Zachary D. Calhoun
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Zachary D. Calhoun has a total of 16 co-authors. Here are the top co-authors by number of shared publications:
- David Carlson
Duke University
6 shared publications Mike Bergin
2 shared publications- Charles T. Wood
Department of Pediatrics, Duke University School of Medicine
1 shared publication - Jordan Milton Malof
Department of Electrical Engineering and Computer Science, University of Missouri, Columbia, MO 65211, USA
1 shared publication - Keyu Li
Department of Electrical and Computer Engineering, Duke University Pratt School of Engineering , Durham, NC 27705
1 shared publication - Kyle Bradbury
Department of Electrical and Computer Engineering, Duke University, Durham, NC 27708, USA
1 shared publication
Latest Publications
- Open Access
- 7 June 2026
- Published by Elsevier in Urban Climate
- Vol. 67
Probabilistic interpolation of crowdsourced meteorological data for higher-resolution gridded estimates of surface air temperature
- 7 June 2026
- Published by Elsevier in Urban Climate
- Vol. 67
Open AccessProbabilistic interpolation of crowdsourced meteorological data for higher-resolution gridded estimates of surface air temperature
- Open Access
- 26 October 2025
- Published by California Digital Library (CDL) in EarthArXiv
Scalable and robust Gaussian processes for reanalysis of urban air temperature with crowdsourced meteorological data
Crowdsourced air temperature data from networks like Weather Underground offer dense spatial coverage and are increasingly used to study the canopy urban heat island (CUHI) effect. However, these observations are noisy: siting conditions, environmental interference, and sensor failures introduce spatially and temporally varying bias. This complicates interpolation, limiting our ability to estimate neighborhood-level air temperature. While interpolation techniques such as kriging account for uncertainty, they do so under the assumption of homoscedasticity. Moreover, they struggle to scale beyond a few thousand observations, which limits their utility on crowdsourced data. To overcome these limitations, we develop a sparse variational Gaussian process model that accounts for heteroscedasticity, allowing us to efficiently produce interpolated air temperature fields with calibrated uncertainty estimates. To test our approach, we apply our model to six years of hourly data across Durham County, North Carolina. This area includes a medium-sized city, so we expect our model to generalize to similarly sized regions with sufficient sensor coverage. Compared to ERA5-Land, it improves estimates of canopy temperature (MAE=0.57°C versus ERA5-Land MAE=3.20°C on held-out locations) and enables high-resolution analysis of CUHI patterns over space and time. We illustrate this by visualizing (1) how CUHI patterns vary with synoptic conditions, (2) differential impacts on heating and cooling demand, and (3) annual hours exceeding 35°C by neighborhood. Our method provides a scalable and statistically rigorous framework for transforming crowdsourced climate data into a gridded reanalysis product. Using these data, we can better quantify urban heat exposure and its impact on health and energy.
- 26 October 2025
- Published by California Digital Library (CDL) in EarthArXiv
Open AccessScalable and robust Gaussian processes for reanalysis of urban air temperature with crowdsourced meteorological data
Crowdsourced air temperature data from networks like Weather Underground offer dense spatial coverage and are increasingly used to study the canopy urban heat island (CUHI) effect. However, these observations are noisy: siting conditions, environmental interference, and sensor failures introduce spatially and temporally varying bias. This complicates interpolation, limiting our ability to estimate neighborhood-level air temperature. While interpolation techniques such as kriging account for uncertainty, they do so under the assumption of homoscedasticity. Moreover, they struggle to scale beyond a few thousand observations, which limits their utility on crowdsourced data. To overcome these limitations, we develop a sparse variational Gaussian process model that accounts for heteroscedasticity, allowing us to efficiently produce interpolated air temperature fields with calibrated uncertainty estimates. To test our approach, we apply our model to six years of hourly data across Durham County, North Carolina. This area includes a medium-sized city, so we expect our model to generalize to similarly sized regions with sufficient sensor coverage. Compared to ERA5-Land, it improves estimates of canopy temperature (MAE=0.57°C versus ERA5-Land MAE=3.20°C on held-out locations) and enables high-resolution analysis of CUHI patterns over space and time. We illustrate this by visualizing (1) how CUHI patterns vary with synoptic conditions, (2) differential impacts on heating and cooling demand, and (3) annual hours exceeding 35°C by neighborhood. Our method provides a scalable and statistically rigorous framework for transforming crowdsourced climate data into a gridded reanalysis product. Using these data, we can better quantify urban heat exposure and its impact on health and energy.
- Open Access
- 24 September 2025
- Published by Oxford University Press (OUP) in AJE Advances: Research in Epidemiology
- Vol. 1 (3)
Neighborhood environmental and contextual factors improve prediction of childhood body mass index: an application of novel graph neural networks
Childhood obesity is a major risk factor for adult cardiovascular disease. Current obesity-prediction models were not developed in diverse populations and do not include heterogeneous social, environmental, and climate factors that may impact body mass index across the full pediatric spectrum. Additionally, they consider only the immediate neighborhood within which a child lives, ignoring contextual factors from expanded (ie, distal) neighborhoods. This study uses expanded neighborhoods’ social, environmental, and climate data to improve individual-level body mass index prediction–from underweight through obesity–using a novel machine learning approach. We obtained demographic and clinical data from the electronic health records of the Duke University Health System, identifying 12,226 children aged 6-18 years in Durham County, North Carolina, with body mass index data from 2014 to 2022. Participants’ data were linked to socioeconomic and environmental information at the census block group level. We captured expanded neighborhood effects with a graph neural network and combined this information with individual-level factors to predict body mass index. Our model predicted body mass index more accurately than simpler models for children aged 6–11 (R2 = 0.234, mean absolute error = 3.352, root mean square error = 4.370) and 12–18 (R2 = 0.147, mean absolute error = 4.980, root mean square error = 6.343) using all features. Key predictive factors identified included rent burden, poverty rate, and tree coverage. This research highlights the value of including broader socioeconomic and environmental factors in body mass index prediction, offering insights that could guide targeted, neighborhood-level interventions.
- 24 September 2025
- Published by Oxford University Press (OUP) in AJE Advances: Research in Epidemiology
- Vol. 1 (3)
Open AccessNeighborhood environmental and contextual factors improve prediction of childhood body mass index: an application of novel graph neural networks
Childhood obesity is a major risk factor for adult cardiovascular disease. Current obesity-prediction models were not developed in diverse populations and do not include heterogeneous social, environmental, and climate factors that may impact body mass index across the full pediatric spectrum. Additionally, they consider only the immediate neighborhood within which a child lives, ignoring contextual factors from expanded (ie, distal) neighborhoods. This study uses expanded neighborhoods’ social, environmental, and climate data to improve individual-level body mass index prediction–from underweight through obesity–using a novel machine learning approach. We obtained demographic and clinical data from the electronic health records of the Duke University Health System, identifying 12,226 children aged 6-18 years in Durham County, North Carolina, with body mass index data from 2014 to 2022. Participants’ data were linked to socioeconomic and environmental information at the census block group level. We captured expanded neighborhood effects with a graph neural network and combined this information with individual-level factors to predict body mass index. Our model predicted body mass index more accurately than simpler models for children aged 6–11 (R2 = 0.234, mean absolute error = 3.352, root mean square error = 4.370) and 12–18 (R2 = 0.147, mean absolute error = 4.980, root mean square error = 6.343) using all features. Key predictive factors identified included rent burden, poverty rate, and tree coverage. This research highlights the value of including broader socioeconomic and environmental factors in body mass index prediction, offering insights that could guide targeted, neighborhood-level interventions.
- Open Access
- 21 May 2025
- Published by Copernicus GmbH
Big, noisy data: how scalable Gaussian processes can leverage personal weather stations to improve spatiotemporal coverage of urban climate networks
Urban temperature varies dramatically across space and time, yet capturing this variability requires a dense, reliable sensor network—something that is rarely available in practice. Spatiotemporal gaps in data coverage make it difficult to connect localized urban heat stress to health outcomes and energy demand. In this work, we demonstrate how personal weather stations (PWSs) and machine learning can bridge these gaps to improve urban climate monitoring.To show this, we analyze PWS data collected in Durham County, North Carolina, from 2019 to 2024—a network of over 200 sensors recording hourly temperature data, totaling more than 15 million observations. This dataset presents two key sources of bias that must be addressed to ensure reliable urban heat estimates. First, it is preferentially sampled, with a higher density of weather stations in wealthier (and often cooler) neighborhoods. Second, faulty radiation shields on low-cost sensors may positively bias sensor measurements on sunny days.To address these challenges, we explore Gaussian Process Regression (GPR), a flexible machine learning technique that, when defined with a carefully designed covariance structure, can account for non-uniform sensor placement and measurement noise. However, exact GPR is computationally intractable for large spatiotemporal datasets (i.e., > 10,000 observations). To overcome this, we leverage the Variational Nearest Neighbor Gaussian Process (VNNGP), a scalable approximation that enables the application of complex covariance structures to arbitrarily large datasets.Our approach demonstrates that the VNNGP model allows for complex spatiotemporal dependencies to be learned, making them well-suited for urban temperature modeling. Additionally, we show that abundant but noisy PWS data, when integrated with these models, can further improve spatial coverage. Together, these advancements highlight how combining large, imperfect datasets with sophisticated modeling techniques can enhance urban climate monitoring, leading to better heat exposure assessments and more informed environmental policies.
- 21 May 2025
- Published by Copernicus GmbH
Open AccessBig, noisy data: how scalable Gaussian processes can leverage personal weather stations to improve spatiotemporal coverage of urban climate networks
Urban temperature varies dramatically across space and time, yet capturing this variability requires a dense, reliable sensor network—something that is rarely available in practice. Spatiotemporal gaps in data coverage make it difficult to connect localized urban heat stress to health outcomes and energy demand. In this work, we demonstrate how personal weather stations (PWSs) and machine learning can bridge these gaps to improve urban climate monitoring.To show this, we analyze PWS data collected in Durham County, North Carolina, from 2019 to 2024—a network of over 200 sensors recording hourly temperature data, totaling more than 15 million observations. This dataset presents two key sources of bias that must be addressed to ensure reliable urban heat estimates. First, it is preferentially sampled, with a higher density of weather stations in wealthier (and often cooler) neighborhoods. Second, faulty radiation shields on low-cost sensors may positively bias sensor measurements on sunny days.To address these challenges, we explore Gaussian Process Regression (GPR), a flexible machine learning technique that, when defined with a carefully designed covariance structure, can account for non-uniform sensor placement and measurement noise. However, exact GPR is computationally intractable for large spatiotemporal datasets (i.e., > 10,000 observations). To overcome this, we leverage the Variational Nearest Neighbor Gaussian Process (VNNGP), a scalable approximation that enables the application of complex covariance structures to arbitrarily large datasets.Our approach demonstrates that the VNNGP model allows for complex spatiotemporal dependencies to be learned, making them well-suited for urban temperature modeling. Additionally, we show that abundant but noisy PWS data, when integrated with these models, can further improve spatial coverage. Together, these advancements highlight how combining large, imperfect datasets with sophisticated modeling techniques can enhance urban climate monitoring, leading to better heat exposure assessments and more informed environmental policies.
- Open Access
Deep Causal Inference for Point-referenced Spatial Data with Continuous Treatments
Causal reasoning is often challenging with spatial data, particularly when handling high-dimensional inputs. To address this, we propose a neural network (NN) based framework integrated with an approximate Gaussian process to manage spatial interference and unobserved confounding. Additionally, we adopt a generalized propensity-score-based approach to address partially observed outcomes when estimating causal effects with continuous treatments. We evaluate our framework using synthetic, semi-synthetic, and real-world data inferred from satellite imagery. Our results demonstrate that NN-based models significantly outperform linear spatial regression models in estimating causal effects. Furthermore, in real-world case studies, NN-based models offer more reasonable predictions of causal effects, facilitating decision-making in relevant applications.
Open AccessDeep Causal Inference for Point-referenced Spatial Data with Continuous Treatments
Causal reasoning is often challenging with spatial data, particularly when handling high-dimensional inputs. To address this, we propose a neural network (NN) based framework integrated with an approximate Gaussian process to manage spatial interference and unobserved confounding. Additionally, we adopt a generalized propensity-score-based approach to address partially observed outcomes when estimating causal effects with continuous treatments. We evaluate our framework using synthetic, semi-synthetic, and real-world data inferred from satellite imagery. Our results demonstrate that NN-based models significantly outperform linear spatial regression models in estimating causal effects. Furthermore, in real-world case studies, NN-based models offer more reasonable predictions of causal effects, facilitating decision-making in relevant applications.
- Open Access
- 17 June 2024
- Published by American Chemical Society (ACS) in Environmental Science & Technology Letters
- Vol. 11 (8) , 845-850
Refining Citizen Climate Science: Addressing Preferential Sampling for Improved Estimates of Urban Heat
- 17 June 2024
- Published by American Chemical Society (ACS) in Environmental Science & Technology Letters
- Vol. 11 (8) , 845-850
Open AccessRefining Citizen Climate Science: Addressing Preferential Sampling for Improved Estimates of Urban Heat
- Open Access
- 1 November 2022
- Published by MDPI AG in Remote Sensing
- Vol. 14 (21) , 5500
Self-Supervised Encoders Are Better Transfer Learners in Remote Sensing Applications
Transfer learning has been shown to be an effective method for achieving high-performance models when applying deep learning to remote sensing data. Recent research has demonstrated that representations learned through self-supervision transfer better than representations learned on supervised classification tasks. However, little research has focused explicitly on applying self-supervised encoders to remote sensing tasks. Using three diverse remote sensing datasets, we compared the performance of encoders pre-trained through both supervision and self-supervision on ImageNet, then fine-tuned on a final remote sensing task. Furthermore, we explored whether performance benefited from further pre-training on remote sensing data. Our experiments used SwAV due to its comparably lower computational requirements, as this method would prove most easily replicable by practitioners. We show that an encoder pre-trained on ImageNet using self-supervision transfers better than one pre-trained using supervision on three diverse remote sensing applications. Moreover, self-supervision on the target data alone as a pre-training step seldom boosts performance beyond this transferred encoder. We attribute this inefficacy to the lower diversity and size of remote sensing datasets, compared to ImageNet. In conclusion, we recommend that researchers use self-supervised representations for transfer learning on remote sensing data and that future research should focus on ways to increase performance further using self-supervision.
- 1 November 2022
- Published by MDPI AG in Remote Sensing
- Vol. 14 (21) , 5500
Open AccessSelf-Supervised Encoders Are Better Transfer Learners in Remote Sensing Applications
Transfer learning has been shown to be an effective method for achieving high-performance models when applying deep learning to remote sensing data. Recent research has demonstrated that representations learned through self-supervision transfer better than representations learned on supervised classification tasks. However, little research has focused explicitly on applying self-supervised encoders to remote sensing tasks. Using three diverse remote sensing datasets, we compared the performance of encoders pre-trained through both supervision and self-supervision on ImageNet, then fine-tuned on a final remote sensing task. Furthermore, we explored whether performance benefited from further pre-training on remote sensing data. Our experiments used SwAV due to its comparably lower computational requirements, as this method would prove most easily replicable by practitioners. We show that an encoder pre-trained on ImageNet using self-supervision transfers better than one pre-trained using supervision on three diverse remote sensing applications. Moreover, self-supervision on the target data alone as a pre-training step seldom boosts performance beyond this transferred encoder. We attribute this inefficacy to the lower diversity and size of remote sensing datasets, compared to ImageNet. In conclusion, we recommend that researchers use self-supervised representations for transfer learning on remote sensing data and that future research should focus on ways to increase performance further using self-supervision.