Au cours de 2 crues survenues le 24 mai 1992 et le 15 mai 1993 sur 2 bassins versants, nous avons étudié la composition isotopique et chimique des précipitations (pluies et pluviolessivats) ainsi que leurs variations temporelle et spatiale. Les bassins étudiés (d'environ 1,5 ha) sont situés près de la ville de Sinnamary (Guyane Française) et sont proches l'un de l'autre (200 m). Un des bassins (bassin B) est recouvert par une forêt primaire, tandis que le second (bassin A) a été défriché et transformé en prairie (Digitaria swazilendensis, programme ÉCÉREX Orstom-CTFT). Le dispositif expérimental est composé de 31 pluviomètres sur le bassin B et de 3 pluviomètres sur le bassin A. Les hauteurs d'eau précipitées lors des événements étudiés sont importantes (environ 60 mm sur le bassin A). La hauteur d'eau précipitée est homogène spatialement sur le bassin A, alors qu'elle est très hétérogène sur le bassin B. La teneur instantanée des précipitations en18O est très variable temporellement, mais reste homogène spatialement, sur les 2 bassins. L'interception de la pluie par la canopée déstructure donc la hauteur d'eau précipitée sous forêt, mais pas sa signature isotopique. Le 24 mai 1992, nous avons observé une dilution de la composition chimique de la pluie et une diminution de son pH au cours du temps. Les pluviolessivats sont généralement plus concentrés que la pluie et leur pH est plus tamponné. Nous n'avons pas observé de corrélation entre la composition chimique de la pluie ou des pluviolessivats et l'intensité des précipitations. La variabilité spatiale de la composition chimique des pluviolessivats, étudiée lors de l'averse principale du 24 mai 1992, est très importante et 31 pluviomètres semblent insuffisants pour estimer précisément les apports au sol. L'effet de masse est respecté le 24 mai 1992, mais n'est pas visible le 15 mai 1993. La comparaison de l'évolution des teneurs intégrées en Cl- et en18O montre que l'événement pluvieux du 24 mai 1992 est issu d'une masse d'air unique, alors que celui du 15 mai 1993 est issu de plusieurs masses d'air différentes. On remarque également que la teneur intégrée en18O des pluviolessivats est légèrement supérieure à celle de la pluie en milieu ouvert. En l'absence d'évaporation (la composition isotopique des pluviolessivats est alignée sur la droite locale des eaux météoriques), cela s'explique par un mélange entre la pluie directe et de l'eau de pluie plus ancienne, retenue sur la canopée et de composition isotopique différente.
Geochemical hydrograph separation methods are frequently employed because they allow one to determine the origin (spatial or temporal) of water that contributes to creating floods. This approach, based on mass balance equations, requires a good knowledge of the geochemical (isotopic and chemical) signals of the reservoirs that contribute to the flood. However, geochemical signals in precipitation, an obvious reservoir, may vary strongly over time. In forested watersheds, throughfall - and not direct rain - make up the input signal. The geochemical signal of throughfall may be different from that of rain and it may vary temporally and spatially. In order to clarify the use of geochemical tracers for hydrograph separation, we studied the isotopic (δ18O, δ2H) and chemical composition of precipitation (rain and throughfall) in two watersheds, as well as the spatial and temporal variations of this precipitation during two runoff events that occurred on May 24, 1992 and on May 15, 1993. The studied watersheds are located near the city of Sinnamary (French Guyana), 120 km south-west of Cayenne. They are small in size (1,5 ha) and close to each another (200 m). One basin (hereafter named B basin) is covered by primary forest, whereas the other (hereafter named A basin) was cleared and turned into grassland (Digitaria swazilendensis, ÉCÉREX program, supported by Orstom-CTFT). The climate is tropical-humid, characterised by high mean annual temperatures (26°C), which slightly varied from month to month, and high mean annual precipitation (3500 to 3900 mm.yr-1). Precipitation occurred primarily during the main wet season, centred around May and June, and during the secondary wet season from December to January. Given the small distance between the watersheds, the differences noted between the rain collected in the A basin and the throughfall collected in the B basin (amount, geochemical signal) were attributed to the forest cover (leaching, interception,...). The monitoring equipment consisted of 31 rain gauges in the B basin and 3 rain gauges in the A basin. Rainfall was important for the two studied rain events (about 60 mm in basin A). Average rainfall in the A basin is characterised by low coefficients of variation, whereas average water inputs in the B basin showed high coefficients of variation. Thus, the amount of incoming water was spatially homogeneous in basin A, but heterogeneous in basin B. In both basins, the instantaneous δ18O value for precipitation varied considerably over time, but it was still spatially homogeneous in both watersheds with the average δ18O value showing a low coefficient of variation. This result means that the interception of the rain by the canopy destroyed the structure of the precipitation amounts under the forest, but not the structure of its isotopic signal. On May 24, 1992, we noted a dilution of the chemical content of the rain and a decrease in its pH over the course of the event. The chemical contents of the throughfall were on the whole more concentrated than in the incident rain and the pH more buffered. We did not note any correlation between the chemical content of rain or throughfall and the intensity of precipitation. The chemical composition of throughfall, studied during the main shower on May 24, 1992, exhibited considerable spatial variation and 31 rain gauges did not seem to be enough to precisely estimate the amount reaching the soil. A continuous depletion in heavy isotopes (18O,2H) and some chemical species (e.g., Cl-) was noted for the first episode (May 24, 1992) but not for the second (May, 15, 1993). This depletion may be explained by water vapour condensation outside the Rayleigh distillation, or by mixing of different air masses. The comparison between the evolution of integrated values of δ18O and the integrated Cl- content versus the amount of accumulated precipitation proved that the rain event of May 24, 1992, was generated by a single air mass whereas the event of May 15, 1993 was generated by several air masses. We also noted that the integrated value of δ18O for throughfall was slightly more concentrated than the content of rain. In the absence of evaporation (the isotopic composition of the throughfall corresponded to the local meteoric line), this enrichment suggests that direct rain mixed with older water that was stored in the canopy and had a different isotopic composition.This study showed that the intensity and the geochemical signal of precipitation (rain and throughfall) vary greatly on a temporal scale in a tropical environment. It also showed that the amount of incoming water varied spatially under a forest cover, as did its geochemical (isotopic and chemical) signal. In order to achieve a stream hydrograph geochemical separation, it is necessary to collect the precipitation (rain and throughfall) with a short time step. It is also necessary to collect the throughfall across a concentrated network of rain gauges.
