Anthropogenic Heat in Urban Climate Systems: Forcing, Sensitivity, and Feedback

Abstract: Anthropogenic heat flux, the heat released to the environment from human activities such as building energy use, transportation, industrial processes, and human metabolism, is a defining feature of the urban climate system. It is an important contributor to the urban heat island (UHI) effect and influences a wide range of urban meteorological processes. Its significance extends beyond urban climatology because of its close connections to energy consumption, greenhouse gas emissions, and climate adaptation. Despite decades of research, anthropogenic heat flux remains one of the least well-constrained components of the urban energy balance. Moreover, its climatic significance has often been assessed from an applied perspective, with less emphasis on developing transferable physical understanding of how anthropogenic heat flux acts as a forcing, how urban temperatures respond, and how feedbacks modify that response. This review develops a forcing-response-feedback framework for synthesizing current understanding of the role of anthropogenic heat flux in the urban climate system and identifies priorities for future research.

• The paper develops a forcing–response–feedback framework showing that anthropogenic heat sensitivity is typically about 0.017 K (W m−2)−1, with baseline heat-loss processes dominating urban temperature response.
• The paper finds that sensitivity increases under weak nighttime and wintertime mixing, reaching median values of 0.040 K (W m−2)−1 on summer nights and 0.063 K (W m−2)−1 on winter nights.
• The paper shows that spatial smoothing, proxy-based downscaling, release pathways, model structure, and unresolved feedbacks can substantially alter estimates, motivating standardized datasets and coordinated model intercomparisons.

This review synthesizes more than five decades of research on anthropogenic heat flux ($Q_F$) — the sensible heat released to the environment by building energy use, transportation, industry, and human metabolism — and reinterprets it through a climate-science lens of forcing, sensitivity, and feedback (2608.17782). The central premise is that the climatic significance of $Q_F$ depends not only on its magnitude but also on the temperature sensitivity of the urban system to that forcing and on feedbacks that amplify or damp the response. The review organizes a corpus of over 500 publications (assembled with LLM-assisted extraction) around a canopy-air energy balance model, yielding a unified framework that connects applied urban meteorology to global climate feedback analysis.

A forcing–response–feedback framework

The theoretical core is a linearized energy budget for a control volume of urban canopy air:

$C\frac{dT}{dt} = Q_F - (\lambda T + c),$

where $C = \rho c_p h$ is an effective heat capacity per unit area and $\lambda$ represents heat-loss efficiency. In the reference case with no feedbacks, the equilibrium sensitivity is $S_0 = 1/\lambda_0$. Using representative conductances ($C_{ha} + C_{hs} \sim 0.1$ m s$^{-1}$), $\lambda$ is of order $100$ W m$Q_F$0 K$Q_F$1, implying a reference sensitivity near $Q_F$2 K (W m$Q_F$3)$Q_F$4 — a value that independently matches both the observational benchmark of Kikegawa et al. (2014) and the median of numerical estimates synthesized in the review.

When feedbacks are included, the total feedback parameter decomposes as $Q_F$5, where restoring feedbacks arise when heat-exchange conductances or boundary temperatures depend on canopy air temperature, and source feedbacks arise when $Q_F$6 itself responds to temperature (e.g., air-conditioning demand increasing with outdoor heat). The framework recovers earlier gain-factor formulations ($Q_F$7) while generalizing them to include restoring feedbacks. A key conceptual contribution is the distinction between forcing-based sensitivity ($Q_F$8, normalized by the imposed forcing) and effective sensitivity ($Q_F$9, normalized by the total realized change in $C\frac{dT}{dt} = Q_F - (\lambda T + c),$0 including source-feedback contributions). This distinction matters because, counterintuitively, studies that include positive source feedback report lower median sensitivities under the effective definition: the feedback amplifies temperature but simultaneously inflates the denominator. The authors note plainly that no generally accepted method exists for constructing the counterfactual $C\frac{dT}{dt} = Q_F - (\lambda T + c),$1 needed to recover $C\frac{dT}{dt} = Q_F - (\lambda T + c),$2 from coupled simulations, and some proposed approaches violate energy conservation.

Anthropogenic heat as a structured forcing

The review traces the evolution of $C\frac{dT}{dt} = Q_F - (\lambda T + c),$3 estimation from Eaton's 1878 coal-based estimate (~12 W m$C\frac{dT}{dt} = Q_F - (\lambda T + c),$4 for London) through inventory methods to remote sensing, data fusion, and machine learning. Global annual mean $C\frac{dT}{dt} = Q_F - (\lambda T + c),$5 is only ~0.01 W m$C\frac{dT}{dt} = Q_F - (\lambda T + c),$6, but this masks extreme concentration: peak values reach 1590 W m$C\frac{dT}{dt} = Q_F - (\lambda T + c),$7 in central Tokyo in winter and up to 6000 W m$C\frac{dT}{dt} = Q_F - (\lambda T + c),$8 for data centers in Phoenix. Eight global datasets are compared, showing clear progress from 0.5° population-based disaggregation toward hourly, sector-resolved products at 500 m resolution. However, increased realism does not imply reduced uncertainty. Comparisons over Kinshasa, London, and Singapore show that datasets agree better at city-scale averages than at hotspots; in Singapore, local inventories report a building-only mean of ~10.1 W m$C\frac{dT}{dt} = Q_F - (\lambda T + c),$9 with commercial peaks of 663 W m$C = \rho c_p h$0, yet global products diverge widely depending on whether nighttime lights, population, or point-source data drive spatial allocation. Two dominant uncertainty sources are identified: spatial smoothing of compact high-emission zones and proxy sensitivity in downscaling assumptions. The review stresses that global datasets are scale-dependent estimates encoding assumptions about human activity, not direct observations.

A further unresolved issue is that sectoral partitioning does not equal pathway representation: heat released at street level by vehicles, ejected from rooftop HVAC units, or emitted from industrial stacks interacts with the urban climate differently, and release location alone can alter thermal effects by up to an order of magnitude (2608.17782).

Sensitivity: magnitude, variability, and uncertainty

From 44 modeling studies, the review extracts 81 sensitivity estimates using strict temporal and spatial consistency criteria (excluding peak-to-peak ratios used in prior syntheses). The study-level median is 0.017 K (W m$C = \rho c_p h$1)$C = \rho c_p h$2, with an interquartile range of 0.008–0.029, supporting a practical rule-of-thumb of order 0.01 K (W m$C = \rho c_p h$3)$C = \rho c_p h$4. Sensitivity varies systematically with timing: medians rise from 0.012 (summer daytime) to 0.040 (summer nighttime) and from 0.018 (winter daytime) to 0.063 K (W m$C = \rho c_p h$5)$C = \rho c_p h$6 (winter nighttime), consistent with weaker turbulent mixing and shallower boundary layers suppressing heat removal. Notably, seven pre-2000 studies yield a statistically indistinguishable median from post-2000 studies, indicating that the basic physics was captured early; what has changed is awareness of definitional issues.

Methodological variability is substantial. Definitions of $C = \rho c_p h$7 differ between inventory- and building-energy perspectives; reference areas are inconsistent across model types (areal fluxes versus line or volumetric sources); and spatial supports may be incompatible — in WRF-SLUCM, $C = \rho c_p h$8 is defined per unit impervious area while canopy temperature applies only to canyon air, so diagnosed sensitivity depends on roof-to-canyon-floor area ratios. One outlier estimate of ~1.2 K (W m$C = \rho c_p h$9)$\lambda$0 from Jacobson (2014) reflects division by a tiny globally averaged forcing and illustrates how internal variability can dominate when the forced signal is small. The review argues that future global studies should use ensembles and report statistical uncertainty explicitly, and that structural uncertainty — differences in UCM structure and release pathways approaching an order of magnitude — requires coordinated model intercomparisons.

Observational constraints remain sparse. Natural experiments exploiting weekday–weekend contrasts, lockdowns, and summits rarely quantify $\lambda$1 consistently, and statistical regressions sometimes yield physically counterintuitive negative coefficients or seasonal patterns opposite to expectations, underscoring dependence on model formulation rather than physics alone. The Osaka/Tokyo weekday–weekend analysis stands out as one of the clearest observational benchmarks, yielding ~0.01 K (W m$\lambda$2)$\lambda$3.

Feedbacks: baseline, restoring, and source

The interquartile range of $\lambda$4 (0.01–0.033 K (W m$\lambda$5)$\lambda$6) corresponds to a total feedback parameter $\lambda$7 of roughly −100 to −30 W m$\lambda$8 K$\lambda$9. Synthesis of the few studies quantifying components explicitly suggests:

Feedback component	Typical magnitude	Key sources
Baseline ($S_0 = 1/\lambda_0$0)	~−120 W m$S_0 = 1/\lambda_0$1 K$S_0 = 1/\lambda_0$2 (CONUS climatology); spans −100 to −10 (max-advection limit) down to ~−1.8 (min-advection)	Wang et al.; Ginzburg & Demchenko
Restoring ($S_0 = 1/\lambda_0$3)	~0 (summer) to ~+10 W m$S_0 = 1/\lambda_0$4 K$S_0 = 1/\lambda_0$5 (winter)	Wang et al.
Source ($S_0 = 1/\lambda_0$6)	~+0.6 to +4.9 (cooling); ~−6.8 to −0.4 (heating) W m$S_0 = 1/\lambda_0$7 K$S_0 = 1/\lambda_0$8	BEM simulations; empirical demand–temperature relations

The baseline feedback dominates: because it is nearly an order of magnitude larger than the restoring term, it largely sets anthropogenic heat sensitivity, explaining why so many studies cluster near 0.01 K (W m$S_0 = 1/\lambda_0$9)$C_{ha} + C_{hs} \sim 0.1$0. Spatial variability of temperature response under uniform forcing is likewise controlled primarily by baseline rather than restoring feedbacks. Source feedback gain factors of roughly 10% (commercial) and 20% (residential) were diagnosed in Osaka via energy-conserving weekday/holiday simulation pairs, implying amplification of the reference response by those fractions. The review cautions that regression-based diagnostics of $C_{ha} + C_{hs} \sim 0.1$1 and $C_{ha} + C_{hs} \sim 0.1$2 should be read as operational estimates rather than strict partial derivatives, since temperature covaries with other meteorological controls on energy demand.

An important emerging complication involves electrification: air-source heat pumps reduce conventional wintertime $C_{ha} + C_{hs} \sim 0.1$3, but colder conditions increase heat extraction from outdoor air, creating an amplifying winter-cooling pathway analogous to a positive source feedback that has not been formally quantified.

Limitations and open questions

The review is candid about the preliminary state of its synthesis. Feedback parameters have been quantified in only a handful of studies with differing models, scales, and diagnostic conventions, so generalization across cities, climates, and modeling systems remains untested. Several structural open problems are identified:

• Scale dependence: diagnosed sensitivity depends on temporal averaging windows relative to adjustment timescales (canopy air adjusts in ~100–1000 s, whereas surface, boundary-layer, and source-feedback adjustments are slower) and on spatial domains that rarely match the true response footprint, which itself is poorly characterized.
• Pattern dependence: whether temporally episodic or spatially concentrated forcing produces different responses than uniform forcing at equal domain means is unknown; idealized experiments varying forcing structure at fixed total input are proposed.
• Counterfactual definition: isolating source feedback without violating energy conservation in coupled building–atmosphere systems remains unsolved.
• Linearity: evidence is mixed — offline simulations show declining sensitivity at large forcings, regional simulations suggest approximate linearity, and large-eddy simulations reveal power-law scaling under calm conditions — so the validity range of the local linearization is unresolved.
• Decomposition ambiguity: the split of restoring feedback into $C_{ha} + C_{hs} \sim 0.1$4 and $C_{ha} + C_{hs} \sim 0.1$5 terms depends on the choice of temperature origin; only the total restoring feedback is physically meaningful.

Conclusion

This review reframes anthropogenic heat research within a forcing–response–feedback architecture adapted from global climate science, distinguishing reference, forcing-based, and effective sensitivities and baseline, restoring, and source feedbacks. Its principal quantitative results — a rule-of-thumb sensitivity of order 0.01 K (W m$C_{ha} + C_{hs} \sim 0.1$6)$C_{ha} + C_{hs} \sim 0.1$7, systematic nocturnal and wintertime enhancement, and baseline feedback dominance — provide transferable physical understanding where the literature previously offered city-specific case studies. The framework also exposes definitional ambiguities that complicate cross-study comparison and identifies concrete priorities: standardized and benchmarked $C_{ha} + C_{hs} \sim 0.1$8 datasets with scenario-based futures, explicit representation of release pathways, ensemble-based sensitivity diagnosis, and consistent feedback-parameter estimation. The broader lesson is that progress requires attending to the forcing–response relationship as much as to the forcing itself, an approach the authors suggest could extend to other urban interventions such as reflective roofs.