Estimating persistent oil contamination in tropical region using vegetation indices and random forest regression

Estimation de la contamination pétrolière persistente en région tropicale à l'aide d'indices de végétation et de la régression Random Forest

Guillaume Lassalle (1, 2) , Anthony Crédoz (2) , Rémy Hédacq (2) , Georges Bertoni (3) , Dominique Dubucq (2) , Sophie Fabre (1) , Arnaud Elger (4)

1 DOTA, ONERA, Université de Toulouse [Toulouse]
2 TOTAL S.A.
3 DYNAFOR - Dynamiques Forestières dans l'Espace Rural
4 LEFE - Laboratoire Ecologie Fonctionnelle et Environnement

The persistence of soil contamination after cessation of oil activities remains a major environmental issue in tropical regions. The assessment of the contamination is particularly difficult on vegetated sites, but promising advances in reflectance spectroscopy have recently emerged for this purpose. This study aimed to exploit vegetation reflectance for estimating low concentrations of Total Petroleum Hydrocarbons (TPH) in soils. A greenhouse experiment was carried out for 42 days on Cenchrus alopecuroides (L.) under realistic tropical conditions. The species was grown on oil-contaminated mud pit soils from industrial sites, with various concentrations of TPH. After 42 days, a significant decrease in plant growth and leaf chlorophyll and carotenoid contents was observed for plants exposed to 5–19 g kg−1 TPH in comparison to the controls (p < 0.05). Conversely, pigment contents were higher for plants exposed to 1 g kg−1 TPH (hormesis phenomenon). These modifications proportionally affected the reflectance of C. alopecuroides at leaf and plant scales, especially in the visible region around 550 and 700 nm. 33 vegetation indices were used for linking the biochemical and spectral responses of the species to oil using elastic net regressions. The established models indicated that chlorophylls a and b and β-carotene were the main pigments involved in the modifications of reflectance (R2 > 0.7). The same indices also succeeded in estimating the concentrations of TPH using random forest regression, at leaf and plant scales (RMSE = 1.46 and 1.63 g kg−1 and RPD = 5.09 and 4.44, respectively). Four out of the 33 indices contributed the most to the models (>75%). This study opens up encouraging perspectives for monitoring the cessation of oil activities in tropical regions. Further researches will focus on the application of our approach at larger scale, on airborne and satellite imagery.

Mots clés: reflectance spectroscopy; RANDOM FOREST; soil contamination; Total Petroleum Hydrocarbons; vegetation indices

Mots clés: HYDROCARBURES PÉTROLIERS TOTAUX; REFLECTANCE SPECTRALE; CONTAMINATION DU SOL; INDICES DE VÉGÉTATION

Domaines: Ingénierie de l'environnement

Guillaume Lassalle, Anthony Crédoz, Rémy Hédacq, Georges Bertoni, Dominique Dubucq, et al.. Estimating persistent oil contamination in tropical region using vegetation indices and random forest regression. Ecotoxicology and Environmental Safety, 2019, 184, pp.109654. ⟨10.1016/j.ecoenv.2019.109654⟩. ⟨hal-02296861⟩