Cities are typical sources of carbon and a focus of climate-change mitigation. Although there is great potential for reducing emissions in cities, constructing low-carbon emission cities under the carbon-emission reduction target of the 2016 Paris Agreement remains challenging, as there is little scientific evidence for use by government decision-makers. To solve this problem, we must understand the urban carbon-cycle process in greater detail, especially the spatiotemporal variations in carbon sources and sinks. This project maps the urban carbon cycle from both directions: accurate carbon-sink mapping driven by forest inventory and machine learning, fine-scale CO₂ emission mapping with uncertainty assessment, the relationships between CO₂ emissions and urban form, and the application-oriented local carbon emission zone construction.

1 Carbon sink mapping with machine learning and spatial statistics

Forests are the only precisely measurable carbon sink. How can we map carbon sinks accurately at low cost? We proposed a combined approach that fuses machine learning with spatial statistics to construct a regional biomass map from non-representative sample units: three machine learning models (SVM, RBF-ANN, and RF) generate plot-level estimates, which are then scaled up through the P-BSHADE spatial statistical model. The combined models outperformed single models in accuracy, providing a practical pathway from sparse field inventory to wall-to-wall biomass mapping.

Figure 1. The framework for estimating (a–c) the machine learning models, (d) the P-BSHADE model, and the three models that combine machine learning with the P-BSHADE model ((a,d), (b,d) and (c,d)).

Figure 2. The accuracy of different methods for biomass estimation.

【动态】城市环境研究所在提升森林样地水平地上生物量的估算精度方面取得进展

The paper was published in Forests

2 Gridded CO₂ emission maps and uncertainty assessment

Cities are the main sources of CO₂ emissions and thus must abate emissions responsibly. We produced fine-scale (residential and transport) CO₂ emission gridded maps by fusing global downscaled and bottom-up elements, and compared them with existing inventories to verify the spatial patterns. Because large uncertainties exist at the urban scale, we further proposed an analytic workflow based on Monte Carlo simulation and bootstrap sampling to quantify the uncertainties propagated from the gridded model and the spatial proxies, finding that fine-resolution (30 m) maps carry a higher degree of uncertainty propagation.

Figure 3. Comparison between our CO₂ emission gridded map and other inventories.

Figure 4. The CO₂ gridded emissions maps and uncertainty maps.

[动态]城市环境研究所在城市尺度二氧化碳排放网格制图的不确定性传播分析取得研究进展

The database paper was published in Data in Brief, and the uncertainty analysis was published in Remote Sensing

3 CO₂ emissions and urban form

Understanding the spatial distribution of CO₂ emissions is the key to designing low-carbon city development policies. We compared the relationships between urban form fragmentation and CO₂ emissions through an analytic framework composed of Pearson correlation analysis, geographically weighted regression (GWR), and the geographical detector method. The results show that urban form indicators—especially the functional mixed entropy—significantly drive CO₂ emissions, with clear spatial heterogeneity across 31 Chinese cities, and the framework can be applied to urban agglomerations, megacities, and small towns.

Figure 5. The analytical framework.

Figure 6. Geographical detector results for fragmentation factors, PUA, and POID for CO₂ emissions from 4 sources.

Figure 7. The relation between the functional mixed entropy and the GWR mean regression coefficients in 31 cities. The red, yellow, light blue and blue dots represent the Type I, II, III and IV city, respectively.

The paper was published in Journal of Cleaner Production

4 Local carbon emission zone construction

To link research findings with planning practice, we constructed the local carbon emission zone (LCEZ), which links highly urbanized morphology profiles with CO₂ emission intensity to summarize the form characteristics for low-emission control in new communities, providing an operational bridge from carbon mapping to urban design guidelines.

Figure 8. The flowchart of the LCEZ construction. Step 1: Luojia1-01 night light image data is the proxy variable of the RTCE map. GWR is geographically weighted regression. Step 2: L is low, M is medium, and H is high; the solid line represents the three key factors; the point is POI; the column is BF; and the line is RD.

The paper was published in Building and Environment