Fig 1.
Research area in Setiu Wetland.
(A) Setiu Wetland as shown by Pleiades-1B satellite imagery (acquired on 16 August 2013 at 11.43 AM). Red box indicates the location of mangroves being considered for vegetation mapping in the present study. (B) Pan-sharpened Pleiades-1B imagery (spatial resolution: 50cm) showing the zoomed-in portion of the mangroves selected for mapping (orange box). (C) DJI-Phantom-2 drone imagery (acquired on 3 July 2015 at 10.00AM) (spatial resolution: 5 cm) showingthe same mangrove coverage area as that of Pleiades for mapping (orange box) (background: Pleiades imagery).
Fig 2.
Stepwise protocol and the technical processes involved in drone and satellite remote sensing data analyses for mangrove mapping at the Setiu Wetland.
The 10 ROI sets were named 1A-1B to 5A-5B. Except the manual rule-set algorithm, the remaining algorithms i.e., automatic, maximum likelihood and spectral angle mapping, were used 10 times (10×) for running the object- and pixel-based classification approaches (grey and white shades are for visualization purposes).
Table 1.
List of bands in a RGB, IR and DEM merged imagery of the DJI-Phantom-2 drone.
Fig 3.
Comparison of cropped true colour composite images from Pleiades-1B satellite and DJI-Phantom-2 drone.
(A-B) The images for mangrove vegetation mapping at the Setiu Wetland. Red boxes show the zoomed-in subsets of–(C) Pleiades-1B and, (D) DJI-Phantom-2 drone, revealing mangrove and non-mangrove details on the ground at 50cm and 5cm spatial resolutions respectively.
Table 2.
Spatial, spectral, radiometric and temporal resolutions of the Pleiades-1B satellite and DJI-Phantom-2 drone images (source for Pleiades-1B information: Pleiades user guide [74]).
Fig 4.
Visual representation of the spatial coverage of Pleiades-1B and DJI-Phantom-2 drone data sets.
While new tasking/purchasing order of Pleiades images requires at least 100km2 coverage, the archived data of each image is available for a minimum of 25 km2. A drone is expected to work efficiently (if it does not crash or have technical problems) for 500 flights. If an efficient drone flight for 15 min (corresponding to average battery run-time) can cover approximately 0.1 km2, the total drone flights would be able to cover ca. 50 km2. However, with an improved battery run-time of up to 20 min these days, the same drone can deliver aerial photos of an area covering up to 75 km2 (box dimensions are arbitrary, and the colours are for visualization purposes only).
Table 3.
Data processing time of DJI-Phantom-2 drone and Pleiades-1B satellite images.
For the drone, the accuracy analysis of pixel-based classification was conducted twice due to there being two different training sites–one representing dominant land-use/cover classes, and another representing both dominant and non-dominant classes visible on the ground.
Fig 5.
Land-cover classes at the Setiu Wetland.
(A) Land-cover classes based on Pleiades-1B satellite (on the left) and DJI-Phantom-2 drone (on the right). Due to the poor demarcation of some features in the Pleiades, only 1–6 land-cover classes were considered for its image classification. (B) Locations of the ten (1–10) land-use/cover classes marked on the Pleiades-1B (top) and DJI-Phantom-2 drone (bottom) images.
Table 4.
Details of the classification approaches and resultant maps (with accuracy iterations) using DJI-Phantom-2 drone and Pleiades-1B satellite images for the Setiu Wetland.
Each classified map was given a unique identification code that starts with ‘D’ for drone and ‘S’ for satellite, followed by a number of the land-use/cover classes used (10 = all ten land-use/cover categories and 6 = dominant six classes), classification approach (O = object-based and P = pixel-based), and the algorithm (MAN = Manual rule-set, AUT = Automatic, MLI = Maximum Likelihood and SAM = Spectral Angle Mapping) (OA = Overall Accuracy) (ROI = Region of Interest) (*training site codes follow those used in Fig 2).
Fig 6.
The Setiu Wetland mangrove maps based on object-based classification.
Masp based on Manual rule-set algorithm (A-C) and the Automatic classifier algorithm (D-F). The images shown have the highest Overall Accuracy from 10 iterations. (Abbreviations for each image follow the map identification codes in Table 4. OA = Overall Accuracy. Genus names: A = Avicennia, B = Bruguiera, L = Lumnitzera, N = Nypa, R = Rhizophora, S = Sonneratia and, C = Casuarina.).
Fig 7.
The Setiu Wetland mangrove maps based on pixel-based classification.
Map sbased on Maximum Likelihood algorithm (A-C) and the Spectral Angle Mapping algorithm (D-F). The images shown have the highest Overall Accuracy from 10 iterations. (Abbreviations and genus names follow those used in Fig 6).
Fig 8.
Accuracy analysis of the mangrove vegetation mapping at the Setiu Wetland, based on DJI-Phantom-2 drone and Pleiades-1B satellite images.
(A) Overall Accuracy, (B) Kappa Index, (C) Specific Accuracy and, (D) Specific Reliability. Each abbreviation along the X-axis follows the map identification codes in Table 4.
Fig 9.
Mean shift (black), exchange (grey) and quantity (white) differences in the first three maps with highest overall accuracy.
Result based on DJI-Phantom-2 drone (A-B) and Pleiades-1B satellite (C) images for the Setiu Wetland. Abbreviations for each image follow the map identification codes in Table 4. Abbreviations follow the identification codes in Table 4.
Table 5.
Various types of drone equipment, camera choices and sensor combinations, with applicable prices, useful for mangrove vegetation mapping (source: DJI Store [84], Specsheet Sequoia [85]) JI, 2017; Parrot, 2016).