About this map

This interactive map shows the local warming induced by tropical deforestation. The warming is shown as an average value per administrative area e.g. a region, district or province.

2020 is the baseline year for the forest cover values i.e. the deforestation percentage is relative to the forest cover in 2020 for any given mapping unit.

This map has been produced by a team from the School of Earth and Environment at the University of Leeds. The data is from Reddington et al. 2025

Methods and data

Temperature Datasets

We used land surface temperature (LST) data from NASA MODIS on-board the Terra satellite, specifically the MOD11A2 8-d LST data version 6.1 at 0.01° x 0.01° spatial resolution. We excluded data where the estimated emissivity error was greater than 0.02 and where the LST error was greater than 1 K.

We first aggregated the 8-day LST data by month ignoring any 8-day period where data were missing due to clouds or as a result of the quality screening process. We then calculated 3-year means for two periods at the start (2001 to 2003) and end (2018 to 2020) of the study period. The temperature change (ΔT) in multi-annual mean LST was then calculated by subtracting the 2001-2003 mean LST from the 2018-2020 mean LST.

Forest Cover Datasets

Forest cover data were taken from the Global Forest Change (GFC) V1.9 dataset at 30 m x 30 m spatial resolution. Annual forest cover for the period 2000 to 2020 was calculated by taking tree cover in the year 2000, defined as canopy closure for all vegetation taller than 5 m in height, and subjecting it to annual forest loss, defined as a disturbance from a forest to non-forest state. Forest cover was calculated at the native 30 m x 30 m spatial resolution and then converted to match the spatial resolution of the ΔT data. The percentage-point change in forest cover between 2001 and 2020 was calculated as the difference between the annual forest cover fractions (in percent) in 2001 and 2020.

To identify and exclude areas where forest regrowth has occurred, we used forest extent data at 30 m x 30 m spatial resolution from the Global Land Analysis and Discovery (GLAD) Global Land Cover and Land Use Change dataset, available for the years 2000 and 2020. In the forest extent dataset, forest presence is indicated for pixels with 5 m or greater forest height. We converted this dataset to match the spatial resolution of the ΔT data and calculated the percentage-point difference in forest extent between 2000 and 2020. We then removed any 30-arc-second pixels with greater than a 50%-point net increase in forest extent between 2000 and 2020 from the ΔT data.

Elevation Dataset

We used elevation data to take into account the thermal lapse rate. Elevation data was taken from Global Multi-resolution Terrain Elevation Data (GMTED2010).

Estimating deforestation-induced temperature change

We estimated the ΔT due to local forest loss between 2001 and 2020 at the ~1 km2 pixel level across the tropics (25°S – 25°N). To remove the influence of global climate change over the study period on ΔT, we used a moving-window nearest-neighbour approach described below.

In our analysis we used all data pixels within the tropics (25°S – 25°N) with 10% or greater forest cover fraction in 2001. Pixels with less than 0.5%-point forest cover loss between 2001 and 2020 were classified as “non-deforested” pixels. Pixels with 2%-point or greater forest cover loss between 2001 and 2020 were classified as “deforested” pixels.

For each deforested pixel, we selected all surrounding pixels that maintained forest cover (“non-deforested” pixels) within a circle of 0.25° radius (approx. 27 km at the equator) and within an elevation of ±50 m. If no surrounding non-deforested pixels were available within the 0.25° radius circle (3% of pixels), the radius of the circle was extended to 0.50° (approx. 55 km at the equator), with the same elevation constraints of within ±50 m. We calculated the mean ΔT over all selected non-deforested (“control”) pixels and subtracted this from the ΔT of the central deforested pixel to get the deforestation-induced ΔT. We then divided the resulting deforestation-induced ΔT data array by the percentage-point change in forest cover fraction to calculate the change in temperature per percentage-point forest cover loss.

To reduce noise in the dataset and to expand the amount of data included to calculate individual pixel-level ΔT-per-percentage-point-forest-cover-loss values we used a moving average smoothing approach. For each deforested pixel, we calculated the mean ΔT per percentage point forest cover loss over all surrounding deforested pixels within a 0.10° rolling circle: excluding any pixel with less than 5%-point forest cover loss or with a z-score greater than 3 from the mean. Finally, we multiplied the “smoothed” pixel-level ΔT-per-percentage-point-forest-cover-loss values by the pixel-level percentage-point change in forest cover to obtain a more robust estimate of the pixel-level deforestation-induced ΔT.

Dataset references and links

MODIS land surface temperature data: https://www.earthdata.nasa.gov/

Global Forest Change (GFC) dataset: https://storage.googleapis.com/earthenginepartners-hansen/GFC-2023-v1.11/download.html

Global Land Analysis and Discovery (GLAD) Global Land Cover and Land Use Change dataset: https://glad.umd.edu/dataset/GLCLUC2020

Global Multi-resolution Terrain Elevation Data (GMTED2010): https://earthexplorer.usgs.gov/