Table 1 Examples of the use of uncrewed aerial vehicles in vector-borne diseases showing ecosystems, purpose of the study, mosquito species targeted and technical characteristics of the drone surveys
From: The use of drones for mosquito surveillance and control
Location | Landscape/ecosystem | Area | Purpose | Feature of interest | Target vector | Equipment | Pixel resolution | Ground-based activities | Image processing | Analysis | Other | Outcomes | References |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Westchester and Suffolk Counties, NY, USA | Suburban neighbourhoods | Westchester, 0.05 km2; Suffolk, 0.14 km2 | Determine capabilities of UAVs to image preferred oviposition sites of Aedes albopictus; validate results by entomological survey | Oviposition containers | Aedes albopictus | DJI Mavic Pro drone (DJI, Shenzhen, China), with custom-built UAV designed around 3DR Pixhawk 1 flight controller (3DR, Berkeley, CA, USA) using a GoPro Hero 5 camera (GoPro, Inc., San Mateo, CA, USA) | DJI Mavic < 1 cm; Go Pro, 1.5—4 cm/pixel | Yes: entomological and container ground survey | Outer 500 pixels cropped to account for lens distortion with GO PRO photos using the open-source ImageMagick (2020); images sliced into 512 × 512-pixel images to translate to SSD network preinitialized weights and bounding box anchors; annotated aerial imagery (LabelImg) to determine container types UAVs could see; piexif in Python used to convert container locations into GPS | CNN based on the SSD300 algorithm using API Keras on top of Tensorflow. Amazon EC2 p2.xlarge instance handled computations using Amazon Machine Image | Flight at altitude of 50 m; pre-programmed flight paths (Ardu Pilot Development Team software suite 2019) for Pixhawk quadricopter and DJI GO 4 for Mavic Pro | CNN trained on UAV imagery detected up to 67% Ae. albopictus habitat; classified whole properties as positive or negative for larvae 80% of the time | [36] |
Cote d'Ivoire | Rural, agricultural landscape | 30.42 km2 | To develop a technical workflow to integrate drone surveys and mosquito larval sampling. To characterize Anopheles funestus breeding sites in an agricultural setting | Large, semi-permanent water bodies | Anopheles, Anopheles funestus | DJI Phantom 4 Pro quadcopter drone, fitted with a DJI 4 K camera (8.8 mm/24 mm; f/2.8; 1″ CMOS sensor; 20 MP) | ~ 4 cm | Yes: GPS coordinates and photos were collected via ODK, and the predominant land cover type was recorded | AgiSoft Metashape Pro (AgiSoft LLC, St. Petersburg, Russian Federation) | Images classified using collaborative online image labelling tool Groundwork (https://groundwork.azavea.com); developing of deep learning approaches (u-Net) integrating drone and satellite data on Amazon Web Services | Flight at an altitude of 150 m Flight plans were programmed using Pix4Dcapture and DJI GS Pro mapping applications | Development of land cover classification scheme with classes relative to An. funestus breeding ecology; developed protocols to integrate drone surveys and statistically rigorous entomological sampling | [113] |
Maynas Province, Peru | Riverine/rainforest | ~ 1 km2 | Use of UAVs to identify breeding sites | Water bodies | Nyssorhynchus darlingi (formerly Anopheles darlingi) | DJI Phantom 4 Pro quadcopter drone, fitted with a DJI 4 K camera (8.8 mm/24 mm; f/2.8; 1'' CMOS; 20 MP) 3DR Solo (3DR) quadcopter fitted with a Parrot Sequoia multispectral sensor (Parrot SA, Paris, France) | DJI Phantom 4 Pro, GSD or spatial resolution of 0.1 m/pixel. 3DR Solo GSD of 0.02 m/pixel | 31 water bodies located within 1 km of each of 4 villages surveyed were identified, characterized and sampled along edges for larvae using standard dippers | AgiSoft Photoscan Pro (AgiSoft LLC) | GEE. Integrated Development Environment at https://code.earthengine.google.com used to implement pixel-based random forest classification algorithm | DJI Phantom 4 Pro—altitude of ~ 100 m.3DR Solo drone was flown to an altitude of approximately 50 m | High-resolution multispectral imagery discriminated water bodies likely to be positive for Ny. darlingi with 86.73–96.98% accuracy (k-fold cross validation), with a moderate differentiation of spectral bands | [22] |
Sabah, Malaysia/Palawan | Forest/agriculture | 3 case study areas of 50–100 km2 | To characterize land-use types and create a spatial sampling frame | Land cover | Habitats of Anopheles balabacensis and non-human primate reservoirs of Plasmodium knowlesi | senseFLY eBee drone (senseFLY, Cheseaux-sur-Lausanne, Switzerland), fitted with 16-megapixel digital camera, eMotion2 software (Ageagle Aerial Systems, Wichita, KS, USA) | Average resolution of 11.22 cm | Yes: ground-truth land cover classes; larval and adult mosquito surveys | Postflight Terra3D software (Terra3D, Paris, France) | Land cover classes manually digitized, used to identify training data for pixel-based classification of Landsat satellite data using random forest algorithms | 350–400 m altitude | Use of UAVs most appropriate for detailed mapping of relatively small geographical areas where high-resolution satellite data are not readily available | [46] |
Kinabatangan, Sabah, Malaysia | Riverine forest | 10 × 1-km transects along 20 km of riverbank | To compare visual counting versus use of thermal cameras for primate census | 8 diurnal primate species | N/A, reservoir | Custom built hexacopter drone with FLIR Thermal Camera (Teledyne FLIR LLC, Wisonville, OR, USA) | Yes: boat surveys of primates in early morning | FLIR Tools (Tledyne FLIR LLC Co.) | Visual identification and counting | Thermal cameras detected 1.78x (P < 0.001) more primates than were detected by eye; ground-truthing must be conducted during or immediately following to verify species, sexes, age classes and closely aggregated animals | [81] | ||
Kasunga town, Malawi | Rural area surrounding artificial lakes in designated "humanitarian drone testing corridor" | 8.9 km2 across 8 sites | Determining the feasibility of drone-led mosquito larval habitat identification in rural environments to inform local malaria control | Water bodies and aquatic vegetation | Anopheles | DJI Phantom 4 Pro quadcopter drone, fitted with NIR sensor(Sentera) vs. eBee SQ fixed-wing drone fitted with Parrot Sequoia multispectral camera (Parrot SA) | RGB camera: 3.3 for Phantom 4 Pro; 3.7 for eBee SQ; NIR sensor, 11 for both | Yes: larval surveys conducted concurrently with drone image capture; GPS co-ordinates recorded using ODK app; photographs taken using smartphone; aerial imagery were captured | Spervised classification of land cover classes (AgiSoft Metashape Pro; AgiSoft LLC) | Land cover and surface water layer classes were manually digitized, used to identify training data for pixel-based classification using random forest algorithms | flying at 120 m | Demonstration of potential for drone imagery as a tool to support identification of mosquito larval habitat in malaria endemic settings Cannot completely replace larval surveys | [71] |