Ahmed Tarek Hammad
Research Subjects
Ahmed Tarek Hammad has a total of 22 co-authors. Here are the top co-authors by number of shared publications:
- Giacomo Falchetta
CMCC Foundation - Euro-Mediterranean Center on Climate Change (Lecce, Italy)
4 shared publications - Soheil Shayegh
CMCC Foundation—Euro-Mediterranean Center on Climate Change, Milan, Italy
2 shared publications Aleksandra Torbica
25 Cergas—Centre for Research on Health and Social Csare Management, SDA Bocconi School of Management, Bocconi University, Milan, Italy
1 shared publication- Alessia Melegaro
Dondena Centre for Research on Social Dynamics and Public Policy, Bocconi University, Milano, Italy
1 shared publication Alvaro Labella
16 Department of Computer Science, University of Jaén, Jaén, Spain
1 shared publicationAndy South
24 Liverpool School of Tropical Medicine, Liverpool, United Kingdom
1 shared publication
Latest Publications
- Open Access
- 5 August 2026
- Published by Elsevier in SSRN Electronic Journal
Ten Rules for Deploying Large Language Models in Official Statistics
Official statistics is increasingly produced with the help of large language models, which are being asked to write the code of its pipelines and to guide the statistics themselves, work whose outputs carry institutional weight and inform policy. The fluency of these models creates a failure mode distinct from ordinary software error: they produce plausible text whether or not it is correct, so a pipeline that relies on them can fail silently while still looking authoritative, which is at odds with the obligations written down in the Fundamental Principles of Official Statistics and the European Statistics Code of Practice. We offer ten rules for deploying these models safely in official statistics at the current state of the technology. The rules follow one architectural principle: the model serves as a monitoring agent that observes,corrects bounded deviations, flags, and reports, while data selection, method choice, processing, causal reasoning, interpretation, and disclosure control remain with human analysts. We place this boundary on an adoption frontier, a decision tool that weighs the impact of a model error against the cost to the office of not adopting, so that each use case can be triaged on its own terms. We ground each rule in the Principles it protects and locate it in the phases of the Generic Statistical Business Process Model.
- 5 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessTen Rules for Deploying Large Language Models in Official Statistics
Official statistics is increasingly produced with the help of large language models, which are being asked to write the code of its pipelines and to guide the statistics themselves, work whose outputs carry institutional weight and inform policy. The fluency of these models creates a failure mode distinct from ordinary software error: they produce plausible text whether or not it is correct, so a pipeline that relies on them can fail silently while still looking authoritative, which is at odds with the obligations written down in the Fundamental Principles of Official Statistics and the European Statistics Code of Practice. We offer ten rules for deploying these models safely in official statistics at the current state of the technology. The rules follow one architectural principle: the model serves as a monitoring agent that observes,corrects bounded deviations, flags, and reports, while data selection, method choice, processing, causal reasoning, interpretation, and disclosure control remain with human analysts. We place this boundary on an adoption frontier, a decision tool that weighs the impact of a model error against the cost to the office of not adopting, so that each use case can be triaged on its own terms. We ground each rule in the Principles it protects and locate it in the phases of the Generic Statistical Business Process Model.
- Open Access
- 29 July 2026
- Published by Elsevier in SSRN Electronic Journal
Measuring Climate Change Adaptation: Are Countries Learning to Cope with Floods and Droughts?
Are countries adapting to climate-related disasters? This paper tests whether the causal economic damage from floods and droughts is declining over time across 158 countries from 2000 to 2024—a direct, empirically grounded measure of adaptation success. Standard panel regression approaches are ill-equipped to answer this question because adaptive capacity—the institutional quality, infrastructure investment, and risk culture that simultaneously affects both disaster probability and economic outcomes—is an unobserved, time-varying confounder. Ignoring it biases estimated treatment effects and compounds that bias when interacted with a time trend. We address this challenge by extending the Time Series Deconfounder (Bica et al., 2020), a causal machine learning framework that infers a substitute confounder from the temporal structure of disaster assignments, bypassing the need to directly observe adaptive capacity. Our extension introduces MC-dropout averaging for stable confounder extraction in small panels and, critically, a within-country time-trend regression on deconfounded individual treatment effects—a test with no analogue in the original framework and our primary methodological contribution. Using a 50-member GRU ensemble that averages over training stochasticity, we find that both outcomes exhibit maladaptation over 2000–2024, though the pace is decelerating sharply. For GDP per capita, the full-sample coefficient is ˆβGDP = −0.00051 (p < 0.001), indicating damage is growing by approximately 0.51% per decade on average. Maladaptation was most rapid in 2000–2009 ( ˆβ = −0.0020, approximately −0.20% per year) and has since slowed to near-zero ( ˆβ = −0.00018, 2010–2024). For agricultural value added, the full-sample coefficient is ˆβagri = −0.0194 percentage points per year (p < 0.001), reflecting a catastrophic deterioration in 2000–2009 ( ˆβ = −0.085) that has substantially decelerated in 2010–2024 ( ˆβ = +0.0018, near stabilisation). The only income group showing GDP improvement is low-income countries ( ˆβ = +0.00032), while high-income countries suffer the fastest-worsening GDP damage ( ˆβ = −0.00124). Agricultural maladaptation is universal across all income groups and regions. Results are robust to alternative treatment definitions, country-block bootstrap, and randomisation inference.
- 29 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessMeasuring Climate Change Adaptation: Are Countries Learning to Cope with Floods and Droughts?
Are countries adapting to climate-related disasters? This paper tests whether the causal economic damage from floods and droughts is declining over time across 158 countries from 2000 to 2024—a direct, empirically grounded measure of adaptation success. Standard panel regression approaches are ill-equipped to answer this question because adaptive capacity—the institutional quality, infrastructure investment, and risk culture that simultaneously affects both disaster probability and economic outcomes—is an unobserved, time-varying confounder. Ignoring it biases estimated treatment effects and compounds that bias when interacted with a time trend. We address this challenge by extending the Time Series Deconfounder (Bica et al., 2020), a causal machine learning framework that infers a substitute confounder from the temporal structure of disaster assignments, bypassing the need to directly observe adaptive capacity. Our extension introduces MC-dropout averaging for stable confounder extraction in small panels and, critically, a within-country time-trend regression on deconfounded individual treatment effects—a test with no analogue in the original framework and our primary methodological contribution. Using a 50-member GRU ensemble that averages over training stochasticity, we find that both outcomes exhibit maladaptation over 2000–2024, though the pace is decelerating sharply. For GDP per capita, the full-sample coefficient is ˆβGDP = −0.00051 (p < 0.001), indicating damage is growing by approximately 0.51% per decade on average. Maladaptation was most rapid in 2000–2009 ( ˆβ = −0.0020, approximately −0.20% per year) and has since slowed to near-zero ( ˆβ = −0.00018, 2010–2024). For agricultural value added, the full-sample coefficient is ˆβagri = −0.0194 percentage points per year (p < 0.001), reflecting a catastrophic deterioration in 2000–2009 ( ˆβ = −0.085) that has substantially decelerated in 2010–2024 ( ˆβ = +0.0018, near stabilisation). The only income group showing GDP improvement is low-income countries ( ˆβ = +0.00032), while high-income countries suffer the fastest-worsening GDP damage ( ˆβ = −0.00124). Agricultural maladaptation is universal across all income groups and regions. Results are robust to alternative treatment definitions, country-block bootstrap, and randomisation inference.
- Open Access
- 20 April 2026
- Published by Elsevier in SSRN Electronic Journal
OnlineCML: A Python Library for Online Causal Machine Learning
We present OnlineCML, the first Python library to provide causal effect estimation under the potential outcomes framework in a purely online setting, processing one observation at a time without storing historical data. Real-world systems-A/B testing platforms, clinical monitoring, and marketing pipelines-generate data as continuous streams where treatment decisions must be made immediately and causal relationships shift over time. Existing causal inference libraries (EconML, CausalML, DoWhy, GRF) all require a complete dataset before estimation. OnlineCML bridges this gap by providing online adaptations of inverse probability weighting, doubly robust estimation, four meta-learner families (S/T/X/R), three matching variants, and a novel CausalHoeffdingTree and OnlineCausalForest with built-in concept drift detection. The library integrates natively with River's online ML ecosystem and ships with streaming benchmark datasets, real-time diagnostics, and an exploration policy layer for adaptive experimentation.
- 20 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessOnlineCML: A Python Library for Online Causal Machine Learning
We present OnlineCML, the first Python library to provide causal effect estimation under the potential outcomes framework in a purely online setting, processing one observation at a time without storing historical data. Real-world systems-A/B testing platforms, clinical monitoring, and marketing pipelines-generate data as continuous streams where treatment decisions must be made immediately and causal relationships shift over time. Existing causal inference libraries (EconML, CausalML, DoWhy, GRF) all require a complete dataset before estimation. OnlineCML bridges this gap by providing online adaptations of inverse probability weighting, doubly robust estimation, four meta-learner families (S/T/X/R), three matching variants, and a novel CausalHoeffdingTree and OnlineCausalForest with built-in concept drift detection. The library integrates natively with River's online ML ecosystem and ships with streaming benchmark datasets, real-time diagnostics, and an exploration policy layer for adaptive experimentation.
- Open Access
- 27 May 2025
- Published by IOP Publishing in Environmental Research: Infrastructure and Sustainability
- Vol. 5 (2) , 025011
Tracking green space along streets of world cities
Street green space (SGS) - the presence of vegetation along streets of cities—is a key piece of urban infrastructure. SGS provides a broad range of functions, such as mitigating the urban heat island effect, reducing the impact of extreme precipitation events, and supporting human and animal well-being. Here we introduce an approach to estimate SGS based on the statistical modeling of a street-based indicator of canopy coverage (the green view index, GVI) with multispectral satellite observations and ancillary spatially granular data. Based on our trained and cross-validated non-parametric model, we conduct spatial sampling and prediction in 190 large cities distributed across twenty regions and estimate local to continental GVI trends between 2016–2023. Jointly considering such global pool of cities, we find evidence of a trend of GVI decrease of 0.3%–0.5% per year ( p < 0.01 ) . Yet, both the direction and magnitude of trends show high heterogeneity across and within regions and cities, which we explore, along with stark inequalities in SGS availability within each city. Our analysis provides an updated estimate of the GVI as a measure of SGS across a global pool of cities and an open-source, validated approach to assess its future changes and support the design of policies for sustainable cities.
- 27 May 2025
- Published by IOP Publishing in Environmental Research: Infrastructure and Sustainability
- Vol. 5 (2) , 025011
Open AccessTracking green space along streets of world cities
Street green space (SGS) - the presence of vegetation along streets of cities—is a key piece of urban infrastructure. SGS provides a broad range of functions, such as mitigating the urban heat island effect, reducing the impact of extreme precipitation events, and supporting human and animal well-being. Here we introduce an approach to estimate SGS based on the statistical modeling of a street-based indicator of canopy coverage (the green view index, GVI) with multispectral satellite observations and ancillary spatially granular data. Based on our trained and cross-validated non-parametric model, we conduct spatial sampling and prediction in 190 large cities distributed across twenty regions and estimate local to continental GVI trends between 2016–2023. Jointly considering such global pool of cities, we find evidence of a trend of GVI decrease of 0.3%–0.5% per year ( p < 0.01 ) . Yet, both the direction and magnitude of trends show high heterogeneity across and within regions and cities, which we explore, along with stark inequalities in SGS availability within each city. Our analysis provides an updated estimate of the GVI as a measure of SGS across a global pool of cities and an open-source, validated approach to assess its future changes and support the design of policies for sustainable cities.
- 27 May 2024
- Published by The R Foundation
pbox: Exploring Multivariate Spaces with Probability Boxes
- Open Access
- 10 August 2023
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 18 (8) , e0275037
Prioritizing COVID-19 vaccine allocation in resource poor settings: Towards an Artificial Intelligence-enabled and Geospatial-assisted decision support framework
Objectives: To propose a novel framework for COVID-19 vaccine allocation based on three components of Vulnerability, Vaccination, and Values (3Vs). Methods: A combination of geospatial data analysis and artificial intelligence methods for evaluating vulnerability factors at the local level and allocate vaccines according to a dynamic mechanism for updating vulnerability and vaccine uptake. Results: A novel approach is introduced including (I) Vulnerability data collection (including country-specific data on demographic, socioeconomic, epidemiological, healthcare, and environmental factors), (II) Vaccination prioritization through estimation of a unique Vulnerability Index composed of a range of factors selected and weighed through an Artificial Intelligence (AI-enabled) expert elicitation survey and scientific literature screening, and (III) Values consideration by identification of the most effective GIS-assisted allocation of vaccines at the local level, considering context-specific constraints and objectives. Conclusions: We showcase the performance of the 3Vs strategy by comparing it to the actual vaccination rollout in Kenya. We show that under the current strategy, socially vulnerable individuals comprise only 45% of all vaccinated people in Kenya while if the 3Vs strategy was implemented, this group would be the first to receive vaccines.
- 10 August 2023
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 18 (8) , e0275037
Open AccessPrioritizing COVID-19 vaccine allocation in resource poor settings: Towards an Artificial Intelligence-enabled and Geospatial-assisted decision support framework
Objectives: To propose a novel framework for COVID-19 vaccine allocation based on three components of Vulnerability, Vaccination, and Values (3Vs). Methods: A combination of geospatial data analysis and artificial intelligence methods for evaluating vulnerability factors at the local level and allocate vaccines according to a dynamic mechanism for updating vulnerability and vaccine uptake. Results: A novel approach is introduced including (I) Vulnerability data collection (including country-specific data on demographic, socioeconomic, epidemiological, healthcare, and environmental factors), (II) Vaccination prioritization through estimation of a unique Vulnerability Index composed of a range of factors selected and weighed through an Artificial Intelligence (AI-enabled) expert elicitation survey and scientific literature screening, and (III) Values consideration by identification of the most effective GIS-assisted allocation of vaccines at the local level, considering context-specific constraints and objectives. Conclusions: We showcase the performance of the 3Vs strategy by comparing it to the actual vaccination rollout in Kenya. We show that under the current strategy, socially vulnerable individuals comprise only 45% of all vaccinated people in Kenya while if the 3Vs strategy was implemented, this group would be the first to receive vaccines.
- Open Access
- 15 May 2023
- Published by Copernicus GmbH
Tracking global urban green space trends
Urban green space - the presence of vegetation-covered area within cities' boundaries - is an increasingly relevant indicator for evaluating sustainable cities. This is because besides providing a set of local services such as mitigating the urban heat island effect and reducing the impact of extreme precipitation events, urban green space has been widely associated with increasing well-being of urban dwellers. Here we present a global analysis of recent trends in urban green space based on modelling of multi-spectral satellite imagery data to reproduce street-based vegetation presence indicators. We estimate local to continental trends over the 2016-2022 period and estimate global scale (considering a large set of the most populated cities) urban green space change statistical trends. We examine heterogeneities in the direction and magnitude of trends across cities and regions, while also analysing within-city inequalities. Our analysis provides an updated picture of urban green space across world cities and an open-source and open data-driven, spatially cross-validated approach to assess changes in near-real-time.
- 15 May 2023
- Published by Copernicus GmbH
Open AccessTracking global urban green space trends
Urban green space - the presence of vegetation-covered area within cities' boundaries - is an increasingly relevant indicator for evaluating sustainable cities. This is because besides providing a set of local services such as mitigating the urban heat island effect and reducing the impact of extreme precipitation events, urban green space has been widely associated with increasing well-being of urban dwellers. Here we present a global analysis of recent trends in urban green space based on modelling of multi-spectral satellite imagery data to reproduce street-based vegetation presence indicators. We estimate local to continental trends over the 2016-2022 period and estimate global scale (considering a large set of the most populated cities) urban green space change statistical trends. We examine heterogeneities in the direction and magnitude of trends across cities and regions, while also analysing within-city inequalities. Our analysis provides an updated picture of urban green space across world cities and an open-source and open data-driven, spatially cross-validated approach to assess changes in near-real-time.
- Open Access
- 27 March 2022
- Published by California Digital Library (CDL) in EarthArXiv
The long-run welfare impact of hydrological extremes in sub-Saharan Africa
The economic impact of dry and wet spells is increasingly investigated, also relative to anthropogenic climate change. Combining high-resolution wealth estimates and long-run (1901-2018) hydrological records for nearly one million settlements , here we empirically estimate the long-run welfare impact of hydrological events in sub-Saharan Africa (SSA). To achieve causal identification, we adopt a generalised propensity score approach allowing for continuous longitudinal treatments. Our results suggest that hydrological events played a long-lasting impact in shaping the current wealth distribution patterns in SSA. While even moderate dry spells diminished wealth (on average -$110 in the local per-capita GDP for each additional dry month), only extreme wet spells show a significant impact, but this is significantly harsher (-$387). We examine the temporal variability of estimates and discuss the significance of adaptive capacity. Our findings support the consideration of adaptation investment to mitigate future impacts and an exacerbation of existing inequalities.
- 27 March 2022
- Published by California Digital Library (CDL) in EarthArXiv
Open AccessThe long-run welfare impact of hydrological extremes in sub-Saharan Africa
The economic impact of dry and wet spells is increasingly investigated, also relative to anthropogenic climate change. Combining high-resolution wealth estimates and long-run (1901-2018) hydrological records for nearly one million settlements , here we empirically estimate the long-run welfare impact of hydrological events in sub-Saharan Africa (SSA). To achieve causal identification, we adopt a generalised propensity score approach allowing for continuous longitudinal treatments. Our results suggest that hydrological events played a long-lasting impact in shaping the current wealth distribution patterns in SSA. While even moderate dry spells diminished wealth (on average -$110 in the local per-capita GDP for each additional dry month), only extreme wet spells show a significant impact, but this is significantly harsher (-$387). We examine the temporal variability of estimates and discuss the significance of adaptive capacity. Our findings support the consideration of adaptation investment to mitigate future impacts and an exacerbation of existing inequalities.
- Open Access
- 28 May 2021
- Published by California Digital Library (CDL) in EarthArXiv
Application of Probabilistic Machine Learning to the prediction of Remotely Sensed Vegetation Health
Every prediction of the future carries some level of uncertainty; making this explicit is challenging. This paper introduces a Probabilistic Machine Learning algorithm, namely the Natural Gradient Boosting algorithm, as a modelling tool for predicting the Vegetation Health Index, a proxy for monitoring vegetation stress in response to changing weather conditions. It then elaborates on the time heterogeneity of the probabilistic predictions generated by the model over one year on West Java, the most populated Indonesia region, providing important consideration for Food Security. The results highlight that while 94% of the observed values fall within the model’s 95% prediction interval, during February and March, the probability of reaching critically low levels of VHI below 40 is above 75%. The paper foster the multidisciplinary literature linking the use of remotely sensed vegetation health data with Food Security by incorporating uncertainty in predicting the impact of changing weather conditions on Food Security.
- 28 May 2021
- Published by California Digital Library (CDL) in EarthArXiv
Open AccessApplication of Probabilistic Machine Learning to the prediction of Remotely Sensed Vegetation Health
Every prediction of the future carries some level of uncertainty; making this explicit is challenging. This paper introduces a Probabilistic Machine Learning algorithm, namely the Natural Gradient Boosting algorithm, as a modelling tool for predicting the Vegetation Health Index, a proxy for monitoring vegetation stress in response to changing weather conditions. It then elaborates on the time heterogeneity of the probabilistic predictions generated by the model over one year on West Java, the most populated Indonesia region, providing important consideration for Food Security. The results highlight that while 94% of the observed values fall within the model’s 95% prediction interval, during February and March, the probability of reaching critically low levels of VHI below 40 is above 75%. The paper foster the multidisciplinary literature linking the use of remotely sensed vegetation health data with Food Security by incorporating uncertainty in predicting the impact of changing weather conditions on Food Security.
- 30 November 2020
- Published by National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 117 (50) , 31760-31769
Planning universal accessibility to public health care in sub-Saharan Africa
Achieving universal health care coverage—a key target of the United Nations Sustainable Development Goal number 3—requires accessibility to health care services for all. Currently, in sub-Saharan Africa, at least one-sixth of the population lives more than 2 h away from a public hospital, and one in eight people is no less than 1 h away from the nearest health center. We combine high-resolution data on the location of different typologies of public health care facilities [J. Maina et al., Sci. Data 6, 134 (2019)] with population distribution maps and terrain-specific accessibility algorithms to develop a multiobjective geographic information system framework for assessing the optimal allocation of new health care facilities and assessing hospitals expansion requirements. The proposed methodology ensures universal accessibility to public health care services within prespecified travel times while guaranteeing sufficient available hospital beds. Our analysis suggests that to meet commonly accepted universal health care accessibility targets, sub-Saharan African countries will need to build ∼6,200 new facilities by 2030. We also estimate that about 2.5 million new hospital beds need to be allocated between new facilities and ∼1,100 existing structures that require expansion or densification. Optimized location, type, and capacity of each facility can be explored in an interactive dashboard. Our methodology and the results of our analysis can inform local policy makers in their assessment and prioritization of health care infrastructure. This is particularly relevant to tackle health care accessibility inequality, which is not only prominent within and between countries of sub-Saharan Africa but also, relative to the level of service provided by health care facilities. Significance Achieving universal health care access is a key development priority and a target of Sustainable Development Goal number 3. The COVID-19 pandemic has only reinforced this urge. A rapid expansion of public, affordable health care infrastructure is particularly crucial in sub-Saharan Africa (SSA): communicable diseases are the first cause of death, infant mortality rates are above 5%, and lengthy journeys to health care facilities undermine the accessibility to basic health care for millions. This paper provides a planning-oriented, inequality-focused analysis of health care accessibility in SSA based on a state-of-the-art georeferenced database of public health care facilities. We devise a spatial optimization framework to identify the optimal location and required bed capacity of public health care facilities in the region.
- 30 November 2020
- Published by National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 117 (50) , 31760-31769
Planning universal accessibility to public health care in sub-Saharan Africa
Achieving universal health care coverage—a key target of the United Nations Sustainable Development Goal number 3—requires accessibility to health care services for all. Currently, in sub-Saharan Africa, at least one-sixth of the population lives more than 2 h away from a public hospital, and one in eight people is no less than 1 h away from the nearest health center. We combine high-resolution data on the location of different typologies of public health care facilities [J. Maina et al., Sci. Data 6, 134 (2019)] with population distribution maps and terrain-specific accessibility algorithms to develop a multiobjective geographic information system framework for assessing the optimal allocation of new health care facilities and assessing hospitals expansion requirements. The proposed methodology ensures universal accessibility to public health care services within prespecified travel times while guaranteeing sufficient available hospital beds. Our analysis suggests that to meet commonly accepted universal health care accessibility targets, sub-Saharan African countries will need to build ∼6,200 new facilities by 2030. We also estimate that about 2.5 million new hospital beds need to be allocated between new facilities and ∼1,100 existing structures that require expansion or densification. Optimized location, type, and capacity of each facility can be explored in an interactive dashboard. Our methodology and the results of our analysis can inform local policy makers in their assessment and prioritization of health care infrastructure. This is particularly relevant to tackle health care accessibility inequality, which is not only prominent within and between countries of sub-Saharan Africa but also, relative to the level of service provided by health care facilities. Significance Achieving universal health care access is a key development priority and a target of Sustainable Development Goal number 3. The COVID-19 pandemic has only reinforced this urge. A rapid expansion of public, affordable health care infrastructure is particularly crucial in sub-Saharan Africa (SSA): communicable diseases are the first cause of death, infant mortality rates are above 5%, and lengthy journeys to health care facilities undermine the accessibility to basic health care for millions. This paper provides a planning-oriented, inequality-focused analysis of health care accessibility in SSA based on a state-of-the-art georeferenced database of public health care facilities. We devise a spatial optimization framework to identify the optimal location and required bed capacity of public health care facilities in the region.