all other components fixed. While the paper does cite UrbanCLIP [48] for prompt design inspiration, this citation is not load-bearing for the central claim—it is a methodological borrowing, not a self-citation that defines the result. The temporal month encoder (Eqs. 1-6) is a standard positional encoding construction, not a renaming of a known empirical result. No 'prediction' or 'first-principles result' reduces by construction to its inputs.

Assumptions & free parameters 8 free parameters · 4 assumptions · 1 invented entities

The paper introduces several hyperparameters whose values are not specified, making exact reproduction difficult. The core axioms are domain assumptions typical of remote-sensing ML papers but are only validated on two cities. The month encoder is a genuinely new component but lacks comparison against simpler temporal encoding baselines.

free parameters (8)
  • Temperature τ (image-text contrastive loss) = Not explicitly stated
    Used in Eq. 7 to scale logits; value not specified in the paper.
  • Temperature τ (temporal alignment loss) = Not explicitly stated
    Used in Eq. 9; value not specified.
  • σ (Gaussian kernel width for temporal similarity) = Not explicitly stated
    Controls temporal similarity window in Eq. 8; value not specified.
  • γ (max log-frequency for month encoder) = Not explicitly stated
    Controls frequency range in Eq. 1; value not specified.
  • n (number of base frequencies) = Not explicitly stated
    Determines dimensionality of temporal base feature in Eq. 2.
  • T and φ (cosine transition parameters) = Not explicitly stated
    Control periodicity and phase shift in Eq. 5 for blending fine/coarse temporal features.
  • Calendar group definitions = Beijing: 4 groups; Singapore: 2 groups
    Manually defined seasonal partitions used in Eq. 4.
  • Label scaling constant = 1000
    Emission labels divided by 1000 for numerical stability (Section 3.4).
assumptions (4)
  • domain assumption Street-view textual descriptions generated by LMMs provide semantically meaningful supervision for satellite image representation learning.
    Core premise of Section 3.1; supported by ablation (Table 8) but not independently verified beyond two cities.
  • domain assumption ODIAC gridded carbon emission data provides accurate ground-truth labels for model training.
    Used as ground truth throughout; ODIAC is model-derived and has known uncertainties not discussed.
  • domain assumption Monthly satellite imagery with <10% cloud cover captures representative surface conditions for that month.
    Section 4.1.1 describes manual selection of one image per month; assumes single image is representative.
  • domain assumption Contrastive alignment between satellite features and street-view text embeddings transfers useful semantic information.
    Foundational assumption of the spatial branch (Section 3.3.1); supported by performance gains but mechanism not independently verified.
invented entities (1)
  • Month encoder with multi-frequency sinusoidal encoding and coarse calendar-group interpolation independent evidence
    purpose: Encode cyclic temporal context for monthly emission prediction
    The month encoder is a novel architectural component (Section 3.2) with falsifiable predictions: temporal embedding similarity patterns (Fig. 6, Table 9) show city-specific structure consistent with climate differences. However, it is not validated as superior to simpler alternatives (e.g., standard month embeddings).

how reviews work

0 comments
Cite this review

Pith. "Pith review of CarbonCLIP: Enhance Carbon Prediction from Satellite Imagery via Integrated Street-View Semantics and Temporal Context Training." pith.science (2026). https://pith.science/paper/EKN447MZ

@misc{pith2026260707292,
  author       = {Pith},
  title        = {Pith review of: CarbonCLIP: Enhance Carbon Prediction from Satellite Imagery via Integrated Street-View Semantics and Temporal Context Training},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EKN447MZ}},
  note         = {Machine review of arXiv:2607.07292}
}
read the original abstract

Accurately estimating urban carbon emissions is critical for sustainable urban planning, yet many existing approaches remain difficult to apply consistently across cities due to data-source heterogeneity and the lack of fine-grained semantic-temporal context in remote sensing data. We propose CarbonCLIP, a task-oriented multimodal distillation framework that improves satellite-based carbon emission prediction by transferring contextual knowledge into a unified satellite representation through dual-branch contrastive learning. Unlike conventional methods that rely on static visual features, CarbonCLIP explicitly bridges the gap between top-down satellite views and ground-level human activities. Specifically, the spatial branch uses fine-grained textual descriptions automatically generated from street-view images by Large Multimodal Models (LMMs) to provide semantic priors reflecting building functions, infrastructure, and urban activities, while the temporal branch employs a month encoder to encode temporal priors associated with monthly emission variation. CarbonCLIP requires multimodal data only during the pretraining phase; during inference, it relies solely on satellite imagery, thereby supporting scalable deployment when ground-level data are unavailable at inference. Experiments on Beijing and Singapore demonstrate that CarbonCLIP outperforms baselines in both study cities. The results validate that our method effectively transfers multimodal knowledge into satellite representations, offering a robust solution for satellite-based urban carbon modeling.

Figures

Figures reproduced from arXiv: 2607.07292 by the authors.

Figure 1
Figure 1. Comparison of frameworks for carbon emission prediction: (a) Simple transfer learning framework based on [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overall framework of our proposed CarbonCLIP. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Monthly carbon emission R2 comparison of different methods on Beijing and Singapore. aligning satellite features with street-view-generated text and temporal embeddings during pretraining. Therefore, with￾out explicitly using street-view imagery at inference, CarbonCLIP learns satellite representations that are semantically richer than traditional CNN or transformer backbones. 4.3 Ablation Studies As shown in [PITH… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Ablation studies on Beijing and Singapore. [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]
Figure 5
Figure 5. Figure 5: UMAP visualization of CarbonCLIP satellite embeddings. Each point represents a 1 km [PITH_FULL_IMAGE:figures/full_fig_p015_5.png]
Figure 6
Figure 6. Figure 6: Similarity of temporal embeddings across months in Beijing (left) and Singapore (right). [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

63 extracted references · 63 canonical work pages

  1. [1]

    Crucial factors of the built environment for mitigating carbon emissions.Science of The Total Environment, 806:150864, 2022

    Yu-Sheng Shen, Ying-Chen Lin, Shenghui Cui, Yanmin Li, and Xingxing Zhai. Crucial factors of the built environment for mitigating carbon emissions.Science of The Total Environment, 806:150864, 2022

  2. [2]

    Global anthropogenic emissions in urban areas: patterns, trends, and challenges.Environmental Research Letters, 16(7):074033, jul 2021

    Monica Crippa, Diego Guizzardi, Enrico Pisoni, Efisio Solazzo, Antoine Guion, Marilena Muntean, Aneta Florczyk, Marcello Schiavina, Michele Melchiorri, and Andres Fuentes Hutfilter. Global anthropogenic emissions in urban areas: patterns, trends, and challenges.Environmental Research Letters, 16(7):074033, jul 2021. 17 CarbonCLIP

  3. [3]

    Enabling technologies and sustainable smart cities.Sustainable Cities and Society, 61:102301, 2020

    Mohd Abdul Ahad, Sara Paiva, Gautami Tripathi, and Noushaba Feroz. Enabling technologies and sustainable smart cities.Sustainable Cities and Society, 61:102301, 2020

  4. [4]

    Using convolutional networks and satellite imagery to identify patterns in urban environments at a large scale

    Adrian Albert, Jasleen Kaur, and Marta C Gonzalez. Using convolutional networks and satellite imagery to identify patterns in urban environments at a large scale. InProceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 1357–1366, 2017

  5. [5]

    Satellite data for the air pollution mapping

    Lucien Wald, Ludovic Basly, and Jean-Michel Baleynaud. Satellite data for the air pollution mapping. In Operational Remote Sensing for Sustainable Development, pages 133–139. CRC Press, 2020

  6. [6]

    A review of satellite-based global agricultural monitoring systems available for africa.Global Food Security, 29:100543, 2021

    Catherine Nakalembe, Inbal Becker-Reshef, Rogerio Bonifacio, Guangxiao Hu, Michael Laurence Humber, Christina Jade Justice, John Keniston, Kenneth Mwangi, Felix Rembold, Shraddhanand Shukla, et al. A review of satellite-based global agricultural monitoring systems available for africa.Global Food Security, 29:100543, 2021

  7. [7]

    Reforestree: A dataset for estimating tropical forest carbon stock with deep learning and aerial imagery

    Gyri Reiersen, David Dao, Björn Lütjens, Konstantin Klemmer, Kenza Amara, Attila Steinegger, Ce Zhang, and Xiaoxiang Zhu. Reforestree: A dataset for estimating tropical forest carbon stock with deep learning and aerial imagery. InProceedings of the AAAI Conference on Artificial Intelligence, volume 36, pages 12119–12125, 2022

  8. [8]

    Planet application program interface: In space for life on earth

    Planet. Planet application program interface: In space for life on earth. https://api.planet.com, 2024. San Francisco, CA

Show all 63 references
  1. [9]

    Position: mission critical–satellite data is a distinct modality in machine learning

    Esther Rolf, Konstantin Klemmer, Caleb Robinson, and Hannah Kerner. Position: mission critical–satellite data is a distinct modality in machine learning. InForty-first International Conference on Machine Learning, 2024

  2. [10]

    Streetvizor: Visual exploration of human-scale urban forms based on street views.IEEE Transactions on Visualization and Computer Graphics, 24(1):1004–1013, 2017

    Qiaomu Shen, Wei Zeng, Yu Ye, Stefan Müller Arisona, Simon Schubiger, Remo Burkhard, and Huamin Qu. Streetvizor: Visual exploration of human-scale urban forms based on street views.IEEE Transactions on Visualization and Computer Graphics, 24(1):1004–1013, 2017

  3. [11]

    Mapping sky, tree, and building view factors of street canyons in a high-density urban environment.Building and Environment, 134:155–167, 2018

    Fang-Ying Gong, Zhao-Cheng Zeng, Fan Zhang, Xiaojiang Li, Edward Ng, and Leslie K Norford. Mapping sky, tree, and building view factors of street canyons in a high-density urban environment.Building and Environment, 134:155–167, 2018

  4. [12]

    Urban visual intelligence: Uncovering hidden city profiles with street view images.Proceedings of the National Academy of Sciences, 120(27):e2220417120, 2023

    Zhuangyuan Fan, Fan Zhang, Becky PY Loo, and Carlo Ratti. Urban visual intelligence: Uncovering hidden city profiles with street view images.Proceedings of the National Academy of Sciences, 120(27):e2220417120, 2023

  5. [13]

    Street view imagery in urban analytics and gis: A review.Landscape and Urban Planning, 215:104217, 2021

    Filip Biljecki and Koichi Ito. Street view imagery in urban analytics and gis: A review.Landscape and Urban Planning, 215:104217, 2021

  6. [14]

    Investigating the associ- ation between streetscapes and human walking activities using google street view and human trajectory data

    Xiaojiang Li, Paolo Santi, Theodore K Courtney, Santosh K Verma, and Carlo Ratti. Investigating the associ- ation between streetscapes and human walking activities using google street view and human trajectory data. Transactions in GIS, 22(4):1029–1044, 2018

  7. [15]

    3d building reconstruction from single street view images using deep learning

    Hui En Pang and Filip Biljecki. 3d building reconstruction from single street view images using deep learning. International Journal of Applied Earth Observation and Geoinformation, 112:102859, 2022

  8. [16]

    Using google street view to reveal environmental justice: Assessing public perceived walkability in macroscale city.Landscape and Urban Planning, 244:104995, 2024

    Yi Lu and Hui-Mei Chen. Using google street view to reveal environmental justice: Assessing public perceived walkability in macroscale city.Landscape and Urban Planning, 244:104995, 2024

  9. [17]

    Evaluating the multi-seasonal impacts of urban blue-green space combination models on cooling and carbon-saving capacities.Building and Environment, 266:112045, 2024

    Fei Yang, Dongdong Yang, Ying Zhang, Ru Guo, Jiaying Li, and Hongcheng Wang. Evaluating the multi-seasonal impacts of urban blue-green space combination models on cooling and carbon-saving capacities.Building and Environment, 266:112045, 2024

  10. [18]

    Estimating carbon dioxide emissions from power plant water vapor plumes using satellite imagery and machine learning.Remote Sensing, 16(7):1290, 2024

    Heather D Couture, Madison Alvara, Jeremy Freeman, Aaron Davitt, Hannes Koenig, Ali Rouzbeh Kargar, Joseph O’Connor, Isabella Söldner-Rembold, André Ferreira, Jeyavinoth Jeyaratnam, et al. Estimating carbon dioxide emissions from power plant water vapor plumes using satellite ...

  11. [19]

    Estimating carbon dioxide emissions in two california cities using bayesian inversion and satellite measurements.Geophysical Research Letters, 51(20):e2024GL111150, 2024

    Sofia D Hamilton, Dien Wu, Matthew S Johnson, Alexander J Turner, Marc L Fischer, Nikhil Dadheech, and Seongeun Jeong. Estimating carbon dioxide emissions in two california cities using bayesian inversion and satellite measurements.Geophysical Research Letters, 51(20):e2024GL1...

  12. [20]

    Estimating carbon emissions in urban functional zones using multi-source data: A case study in beijing.Building and Environment, 212:108804, 2022

    Yunqiang Zheng, Shihong Du, Xiuyuan Zhang, Lubin Bai, and Haoyu Wang. Estimating carbon emissions in urban functional zones using multi-source data: A case study in beijing.Building and Environment, 212:108804, 2022

  13. [21]

    Uncovering the spatiotemporal impacts of built environment on traffic carbon emissions using multi-source big data.Land Use Policy, 129:106621, 2023

    Jishi Wu, Peng Jia, Tao Feng, Haijiang Li, Haibo Kuang, and Junyi Zhang. Uncovering the spatiotemporal impacts of built environment on traffic carbon emissions using multi-source big data.Land Use Policy, 129:106621, 2023

  14. [22]

    Carbon emission estimation at the urban functional zone scale: Integrating multi-source data and machine learning approach.Energy and Buildings, page 115832, 2025

    Cui Wang, Liuchang Xu, Xinyu Zheng, Yiming Hua, and Xingyu Xue. Carbon emission estimation at the urban functional zone scale: Integrating multi-source data and machine learning approach.Energy and Buildings, page 115832, 2025. 18 CarbonCLIP

  15. [23]

    Urbanmllm: Joint learning of cross-view imagery for urban understanding

    Xin Zhang, Tianjian Ouyang, Yu Shang, Qingmin Liao, and Yong Li. Urbanmllm: Joint learning of cross-view imagery for urban understanding

  16. [24]

    Urbanvlp: Multi-granularity vision-language pretraining for urban socioeconomic indicator prediction

    Xixuan Hao, Wei Chen, Yibo Yan, Siru Zhong, Kun Wang, Qingsong Wen, and Yuxuan Liang. Urbanvlp: Multi-granularity vision-language pretraining for urban socioeconomic indicator prediction. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 28061–2...

  17. [25]

    Boan Chen, Quanlong Feng, Bowen Niu, Fengqin Yan, Bingbo Gao, Jianyu Yang, Jianhua Gong, and Jiantao Liu. Multi-modal fusion of satellite and street-view images for urban village classification based on a dual-branch deep neural network.International Journal of Applied Earth O...

  18. [26]

    A neural network model for forecasting co2 emission.AGRIS on-line Papers in Economics and Informatics, 6(2):31–36, 2014

    Crescenzio Gallo, Francesco Conto, and Mariantonietta Fiore. A neural network model for forecasting co2 emission.AGRIS on-line Papers in Economics and Informatics, 6(2):31–36, 2014

  19. [27]

    Tomohiro Oda, Shamil Maksyutov, and Robert J Andres. The open-source data inventory for anthropogenic co 2, version 2016 (odiac2016): a global monthly fossil fuel co 2 gridded emissions data product for tracer transport simulations and surface flux inversions.Earth System Scie...

  20. [28]

    Wei Wei, Dibo Chen, Xing Zhang, Libang Ma, Binbin Xie, Junju Zhou, Mintong Wang, Xiaoxu Wei, Shilong Yang, and Peng Yan. Estimation of carbon emissions from different industrial categories integrated nighttime light and poi data—a case study in the yellow river basin.Journal o...

  21. [29]

    Xuezhu Cui, Caigang Zhuang, Zhenzhi Jiao, Zhangzhi Tan, and Shaoying Li. How can urban built environment (be) influence on-road (or) carbon emissions? a road segment scale quantification based on massive vehicle trajectory big data.Journal of Transport Geography, 111:103669, 2023

  22. [30]

    Exploring spatio- temporal carbon emission across passenger car trajectory data.IEEE Transactions on Intelligent Transportation Systems, 2024

    Zhu Xiao, Bo Liu, Linshan Wu, Hongbo Jiang, Beihao Xia, Tao Li, and Cassandra C Wang. Exploring spatio- temporal carbon emission across passenger car trajectory data.IEEE Transactions on Intelligent Transportation Systems, 2024

  23. [31]

    Real time estimation of carbon emissions for industrial users based on load monitoring in advanced metering infrastructure.Journal of Cleaner Production, 483:144226, 2024

    Yunpeng Gao, Jiangzhao Wang, Yanqing Zhu, Wei Zhang, Fei Teng, and Yunfeng Li. Real time estimation of carbon emissions for industrial users based on load monitoring in advanced metering infrastructure.Journal of Cleaner Production, 483:144226, 2024

  24. [32]

    The estimation of building carbon emission using nighttime light images: A comparative study at various spatial scales.Sustainable Cities and Society, 101:105066, 2024

    Gengzhe Wang, Qing Hu, Linghao He, Jialong Guo, Jin Huang, and Lijin Zhong. The estimation of building carbon emission using nighttime light images: A comparative study at various spatial scales.Sustainable Cities and Society, 101:105066, 2024

  25. [33]

    What drives urban carbon emission efficiency?–spatial analysis based on nighttime light data.Applied Energy, 312:118772, 2022

    Guochang Fang, Zhengye Gao, Lixin Tian, and Min Fu. What drives urban carbon emission efficiency?–spatial analysis based on nighttime light data.Applied Energy, 312:118772, 2022

  26. [34]

    Correcting the saturation effect in dmsp/ols stable nighttime light products based on radiance-calibrated data.IEEE Transactions on Geoscience and Remote Sensing, 60:1–11, 2021

    Yang Hu, Jin Chen, Xin Cao, Xuehong Chen, Xihong Cui, and Liqin Gan. Correcting the saturation effect in dmsp/ols stable nighttime light products based on radiance-calibrated data.IEEE Transactions on Geoscience and Remote Sensing, 60:1–11, 2021

  27. [35]

    Jingyan Sun, Yanbing Qi, Jinwei Guo, Jiale Zheng, Luhao Zhang, and Xiangyun Yang. Impact of nighttime light data saturation correction on the application of carbon emissions spatialization: A comparative study of the correction effect and application effect based on five metho...

  28. [36]

    Phenological classification using deep learning and the sentinel-2 satellite to identify priority afforestation sites in north korea.Remote Sensing, 13(15):2946, 2021

    Joon Kim, Chul-Hee Lim, Hyun-Woo Jo, and Woo-Kyun Lee. Phenological classification using deep learning and the sentinel-2 satellite to identify priority afforestation sites in north korea.Remote Sensing, 13(15):2946, 2021

  29. [37]

    Inferring carbon dioxide emissions from power plants using satellite imagery and machine learning

    Madison Hobbs, Ali Rouzbeh Kargar, Heather Couture, Jeremy Freeman, Isabella Söldner-Rembold, André Ferreira, Jeyavinoth Jeyaratnam, Joseph O’Connor, Jordan Lewis, Hannes Koenig, Colin McCormick, Tiffany Nakano, Charmaine Dalisay, Aaron Davitt, Lee Gans, Christy Lewis, Gabriel...

  30. [38]

    Ai-powered computer vision for remote sensing and carbon emission detection in industrial and urban environments.Iconic Research and Engineering Journals, 7(10):490–505, 2024

    Jessica Obianuju Ojadi, Chinekwu Somtochukwu Odionu, Ekene Cynthia, and Olumide Akindele Owulade Onukwu. Ai-powered computer vision for remote sensing and carbon emission detection in industrial and urban environments.Iconic Research and Engineering Journals, 7(10):490–505, 2024

  31. [39]

    Ziyan Liu, Ling Han, and Ming Liu. High-resolution carbon emission mapping and spatial-temporal analysis based on multi-source geographic data: A case study in xi’an city, china.Environmental Pollution, 361:124879, 2024

  32. [40]

    Impact of building materials for the facade on energy consumption and carbon emissions (case study of residential buildings in tehran).Energy Engineering, 122(9), 2025

    Amir Sina Darabi and Mehdi Ravanshadnia. Impact of building materials for the facade on energy consumption and carbon emissions (case study of residential buildings in tehran).Energy Engineering, 122(9), 2025. 19 CarbonCLIP

  33. [41]

    Urban region representation learning with openstreetmap building footprints

    Yi Li, Weiming Huang, Gao Cong, Hao Wang, and Zheng Wang. Urban region representation learning with openstreetmap building footprints. InProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 1363–1373, 2023

  34. [42]

    Flexireg: Flexible urban region representation learning

    Fengze Sun, Yanchuan Chang, Egemen Tanin, Shanika Karunasekera, and Jianzhong Qi. Flexireg: Flexible urban region representation learning. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V . 2, pages 2702–2713, 2025

  35. [43]

    Geoclip: Clip-inspired alignment between locations and images for effective worldwide geo-localization.Advances in Neural Information Processing Systems, 36:8690–8701, 2023

    Vicente Vivanco Cepeda, Gaurav Kumar Nayak, and Mubarak Shah. Geoclip: Clip-inspired alignment between locations and images for effective worldwide geo-localization.Advances in Neural Information Processing Systems, 36:8690–8701, 2023

  36. [44]

    Satclip: Global, general-purpose location embeddings with satellite imagery

    Konstantin Klemmer, Esther Rolf, Caleb Robinson, Lester Mackey, and Marc Rußwurm. Satclip: Global, general-purpose location embeddings with satellite imagery. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 4347–4355, 2025

  37. [45]

    Img2loc: Revisiting image geolocalization using multi-modality foundation models and image-based retrieval- augmented generation

    Zhongliang Zhou, Jielu Zhang, Zihan Guan, Mengxuan Hu, Ni Lao, Lan Mu, Sheng Li, and Gengchen Mai. Img2loc: Revisiting image geolocalization using multi-modality foundation models and image-based retrieval- augmented generation. InProceedings of the 47th international acm sigi...

  38. [46]

    Towards interpretable geo-localization: a concept-aware global image-gps alignment framework.arXiv preprint arXiv:2509.01910, 2025

    Furong Jia, Lanxin Liu, Ce Hou, Fan Zhang, Xinyan Liu, and Yu Liu. Towards interpretable geo-localization: a concept-aware global image-gps alignment framework.arXiv preprint arXiv:2509.01910, 2025

  39. [47]

    Remoteclip: A vision language foundation model for remote sensing.IEEE Transactions on Geoscience and Remote Sensing, 62:1–16, 2024

    Fan Liu, Delong Chen, Zhangqingyun Guan, Xiaocong Zhou, Jiale Zhu, Qiaolin Ye, Liyong Fu, and Jun Zhou. Remoteclip: A vision language foundation model for remote sensing.IEEE Transactions on Geoscience and Remote Sensing, 62:1–16, 2024

  40. [48]

    Urbanclip: Learning text-enhanced urban region profiling with contrastive language-image pretraining from the web

    Yibo Yan, Haomin Wen, Siru Zhong, Wei Chen, Haodong Chen, Qingsong Wen, Roger Zimmermann, and Yuxuan Liang. Urbanclip: Learning text-enhanced urban region profiling with contrastive language-image pretraining from the web. InProceedings of the ACM Web Conference 2024, pages 40...

  41. [49]

    Xiang Li, Congcong Wen, Yuan Hu, and Nan Zhou. Rs-clip: Zero shot remote sensing scene classification via contrastive vision-language supervision.International Journal of Applied Earth Observation and Geoinformation, 124:103497, 2023

  42. [50]

    Geochat: Grounded large vision-language model for remote sensing

    Kartik Kuckreja, Muhammad Sohail Danish, Muzammal Naseer, Abhijit Das, Salman Khan, and Fahad Shahbaz Khan. Geochat: Grounded large vision-language model for remote sensing. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 27831–27840, 2024

  43. [51]

    Earthgpt: A universal multimodal large language model for multisensor image comprehension in remote sensing domain.IEEE Transactions on Geoscience and Remote Sensing, 62:1–20, 2024

    Wei Zhang, Miaoxin Cai, Tong Zhang, Yin Zhuang, and Xuerui Mao. Earthgpt: A universal multimodal large language model for multisensor image comprehension in remote sensing domain.IEEE Transactions on Geoscience and Remote Sensing, 62:1–20, 2024

  44. [52]

    Earthgpt-x: A spatial mllm for multilevel multisource remote sensing imagery understanding with visual prompting.IEEE Transactions on Geoscience and Remote Sensing, 63:1–21, 2025

    Wei Zhang, Miaoxin Cai, Yaqian Ning, Tong Zhang, Yin Zhuang, Shijian Lu, He Chen, Jun Li, and Xuerui Mao. Earthgpt-x: A spatial mllm for multilevel multisource remote sensing imagery understanding with visual prompting.IEEE Transactions on Geoscience and Remote Sensing, 63:1–21, 2025

  45. [53]

    Addressvlm: Cross-view alignment tuning for image address localization using large vision-language models

    Shixiong Xu, Chenghao Zhang, Lubin Fan, Yuan Zhou, Bin Fan, Shiming Xiang, Gaofeng Meng, and Jieping Ye. Addressvlm: Cross-view alignment tuning for image address localization using large vision-language models. arXiv preprint arXiv:2508.10667, 2025

  46. [54]

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, et al. Qwen2. 5-vl technical report.arXiv preprint arXiv:2502.13923, 2025

  47. [55]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. InInternational Conference on Machine Learning, pag...

  48. [56]

    Attention is all you need.Advances in Neural Information Processing Systems, 30, 2017

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in Neural Information Processing Systems, 30, 2017

  49. [57]

    An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:...

  50. [58]

    Google Maps Platform.https://maps.google.com, 2025

    Google Inc. Google Maps Platform.https://maps.google.com, 2025. Accessed via Google Maps API

  51. [59]

    Baidu Maps API.https://lbsyun.baidu.com, 2025

    Baidu Inc. Baidu Maps API.https://lbsyun.baidu.com, 2025. Accessed via Baidu Map Open Platform

  52. [60]

    The open-source data inventory for anthropogenic co2 (odiac) 2023, 2023

    Tomohiro Oda, R Bun, V Kinakh, P Topylko, M Halushchak, G Marland, T Lauvaux, M Jonas, and R Zhuravlev. The open-source data inventory for anthropogenic co2 (odiac) 2023, 2023. 20 CarbonCLIP

  53. [61]

    Effects of 3d urban morphology on co2 emissions using machine learning: Towards spatially tailored low-carbon strategies in central wuhan, china.Urban Climate, 57:102122, 2024

    Peng Tian, Meng Cai, Zhihao Sun, Sheng Liu, Hao Wu, Lingbo Liu, and Zhenghong Peng. Effects of 3d urban morphology on co2 emissions using machine learning: Towards spatially tailored low-carbon strategies in central wuhan, china.Urban Climate, 57:102122, 2024

  54. [62]

    Impact of compact city on carbon emission reduction based on urban size: A spatial analysis using satellite imagery.Sustainable Cities and Society, 126:106326, 2025

    Hansol Mun and Juchul Jung. Impact of compact city on carbon emission reduction based on urban size: A spatial analysis using satellite imagery.Sustainable Cities and Society, 126:106326, 2025

  55. [63]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2016. 21

Pith tools

Reviewed July 9, 2026 · model on record in the stance chip above.