Repository Documentation

Overview

This repository contains a Google Earth Engine (GEE) script designed to extract spatially explicit climate indicators, atmospheric circulation proxies, sea surface temperature anomalies, vegetation indices, and a wide range of geographic, soil, infrastructure, and market-access covariates at Demographic and Health Survey (DHS) cluster locations in Sierra Leone. The script supports the DHS survey waves 2008, 2013, and 2019.

The code represents a substantially expanded and refined version (Version 6) of earlier iterations. In contrast to previous versions, all lagged exposure windows for rainfall have been removed. Climate variables are therefore extracted for the survey year only. The script incorporates several new environmental and climatic data products that facilitate instrumental-variable strategies and robustness checks in econometric analyses of climate impacts on household welfare.

Climate indicators are primarily derived from daily CHIRPS precipitation data, with cross-validation using ERA5-Land daily precipitation and PERSIANN-CDR datasets. Baseline climate statistics are computed using a survey-specific historical reference period, defined as 1990 to the year preceding the survey wave. Specifically:

• 1990–2007 for the 2008 DHS wave

• 1990–2012 for the 2013 DHS wave

• 1990–2018 for the 2019 DHS wave

For each survey wave, the baseline period is used to compute long-term median and standard deviation statistics, which serve as reference values for rainfall anomaly calculations.

Rainfall Indicators

Using CHIRPS daily precipitation, the script constructs a comprehensive set of rainfall indicators for each survey year. These include:

• Total annual precipitation

• Median daily precipitation

• Maximum daily precipitation

• Standard deviation of daily precipitation

In addition, the script calculates rainfall anomalies relative to the survey-specific baseline, including:

• Absolute precipitation anomaly

• Standardised precipitation anomaly

• Percentage precipitation anomaly

The script also produces several indicators describing rainfall frequency and intensity:

• Wet day count (days with rainfall ≥ 1 mm)

• Dry day count

• Heavy rain days (≥ 10 mm) and total rainfall on heavy rain days

• Very heavy rain days (≥ 20 mm) and total rainfall on those days

Hydrological intensity metrics are also calculated. These include the Simple Daily Intensity Index (SDII), defined as total annual precipitation divided by the number of wet days.

Finally, the script measures rainfall persistence through:

• Consecutive dry days

• Consecutive wet days

Atmospheric Circulation and Climate Dynamics

To capture broader atmospheric processes affecting rainfall variability, the script extracts several indicators from ERA5-Land reanalysis data.

These include:

• 10 m eastward wind component (u)

• 10 m northward wind component (v)

• Mean wind speed

• Meteorological wind direction (0–360° clockwise from North)

• Southerly flow indicator, defined where the northward wind component is positive (v > 0)

Surface atmospheric conditions are represented using:

• Annual mean surface pressure for the survey year

• Baseline mean surface pressure calculated over the historical reference period

Annual precipitation totals from ERA5-Land are also extracted to provide a cross-validation measure relative to the CHIRPS rainfall dataset.

Radiation and Cloud Cover Proxies

Large-scale atmospheric convection is approximated using radiation-based indicators.

The script extracts:

• Surface net thermal radiation from ERA5-Land (W/m²), where negative values indicate net upward radiation

• Annual median 11 µm brightness temperature from the GRIDSAT-B1 dataset

Brightness temperature is further used to construct an outgoing longwave radiation (OLR) proxy using the Stefan–Boltzmann law, providing an indirect measure of convective cloud activity.

Inter-Tropical Convergence Zone (ITCZ) Position

The script calculates the annual latitudinal position of the Inter-Tropical Convergence Zone (ITCZ) using CHIRPS precipitation data.

The ITCZ latitude is defined as the precipitation-weighted mean latitude within the Guinea Coast band (16°W–8°W, 0°–20°N). The resulting value is used to compute the distance between the ITCZ position and Sierra Leone's approximate geographic centre (8.5°N), providing an indicator of monsoon displacement during each survey year.

Sea Surface Temperature Indicators

To capture oceanic drivers of regional climate variability, the script extracts sea surface temperature (SST) anomalies from NOAA OISST Version 2.1.

SST anomalies are calculated relative to the 1971–2000 climatological baseline for the following regions:

• Niño 3.4 region (Pacific ENSO index)

• ATL3 tropical Atlantic cold tongue

• Guinea Gulf region (off the Sierra Leone coast)

• Tropical Indian Ocean basin

In addition, the script extracts the mean SST level in the Niño 3.4 region.

Vegetation Indicators

Vegetation conditions are measured using MODIS MOD13Q1 NDVI data, which are quality-assured and rescaled to physical units.

The script calculates several lagged vegetation indicators:

• Annual median NDVI (lag-1 year)

• Three-year average NDVI (lag-3 to lag-1 years)

• Growing season median NDVI (July–September, lag-1 year)

• Dry season median NDVI (December–February spanning lag-2 and lag-1 years)

These measures capture vegetation responses to prior climatic conditions.

Geographic and Socioeconomic Controls

Several geographic and socioeconomic control variables are extracted for each DHS cluster.

Topographic indicators include:

• Elevation

• Terrain slope

• Terrain aspect

• Windward exposure indicator

• Monsoon exposure intensity index

Accessibility indicators include:

• Distance to the coastline

• Travel time to urban centres

• Population density

Soil Characteristics

Soil quality variables are derived from the SoilGrids ISRIC database at the 0–5 cm depth layer. Extracted variables include:

• Sand content

• Clay content

• Soil organic carbon

• Soil pH

These variables are combined into a composite soil quality index.

Market Access and Road Infrastructure

Road infrastructure indicators are derived from the Global Roads Inventory Project (GRIP4) Africa dataset.

The script calculates:

• Distance to any road

• Distance to major roads (highway, primary, or secondary)

• Road density index

Road surface quality indicators include:

• Distance to paved roads

• Distance to unpaved roads

• Share of paved roads in the surrounding area

Terrain and Flood Risk Indicators

Additional geographic indicators include:

• Terrain Ruggedness Index

• Terrain relief (difference between maximum and minimum elevation within a 500 m radius)

Flood exposure variables are derived from the JRC Global Surface Water dataset, including:

• Flood occurrence percentage

• Flood recurrence percentage

• Flood seasonality (months)

• Flood-prone indicator

Historical Infrastructure and Development Indicators

Historical infrastructure indicators are constructed using DMSP-OLS nighttime lights data. These include:

• Stable nighttime lights in 1992

• Mean nighttime light intensity in the five years prior to each survey

• Change in nighttime lights since 1992

• Cumulative nighttime lights from 1992 to the survey year

Historical transport infrastructure is also captured through distance to colonial-era roads, defined as GRIP4 road types 1 and 2.

River Proximity Indicators

Hydrological accessibility is measured using HydroSHEDS river network data, from which the script calculates:

• Distance to major rivers

• River density index

The dataset also includes distance to colonial administrative headquarters, capturing historical administrative geography.

Spatial Aggregation Procedure

All environmental and geographic variables are spatially aggregated using a median reducer within a 10 km buffer surrounding each DHS cluster centroid. Extraction is performed at a 5 km spatial resolution.

This buffer accounts for the random spatial displacement applied to DHS cluster coordinates, ensuring that extracted environmental conditions reflect the broader local area surrounding each survey cluster.

Output Dataset

The script produces a harmonised cluster-level CSV dataset suitable for econometric analysis. The dataset integrates climate indicators, environmental variables, and infrastructure controls for each DHS cluster.

The resulting dataset is designed for causal analyses of climate impacts on household welfare, with particular relevance for gender-disaggregated development research in data-scarce and post-conflict settings.

Authorship and Development

The script was developed with the assistance of artificial intelligence tools and subsequently reviewed, modified, and fully supervised by the author. All methodological decisions—including variable definitions, baseline construction, anomaly thresholds, wind direction verification, scaling procedures, and analytical logic—were designed and validated by the author.