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

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

cs.CVApr 14, 2026

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

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

Organizations: Computer Science Purdue University, West Lafayette, IN, USA · Forestry and Natural Resources Purdue University, West Lafayette, IN, USA

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.

Figures & tables

Figure 1: Our conflated framework 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 (left pair temperature decrease and right pair temperature increase). Higher NDVI values (closer to 1) indicate denser vegetation, while lower NDVI values indicate sparser vegetation.

Figure 2: We show our system pipeline, including inverse and forward model, and the training and inference processes.

Figure 3: Diversity. Diverse NDVI generation under identical conditioning ( $\Delta$ = 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.

Figure 4: Chicago Specificity. Images show the ability to alter regional temperature to specific changes in a Chicago tile. Top: generated NDVI for $\Delta_{\text{target}}\in\{-2,-1,0,+1,+2\}^{\circ}$ C. Bottom: forward-model predicted LST. Numbers indicate ROI mean $\Delta_{\text{pred}}$ .

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 $\Delta_{\text{target}}\in\{-2,-1,0,+1,+2\}^{\circ}$ C. Bottom: forward-model predicted LST. Numbers indicate ROI mean $\Delta_{\text{pred}}$ .

$\lambda$	Best $w$	CtrlErr ${}_{\text{avg}}$	BaseErr ${}_{\text{ROI}}$ $(\Delta=0)$	Diversity
1	5	$0.772\pm 0.350$	1.49	$0.707\pm 0.228$
2	3	$0.715\pm 0.301$	1.51	$0.830\pm 0.118$
4	3	$0.637\pm 0.230$	1.53	$0.879\pm 0.070$
6	3	$0.609\pm 0.251$	1.55	$0.888\pm 0.064$
8	3	$0.573\pm 0.257$	1.59	$0.889\pm 0.062$
16	3	$0.559\pm 0.310$	1.63	$0.905\pm 0.052$

Table 1: Training. Results at ROI scale (

k=32

) and for various values for

\lambda

. For each configuration, we report the best temperature control error (CtrlErr

{}_{\text{avg}}

) over

w\in\{1,2,3,5,8\}

. BaseErr

{}_{\text{ROI}}

(

\Delta

= 0) denotes the absolute difference between the predicted and ground-truth mean ROI temperature for samples generated with

\Delta

= 0.

Model	LST Coarse	Physics	Best $w$	CtrlErr ${}_{\text{avg}}$	BaseErr ${}_{\text{ROI}}$ ( $\Delta$ = 0)	Diversity
U-Net	✗	✗	2	$0.211\pm 1.097$	3.267	-
Fine LST	✗	✗	3	$0.746\pm 0.341$	1.50	$0.265\pm 0.183$
Coarse LST only	✓	✗	8	$0.882\pm 0.331$	1.57	$0.341\pm 0.240$
Coarse + Physics	✓	✓	3	$0.559\pm 0.310$	1.633	$0.905\pm 0.052$

Table 2: Comparisons. We compare several configurations, each at the best

w

and

\lambda

value (if applicable). As observed, our method (last row) produces the lowest

CtrlErr

and highest

Diversity

values.

Figure 6: Gain control. Controllability as a function of $w$ gain. Y-axis shows ROI Mean $\Delta T$ Absolute Error (°C). Lower is better.

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.

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.

Panel of Figure 9 — Figure 9: Additional city-level qualitative results. Each row shows NDVI generation under $\Delta\in\{-2,0,+2\}^{\circ}$ C with the ROI mean predicted change $\Delta_{\text{pred}}$ .

Architecture
Parameter	Value
Backbone	NCSN++
Image size	$128\times 128$
Base channels ( $nf$ )	128
Channel multipliers	$(1,2,2,2)$
Residual blocks / resolution	4

Table 3: Inverse diffusion model hyperparameters.

Explore similar work

Aug 24, 2026

UHI-Bench: Benchmarking Dual-Source Urban Heat Island Modeling Across Cities in Diverse Climate Regimes

Urban heat islands (UHIs) are intensifying under climate change, exacerbating thermal exposure risks. Their two primary observations, land surface temperature UHI (LST-UHI) and near-surface air temperature UHI (AirT-UHI), capture physically distinct aspects of urban heat. However, most studies rely on a single source, and substituting one for the other can substantially bias the magnitude and spatial variability of human heat exposure. Accurate UHI modeling also requires dynamic meteorological drivers and static urban morphology features, but spatiotemporal incompatibilities hinder their alignment. Cloud gaps in LST observations and sparse AirT station networks further limit dual-source UHI modeling, motivating cross-city transfer across diverse climates. To bridge these gaps, we introduce UHI-Bench, the first UHI benchmark for dual-source UHI modeling that integrates dynamic and static environmental context. Following a unified signal, mechanism, and transfer framework, it evaluates over 20 baselines from four model families on five tasks across 20 cities and nine Köppen climate classes. Results show that no model is uniformly best, although foundation models remain consistently competitive and stable. Environmental covariates generally improve performance, but their utility varies across sources and tasks. Cross-city transferability is better explained by overlap in UHI regimes than by climate-zone similarity. With the dataset and standardized pipeline, our work provides practical guidance for urban heat modeling, promotes climate data equity, and supports future advances in climate research.

Wanyun Ling, Chenxi Liu, Yi Xie +3

Department of Operations and Technology Technical University of Munich Heilbronn, Germany · Hong Kong Institute of Science & Innovation, Chinese Academy of Sciences Hong Kong SAR, China · Faculty of Computer Science RWTH Aachen University Aachen, Germany +2

Benchmark Design Climate Science Transfer Learning

Ask PDF

Feb 23, 2026

HeatPrompt: Zero-Shot Vision-Language Modeling of Urban Heat Demand from Satellite Images

Accurate heat-demand maps play a crucial role in decarbonizing space heating, yet most municipalities lack detailed building-level data needed to calculate them. We introduce HeatPrompt, a zero-shot vision-language energy modeling framework that estimates annual heat demand using semantic features extracted from satellite images, basic Geographic Information System (GIS), and building-level features. We feed pretrained Large Vision Language Models (VLMs) with a domain-specific prompt to act as an energy planner and extract the visual attributes such as roof age, building density, etc, from the RGB satellite image that correspond to the thermal load. A Multi-Layer Perceptron (MLP) regressor trained on these captions shows an

R^2

uplift of 93.7% and shrinks the mean absolute error (MAE) by 30% compared to the baseline model. Qualitative analysis shows that high-impact tokens align with high-demand zones, offering lightweight support for heat planning in data-scarce regions.

Kundan Thota, Xuanhao Mu, Thorsten Schlachter +1

Institute for Automation and Applied Informatics (IAI), Karlsruhe Institute of Technology (KIT), Germany

Building Energy Management Vision-Language Models

Ask PDF

Jun 10, 2026

Urban Heat MiniCubes: An AI-Ready dataset for urban heat research

Urban heat is amplified by impermeable surfaces and heterogeneous built environments, yet street-level variability remains difficult to quantify because multi-sensor observations are rarely available in consistent, analysis-ready form at the necessary spatiotemporal scales. We present "Urban Heat MiniCubes," a publicly available, FAIR-oriented dataset designed for machine learning applications in urban heat research. The dataset provides harmonized 90 x 90 km gridded data cubes for 48 cities in the Western Hemisphere spanning 2022-2023, with variables reprojected and collocated to a common grid to reduce preprocessing (e.g., reprojection, resampling, and spatiotemporal alignment). Urban Heat MiniCubes includes two complementary modalities: (i) higher-spatial-resolution, lower-frequency observations from Landsat 8/9 (e.g., surface reflectances) and Sentinel-1 (e.g., synthetic aperture radar backscatter), and (ii) higher-temporal-frequency, coarser observations from GOES-R (e.g., longwave infrared brightness temperatures) and a microwave land surface temperature product. We document variables and metadata and provide technical assessment using inter-variable analyses and autoencoder-based reconstruction-error summaries across pixel classes (e.g., water and cloud). Potential use cases and limitations are also discussed.

Jonathan Starfeldt, Maria J. Molina, Alexander Kerr +3

Department of Atmospheric and Oceanic Science, University of Maryland, College Park, MD, USA. · Department of Computer Science, University of Maryland, College Park, MD, USA. · NASA Goddard Space Flight Center, Greenbelt, MD, USA. +1

Remote Sensing

Ask PDF