Abstract
Remotely piloted aircrafts (RPAs) are becoming well-known for monitoring forest restoration (FR). Canopy cover percentage on a restoration site is a major indicator for the early phases of FR. We present a protocol to evaluate canopy cover using low-cost RPA and a friendly computational environment. We tested the protocol in six FR projects in the Brazilian Atlantic Forest. After processing the images automatically to generate the digital surface model and the digital terrain model, the canopy cover is measured in QGIS by considering height thresholds in the canopy height model. The results had F1 values higher than 0.9, which confirms a robust methodology that does not require specific computer science or remote sensing skills. Thus, low-cost RPA proved to be accurate and user-friendly in measuring the canopy cover, which improves monitoring during most FR stages. Such finding plays an important role in FR policies and future works must study a user-friendly protocol for measuring tree height via low-cost drones.
1 Introduction
Forest restoration (FR) monitoring is critical to ensure the estimation of the restoration outcomes [1,2,3]. FR is the process of recreating the ecological function of forests that have been degraded due to timber harvest and fire suppression [1].
FR monitoring consists of a set of indicators where each indicator evaluates a specific forest attribute. A major ecological indicator of early FR success is canopy cover, which is the percentage of soil covered by tree crowns. During early FR, canopy cover must reach at least 70–80% [4]. When a site closes the canopy, a restored forest shades out invasive grasses, attracts seed dispersers, and modifies the microclimate and soil for tree seedling establishment [4,5]. After closing the canopy, the next phase is the density and diversity of recruits that will replace canopy trees in the next years, guaranteeing a long-term forest structure and species diversity [4,5].
Remotely piloted aircraft (RPA), popularly known as drones, have a high potential to monitor FR [6,7]. RPA can register whole areas of interest with a georeferenced panoramic high spatial resolution image and point cloud data [8,9,10]. In FR projects, this technology could assess relevant information on areas not covered by traditional field plots [7,11].
RPA has accurately measured canopy cover in open canopy conditions [7,12,13,14], but assessing and publishing its accuracy on a wider spectrum of open and closed canopy conditions is timely given. Thus, considering the RPA potential for FR monitoring, it is worth publicizing a friendly protocol for measuring canopy in a wide variety of FR situations. In such a protocol, society should be able to monitor canopy cover accurately and without specific knowledge requirements, as Mohan et al. [15] did, but for individual tree detection. For canopy cover in FR projects, machine learning processes accurately measured it [16], but a remote sensing specialist is required to conduct this kind of methodology.
This study aims to present a low-cost RPA canopy cover monitoring protocol where neither specific field of knowledge is required, nor any remote sensing samples. To test and illustrate the protocol, we evaluated its effectiveness in six different FR situations.
2 Materials and methods
2.1 Materials
The RPA must couple RGB sensors that have at least 1/2.3″ and 12 MP CMOS. Within such sensor specs, the RPA takes pictures that will be processed by structure from motion – multi view stereo (SfM-MVS) [17,18] to generate cartographic information. For the case studies of this work, we used a Phantom 4 Pro (a rotary wing RPA), which is coupled with a RGB 1″ and 20 MP CMOS. For more information about this model, see [19].
Regarding software, we mentioned the ones that we applied in the case studies of this work, but other software with similar functions can be used. Thus, in the case studies, we drafted the flight plan using the Map Pilot software [20], while we obtained the digital surface model (DSM), digital terrain model (DTM), and orthorectified mosaics using the Agisoft Metashape [21] software. We used the QGIS [22] software version 3.12 to perform the image processing workflow and map layouts.
2.2 Methods
2.2.1 Flight patterns
All flights must comply with the local authority’s laws, and thus the heights of the flights are recommended to be less than 120 m in general. The RPA pilot must choose the flight height according to: the desired ground sampling distance (GSD); and the height of the surface objects (mainly trees) to avoid RPA collisions.
The flights must be planned according to the RPA characteristics. Since this protocol considers low-cost RPA, the battery of this kind of equipment usually enables 15 min flights. Low-cost RPA can thus be recommended for areas smaller than 100 ha [23], but it is possible to use more than one RPA during the fieldwork for mapping large sites. Despite this battery limitation, RPA engines fed by solar energy may solve this flight time limitation in the near future [24].
In the RPA flight missions, the camera angle must be equal to zero degrees (Nadir), and the minimum front and side overlap between photos must be equal to 90 and 80%, respectively. Although these overlapping values increase the SfM-MVS processing time for generating the cartographic data, it provides enough details that are required for processing the point cloud data to FR projects [25].
2.2.2 Canopy cover measured by RPA
Figure 1 synthetizes the canopy cover measurement method via low-cost RPA.
Methods for the canopy cover measurement via low-cost RPA.
Ground control points (GCPs) are not mandatory for the canopy cover measurement, and thus the cartographic precision of the RPA data gets close to 3 m in these situations [26]. If each RPA image without GCP accurately provides canopy cover, it is possible to monitor this indicator’s evolution through time. However, the lack of GCP makes it not possible to overlap the images or the canopy cover maps between them (cartographic precision is not enough to overlap maps in this case). Therefore, if the objective is to measure canopy cover, which is a number in percentage (%), and not to overlap maps, GCP is not mandatory. The lack of GCP is an advantage because it reduces costs due to a shorter time spent in fieldworks [27] and the considerable price of precise global navigation satellite system (GNSS) equipment [28]. If one needs to overlap maps, GCPs have to be measured at least in the first flight, as Albuquerque et al. [7] recommends.
After flying the RPA and generating the orthomosaic and the DSM via SfM-MVS, canopy cover is measured using the canopy height model (CHM) data, which is the difference between DSM and DTM [29]. Then, each FR site must have the values of the algorithm parameters for DTM generation calibrated according to its characteristics and to the experience of the analyst. To illustrate some examples of values that can be used on the DTM algorithm parameters, Table 1 gives the “classify ground points” tool parameters (which are maximum angle, maximum distance, and cell size in Agisoft Metashape) that we used in the case studies of this work.
Values of the “Classify ground points” algorithm parameters that were used and the canopy cover accuracy of the case studies
| “Classify ground points” algorithm parameters | Canopy cover accuracy | |||||||
|---|---|---|---|---|---|---|---|---|
| FR site | Max angle (°) | Max distance (m) | Cell size (m) | Canopy cover | Overall accuracy | Precision | Recall | F1 |
| 6 years old active FR with a 1 year active FR patch | 11 | 0.6 | 35 | 0.417 | 0.935 | 0.892 | 0.99 | 0.938 |
| 4–16 years-old active FR varying in different patches | 2.5 | 0.5 | 10 | 0.735 | 0.97 | 0.99 | 0.95 | 0.969 |
| 1 year active and passive FR with some mature forest patches | 1 | 0.1 | 3 | 0.353 | 0.965 | 1 | 0.93 | 0.964 |
| 40 years old active FR with grass patches | 5 | 0.2 | 20 | 0.698 | 0.985 | 0.98 | 0.99 | 0.985 |
| Mature forest with grass patches | 11 | 0.6 | 35 | 0.646 | 0.985 | 0.971 | 1 | 0.985 |
| Mature forest with 1 year old active FR patches | 15 | 1 | 50 | 0.518 | 0.975 | 0.952 | 1 | 0.976 |
After generating the DTM, the CHM is obtained by the difference between the DSM layer and the DTM layer. In the case studies of this work, we performed this operation using the “Raster Calculator” tool in QGIS. Then, after the CHM acquisition, a height value threshold has to be defined in the CHM layer to avoid confusions with bulky grass. According to Viani et al. [4], canopy cover in the Atlantic Forest biome must separate trees (woody vegetation) from grasses (herbaceous vegetation). Since some sites may present conditions that favor grass development, it is relevant to check how much tall the grass is at each site. A CHM threshold may perform such a task by involving only trees when they are taller than grasses. In QGIS, this CHM threshold can be applied using the “Raster Calculator” tool by entering, for instance, “CHM_layer > 0.5,” where “CHM_layer” is the CHM raster layer and the grass height is 0.5 m tall. The grass height varies in each FR project, and thus the user must check which threshold value generates feasible results. In the case studies of this work, the grass height ranged from 0 to 1 m along with the FR sites.
After choosing the threshold value that includes only woody cover, the “Raster Calculator” tool in QGIS generates a binary mask where tree pixels have value 1, and non-tree pixels have value 0. To check the area (in square meters) covered by tree and non-tree pixels, the “Raster layer unique values report” tool in QGIS can be used. Canopy Cover is then calculated according to equation (1).
where CC = canopy cover (ranges from 0 to 1), ATP = area of tree pixels, and ANTP = area of non-tree pixels.
2.2.3 Case studies
We evaluated six FR sites in the Brazilian Atlantic Forest biome. Three sites are located in the municipality of Extrema in the state of Minas Gerais (MG) – Brazil and are part of the Fernão Dias Environmental Protection Area. The other three sites are in the municipality of São José dos Campos in the state of São Paulo (SP) – Brazil and inside the Paraíba do Sul Basin Environmental Protection Area (Figure 2).
Location of the FR case studies on Brazilian Atlantic Forest.
The case studies were active and passive restoration [30,31] sites. Site 1 (Figure 3a) has a 6 years old active restoration patch and a 1 year old active restoration patch in a hilltop area. Site 2 (Figure 3b) has active restoration patches varying from 4 to 16 years old. Site 3 (Figure 3c) has a 1 year old active and passive restoration patch and some mature forest patches. Site 4 (Figure 3d) is a 40 year old active restoration site with some grass patches. Site 5 (Figure 3e) is a mature forest with grass patches. Site 6 (Figure 3f) is a mature forest with patches of 1 year old active restoration.
RPA orthomosaics of the 6 FR sites: 6 years old active restoration patch and a 1 year old active restoration patch in a hilltop area (a); active restoration patches varying from 4 to 16 years old (b); 1 year old active and passive restoration patch and some mature forest patches (c); 40 years old active restoration with grass patches (d); mature forest with grass patches (e); and mature forest with patches of 1 year old active restoration (f).
The flights over the study areas in Extrema Municipality happened in January 2019, while in São José dos Campos in June 2019. All flights complied with Brazil’s RPAs laws [32] at 100 m height, generating approximately 3 cm GSD with front and side overlap equal to 90% and 80%, respectively.
When evaluating canopy cover accuracy, each site had 200 test samples collected by photointerpretation for the binary classification of trees (100 samples) and non-trees (100 samples). Each sample consisted of a pixel because it is considered more flexible than blocks of pixels [33] and can be applied when there is no segmentation process to detect objects [34]. After all, measuring canopy cover is not the same as detecting each tree crown via Object-based image analysis (OBIA) classification [35] or instance segmentation [36]. The amount of 200 test samples was enough because each hectare must contain at least 10 test samples [37], and the largest study area of the manuscript has 17.5 ha.
After collecting 200 test samples via photointerpretation of the RPA orthomosaic of each site, we then calculated the overall accuracy [38], recall, precision, and F1 indexes [39] according to equations (2)–(5), respectively.
where OA = overall accuracy, TP = true positive, FN = false negative, FP = false positive, r = recall, and p = precision.
3 Results
Figure 4 shows the canopy cover mapping results of all the case studies. By looking at Figure 4, the canopy cover results seem to be very accurate when assessed qualitatively.
: Canopy cover results on the RPA orthomosaics of the 6 FR sites: 6 years old active restoration patch and a 1 year old active restoration patch in a hilltop area (a); active restoration patches varying from 4 to 16 years old (b); 1 year old active and passive restoration patch and some mature forest patches (c); 40 years old active restoration with grass patches (d); mature forest with grass patches (e); and mature forest with patches of 1 year old active restoration (f).
Table 1 shows the parameter values used by the “Classify ground points” algorithm in Agisoft Metashape to generate the DTM of each case study site. Table 1 also shows the canopy cover result’s quantitative accuracy, which proves that the low-cost RPA accurately mapped it in all case studies. All F1 values are higher than 0.9, while the mean overall accuracy, precision, recall, and F1 of all the case studies were equal to 0.97, 0.96, 0.98, and 0.97, which shows the robustness of the method. Some grass patches were included in some of the FR surrounding areas (Figures 3 and 4) to assess the proposed method’s capability of separating woody and herbaceous vegetation.
4 Discussion
The low-cost RPA canopy cover monitoring protocol accurately mapped the canopy cover by using a friendly computational environment. No advanced skills are required to run the proposed method, so one must know how to fly an RPA, generate an orthomosaic, a DSM, a DTM, and a CHM. These are widespread automatic procedures that can be learned after a few days of proper training without specific background as a pre-requisite.
GCP are not mandatory. This is relevant because GCP increases operational costs due to the amount of time spent in fieldwork [27] along with the high price of precise geodetic GNSS equipment [28]. Besides increasing the costs, managing geodetic GNSS equipment is a specific field of knowledge that requires intensive training for proper operation.
Moreover, no machine learning processes are involved, no coding abilities are required and the canopy cover calculation can be done in a free and open-source software. Other solutions mentioned in the literature require costly hardware (multispectral sensor and LiDAR – light detection and ranging) and supervised classification of land cover classes [7,16,40,41,42,43]. However, this work demonstrated a low-cost RPA methodology that does not require land cover samples, which is relevant because these methods (where samples are not mandatory) are more automated than the supervised classification ones [44,45].
Another advantage of the protocol described in this work is that it uses the point cloud data of a single flight. This flight generates a single image that does not depend on sunlight conditions, and thus it does not require domain adaptation [46], which is a remote sensing disadvantage when involving machine learning processes [47]. Also, in this work, the closed canopy conditions did not degrade the canopy cover accuracy, as it does to SfM-MVS tree height measurement, as stated in the literature [12,13,14].
We generated the orthomosaic and the CHM using commercial software in this study, but open-source software is more democratic [48]. If open license initiatives, like OpenDroneMap [49], prove to be capable of accurately generating these cartographic data, RPA usage in FR monitoring would become even more accessible. In this case, only the acquisition of the hardware (RPA and computer with processing capacity) would be necessary. Also, it is relevant to highlight that some initiatives, like OpenDroneMap, also offer processing data in the cloud and, therefore, could avoid the high processing computer requirement.
Despite using a user-friendly hardware and software environment, the canopy cover method applied in this study works only after the tree height becomes higher than grass height. Albuquerque et al. [7] used a non-user-friendly method to map canopy cover in patches where trees were a bit taller than grasses, which means that, in FR areas in biomes such as the Brazilian Cerrado, where invasive grass significantly jeopardizes trees’ growth [50], our methodology may not work during the first years of restoration. However, other relevant Biomes, such as the Amazon and the Brazilian Atlantic Forest, benefit from our RPA remote sensing methodology.
Besides 90% of the front and side overlap between photos, Swayze et al. [51] recommends crosshatch flights for a more refined point cloud data of vegetated areas. The authors stated that crosshatch flights improve the accuracy of relevant parameters measured by RPA, such as diameter breast height (DBH) and tree height. However, this study evaluated the canopy cover only. In closed-canopy conditions, although canopy gaps are a relevant biodiversity metric [52] that can be assessed by canopy cover measurements, tree height also becomes a relevant structural parameter for FR monitoring because it estimates the community ecology sustainability [53] and the forest biomass/carbon [54], which is related to species diversity [55]. However, the canopy height measurement via low-cost RPA requires GCP [7]. Thus, future works must evaluate user-friendly protocols for tree height measurements using low-cost RPA in the FR context.
5 Conclusion
This study shows a canopy cover monitoring protocol for FR projects using low-cost RPA and a user-friendly computational environment. The method was accurate in all the six case studies, which involved the whole range of canopy cover. Thus, the low-cost RPA protocol presented in this work can improve the monitoring policies such as the ones of the Atlantic FR Pact. Since no specific field of knowledge is required, restoration technicians can use this protocol for canopy cover monitoring of FR projects without intense training requirements, as hardware acquisition, flights, and time for processing the RPA data are the main issues to be planned. The protocol can be useful in the initial FR stage, when canopy cover is less than 70%, and in the advanced stage, when canopy gaps play a relevant role.
Acknowledgments
The authors thank World Resources Institute (M.H.M. and M.C. were WRI employees when we conducted the fieldwork), the field support of Extrema’s City Hall in Extrema Municipality and of Instituto de Pesquisas Espaciais (INPE) graduate students in São José dos Campos Municipality, Dr. Laura Borma, for providing access to São José dos Campos FR areas, and the Federal University of Goiás/LAPIG/Pro-Vant for encouraging RPA projects. We also thank the SPAMLab at IEE-USP for the provision of the RPA, data processing and analysis infrastructure and Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) for financial support on PhD scholarships. We are thankful for the owners of the FR areas. M.E.F. (grant #315699/2020-5) and C.H.G. (grant #423481/2018-5 and #304413/2018-6) are CNPq Research Fellowship.
-
Funding information: This study was financed in part by the World Resources Institute (WRI) and in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior – Brasil (CAPES) – Finance Code 001. M.E.F. (grant #315699/2020- 5) and C.H.G. (grant #423481/2018-5 and #304413/2018-6) are CNPq Research Fellows. The Article Processing Charge (APC) was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior – Brasil (CAPES) – Finance Code 001 and in part by WRI.
-
Author contributions: R.W.A., C.H.G., and M.H.M. designed the experiments and R.W.A. carried them out. M.H.M. and M.C. provided funding and helped the fieldworks. M.C., M.E.F. and D.L.M.V. pointed improvements to the design of the manuscript. R.W.A. prepared the manuscript with contributions from all co-authors. Authors applied the FLAE approach for the sequence of authors.
-
Conflict of interest: Authors state no conflict of interest.
References
[1] DeLuca TH, Aplet GH, Wilmer Bo, Burchfield J. The unknown trajectory of forest restoration: A call for ecosystem monitoring. J Forestry. 2010;108(6):288–95.Search in Google Scholar
[2] Chaves RB, Durigan G, Brancalion PHS, Aronson J. On the need of legal frameworks for assessing restoration projects success: New perspectives from são paulo state (Brazil). Restor Ecol. 2015;23(6):754–9.10.1111/rec.12267Search in Google Scholar
[3] McDonald T, Gann G, Jonson J, Dixon K. International standards for the practice of ecological restoration–including principles and key concepts. Technical report. Washington, DC, USA: Society for Ecological Restoration; 2016. http://www.seraustralasia.com/wheel/image/SER_International_Standards.pdf, Accessed on 2019-08-09.10.1111/rec.12359Search in Google Scholar
[4] Viani RAG, Holl KD, Padovezi A, Strassburg BBN, Farah FT, Garcia LC, et al. Protocol for monitoring tropical forest restoration: Perspectives from the atlantic forest restoration pact in brazil. Tropical Conserv Sci. 2017;10:1940082917697265.10.1177/1940082917697265Search in Google Scholar
[5] Freitas MG, Rodrigues SB, Campos-Filho EM, do Carmo GHP, da Veiga JM, Junqueira RGP, et al. Evaluating the success of direct seeding for tropical forest restoration over ten years. For Ecol Manag. 2019;438:224–32.10.1016/j.foreco.2019.02.024Search in Google Scholar
[6] Zahawi RA, Dandois JP, Holl KD, Nadwodny D, Reid JL, Ellis EC. Using lightweight unmanned aerial vehicles to monitor tropical forest recovery. Biol Conserv. 2015;186:287–95.10.1016/j.biocon.2015.03.031Search in Google Scholar
[7] Albuquerque RW, Ferreira ME, Olsen SI, Tymus JRC, Balieiro CP, Mansur H, et al. Forest restoration monitoring protocol with a low-cost remotely piloted aircraft: Lessons learned from a case study in the brazilian atlantic forest. Remote Sens. 2021;13(12):2401. 10.3390/rs13122401. URL https://www.mdpi.com/2072-4292/13/12/2401. ISSN 2072-4292.Search in Google Scholar
[8] Berni JAJ, Zarco-Tejada PJ, Suárez L, Fereres. E. Thermal and narrowband multispectral remote sensing for vegetation monitoring from an unmanned aerial vehicle. IEEE Trans Geosci Remote Sens. 2009;47(3):722–38.10.1109/TGRS.2008.2010457Search in Google Scholar
[9] d’Oleire Oltmanns S, Marzolff I, Peter K, Ries J. Unmanned aerial vehicle (UAV) for monitoring soil erosion in morocco. Remote Sens. 2012;4(11):3390–3416.10.3390/rs4113390Search in Google Scholar
[10] Aasen H, Burkart A, Bolten A, Bareth G. Generating 3D hyperspectral information with lightweight UAV snapshot cameras for vegetation monitoring: From camera calibration to quality assurance. ISPRS J Photogramm Remote Sens. 2015;108(10):245–59.10.1016/j.isprsjprs.2015.08.002Search in Google Scholar
[11] Almeida DRA, Broadbent EN, Zambrano AMA, Wilkinson BE, Ferreira ME, Chazdon R, et al. Monitoring the structure of forest restoration plantations with a drone-lidar system. Int J Appl Earth Obs Geoinf. 2019;79:192–8. 10.1016/j.jag.2019.03.014. URL https://www.sciencedirect.com/science/article/pii/S0303243418311954 ISSN 0303-2434.Search in Google Scholar
[12] Chen S, McDermid G, Castilla G, Linke J. Measuring vegetation height in linear disturbances in the boreal forest with UAV photogrammetry. Remote Sens. 2017;9(12):1257.10.3390/rs9121257Search in Google Scholar
[13] Wu X, Shen X, Cao L, Wang G, Cao F. Assessment of individual tree detection and canopy cover estimation using unmanned aerial vehicle based light detection and ranging (UAV LIDAR) data in planted forests. Remote Sens. 2019;11(8):908.10.3390/rs11080908Search in Google Scholar
[14] Belmonte A, Sankey T, Biederman JA, Bradford J, Goetz SJ, Kolb T, et al. UAV‐derived estimates of forest structure to inform ponderosa pine forest restoration. Remote Sens Ecol Conserv. 2020;6(2):181–97.10.1002/rse2.137Search in Google Scholar
[15] Mohan M, Leite RV, Broadbent EN, Jaafar WSWM, Srinivasan S, Bajaj S, et al. Individual tree detection using UAV-LIDAR and UAV-SfM data: A tutorial for beginners. Open Geosci. 2021;13(1):1028–39. 10.1515/geo-2020-0290.Search in Google Scholar
[16] Reis BP, Martins SV, Filho EIF, Sarcinelli TS, Gleriani JM, Leite HG, et al. Forest restoration monitoring through digital processing of high resolution images. Ecol Eng. 2019;127:178–86.10.1016/j.ecoleng.2018.11.022Search in Google Scholar
[17] Colomina I, Molina P. Unmanned aerial systems for photogrammetry and remote sensing: A review. ISPRS J Photogramm Remote Sens. 2014;92:79–97.10.1016/j.isprsjprs.2014.02.013Search in Google Scholar
[18] Schonberger JL, Frahm J-M. Structure-from motion revisited. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, NV, USA: Computer Vision Foundation (CVF); 2016. p. 4104–13.10.1109/CVPR.2016.445Search in Google Scholar
[19] DJI Phantom 4pro; 2022. https://www.dji.com/br/phantom-4-pro. Accessed 2022-01-12.Search in Google Scholar
[20] DRONESMADEEASY. Map pilot for DJI; 2020. URL https://support.dronesmadeeasy.com/hc/en-us/categories/200739936-Map-Pilot-for-iOS.Search in Google Scholar
[21] AGISOFT. Discover intelligent photogrammetry with metashape; 2020. URL https://www.agisoft.com/.Search in Google Scholar
[22] QGIS Development Team. QGIS geographic information system. Technical report; 2021. URL https://www.qgis.org Accessed on 2021-06-17.Search in Google Scholar
[23] Albuquerque RW, Costa MO, Ferreira ME, Jorge LAC, Sarracini LH, Rosa EO, et al. Qualitative effectiveness of unmanned aerial vehicles for monitoring forest restoration in Brazil: A brief review. Int J Curr Res. 2017;9(5):50802–6.Search in Google Scholar
[24] Hasan Shaheed M, Abidali A, Ahmed J, Ahmed S, Burba I, Fani PJ, et al. Flying by the sun only: The solarcopter prototype. Aerosp Sci Technol. 2015;45:209–14. ISSN 1270-9638 10.1016/j.ast.2015.05.016. URL https://www.sciencedirect.com/science/article/pii/S1270963815001637.Search in Google Scholar
[25] Albuquerque RW, Costa MO, Ferreira ME, Carrero GC, Grohmann CH. Remotely piloted aircraft imagery for automatic tree counting in forest restoration areas: A case study in the Amazon. J Unmanned Veh Syst. 2020;8(3):207–23.10.1139/juvs-2019-0024Search in Google Scholar
[26] Júnior LRA, Ferreira ME, Côrtes JBR, de Castro Jorge LA. High accuracy mapping with cartographic assessment for a fixed-wing remotely piloted aircraft system. J Appl Remote Sens. 2018;12(1):014003.Search in Google Scholar
[27] Ferrer-González E, Agüera-Vega F, Carvajal-Ramírez F, Martínez-Carricondo P. UAV photogrammetry accuracy assessment for corridor mapping based on the number and distribution of ground control points. Remote Sens. 2020;12(15):2447.10.3390/rs12152447Search in Google Scholar
[28] Nie Z, Liu F, Gao Y. Real-time precise point positioning with a low-cost dual-frequency GNSS device. GPS Solut. 2020;24(1):1–11.10.1007/s10291-019-0922-3Search in Google Scholar
[29] Liu H, Dong P. A new method for generating canopy height models from discrete-return LIDAR point clouds. Remote Sens Lett. 2014;5(6):575–82.10.1080/2150704X.2014.938180Search in Google Scholar
[30] Morrison EB, Lindell CA. Active or passive forest restoration? assessing restoration alternatives with avian foraging behavior. Restor Ecol. 2011;19(201):170–7.10.1111/j.1526-100X.2010.00725.xSearch in Google Scholar
[31] Brancalion PHS, Schweizer D, Gaudare U, Mangueira JR, Lamonato F, Farah FT, et al. Balancing economic costs and ecological outcomes of passive and active restoration in agricultural landscapes: The case of Brazil. Biotropica. 2016;48(6):856–67.10.1111/btp.12383Search in Google Scholar
[32] ANAC. Agência Nacional de Aviação Civil. Requisitos gerais para aeronaves não tripuladas de uso civil. Resolução número 419, de 2 de maio de 2017. Regulamento Brasileiro da Aviação Civil Especial, RBAC-E número 94; 2017. https://www.anac. gov.br/assuntos/legislacao/legislacao-1/rbha-e-rbac/rbac/rbac-e-94/@@display-file/arquivo_norma/RBACE94EMD00.pdf. Accessed on 2021-06-17.Search in Google Scholar
[33] Stehman SV, Wickham JD. Pixels, blocks of pixels, and polygons: Choosing a spatial unit for thematic accuracy assessment. Remote Sens Environ. 2011;115(12):3044–55. ISSN 0034-4257. 10.1016/j.rse.2011.06.007. URL https://www.sciencedirect.com/science/article/pii/S0034425711002318.Search in Google Scholar
[34] Blaschke T. Object based image analysis for remote sensing. IS-PRS J Photogramm Remote Sens. 2010;65(1):2–16.10.1016/j.isprsjprs.2009.06.004Search in Google Scholar
[35] Yurtseven H, Akgul M, Coban S, Gulci S. Determination and accuracy analysis of individual tree crown parameters using UAV based imagery and OBIA techniques. Measurement. 2019;145:651–64.10.1016/j.measurement.2019.05.092Search in Google Scholar
[36] Albuquerque RW, Vieira DLM, Ferreira ME, Soares LP, Olsen SI, Araujo LS, et al. Mapping key indicators of forest restoration in the amazon using a low-cost drone and artificial intelligence. Remote Sens. 2022;14(4):830. ISSN 2072-4292. 10.3390/rs14040830. URL https://www.mdpi.com/2072-4292/14/4/830.Search in Google Scholar
[37] Lyons MB, Keith DA, Phinn SR, Mason TJ, Elith J. A comparison of resampling methods for remote sensing classification and accuracy assessment. Remote Sens Environ. 2018;208:145–53. ISSN 0034-4257. 10.1016/j.rse.2018.02.026. URL https://www.sciencedirect.com/science/article/pii/S0034425718300324.Search in Google Scholar
[38] Congalton RG. A review of assessing the accuracy of classifications of remotely sensed data. Remote Sens Environ. 1991;37(1):35–46.10.1016/0034-4257(91)90048-BSearch in Google Scholar
[39] Goutte C, Gaussier E. A probabilistic interpretation of precision, recall and f-score, with implication for evaluation. European Conference on Information Retrieval. Heidelberg: Springer; 2005. p. 345–59.10.1007/978-3-540-31865-1_25Search in Google Scholar
[40] Resop JP, Lehmann L, Hession. WC. Drone laser scanning for modeling riverscape topography and vegetation: Comparison with traditional aerial LIDAR. Drones. 2019;3(2):35. 10.3390/drones3020035. ISSN 2504446X. URL https://www.mdpi.com/2504-446X/3/2/35.Search in Google Scholar
[41] Wang C, Morgan G, Hodgson ME. sUAS for 3D tree surveying: Comparative experiments on a closed-canopy earthen dam. Forests. 2021;12(6):659. 10.3390/f12060659. ISSN 1999-4907. URL https://www.mdpi.com/1999-4907/12/6/659.Search in Google Scholar
[42] Miller Z, Hupy J, Hubbard S, Shao G. Precise quantification of land cover before and after planned disturbance events with UAS-derived imagery. Drones. 2022;6(2):52. 10.3390/drones6020052. ISSN 2504-446X. URL https://www.mdpi.com/2504-446X/6/2/52.Search in Google Scholar
[43] Reis BP, Martins SV, Filho EIF, Sarcinelli TS, Gleriani JM, Marcatti GE, et al. Management recommendation generation for areas under forest restoration process through images obtained by UAV and LIDAR. Remote Sens. 2019;11(13):1508.10.3390/rs11131508Search in Google Scholar
[44] Belgiu M, Drǎguţ L. Comparing supervised and unsupervised multiresolution segmentation approaches for extracting buildings from very high resolution imagery. ISPRS J Photogramm Remote Sens. 2014;96:67–75.10.1016/j.isprsjprs.2014.07.002Search in Google Scholar PubMed PubMed Central
[45] Duarte L, Silva P, Teodoro A. Development of a QGIS plugin to obtain parameters and elements of plantation trees and vineyards with aerial photographs. ISPRS Int J Geo Inf. 2018;7(3):109.10.3390/ijgi7030109Search in Google Scholar
[46] Venator M, Aklanoglu S, Bruns E, Maier A. Enhancing collaborative road scene reconstruction with unsupervised domain alignment. Mach Vis Appl. 2021;32(1):1–16.10.1007/s00138-020-01144-8Search in Google Scholar
[47] Zhang J, Liu J, Pan B, Shi Z. Domain adaptation based on correlation subspace dynamic distribution alignment for remote sensing image scene classification. IEEE Trans Geosci Remote Sens. 2020;58(11):7920–30. 10.1109/TGRS.2020.2985072.Search in Google Scholar
[48] Powell A. Democratizing production through open source knowledge: From open software to open hardware. Media Cult Soc. 2012;34(6):691–708.10.1177/0163443712449497Search in Google Scholar
[49] ODM. Opendronemap a command line toolkit to generate maps, point clouds, 3d models and dems from drone, balloon or kite images; 2022. https://www.opendronemap.org/. Accessed: 2022-01-12.Search in Google Scholar
[50] Damasceno G, Souza L, Pivello VR, Gorgone-Barbosa E, Giroldo PZ, Fidelis A. Impact of invasive grasses on cerrado under natural regeneration. Biol Invasions. 2018;20(12):3621–9.10.1007/s10530-018-1800-6Search in Google Scholar
[51] Swayze NC, Tinkham WT, Vogeler JC, Hudak AT. Influence of flight parameters on UAS-based monitoring of tree height, diameter, and density. Remote Sens Environ. 2021;263:112540.10.1016/j.rse.2021.112540Search in Google Scholar
[52] Getzin S, Wiegand K, Schöning I. Assessing biodiversity in forests using very high-resolution images and unmanned aerial vehicles. Methods Ecol Evol. 2012;3(2):397–404.10.1111/j.2041-210X.2011.00158.xSearch in Google Scholar
[53] Martins ACM, Willig MR, Presley SJ, Marinho-Filho J. Effects of forest height and vertical complexity on abundance and biodiversity of bats in Amazonia. For Ecol Manag. 2017;391:427–35.10.1016/j.foreco.2017.02.039Search in Google Scholar
[54] Krause S, Sanders TGM, Mund J-P, Greve. K. UAV-based photogrammetric tree height measurement for intensive forest monitoring. Remote Sens. 2019;11(7):758.10.3390/rs11070758Search in Google Scholar
[55] Ferreira J, Lennox GD, Gardner TA, Thomson JR, Berenguer E, Lees AC, et al. Carbon-focused conservation may fail to protect the most biodiverse tropical forests. Nat Clim Change. 2018;8(8):744–9.10.1038/s41558-018-0225-7Search in Google Scholar
© 2022 Rafael Walter Albuquerque et al., published by De Gruyter
This work is licensed under the Creative Commons Attribution 4.0 International License.
