Streamflow data are very important in assessing the groundwater and surface water resources of a given region. In northern Côte d’Ivoire, particularly in the Bagoé region, although there are long series of rainfall data, streamflow data are still scarce. The few chronicles available are very short and incomplete. The aim of this study is to obtain a long flow chronicle for the period 1996-2016. It aims to estimate flows in the Bagoé River at the Kouto hydrometric station using neural networks. To this end, two neural models were developed to estimate variations in monthly flows of the Bagoé River from 1996 to 2016. The modeling was validated using the Nash criterion (%), the Pearson coefficient (R), the maximum flow ratio and the robustness criterion. The results showed that the validation criteria for these models are optimal. The Nash criterion is greater than 84% for both calibration and validation. The Pearson coefficient ranged from 92% to 96% in calibration and validation. The maximum flow ratio ranges from 93% to 118% in calibration and validation. The robustness criterion ranged from 2.91% to 7.62%. All these results reflect the good performance and stability of neural network-based models for estimating flows in the Bagoé river.

The objective of this study performed in the Abidjan District is to map land cover units using the Google Earth Engine (GEE) platform and Machine Learning algorithms such as Random Forest (RF), Support Vector Machine (SVM), Classification and Regression Tree (CART), Naive Bayes (NB), and Minimum Distance (MD). The data used include optical Multispectral Sentinel 2A satellite images with a 10-meter resolution, a 12.5-meter Alos Polsar digital terrain model (DTM) resampled to a 10-meter resolution, as well as cartographic data. The implemented methodology starts with the preprocessing and normalization of the composite image. The final composite image is created using eight spectral indices: NDVI, NDWI, MNDWI, VARI, SBI, SAVI, GCI, RGR, along with the first three bands of Principal Component Analysis and slope information. Subsequently, training and validation points are collected and coded based on image reflectance and ground truth data. The different classifiers SVM, RF, CART, MD, and ND are then trained and evaluated using various metrics such as confusion matrix, overall accuracy, producer’s accuracy, consumer’s accuracy (reliability), and Kappa coefficient. The classification performed with the RF algorithm achieved the highest overall accuracy of 83.28%, with a Kappa coefficient of 0.78. The statistics reveal that the Abidjan District is composed of 28.07% urban areas, 25.35% agricultural and other cultivated areas, 12.39% oil palm plantations, 10.05% rubber plantations, 4.66% banana plantations, 2.53% forests, 3.96% mangroves, 3.80% forest plantations (reforestation), and 9.2% water bodies in 2020. This study has led to an improved mapping of the distribution and proportions of land cover classes in the Abidjan District.