Abstract

Extreme heat is an emerging challenge to well-being in the Global South, where climatic exposure intersects with unequal access to housing, essential services, healthcare and opportunities for adaptation. Systemic cooling poverty captures these intersecting deprivations through a multidimensional perspective extending beyond access to air conditioning. However, national and regional indicators can conceal territorial inequalities, while uncertainty in their constituent estimates can affect the identification of priority areas. This contribution proposes a small area estimation approach to measuring systemic cooling poverty at district level and assessing the reliability of territorial priorities.

Building on an existing multidimensional assessment, two candidate applications are considered: Tanzania, combining the 2022 Demographic and Health Survey (DHS) with the 2022 Population and Housing Census; and Zambia, combining the 2024 DHS with the 2022 Census. Survey microdata provide information on household composition, housing materials and crowding, water and sanitation, electricity, cooling and communication assets, education, employment and health. Census aggregates provide district auxiliary variables and population and household counts. Additional sources include ERA5-Land climate records, Local Climate Zones, Dynamic World land cover, healthcare accessibility surfaces and administrative boundaries. Gridded population products may complement census information for spatial aggregation.

Indicators cover five dimensions: climate exposure; infrastructure and assets; social and thermal inequalities; health; and education and working standards. Household deprivation statuses are constructed using explicit thresholds and aggregation rules. District indicators are defined as dimension-specific deprivation prevalences, whose weighted average provides the systemic cooling poverty index.

The estimation strategy starts from direct survey estimates and their sampling variance–covariance matrices, accounting for stratification, clustering and survey weights. Area-level models combine these estimates with auxiliary information available across districts, borrowing strength across territories and, where supported by the data, across deprivation dimensions. Spatial effects are considered to capture residual geographical dependence. Particular attention is given to districts with few sampled clusters, unstable direct estimates or incomplete information. The modelling strategy will balance flexibility with the information available for estimating sampling and model covariance structures.
Uncertainty assessment is central to the analysis. Dependence between dimension-specific estimation errors will be retained when propagating uncertainty to the composite index. Inference will be conditional on the selected environmental data products, while sensitivity to alternative deprivation thresholds and dimension weights will be examined separately. Beyond point estimates and uncertainty intervals, the analysis will quantify the probability that a district exceeds a deprivation threshold or belongs to a priority group.

The approach will be evaluated through simulation experiments informed by the empirical survey design and comparisons with informative direct estimates. Evaluation will consider bias, mean squared error, interval coverage and the stability of territorial prioritisation under sparse sampling and alternative dependence structures. By providing district estimates alongside their uncertainty and dimensional composition, the contribution aims to improve the geographical detail, interpretability and policy relevance of emerging well-being indicators. It will examine how accounting for estimation uncertainty changes territorial priorities and supports locally differentiated, equitable planning for heat adaptation in data-limited settings.

Authors: Bertarelli, G., Falchetta, G.; Mazzone, A., Pistorio, A. and De Cian, E.;