Dry air episodes in the northern margins of basin Congo forests: detection and climatology

The northern margins of Congo basin rainforests are affected by a dry season, from December to March. During this season, dry-air events are observed, characterized by a marked drop in dew point temperature (Td) below 15°C, a threshold currently used to determine the intertropical discontinuity. These episodes, linked to the Harmattan winds, penetrate the forest via the Sangha interval until they reach Libreville or Brazzaville in the most extreme cases, as in 1983. Given that the detection and characterization of these events of dry air have never been addressed in the scientific literature, the question of their impact on forests arises. Therefore, this study explores methodological approaches to detect these dry-air events and provides a first climatology and interannual variability.

Two approaches have been developed to detect dry-air events (over the area 0°-8.5°N/11°-27.5°E): on the one hand, by applying a clustering algorithm (K-Means) to the diurnal cycles of Td and, on the other hand, by setting a threshold on the distribution of daily values. As in-situ observations are sparse in Central Africa, ERA5 reanalyses are used to supplement the GSOD, GHCHh and ISD databases and capture the spatiotemporal variability of dry-air events over 1970-2024.

In the first approach, days are standardized across all the stations and then separated into two clusters by the algorithm. This leads to two distinct diurnal profiles of Td : during dry-air events, the diurnal cycle shows a characteristic negative bell-shaped curve, with lower Td values, whereas on other days Td remains higher and relatively constant throughout the day. In order to take into account the bias in reanalyses, the two clusters for ERA5 are calculated for pixels/days corresponding to stations and based on their classification as dry-air or non-dry-air events in observations. The two classifications (observations and ERA5) result in a relatively similar total number of dry-air days, with a large majority of events detected jointly (Critical Success Index: 0.72).

In the second approach, daily Td values are also aggregated and standardized across all stations. A set of thresholds (0, -0.25, ... -1.5 std) is tested over the Td distribution to refine the detection of dry air events in both observations and reanalyses. Increasing the threshold leads to an inflation in the occurrence of dry-air days, particulary in the observations; however, the large majority of events are detected jointly by the observations and ERA5 (Critical Success Index>0.7 for nearly all the thresholds). Above −0.5 std, the correlation between the interannual variability in observations and ERA5 decreases (≈0.9 below this threshold and <0.81 above).

Both approaches show similar interannual variability in dry air events with years characterized by exceptionally long and extensive episodes, such as 1983, 1989, and 1997. The climatology of dry-air event frequency is characterized by a marked latitudinal gradient, in agreement with the gradient observed in Td. The northern limit of the current forest extent lies between the 25% and 30% isofrequency contours. Beyond one-third of abnormally dry-air days during the dry season, forest gives way to savanna.