We are updating the TMF dataset every year. The update generally comes around the month of March-April. Note that the yearly update does not only add one year of additional data on forest cover changes but it also helps consolidating change trajectories over the past 3 years (mainly the separation between forest degradation and deforestation). We also constantly improve the products by correcting any sensor artefacts and by visually interpreting new areas of plantations (see note on updates integrated in the new version). Always make sure that you use the latest version of the TMF products.

We recommend to always make sure to use the latest version of the TMF dataset available on our website. Every year’s update not only adds one extra year to the time series, it also impacts the recent dynamic of forest degradation, deforestation and forest regrowth. Additionally, we continuously improve the dataset, see https://forobs.jrc.ec.europa.eu/TMF/data#update.

Forest area in the TMF dataset is not defined by any percentage of tree cover. Our study covers the tropical moist forests, which include all closed forests in the humid tropics with two main forest types: the tropical rain forest and the tropical moist deciduous forest (see Section “Study area and forest types” of Vancutsem et al. 2021). All the classes 1-69 of our transition map correspond to the TMF domain, i.e. the tropical moist forests that are currently undisturbed (classes 10-12) and the initially undisturbed TMFs (during our baseline period 1982-1989) and that have been disturbed over the past 32 years (1990-2021) (classes 20-69). Our mapping approach consists of observing the evolution of spectral signatures over time and of identifying potential disruption observations (i.e. detection of an absence of tree foliage cover within a Landsat pixel for a single-date observation) for each single-date image of the time series. The temporal sequence of those disruption observations at pixel scale was analyzed to first determine the initial extent of the TMF domain (period 1982-1989) and then to identify the change trajectories from this initial forest extent (from 1990 up to now).

The initial tropical moist forest domain can be derived from the Transition Map - Sub types map by selecting all pixels belonging to classes 10 to 89 excluding classes 71 and 72.

Indeed, there is uncertainty in the differentiation between deforestation and forest degradation during the past 3 years of analysis. Therefore, we have additional rules (only for the past 3 years) to distinguish these two processes combining the duration and intensity of the disturbance. Taking the example of a new clear-cut in 2019, this can be classified as degraded in version2019 of TMF if the intensity of the disturbance is low (less than 45 % of total valid observations of the year) but will be classified as deforested in version2021 if the disturbance lasted more than 900 days.

Indeed, there is uncertainty in the differentiation between deforestation and forest degradation during the past 3 years of analysis. Therefore, we have additional rules (only for the past 3 years) to distinguish these two processes combining the duration and intensity of the disturbance. Taking the example of a new clear-cut in 2019, this can be classified as degraded in version2019 of TMF if the intensity of the disturbance is low (less than 45 % of total valid observations of the year) but will be classified as deforested in version2021 if the disturbance lasted more than 900 days.

For the regrowth class (or secondary forest), we separate it from the other forest types (so-called undisturbed or degraded) because the historical trajectory is different (between a trajectory where no disturbance has been observed over the whole study period and a trajectory where deforestation and regrowth are observed). It may potentially be interesting to distinguish this in the framework of applications on carbon, biodiversity or other ecosystem services. It is therefore left to the user (depending on the definition of forest he/she is using) to decide whether or not to group these two types of forest trajectory together.

The commodities mask is a combination of external data sources (see data user guides for more details) which is updated every year with new visually interpreted areas of plantations. You can retrieve the commodity areas by selecting classes 81-86 from the Transition map - Sub types product.

The GMW dataset is the first globally consistent and automated method for mapping mangroves (94% overall accuracy) at different periods. The GMW dataset covers the years 1996 and 2016 so we produced a maximum extent mask of the mangroves during the period 1996-2016.

The identification of changes within mangroves follows the same methodology as for non-mangrove forests (see in methods in article/technical report). Single-date classification of Landsat images were based on multispectral cluster analysis of 38326 sampled pixels that covered mangrove forests and their spectral attributes.

The tropical moist forest represents 60% of the extent of tropical natural forests - based on FAO estimates (FAO, FRA 2015). See also Figure 8 in SOFO (FAO and UNEP. 2020. The State of the World’s Forests 2020. Forests, biodiversity and people. Rome. https://doi.org/10.4060/ca8642en

We provide the statistics of annual forest cover changes for the period 1990-2021 at the country level for countries with more than 1 Mha TMF area in 1990 (see Section X). We also provide a GEE tutorial that explains how to extract statistics over your own study area.

The statistics provide the area (in Mha) of undisturbed forest, forest disturbances and forest regrowth. Forest disturbances are separated into forest degradation, deforestation and deforestation of forest regrowth. Regarding deforestation, we have added the sub classes of direct deforestation in order to separate deforestation of a degraded forest from deforestation of undisturbed forest. We have also added the conversion of forest due to plantation or to water expansion. Regarding forest degradation, we include the area of degradation not followed by deforestation to separate degradation that remains classified as degraded forest from degradation followed by deforestation in the following years.

The classes "Total degradation" and "Direct deforestation" have to be grouped to know all the disturbances that occur in one year (which leads to a conversion of the undisturbed forest into a degraded forest or deforested land). The "total deforestation" class represents all the forest areas (degraded or undisturbed) that have been converted into deforested land in one year.

Note that over the past 3 years, the discrimination between degradation and deforestation is not so robust compared to the historical estimations as 3 years are needed to confirm the type of disturbances. To consolidate our statistics over the past 3 years, we corrected the average proportion of disturbance types within total disturbances using the average proportion observed for the period 2007–2016.

TMF data identify several forest cover change classes (degradation, deforestation, regrowth) which helps to nuance between a light (logging or fire) and a strong event (deforestation) whereas GFC includes all disturbance in losses. The number of disturbances per year is also provided as an indicator of the intensity of the disturbance. TMF may capture disturbances earlier (one year before) than GFC, particularly when the disturbance occurred at the end of the year because of the method (single-date vs composites). GFC data show some inconsistencies between the period 2000-2014 and the period 2015-2021, due to an enhanced Landsat processing algorithm that has “resulted in enhanced detection of loss— particularly from 2015 onwards” (GFW, 2021). Overall the main differences between TMF and GFC rely on different periods of analysis (2001-2020 for GFC) and single date classification for TMF which enables the detection of short duration events), differentiation between degradation, deforestation and regrowth.

We observed that most of the mangroves have a very low tree cover density (0-50%) on GFC, which could prevent the GFC approach’s ability to detect losses (as a minimum tree cover threshold is required).

Continental-level results of the accuracy assessment of the single-date classification algorithm are available in table S3 of the supplementary materials of Vancutsem article. At continental level, overall accuracy is higher for Africa (94.6%) than for Latin America (91.0%) or Asia (89.0%).

The accuracy (presented in Vancutsem et al 2021) depends on the ability of our approach to detect a change from one single-date Landsat image. Consequently, accuracy depends mostly on the type of event observed (lower for shorter events, e.g. selective logging) and on the availability of valid observations (this meta-information is provided to the user as a proxy measure of confidence). Hence, our estimates of changes in the regions where the total number of valid observations is particularly low and/or the start year of the monitoring period is late, e.g., Gabon, Solomon Islands, and La Réunion, should be considered with lower confidence. However, considering the geographic completeness of Landsat 8 coverage after the year 2013, there is high confidence for the contemporary reported estimates.

Our methodology and algorithms for detecting the disturbances did not change over time and the accuracy assessment did not show significant differences in performance among sensors.

However, as mentioned in our paper (Section on "Known limitations and future improvements") the number of valid observations may vary over time and location, and this may have an impact on the detection of short-duration events. The risk of under-estimating the short-duration degradation mostly concerns the period of 2011-2013 (between the decommissioning of Landsat 5 and the launch of Landsat 8), and the full period before 2013 for Central and West Africa where there was low coverage of Landsat 4 and 5 data (unlike South America and Asia that were well covered). The meta-information provided with the TMF maps, i.e. the start of the monitoring period and the number of annual valid observations, allow the user to know when the monitoring period starts and estimate this risk of low coverage during the observation period in their specific region of interest.

One advantage of the TMF product is that we separate the different types of disturbances. Consequently, the user can analyze deforestation trends separately from long-duration and short-duration degradation trends and be more careful in interpreting changes for the short-duration degradation in risk areas.

However, to get unbiased estimates of change areas and related uncertainties (for trend analysis) it is recommended to combine any spatial wall to wall map with a sample-based reference dataset.