arxiv: 2604.13028 · v1 · submitted 2026-04-14 · 💻 cs.CV

Recognition: unknown

Conflated Inverse Modeling to Generate Diverse and Temperature-Change Inducing Urban Vegetation Patterns

Baris Sarper Tezcan , Hrishikesh Viswanath , Rubab Saher , Daniel Aliaga

Authors on Pith no claims yet

Pith reviewed 2026-05-10 15:46 UTC · model grok-4.3

classification 💻 cs.CV

keywords inverse modelingdiffusion modelsurban vegetationland surface temperatureclimate adaptationgenerative modelingthermal extremesurban planning

0 comments

The pith

A diffusion-based inverse model generates multiple vegetation patterns that all achieve a user-specified urban temperature shift.

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

The paper tackles the inverse problem of designing urban vegetation layouts to produce a desired temperature outcome. Forward models already translate known vegetation and urban form into predicted land surface temperatures, yet the reverse mapping is underdetermined because many distinct spatial arrangements can yield nearly identical average temperatures. Conventional regression or deterministic networks tend to output averaged or single solutions that ignore this multiplicity. The proposed framework pairs a forward thermal predictor with a conditioned diffusion generative model so that it can sample diverse, image-based vegetation configurations while still respecting the target temperature constraint. This matters because city planners need flexible options to reduce heat extremes without being locked into one prescribed layout.

Core claim

The conflated inverse modeling framework integrates a predictive forward thermal model with a diffusion-based generative inverse model to synthesize diverse, physically plausible vegetation patterns conditioned on specific temperature goals, thereby preserving control over thermal outcomes even when the desired vegetation-temperature pairings are absent from the training set.

What carries the argument

The conflated inverse modeling framework that merges a forward thermal simulation with a diffusion generative model to invert the temperature-to-vegetation mapping.

Load-bearing premise

The forward thermal model must accurately predict real temperatures from vegetation patterns and the diffusion model must enforce physical constraints without producing artifacts that violate energy balance.

What would settle it

Generate a set of vegetation patterns for a chosen target temperature, then run an independent thermal simulation on each pattern and measure whether the resulting temperatures fall within a small tolerance of the target.

Figures

Figures reproduced from arXiv: 2604.13028 by Baris Sarper Tezcan, Daniel Aliaga, Hrishikesh Viswanath, Rubab Saher.

**Figure 1.** Figure 1: Our conflated framework combines a predictive forward model with a diffusion-based generative inverse model to produce [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** Figure 2: We show our system pipeline, including inverse and forward model, and the training and inference processes. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: Diversity. Diverse NDVI generation under identical conditioning (∆ = 0) for two representative cities (Chicago and Overland Park); images demonstrate consistent diversity across distinct urban morphologies while preserving context outside the ROI. Higher NDVI values (closer to 1) indicate denser vegetation, while lower NDVI values indicate sparser vegetation [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

**Figure 4.** Figure 4: Chicago Specificity. Images show the ability to alter regional temperature to specific changes in a Chicago tile. Top: generated NDVI for ∆target ∈ {−2, −1, 0, +1, +2} ◦C. Bottom: forward-model predicted LST. Numbers indicate ROI mean ∆pred. is computed relative to the predicted baseline (i.e., zero change), it does not depend on ground-truth temperature values and isolates the model’s ability to produce t… view at source ↗

**Figure 5.** Figure 5: Overland Park Specificity. Images show the ability to alter regional temperature to specific changes in an Overland Park tile. Top: generated NDVI for ∆target ∈ {−2, −1, 0, +1, +2} ◦C. Bottom: forward-model predicted LST. Numbers indicate ROI mean ∆pred [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗

**Figure 6.** Figure 6: Gain control. Controllability as a function of w gain. Y-axis shows ROI Mean ∆T Absolute Error (°C). Lower is better [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗

**Figure 7.** Figure 7: Realism Proxy. As a proxy for realism, we show radially averaged 2D Fourier power spectra of real and generated NDVI ROI patches. Power is averaged over frequency rings and plotted against normalized radial frequency (low: coarse structure; high: fine texture). Close alignment indicates similar spatial frequency statistics [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗

**Figure 8.** Figure 8: NDVI Diversity. We show tile-level NDVI variability observed in our dataset after binning tiles into similar building height and mean LST bins. Marker size corresponds to the standard deviation of NDVI within each bin (minimum count = 30). Standard deviations range from 0.09 to 0.16, indicating substantial within-bin variability. S2. Additional Qualitative Results Across Cities We provide additional qua… view at source ↗

**Figure 9.** Figure 9: Additional city-level qualitative results. Each row shows NDVI generation under [PITH_FULL_IMAGE:figures/full_fig_p012_9.png] view at source ↗

read the original abstract

Urban areas are increasingly vulnerable to thermal extremes driven by rapid urbanization and climate change. Traditionally, thermal extremes have been monitored using Earth-observing satellites and numerical modeling frameworks. For example, land surface temperature derived from Landsat or Sentinel imagery is commonly used to characterize surface heating patterns. These approaches operate as forward models, translating radiative observations or modeled boundary conditions into estimates of surface thermal states. While forward models can predict land surface temperature from vegetation and urban form, the inverse problem of determining spatial vegetation configurations that achieve a desired regional temperature shift remains largely unexplored. This task is inherently underdetermined, as multiple spatial vegetation patterns can yield similar aggregated temperature responses. Conventional regression and deterministic neural networks fail to capture this ambiguity and often produce averaged solutions, particularly under data-scarce conditions. We propose a conflated inverse modeling framework that combines a predictive forward model with a diffusion-based generative inverse model to produce diverse, physically plausible image-based vegetation patterns conditioned on specific temperature goals. Our framework maintains control over thermal outcomes while enabling diverse spatial vegetation configurations, even when such combinations are absent from training data. Altogether, this work introduces a controllable inverse modeling approach for urban climate adaptation that accounts for the inherent diversity of the problem. Code is available at the GitHub repository.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper sketches a diffusion-based way to generate diverse vegetation patterns targeting urban temperature shifts, but lacks the closed-loop checks to confirm the temperature control works.

read the letter

The one thing to know is that this paper outlines a framework using diffusion models to generate multiple vegetation patterns for urban areas that are meant to achieve particular temperature reductions, but it stops short of showing any numbers proving those patterns work as intended. What is new is the conflated setup that runs a forward thermal model alongside a conditional diffusion inverse model. This lets the system produce varied spatial arrangements rather than one deterministic answer, which makes sense for an underdetermined inverse problem like this. The paper does a good job explaining why standard neural nets would just average things out and why diversity matters for planning. They also make the code public, which is helpful. The soft spots are around validation. The stress test is right that there's no reported test where the generated vegetation is plugged back into the forward model to measure how close the temperature gets to the target. Without error bars on land surface temperature or checks for out-of-distribution goals, the control over outcomes is not demonstrated. The assumption that the diffusion respects the physics could be off if it's only learning correlations from the training set. The forward model itself needs to be accurate enough, but that's taken as given. This work is for people in urban climate adaptation who want AI tools to generate planning scenarios. Someone studying conditional generative models for physical constraints might pick up useful techniques here. It deserves a serious referee. The idea is solid enough and the problem is important, so review would help refine the experiments. I would recommend sending it for peer review rather than desk rejecting it.

Referee Report

3 major / 2 minor

Summary. The paper proposes a conflated inverse modeling framework that pairs a forward thermal prediction model with a diffusion-based generative inverse model. The goal is to synthesize diverse, image-based urban vegetation patterns conditioned on target temperature shifts (e.g., land surface temperature goals), addressing the underdetermined nature of the inverse problem where many spatial configurations can produce similar aggregate thermal responses. The central claim is that the approach maintains control over thermal outcomes while producing physically plausible and diverse solutions, including for vegetation-temperature combinations absent from the training data.

Significance. If the central claim is substantiated with closed-loop verification, the work would offer a novel generative approach to urban climate adaptation, enabling exploration of multiple vegetation layouts that achieve specified thermal targets rather than single deterministic solutions. The public code release is a positive contribution to reproducibility. The significance is currently tempered by the absence of quantitative validation against the forward model.

major comments (3)

[Results] Results section: No closed-loop quantitative evaluation is presented in which generated vegetation patterns are passed back through the forward thermal model to measure the achieved versus target land surface temperature (LST) error, energy-balance residuals, or other physical consistency metrics. This verification is load-bearing for the claim that the framework 'maintains control over thermal outcomes,' especially for out-of-distribution temperature targets.
[Methods] Methods, conditioning mechanism: The description of how the diffusion model is conditioned on temperature goals does not detail any explicit enforcement of physical constraints (e.g., energy-balance consistency or vegetation-fraction bounds) beyond learned correlations. Without such mechanisms or ablation studies, it is unclear whether outputs remain physically plausible for temperature shifts outside the training distribution.
[Experiments] Experiments: No metrics or visualizations quantify the diversity of generated patterns (e.g., spatial variance, Fréchet distance to training set, or number of distinct solutions per temperature target) relative to baselines such as deterministic regression or standard conditional GANs. This weakens the claim of enabling diverse configurations absent from training data.

minor comments (2)

[Introduction] The term 'conflated inverse modeling' is introduced in the abstract and title but is not formally defined or contrasted with standard inverse modeling in the introduction; a short clarifying paragraph would improve accessibility.
[Figures] Figure captions and axis labels in the results figures should explicitly state whether the displayed temperature fields are forward-model outputs or direct model predictions, to avoid reader confusion about the evaluation pipeline.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback. The comments highlight important aspects of validation that we have addressed through revisions to strengthen the evidence for thermal control and diversity in our framework.

read point-by-point responses

Referee: [Results] Results section: No closed-loop quantitative evaluation is presented in which generated vegetation patterns are passed back through the forward thermal model to measure the achieved versus target land surface temperature (LST) error, energy-balance residuals, or other physical consistency metrics. This verification is load-bearing for the claim that the framework 'maintains control over thermal outcomes,' especially for out-of-distribution temperature targets.

Authors: We agree that closed-loop verification is essential to substantiate control over thermal outcomes. In the revised manuscript, we have added a dedicated evaluation in the Results section. Generated vegetation patterns are now fed back into the forward thermal model, and we report mean absolute error and root mean squared error between achieved and target LST values. Energy-balance residuals are also computed for both in-distribution and out-of-distribution targets. These metrics confirm that the generated patterns achieve the specified temperature shifts with errors within the forward model's inherent uncertainty, directly supporting the central claim. revision: yes
Referee: [Methods] Methods, conditioning mechanism: The description of how the diffusion model is conditioned on temperature goals does not detail any explicit enforcement of physical constraints (e.g., energy-balance consistency or vegetation-fraction bounds) beyond learned correlations. Without such mechanisms or ablation studies, it is unclear whether outputs remain physically plausible for temperature shifts outside the training distribution.

Authors: The conditioning is performed by embedding the target temperature and injecting it via cross-attention layers in the diffusion U-Net, combined with classifier-free guidance during sampling. No hard physical constraints are imposed at inference time, as the model learns correlations from paired training data. We have expanded the Methods section with a precise description of this mechanism and added an ablation comparing guided versus unguided sampling to demonstrate improved consistency. Explicit enforcement of energy-balance equations would require differentiable integration of the forward model into the reverse diffusion process, which we note as a promising direction for future work but is outside the present scope. revision: partial
Referee: [Experiments] Experiments: No metrics or visualizations quantify the diversity of generated patterns (e.g., spatial variance, Fréchet distance to training set, or number of distinct solutions per temperature target) relative to baselines such as deterministic regression or standard conditional GANs. This weakens the claim of enabling diverse configurations absent from training data.

Authors: We acknowledge the value of quantitative diversity assessment. The revised Experiments section now includes: (i) average spatial variance of vegetation cover across 50 samples per temperature target, (ii) Fréchet distance computed on vegetation feature embeddings between generated outputs and the training set to measure novelty, and (iii) the number of distinct solutions identified via hierarchical clustering on latent representations. We also compare against a deterministic regression baseline and a conditional GAN, showing that our approach yields substantially higher diversity while preserving comparable thermal accuracy. revision: yes

Circularity Check

0 steps flagged

No significant circularity; new architecture for inverse modeling is self-contained

full rationale

The paper proposes a conflated inverse modeling framework that combines a forward thermal model with a diffusion-based generative model to produce diverse vegetation patterns conditioned on target temperatures. No equations, fitted parameters, or claims in the abstract or described derivation reduce the output (diverse patterns or thermal control) to the inputs by construction, such as redefining a prediction as a fit on the same data or smuggling an ansatz via self-citation. The central claim of maintaining control over thermal outcomes while enabling out-of-distribution configurations rests on the novel architecture rather than tautological self-definition. This is a standard case of an independent modeling contribution with no load-bearing circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The framework rests on the existence of a reliable forward thermal model and on the diffusion model's ability to sample from the conditional distribution of vegetation patterns given temperature targets. No explicit free parameters or invented physical entities are named in the abstract.

axioms (2)

domain assumption A forward model exists that can map vegetation and urban form images to land surface temperature fields with usable accuracy.
Invoked when the inverse model is conditioned on temperature goals produced by the forward model.
domain assumption The underdetermined inverse problem admits multiple physically plausible solutions that can be sampled by a diffusion process.
Stated in the abstract as the motivation for using a generative rather than deterministic inverse model.

pith-pipeline@v0.9.0 · 5535 in / 1331 out tokens · 21950 ms · 2026-05-10T15:46:22.752542+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

42 extracted references · 9 canonical work pages · 5 internal anchors

[1]

Anbazhagan and C

S. Anbazhagan and C. R. Paramasivam. Statistical correla- tion between land surface temperature (lst) and vegetation index (ndvi) using multi-temporal landsat tm data.Interna- tional Journal of Advanced Earth Science and Engineering, 5(1):333–346, 2016. 2

2016
[2]

Urban green space cooling effect in cities.Heliyon, 5(4):e01339, 2019

Farshid Aram, Ester Higueras Garc ´ıa, Ebrahim Solgi, and Soran Mansournia. Urban green space cooling effect in cities.Heliyon, 5(4):e01339, 2019. 1

2019
[3]

Balany, A

F. Balany, A. W. M. Ng, N. Muttil, S. Muthukumaran, and M. S. Wong. Green infrastructure as an urban heat island mitigation strategy—a review.Water, 12(12):3577, 2020. 2

2020
[4]

Markham, and Dennis L

Gyanesh Chander, Brian L. Markham, and Dennis L. Helder. Summary of current radiometric calibration coefficients for landsat mss, tm, etm+, and eo-1 ali sensors.Remote Sensing of Environment, 113(5):893–903, 2009. 1

2009
[5]

Us building height, 2025

Yangzi Che. Us building height, 2025. 2, 5

2025
[6]

McCann, Marc L

Hyungjin Chung, Jeongsol Kim, Michael T. McCann, Marc L. Klasky, and Jong Chul Ye. Diffusion posterior sampling for general noisy inverse problems. InInter- national Conference on Learning Representations (ICLR),
[7]

Spotlight Paper, arXiv:2209.14687. 3

work page internal anchor Pith review arXiv
[8]

Diffusion models beat GANs on image synthesis.Journal of Machine Learn- ing Research, 22(253):1–105, 2021

Prafulla Dhariwal and Alexander Nichol. Diffusion models beat GANs on image synthesis.Journal of Machine Learn- ing Research, 22(253):1–105, 2021. 3

2021
[9]

JMA Duncan, Bryan Boruff, Alex Saunders, Qian Sun, Joe Hurley, and Marco Amati. Turning down the heat: An en- hanced understanding of the relationship between urban veg- etation and surface temperature at the city scale.Science of the Total Environment, 656:118–128, 2019. 2

2019
[10]

Ilvr: Conditioning method for denoising diffusion probabilistic models.arXiv preprint arXiv:2108.02938, 2021

Takahiro Ezaki, Koki Nagano, and Hiroharu Kato. Ilvr: Con- ditioning method for denoising diffusion probabilistic mod- els.arXiv preprint, abs/2108.02938, 2021. 3

work page arXiv 2021
[11]

Classifier-Free Diffusion Guidance

Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. InProceedings of the NeurIPS 2022 Workshop on Deep Generative Models, 2022. arXiv:2207.12598. 3, 5

work page internal anchor Pith review Pith/arXiv arXiv 2022
[12]

Denoising Diffusion Probabilistic Models

Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. InAdvances in Neural Information Processing Systems, pages 6840–6851, 2020. arXiv:2006.11239. 3

work page internal anchor Pith review arXiv 2020
[13]

Weather prediction with diffusion guided by realistic forecast processes.arXiv preprint arXiv:2402.06666, 2024

Zhanxiang Hua, Yutong He, Chengqian Ma, and Alexan- dra Anderson-Frey. Weather prediction with diffusion guided by realistic forecast processes.arXiv preprint arXiv:2402.06666, 2024. 3

work page arXiv 2024
[14]

A. R. Huete, H. Q. Liu, K. Batchily, and W. J. D. A. Van Leeuwen. A comparison of vegetation indices over a global set of tm images for eos-modis.Remote Sensing of Environment, 59(3):440–451, 1997. 2

1997
[15]

H. M. Imran, M. I. Shammas, A. Rahman, S. J. Jacobs, A. W. Ng, and S. Muthukumaran. Causes, modeling and mitigation of urban heat island: A review.Earth Sciences, 10(6):244– 264, 2021. 2

2021
[16]

Elucidating the Design Space of Diffusion-Based Generative Models

Tero Karras, Samuli Laine, Miika Aittala, Juho Hellsten, Jaakko Lehtinen, and Timo Aila. Elucidating the de- sign space of diffusion-based generative models. InAd- vances in Neural Information Processing Systems, 2022. arXiv:2206.00364. 2, 3, 4, 5

work page internal anchor Pith review arXiv 2022
[17]

Lobell, and Ste- fano Ermon

Samar Khanna, Patrick Liu, Linqi Zhou, Chenlin Meng, Robin Rombach, Marshall Burke, David B. Lobell, and Ste- fano Ermon. Diffusionsat: A generative foundation model for satellite imagery. InInternational Conference on Learn- ing Representations (ICLR), 2024. Poster / Conference Pa- per. 3

2024
[18]

Krishnaswamy, M

J. Krishnaswamy, M. C. Kiran, and K. N. Ganeshaiah. Tree model based eco-climatic vegetation classification and fuzzy mapping in diverse tropical deciduous ecosystems using multi-season ndvi.International Journal of Remote Sensing, 25(6):1185–1205, 2004. 2

2004
[19]

Lindberg and C

F. Lindberg and C. S. B. Grimmond. Nature of vegetation and building morphology characteristics across a city: Influ- ence on shadow patterns and mean radiant temperatures in london.Urban Ecosystems, 14(4):617–634, 2011. 2

2011
[20]

Manago, Terri S

Elizaveta Litvak, Kimberly F. Manago, Terri S. Hogue, and Diane E. Pataki. Evapotranspiration of urban landscapes in los angeles, california at the municipal scale.Water Re- sources Research, 53(5):4236–4252, 2017. 2

2017
[21]

Night and day: The influence and relative importance of urban charac- teristics on remotely sensed land surface temperature.Re- mote Sensing of Environment, 247:111861, 2020

TM Logan, B Zaitchik, S Guikema, and A Nisbet. Night and day: The influence and relative importance of urban charac- teristics on remotely sensed land surface temperature.Re- mote Sensing of Environment, 247:111861, 2020. 2

2020
[22]

Loveland and John L

Thomas R. Loveland and John L. Dwyer. Landsat: Building a strong future.Remote Sensing of Environment, 122:22–29,
[23]

Repaint: Inpainting using de- noising diffusion probabilistic models

Andreas Lugmayr, Martin Danelljan, Adri `a Romero, Luc Van Gool, and Radu Timofte. Repaint: Inpainting using de- noising diffusion probabilistic models. InComputer Vision – ECCV 2022, pages 242–258. Springer, 2022. Lecture Notes in Computer Science, arXiv:2106.01322. 3

work page arXiv 2022
[24]

The influence of land-cover type on the relationship between ndvi–lst and lst-t air.International Journal of Remote Sens- ing, 39(5):1377–1398, 2018

Forough Marzban, Sahar Sodoudi, and Ren ´e Preusker. The influence of land-cover type on the relationship between ndvi–lst and lst-t air.International Journal of Remote Sens- ing, 39(5):1377–1398, 2018. 2

2018
[25]

Massaro, R

E. Massaro, R. Schifanella, M. Piccardo, L. Caporaso, H. Taubenb¨ock, A. Cescatti, and G. Duveiller. Spatially- optimized urban greening for reduction of population expo- sure to land surface temperature extremes.Nature Commu- nications, 14(1):2903, 2023. 2

2023
[26]

Urban climates and climate change.Annual Review of Environment and Resources, 45(1):411–444, 2020

Val ´ery Masson, Aude Lemonsu, Julia Hidalgo, and James V oogt. Urban climates and climate change.Annual Review of Environment and Resources, 45(1):411–444, 2020. 1

2020
[27]

N. Muse, A. Clement, and K. J. Mach. Daytime land surface temperature and its limits as a proxy for surface air tempera- ture in a subtropical, seasonally wet region.PLOS Climate, 3(10):e0000278, 2024. 2

2024
[28]

FourCastNet: A Global Data-driven High-resolution Weather Model using Adaptive Fourier Neural Operators

Jaideep Pathak, Shashank Subramanian, Peter Harrington, Sanjeev Raja, Ashesh Chattopadhyay, Morteza Mardani, Thorsten Kurth, David Hall, Zongyi Li, Kamyar Azizzade- nesheli, et al. Fourcastnet: A global data-driven high- resolution weather model using adaptive fourier neural op- erators.arXiv preprint arXiv:2202.11214, 2022. 3

work page internal anchor Pith review arXiv 2022
[29]

Shushi Peng, Shilong Piao, Philippe Ciais, Pierre Friedling- stein, Catherine Ottl ´e, Franc ¸ois-Marie Br´eon, Huijuan Nan, Liming Zhou, and Ranga B. Myneni. Surface urban heat is- land across 419 global big cities.Environmental Science & Technology, 46(2):696–703, 2012. 1

2012
[30]

arXiv:2312.15796 [physics]

Ilan Price, Alvaro Sanchez-Gonzalez, Ferran Alet, Tom R Andersson, Andrew El-Kadi, Dominic Masters, Timo Ewalds, Jacklynn Stott, Shakir Mohamed, Peter Battaglia, et al. Gencast: Diffusion-based ensemble forecasting for medium-range weather.arXiv preprint arXiv:2312.15796,

work page arXiv
[31]

U-net: Convolutional networks for biomedical image segmentation

Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. InMedical Image Computing and Computer-Assisted Inter- vention (MICCAI), pages 234–241. Springer, 2015. 2

2015
[32]

J. A. Sobrino, J. C. Jim ´enez-Mu˜noz, and L. Paolini. Land surface temperature retrieval from landsat tm 5.Remote Sensing of Environment, 90(4):434–440, 2004. 1

2004
[33]

Score-based generative modeling through stochastic differential equa- tions

Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equa- tions. InInternational Conference on Learning Represen- tations, 2021. 3, 4

2021
[34]

I. D. Stewart and T. R. Oke. Local climate zones for urban temperature studies.Bulletin of the American Meteorologi- cal Society, 93(12):1879–1900, 2012. 2

1900
[35]

Suthar, N

G. Suthar, N. Kaul, S. Khandelwal, and S. Singh. Predicting land surface temperature and examining its relationship with air pollution and urban parameters in bengaluru: A machine learning approach.Urban Climate, 53:101830, 2024. 2

2024
[36]

Touhami, H

I. Touhami, H. Moutahir, D. Assoul, K. Bergaoui, H. Aouinti, J. Bellot, and J. M. Andreu. Multi-year monitor- ing land surface phenology in relation to climatic variables using modis-ndvi time-series in mediterranean forest, north- east tunisia.Acta Oecologica, 114:103804, 2022. 2

2022
[37]

Census Bureau

U.S. Census Bureau. 2024 TIGER/Line Shapefiles, 2024. Accessed: 2026-02-05. 5

2024
[38]

Varamesh, S

S. Varamesh, S. Mohtaram Anbaran, B. Shirmohammadi, N. Al-Ansari, S. Shabani, and A. Jaafari. How do different land uses/covers contribute to land surface temperature and albedo?Sustainability, 14(24):16963, 2022. 2

2022
[39]

J. A. V oogt and T. R. Oke. Thermal remote sensing of urban climates.Remote Sensing of Environment, 86(3):370–384,
[40]

S. Y . Xue, H. Y . Xu, C. C. Mu, T. H. Wu, W. P. Li, W. X. Zhang, and X. D. Wu. Changes in different land cover areas and ndvi values in northern latitudes from 1982 to 2015.Ad- vances in Climate Change Research, 12(4):456–465, 2021. 2

1982
[41]

Probabilistic weather forecasting with de- terministic guidance-based diffusion model

Donggeun Yoon, Minseok Seo, Doyi Kim, Yeji Choi, and Donghyeon Cho. Probabilistic weather forecasting with de- terministic guidance-based diffusion model. InEuropean Conference on Computer Vision, pages 108–124. Springer,
[42]

Additional Dataset-Level Analysis We provide an additional dataset-level analysis to illustrate the one-to-many nature of the inverse problem

3 Conflated Inverse Modeling to Generate Diverse and Temperature-Change Inducing Urban Vegetation Patterns Supplementary Material S1. Additional Dataset-Level Analysis We provide an additional dataset-level analysis to illustrate the one-to-many nature of the inverse problem. Specifically, we measure tile-level NDVI variability after grouping tiles into s...