Arid and semi-arid regions of Sub-Saharan Africa, heavily reliant on rainfall, are experiencing increasingly severe droughts due to climate change, significantly impacting agropastoral communities. Water pans, which are small depressions designed to capture surface water runoff, are crucial for livestock and wildlife but are prone to drying up and are not well-documented by national governments. Effective monitoring is essential to support water security for these communities. Although water pans are usually too small to be observed by most freely available satellite remote sensing datasets, high-resolution remotely sensed data can provide comprehensive information about them. This study evaluated high-resolution remote sensing data types for detecting and mapping water pans in Taita Taveta County, Kenya. We trained computer vision models to detect water pans from red, green, and blue (RGB) and colour-infrared aerial images, digital terrain models (DTM) based on airborne LiDAR, and high-resolution Pleiades and Planet satellite data. YOLOv5, YOLOv8, and Faster R-CNN models achieved high accuracy with the highest spatial resolution data. Aerial imagery, DTM, and Pleiades data achieved mAP50 values greater than 70% with all models. Much lower accuracy was recorded with coarser resolution Planet data, which achieved mAP50 values below 20% with YOLO models and an average precision of less than 10% with Faster R-CNN. Despite this, Planet data was reliable for larger water pans and was available at no cost. Our results demonstrate that remote sensing data and computer vision models can enhance the understanding of water pan distribution in agropastoral areas in Kenya, marking the first step towards effective water pan monitoring in Sub-Saharan Africa.