Pith. sign in

REVIEW 4 major objections 6 minor 45 references

OpenCarbon: A Contrastive Learning-based Cross-Modality Neural Approach for High-Resolution Carbon Emission Prediction Using Open Data

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read OpenCarbon predicts high-resolution carbon emissions from open satellite and POI data, with average $R^2$ = 0.6997.

desk verdict A credible first fusion of satellite imagery and POI data with contrastive learning for ODIAC-style 1 km carbon mapping, but the absolute accuracy claim is undercut by modeled labels and a boundary-contiguous split; worth refereeing with major revision. read the letter →

arxiv 2506.03224 v1 pith:37SWFSGP submitted 2025-06-03 cs.CV cs.AIphysics.soc-ph

Jinwei Zeng , Yu Liu , Guozhen Zhang , Jingtao Ding , Yuming Lin , Jian Yuan , Yong Li This is my paper
classification cs.CVcs.AIphysics.soc-ph
keywords carbonemissionpredictionsatelliteimagerypoint-of-interestdatacontrastivelearningcross-modalityfusionspatialagglomeration1kmgridODIACinventory
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

OpenCarbon is a neural model that tries to establish that yearly carbon emissions at 1 km grid resolution can be predicted from two open data sources alone: satellite imagery and point-of-interest (POI) facility data. The motivation is practical: once trained, such a model can estimate emissions for new city areas without the costly sensor networks and activity surveys that traditional bottom-up accounting requires. The paper reports an average $R^2$ of 0.6997 across Greater London, Beijing, and Yinchuan, outperforming the best baseline by 26.6% on average, and reports that the model preserves the rank order of emissions when transferred to new English regions. The labels come from the ODIAC inventory, a globally gridded fossil-fuel carbon dataset built by allocating regional totals with proxy data, so the accuracy of that inventory is a load-bearing assumption.

What carries the argument

The machinery is a two-module architecture. The cross-modality information extraction and fusion module encodes satellite images with a residual network, encodes POI distributions with convolutional layers plus squeeze-and-excitation channel attention, aligns the two representations with the NT-Xent contrastive loss, and merges them with an aggregate-attention weighted sum. The neighborhood-informed aggregation module then takes the fused representations of the surrounding $M\times M$ grids, runs them through convolutional layers to capture spatial contiguity, and uses cross-attention with the target grid's representation to model the agglomeration effect. The neighborhood module carries the largest measured weight: ablating it lowers $R^2$ by 31.8% on average.

What would settle it

Take one test region and compare OpenCarbon's 1 km predictions against an independent high-resolution inventory built from local point sources, metered fuel use, or direct monitoring; if the predictions track ODIAC more closely than they track the independent measurements, especially in specific facility categories, the claim of accurate carbon estimation would be falsified.

Watch

Extended reading notes

Core claim

The central claim is that the spatial pattern of urban carbon emissions is learnable from the visual layout of a city plus the distribution of its facilities, without direct emission statistics beyond the training labels. The paper identifies two obstacles: the functional effect, where different facility types contribute to emissions in implicit and coupled ways, and the spatial agglomeration effect, where neighboring grids show continuous emission levels. OpenCarbon addresses the first with a cross-modality extraction module that aligns satellite and POI inputs, uses squeeze-and-excitation attention along the facility-category dimension, and applies a contrastive loss so each modality borrows missing information from the other. It addresses the second with a neighborhood-informed module that convolves the fused representations of surrounding grids and applies cross-attention to the target grid. On the three test regions the model reaches $R^2$ values of 0.786, 0.691, and 0.622, and on four English regions it achieves an average Spearman rank correlation of 0.6204 in direct transfer, which the paper presents as evidence that the learned functionality-emission relation generalizes.

Load-bearing premise

The load-bearing premise is that ODIAC's gridded carbon emission values are accurate enough to serve as ground truth for training and evaluation, even though ODIAC is itself a modeled inventory that spreads fuel-based totals across grids with proxy data rather than measuring emissions directly.

Editorial extensions

If this is right

  • A trained OpenCarbon model can estimate 1 km emissions for any city with open satellite imagery and POI data, removing the need for local fuel statistics or sensor networks in initial screening.
  • Because the model preserves emission rank order across regions, city governments can use it to locate high-emission grids for targeted mitigation even where absolute emission factors differ.
  • Spatial context is the dominant driver of accuracy: the 31.8% drop in $R^2$ when the neighborhood module is removed indicates that agglomeration information, not single-grid appearance, carries most of the predictive signal.
  • The case studies show that predictions are interpretable: high-emission grids correspond to denser urbanized land use, and the POI modality becomes more influential as emissions rise.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: because the labels come from ODIAC, a strong $R^2$ may partly mean the model has learned ODIAC's proxy allocation rules; validation against independent bottom-up inventories, such as local metered fuel data or point-source monitoring, would clarify how much of the accuracy is real.
  • Editorial inference: the model is yearly-resolution only, and the same contrastive fusion and neighborhood modules could be retrained on monthly or seasonal emission products to support ongoing mitigation tracking, which the paper itself lists as future work.
  • Editorial inference: the aggregate-attention observation that POI data gains weight in high-emission areas suggests an actionable policy lever: within an unchanged built layout, altering the mix of facilities may shift emissions more than altering building form itself.
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

4 major / 6 minor

Summary. The paper introduces OpenCarbon, a neural network that predicts 1 km × 1 km urban carbon emissions from open satellite imagery and point-of-interest (POI) data. The architecture combines a ResNet-18 image encoder, a CNN/SE-block POI encoder, an NT-Xent contrastive loss for cross-modal fusion, aggregate attention for modality weighting, and a neighborhood-informed convolution/cross-attention module to capture spatial agglomeration. The model is trained and evaluated on ODIAC gridded emissions for Greater London, Beijing, and Yinchuan, and the paper reports an average R² of 0.6997, outperforming baselines by an average of 26.6%, plus ablation studies, a cross-county rank-transfer experiment, and qualitative case studies. The central claim is that a trained OpenCarbon model can infer high-resolution emissions for new areas using only open data, avoiding costly bottom-up accounting.

Significance. If the performance claims held, OpenCarbon would be a practically valuable tool for scalable, low-cost carbon-emission mapping, and the combination of satellite imagery and POI data with contrastive fusion is a sensible design direction. The paper has concrete strengths: it uses publicly available data, releases code and data, compares against multiple baselines, includes ablations for each architectural component, and performs a direct-transfer ranking experiment across four English counties. However, the central claim of 'accurate' emission prediction is not yet established because the training labels are themselves a modeled inventory (ODIAC), the within-city evaluation uses spatially adjacent districts without leakage control, no uncertainty or significance measures are reported, and the cross-city generalization experiment only evaluates rank correlation, not absolute error. These issues are load-bearing for the paper's main assertion, but they are addressable in a revision.

major comments (4)
  1. [§5.1, Datasets] The target labels are ODIAC 2018/2019 gridded emissions, which are a hybrid inventory constructed from point-source data and top-down proxy allocation, not direct measurements of emissions. Treated as ground truth, the reported R² measures how well OpenCarbon reproduces ODIAC's spatial allocation patterns. This does not by itself invalidate the method, but it does invalidate the wording 'accurate carbon estimation' used in the abstract and conclusion. The authors should either evaluate against an independent high-resolution inventory with actual measurements (e.g., CHRED, VULCAN, or local bottom-up accounts) or explicitly reframe the contribution as 'reproducing ODIAC emissions from open data' and discuss what that implies for real-world use. As written, this is a load-bearing gap for the central claim.
  2. [§5.1 and §4.2] The held-out test districts are spatially adjacent to training districts, and the neighborhood-informed aggregation module uses an M×M context with M up to 7. For example, in Beijing the Fengtai test district borders training districts such as Xicheng and Chaoyang, so a large fraction of test-grid neighborhoods include grids whose representations were learned from labels seen during training. Because those neighboring representations are trained to encode emission-relevant information, the model can exploit label-informed spatial context at test time, inflating the reported R². The authors should evaluate on spatially separated holdout regions (e.g., with a buffer zone of at least M/2 grids removed) or report results with M=1 as a no-context control. Without this, the within-city performance numbers in Table 2 are not trustworthy as evidence of generalization to new areas.
  3. [§5.2, Table 2] All reported metrics are point estimates from a single split with no standard deviations, confidence intervals, or significance tests. Given that each test set is a single district (e.g., Fengtai for Beijing) and that the improvement over the best baseline is as small as 11.3% in R², the reader cannot assess whether the gains are stable across random seeds or initializations. At minimum, the authors should run each experiment multiple times and report mean ± std, and perform a paired significance test where appropriate. This is necessary to support the headline claim of a 26.6% average improvement.
  4. [§5.4] The generalizability study reports only Spearman's rank correlation coefficients (average 0.6204) for direct transfer across regions. Rank correlation shows that the model can order high- and low-emission grids, but it does not demonstrate accurate absolute emission estimation for new areas, which is the capability claimed in the introduction. The authors should additionally report MAE, RMSE, and R² (or a calibration plot) on the transferred regions. Without absolute-error metrics, the conclusion that OpenCarbon can 'infer emissions for new areas' is not supported.
minor comments (6)
  1. [§4.1, Eq. (4)] The index notation is inconsistent: the equation first says k ∈ {s, p} and then normalizes over k ∈ {s, i}; the set should be {s, p} throughout, and the variable 'i' appears to be a typo for 'p'.
  2. [§4.2, Eq. (6)] The cross-attention formulation is unclear: 'Attention(query = Xn, value = Xg)' does not specify the key, so the reader cannot tell whether key and value are both Xg or whether there is an additional projection. Please clarify the exact attention operation.
  3. [§5.4] The text says 'As shown in Figure 3' when referring to the generalizability results, but the cited figure is Figure 4; please correct the cross-reference.
  4. [§5.3] There is a typo: 'we remove the see attention' should be 'the SE attention'. Also, the ablation percentages (e.g., '3.8% on R2') should state whether they are relative or absolute changes.
  5. [§5.2, Table 2] The 'Improv.' row does not state which baseline the improvement is computed against; please specify (presumably the best baseline per metric).
  6. [Global] The paper does not explicitly discuss the limitations of using ODIAC as a label source or the potential for spatial autocorrelation to inflate metrics; a Limitations subsection would greatly improve the clarity and honesty of the claims.

Circularity Check

0 steps flagged · score 0.0 of 10

No meaningful circularity: OpenCarbon is trained and evaluated on held-out ODIAC subregions, and no reported prediction reduces to a fitted parameter or self-citation.

full rationale

The paper's derivation chain is self-contained. The model takes satellite images and POI distributions as inputs and is trained with a mean absolute error loss plus a contrastive loss against ODIAC 1 km grid emission labels (Section 4.3, Equation 8). Evaluation is performed on districts held out by regional division, e.g., training on five Beijing districts and testing on Fengtai (Section 5.1), so the reported R2 values are not computed on training labels. No parameter is fitted to the test target and then renamed as a prediction; the cross-city generalizability test is a direct transfer without fine-tuning and is explicitly reported as Spearman rank correlation (Section 5.4), which is an honest and narrower claim. The acknowledged limitation that ODIAC is a hybrid inventory with top-down proxy allocations is an external-validity concern about the ground truth, not a circularity in the model's derivation. Self-citations, such as the PG-SimCLR baseline from Xi et al. 2022, are used only as comparison methods and are not load-bearing for OpenCarbon's architecture or claims. No equation or construction in the paper makes the output equivalent to the input by definition.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim depends on several domain assumptions about the data and target. The model itself introduces learned weights and a few grid-searched hyperparameters, but no new physical entities.

free parameters (4)
  • neighborhood scale M = 3, 5, or 7 (grid-searched)
    Controls the extent of spatial context in the agglomeration module; chosen on validation.
  • contrastive loss weight alpha = 0.1, 0.01, or 0.001 (grid-searched)
    Balances prediction and contrastive losses; selected on validation.
  • temperature tau in NT-Xent = not reported
    Temperature in the contrastive loss; value not specified in the paper.
  • SE compression ratio n = not reported
    Compression ratio in squeeze-and-excitation blocks; value not disclosed.
assumptions (4)
  • domain assumption Satellite images and POI distributions contain sufficient information to predict 1 km carbon emissions.
    Core premise of the method; Section 4.1.
  • domain assumption ODIAC inventory values are an adequate target for training and evaluation.
    Section 5.1 treats ODIAC as the ground truth without independent validation.
  • domain assumption Spatial contiguity implies emission agglomeration that can be captured by convolutions over neighborhood representations.
    Section 4.2 assumes convolution captures spatial continuity.
  • standard math Contrastive learning with NT-Xent enforces complementary information across modalities.
    Standard technique from Chen et al. 2020; used as a building block.

how reviews work

0 comments
Cite this review

Pith. "Pith review of OpenCarbon: A Contrastive Learning-based Cross-Modality Neural Approach for High-Resolution Carbon Emission Prediction Using Open Data." pith.science (2026). https://pith.science/paper/37SWFSGP

@misc{pith2026250603224,
  author       = {Pith},
  title        = {Pith review of: OpenCarbon: A Contrastive Learning-based Cross-Modality Neural Approach for High-Resolution Carbon Emission Prediction Using Open Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/37SWFSGP}},
  note         = {Machine review of arXiv:2506.03224}
}
read the original abstract

Accurately estimating high-resolution carbon emissions is crucial for effective emission governance and mitigation planning. While conventional methods for precise carbon accounting are hindered by substantial data collection efforts, the rise of open data and advanced learning techniques offers a promising solution. Once an open data-based prediction model is developed and trained, it can easily infer emissions for new areas based on available open data. To address this, we incorporate two modalities of open data, satellite images and point-of-interest (POI) data, to predict high-resolution urban carbon emissions, with satellite images providing macroscopic and static and POI data offering fine-grained and relatively dynamic functionality information. However, estimating high-resolution carbon emissions presents two significant challenges: the intertwined and implicit effects of various functionalities on carbon emissions, and the complex spatial contiguity correlations that give rise to the agglomeration effect. Our model, OpenCarbon, features two major designs that target the challenges: a cross-modality information extraction and fusion module to extract complementary functionality information from two modules and model their interactions, and a neighborhood-informed aggregation module to capture the spatial contiguity correlations. Extensive experiments demonstrate our model's superiority, with a significant performance gain of 26.6\% on R2. Further generalizability tests and case studies also show OpenCarbon's capacity to capture the intrinsic relation between urban functionalities and carbon emissions, validating its potential to empower efficient carbon governance and targeted carbon mitigation planning. Codes and data are available: https://github.com/JinweiZzz/OpenCarbon.

Figures

Figures reproduced from arXiv: 2506.03224 by the authors.

Figure 1
Figure 1. Illustrations of the (a)functional effect and (b)spatial ag [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The overall framework of our OpenCarbon model. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Ablation results on (a) London and (b) Yinchuan dataset. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (1 more)
Figure 6
Figure 6. Figure 6: Visualizations of the aggregate-attention weights of the two [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

45 extracted references · 43 canonical work pages

  1. [1]

    Paris agreement

    [Agreement, 2015] Paris Agreement. Paris agreement. In Re- port of the Conference of the Parties to the United Nations Framework Convention on Climate Change (21st Session, 2015: Paris). Retrived December , volume 4, page

    work page 2015

  2. [13]

    High resolution fossil fuel combustion co2 emission fluxes for the united states

    [Gurney et al., 2009] Kevin R Gurney, Daniel L Mendoza, Yuyu Zhou, Marc L Fischer, Chris C Miller, Sarath Geethakumar, and Stephane de la Rue du Can. High resolution fossil fuel combustion co2 emission fluxes for the united states. Environmental science & technology , 43(14):5535–5541,

    work page 2009

  3. [14]

    The vulcan version 3.0 high-resolution fossil fuel co2 emis- sions for the united states

    [Gurney et al., 2020] Kevin R Gurney, Jianming Liang, Risa Patarasuk, Yang Song, Jianhua Huang, and Geoffrey Roest. The vulcan version 3.0 high-resolution fossil fuel co2 emis- sions for the united states. Journal of Geophysical Re- search: Atmospheres, 125(19):e2020JD032974,

    work page 2020

  4. [15]

    The effects of urban agglomeration economies on carbon emissions: Evidence from chinese cities

    [Han et al., 2018] Feng Han, Rui Xie, Jiayu Fang, Yu Liu, et al. The effects of urban agglomeration economies on carbon emissions: Evidence from chinese cities. Journal of Cleaner Production, 172:1096–1110,

    work page 2018

  5. [16]

    Lightweight and robust representation of economic scales from satellite imagery

    [Han et al., 2020] Sungwon Han, Donghyun Ahn, Hyunji Cha, Jeasurk Yang, Sungwon Park, and Meeyoung Cha. Lightweight and robust representation of economic scales from satellite imagery. In Proceedings of the AAAI Confer- ence on Artificial Intelligence, volume 34, pages 428–436,

    work page 2020

  6. [17]

    Deep residual learning for image recog- nition

    [He et al., 2016] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog- nition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778,

    2016

  7. [18]

    Squeeze- and-excitation networks

    [Hu et al., 2018] Jie Hu, Li Shen, and Gang Sun. Squeeze- and-excitation networks. In Proceedings of the IEEE con- ference on computer vision and pattern recognition, pages 7132–7141,

    work page 2018

  8. [19]

    A comparison of five high-resolution spatially-explicit, fossil-fuel, carbon diox- ide emission inventories for the united states

    [Hutchins et al., 2017] Maya G Hutchins, Jeffrey D Colby, Gregg Marland, and Eric Marland. A comparison of five high-resolution spatially-explicit, fossil-fuel, carbon diox- ide emission inventories for the united states. Mitigation and Adaptation Strategies for Global Change, 22(6):947– 972,

    work page 2017

Show all 45 references
  1. [22]

    Predicting transportation carbon emission with urban big data

    [Lu et al., 2017] Xiangyong Lu, Kaoru Ota, Mianxiong Dong, Chen Yu, and Hai Jin. Predicting transportation carbon emission with urban big data. IEEE Transactions on Sustainable Computing, 2(4):333–344,

    2017

  2. [23]

    Management and estimation of thermal comfort, carbon dioxide emission and economic growth by support vector machine

    [Mladenovi´c et al., 2016] Igor Mladenovi ´c, Svetlana Sokolov-Mladenovi´c, Milos Milovan ˇcevi´c, Dušan Markovi´c, and Nenad Simeunovi ´c. Management and estimation of thermal comfort, carbon dioxide emission and economic growth by support vector machine. Renewable and sustai...

    2016

  3. [24]

    A global review of energy consumption, co2 emissions and policy in the residential sector (with an overview of the top ten co2 emitting countries)

    [Nejat et al., 2015] Payam Nejat, Fatemeh Jomehzadeh, Mo- hammad Mahdi Taheri, Mohammad Gohari, and Muhd Zaimi Abd Majid. A global review of energy consumption, co2 emissions and policy in the residential sector (with an overview of the top ten co2 emitting countries). Renewab...

    2015

  4. [25]

    [Oda et al., 2018] Tomohiro Oda, Shamil Maksyutov, and Robert J Andres. The open-source data inventory for anthro- pogenic co2, version 2016 (odiac2016): a global monthly fossil fuel co2 gridded emissions data product for tracer transport simulations and surface flux inversion...

    2018

  5. [26]

    Emission database for global atmospheric research (edgar)

    [Olivier et al., 1994] JGJ Olivier, AF Bouwman, CWM Van der Maas, and JJM Berdowski. Emission database for global atmospheric research (edgar). Environmental Monitoring and Assessment, 31:93–106,

    1994

  6. [29]

    Carbon accounting: a systematic literature review

    [Stechemesser and Guenther, 2012] Kristin Stechemesser and Edeltraud Guenther. Carbon accounting: a systematic literature review. Journal of Cleaner Production, 36:17–38,

    2012

  7. [31]

    Carbon emission from urban passenger transporta- tion in beijing

    [Wang et al., 2015] Zijia Wang, Feng Chen, and Taku Fu- jiyama. Carbon emission from urban passenger transporta- tion in beijing. Transportation Research Part D: Transport and Environment, 41:217–227,

    2015

  8. [32]

    Crime rate inference with big data

    [Wang et al., 2016] Hongjian Wang, Daniel Kifer, Corina Graif, and Zhenhui Li. Crime rate inference with big data. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, pages 635–644,

    2016

  9. [33]

    Agglomeration effect of co2 emissions and emissions reduc- tion effect of technology: A spatial econometric perspective based on china’s province-level data

    [Wang et al., 2018] Bo Wang, Yefei Sun, and Zhaohua Wang. Agglomeration effect of co2 emissions and emissions reduc- tion effect of technology: A spatial econometric perspective based on china’s province-level data. Journal of cleaner production, 204:96–106,

    2018

  10. [34]

    The effect of urbanization and spatial agglomeration on carbon emissions in urban ag- glomeration

    [Wang et al., 2020] Feng Wang, Wenna Fan, Juan Liu, Ge Wang, and Wei Chai. The effect of urbanization and spatial agglomeration on carbon emissions in urban ag- glomeration. Environmental science and pollution research, 27:24329–24341,

    2020

  11. [35]

    High-resolution mapping of regional traffic emissions using land-use ma- chine learning models

    [Wu et al., 2022] Xiaomeng Wu, Daoyuan Yang, Ruoxi Wu, Jiajun Gu, Yifan Wen, Shaojun Zhang, Rui Wu, Renjie Wang, Honglei Xu, K Max Zhang, et al. High-resolution mapping of regional traffic emissions using land-use ma- chine learning models. Atmospheric Chemistry and Physics, 2...

    2022

  12. [36]

    Beyond the first law of geography: Learning representations of satellite imagery by leveraging point-of-interests

    [Xi et al., 2022] Yanxin Xi, Tong Li, Huandong Wang, Yong Li, Sasu Tarkoma, and Pan Hui. Beyond the first law of geography: Learning representations of satellite imagery by leveraging point-of-interests. In Proceedings of the ACM Web Conference 2022, pages 3308–3316,

    2022

  13. [37]

    Ar2net: An attentive neural approach for business location selection with satellite data and urban data

    [Xu et al., 2020] Yanan Xu, Yanyan Shen, Yanmin Zhu, and Jiadi Yu. Ar2net: An attentive neural approach for business location selection with satellite data and urban data. ACM Transactions on Knowledge Discovery from Data (TKDD), 14(2):1–28,

    2020

  14. [38]

    Modeling and spatio-temporal analysis of city-level carbon emissions based on nighttime light satellite imagery

    [Yang et al., 2020] Di Yang, Weixin Luan, Lu Qiao, and Ma- hardhika Pratama. Modeling and spatio-temporal analysis of city-level carbon emissions based on nighttime light satellite imagery. Applied Energy, 268:114696,

    2020

  15. [39]

    Using publicly available satellite imagery and deep learning to understand economic well-being in africa

    [Yeh et al., 2020] Christopher Yeh, Anthony Perez, Anne Driscoll, George Azzari, Zhongyi Tang, David Lobell, Ste- fano Ermon, and Marshall Burke. Using publicly available satellite imagery and deep learning to understand economic well-being in africa. Nature communications, 11...

    2020

  16. [40]

    Smartphone app usage pre- diction using points of interest

    [Yu et al., 2018] Donghan Yu, Yong Li, Fengli Xu, Pengyu Zhang, and Vassilis Kostakos. Smartphone app usage pre- diction using points of interest. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 1(4):1–21,

    2018

  17. [41]

    Household carbon emission research: an analyt- ical review of measurement, influencing factors and miti- gation prospects

    [Zhang et al., 2015] Xiaoling Zhang, Lizi Luo, and Martin Skitmore. Household carbon emission research: an analyt- ical review of measurement, influencing factors and miti- gation prospects. Journal of Cleaner Production, 103:873– 883,

    2015

  18. [42]

    Towards low carbon cities: A machine learning method for predicting urban blocks carbon emissions (ubce) based on built environment factors (bef) in changxing city, china

    [Zhang et al., 2021] Xiaoping Zhang, Fengying Yan, Hongjiang Liu, and Zhi Qiao. Towards low carbon cities: A machine learning method for predicting urban blocks carbon emissions (ubce) based on built environment factors (bef) in changxing city, china. Sustainable Cities and So...

    2021

  19. [43]

    Estimating global anthropogenic co2 gridded emissions using a data-driven stacked random for- est regression model

    [Zhang et al., 2022] Yucong Zhang, Xinjie Liu, Liping Lei, and Liangyun Liu. Estimating global anthropogenic co2 gridded emissions using a data-driven stacked random for- est regression model. Remote Sensing, 14(16):3899,

    2022

  20. [44]

    Identifying urban functional re- gions from high-resolution satellite images using a context- aware segmentation network.Remote Sensing, 14(16):3996,

    [Zhao et al., 2022] Wufan Zhao, Mengmeng Li, Cai Wu, Wen Zhou, and Guozhong Chu. Identifying urban functional re- gions from high-resolution satellite images using a context- aware segmentation network.Remote Sensing, 14(16):3996,

    2022

  21. [45]

    Assessing osm building complete- ness for almost 13,000 cities globally.International Journal of Digital Earth, 15(1):2400–2421, 2022

    [Zhou et al., 2022] Qi Zhou, Yuheng Zhang, Ke Chang, and Maria Antonia Brovelli. Assessing osm building complete- ness for almost 13,000 cities globally.International Journal of Digital Earth, 15(1):2400–2421, 2022

    2022

  22. [1994]

    A 1990 global emission inventory of anthro- pogenic sources of carbon monoxide on 1× 1 developed in the framework of edgar/geia

    [Olivier et al., 1999] Jos GJ Olivier, Jan Pieter J Bloos, Jan JM Berdowski, Antoon JH Visschedijk, and Alex F Bouwman. A 1990 global emission inventory of anthro- pogenic sources of carbon monoxide on 1× 1 developed in the framework of edgar/geia. Chemosphere-Global Change Sc...

    1999

  23. [1998]

    China high resolution emission database (chred) with point emission sources, gridded emission data, and supplemen- tary socioeconomic data

    [Cai et al., 2018] Bofeng Cai, Sai Liang, Jiong Zhou, Jinnan Wang, Libin Cao, Shen Qu, Ming Xu, and Zhifeng Yang. China high resolution emission database (chred) with point emission sources, gridded emission data, and supplemen- tary socioeconomic data. Resources, Conservation...

    2018

  24. [1999]

    Poverty prediction with public landsat 7 satellite imagery and machine learning

    [Perez et al., 2017] Anthony Perez, Christopher Yeh, George Azzari, Marshall Burke, David Lobell, and Stefano Ermon. Poverty prediction with public landsat 7 satellite imagery and machine learning. arXiv preprint arXiv:1711.03654,

    2017 arXiv

  25. [2008]

    The synthesis of bottom-up and top-down in energy policy modeling

    [Böhringer, 1998] Christoph Böhringer. The synthesis of bottom-up and top-down in energy policy modeling. En- ergy economics, 20(3):233–248,

    1998

  26. [2009]

    Understanding urban functionality from poi space

    [Fan et al., 2018] Dongwan Fan, Kun Qin, and Chaogui Kang. Understanding urban functionality from poi space. In 2018 26th International Conference on Geoinformatics, pages 1–6. IEEE,

    2018

  27. [2011]

    A simple framework for con- trastive learning of visual representations

    [Chen et al., 2020] Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for con- trastive learning of visual representations. In Proceedings of the 37th International Conference on Machine Learning (ICML 2020),

    2020

  28. [2012]

    Visualizing data using t-sne

    [Van der Maaten and Hinton, 2008] Laurens Van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of machine learning research, 9(11),

    2008

  29. [2015]

    Machine learn- ing and phone data can improve targeting of humanitarian aid

    [Aiken et al., 2022] Emily Aiken, Suzanne Bellue, Dean Kar- lan, Chris Udry, and Joshua E Blumenstock. Machine learn- ing and phone data can improve targeting of humanitarian aid. Nature, 603(7903):864–870,

    2022

  30. [2016]

    Urban energy use and car- bon emissions from cities in china and policy implications

    [Dhakal, 2009] Shobhakar Dhakal. Urban energy use and car- bon emissions from cities in china and policy implications. Energy policy, 37(11):4208–4219,

    2009

  31. [2017]

    Tile2vec: Unsupervised representation learning for spatially dis- tributed data

    [Jean et al., 2019] Neal Jean, Sherrie Wang, Anshul Samar, George Azzari, David Lobell, and Stefano Ermon. Tile2vec: Unsupervised representation learning for spatially dis- tributed data. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 33, pages 3967–3974,

    2019

  32. [2018]

    Introduction to remote sensing

    [Campbell and Wynne, 2011] James B Campbell and Ran- dolph H Wynne. Introduction to remote sensing. Guilford Press,

    2011

  33. [2019]

    Carbon monitor, a near-real- time daily dataset of global co2 emission from fossil fuel and cement production

    [Liu et al., 2020] Zhu Liu, Philippe Ciais, Zhu Deng, Steven J Davis, Bo Zheng, Yilong Wang, Duo Cui, Biqing Zhu, Xinyu Dou, Piyu Ke, et al. Carbon monitor, a near-real- time daily dataset of global co2 emission from fossil fuel and cement production. Scientific data, 7(1):392,

    2020

  34. [2020]

    Spatiotemporal prediction of carbon emissions using a hybrid deep learn- ing model considering temporal and spatial correlations

    [Chen et al., 2024] Yixiang Chen, Yuxin Xie, Xu Dang, Bo Huang, Chao Wu, and Donglai Jiao. Spatiotemporal prediction of carbon emissions using a hybrid deep learn- ing model considering temporal and spatial correlations. Environmental Modelling & Software, 172:105937,

    2024

  35. [2021]

    Combining bottom-up and top-down

    [Böhringer and Rutherford, 2008] Christoph Böhringer and Thomas F Rutherford. Combining bottom-up and top-down. Energy Economics, 30(2):574–596,

    2008

  36. [2022]

    Advancing climate action and resilience through an urban lens,

    [Bank, 2021] The World Bank. Advancing climate action and resilience through an urban lens,

    2021

  37. [2024]

    Closing the gap? top-down versus bottom-up projections of china’s regional energy use and co2 emissions

    [Dai et al., 2016] Hancheng Dai, Peggy Mischke, Xuxuan Xie, Yang Xie, and Toshihiko Masui. Closing the gap? top-down versus bottom-up projections of china’s regional energy use and co2 emissions. Applied Energy, 162:1355– 1373,

    2016

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.