Pre- and Post-Pruning Assessment of Lychee Tree Crop Structure Using Multi-Spectral RPAS Imagery Dr Kasper Johansen & Tri Raharjo Remote Sensing Research Centre The University of Queensland 1
Outline 1. Introduction and Objectives 2. Study Area 3. Data 4. Methods 5. Results 6. Conclusions and Future Work 2
1. Introduction Tropical tree native to China, producing small fleshy fruit. China and India account for > 80% of total production. Lychee production in Australia worth > $20m annually. Australian lychees in northern hemisphere off-season. 3
1. Introduction Pruning of trees: encourages new growth; has a positive effect on fruiting of lychee; makes fruit-picking easier; and increases yield, as it increases light interception and tree crown surface area. Objectives: to assess changes in tree structure (tree crown height, area, circumference and width, and plant projective cover) using multi-spectral RPAS imagery collected before and after pruning of a lychee plantation; and to assess any variations in the results as a function of three different flying heights. 4
2. Study Area Shailer Park, QLD Shailer Park located 25 km southeast of Brisbane Site context 5
2. Study Area Shailer Park, QLD Shailer Park located 25 km southeast of Brisbane 6
Green, Red, Red Edge, NIR bands 11 Feb 2017 (prepruning 4 March 2017 (postpruning) 3. RPAS Data 7
3. RPAS Data Tower Beta app, 80% overlap, 1 photo / sec, 5 m/s 30 m = 4.1 cm pixels, 50 m = 6.5 cm pixels, 70 m = 8.8 cm pixels 8
3. Field Data 9
3. Field Data 89 trees sampled, 4 March Tree height Tree crown circumference Tree crown width Plant projective crown cover 78.25% 10
4. Methods Image Processing Generation of geometrically corrected orthomosaics, DSM and DTM in Pix4D Mapper 11 February 2017 4 March 2017 DSM DTM 40 20 0 40 Meters 11
4. Methods Radiometric Correction 12
4. Methods Radiometric Correction Green Red Red Edge NIR 13
4. Methods Radiometric Correction 14
4. Methods GEOBIA 50 m 50 m Map lychee tree extent CHM and spectral info Identify tree crown centres Grow tree crown centres based on CHM Adjust shape and tree crown dimensions 15
4. Methods RPAS Derived Tree Crown Parameter Extraction Spectral Information: Green, Red, Red Edge, NIR; Brightness, NIR+Red Edge, NDRE, NDVI Height Information: Average CHM, Max CHM, 90 Percentile per tree crown Geometry: Area, Perimeter length, Length, Width Texture: GLCM Homogeneity, Contrast, Dissimilarity, Standard deviation using the Green, Red, Red Edge and NIR bands 16
5. Results Tree Crown Delineation 50 m 17
5. Results Tree Crown Perimeter RMSE = 4.57 m RMSE = 3.63 m RMSE = 3.42 m 18
5. Results Tree Crown Width Similar results for all flying heights Measurements of tree crown width not affected by pixel size 19
5. Results Tree Crown Height The DTM had higher elevation (up to 60 cm) in some locations for the data set collected at 70 m The DSM showed a lowering of relative height of features above ground with increasing flying height. 20
5. Results Plant Projective Cover RedEdge and NIR bands produced highest correlation Very low correlation between red band and PPC Texture explains some variation in PPC 21
5. Results Plant Projective Cover NDVI insensitive to PPC variation due to red band Brightness under- and Red Edge over-estimated PPC 22
5. Results Pre- and Post Pruning NIR used to predict PPC for 11 Feb 2017 (pre-pruning) An average of 14.8% decrease in PPC, using the two data sets collected at 30 m height. Smaller trees not pruned 23
5. Results Pre- and Post Pruning Height differences based DTM and DSM quality An average of 61.6 cm decrease in height, using the two data sets collected at 30 m height. Smaller trees not pruned 24
5. Results Pre- and Post Pruning Area differences based on quality of tree crown delineation An average of 3.5 m 2 decrease in area, using the two data sets collected at 30 m height Smaller trees not pruned and showed slight area increase 25
5. Results Pre- and Post Pruning Perimeter differences based on quality of tree crown delineation An average of 1.94 m decrease in perimeter, using the two data sets collected at 30 m height Smaller trees not pruned and showed slight increase 26
5. Results Pre- and Post Pruning Tree crown dimension differences based on quality of tree crown delineation An average of 65.3 cm decrease in length, using the two data sets collected at 30 m height Smaller trees not pruned and showed slight increase 27
5. Results Pre- and Post Pruning Tree crown dimension differences based on quality of tree crown delineation An average of 56.7 cm decrease in width, using the two data sets collected at 30 m height Smaller trees not pruned and showed slight increase 28
5. Results Pre- and Post Pruning Difference between 30 m, 50 m and 70 m flying height Plant Projective Cover, n = 89 29
5. Results Pre- and Post Pruning Average differences between 30 m, 50 m and 70 m flying height Height, Area, Perimeter, Length and Width, n = 89 30
6. Conclusions ecognition Developer could be used to map individual lychee tree crowns The Sequoia image data were found suitable for assessing pre- and post-pruned tree crown structure: Plant projective cover (best predicted with NIR band) Tree height (most accurately mapped at 30 m height) Tree crown perimeter, area, and dimensions Tree crown perimeter most accurately mapped at 70 m Tree crown width and length similar for all flying heights PPC accurately predicted at all three flying heights, although data collected at 70 m produced slightly higher correlation for most predictor variables. The developed approach may be used to assess pruning efforts undertaken by contractors 31
Pre- and Post-Pruning Assessment of Lychee Tree Crop Structure Using Multi-Spectral RPAS Imagery Dr Kasper Johansen & Tri Raharjo Remote Sensing Research Centre The University of Queensland 32