Science of the Total Environment 657 (2019) 1602–1614 Contents lists available at ScienceDirect Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv Monitoring opencast mine restorations using Unmanned Aerial System (UAS) imagery Joan-Cristian Padró a,⁎, Vicenç Carabassa b, Jaume Balagué c, Lluís Brotons d, Josep M. Alcañiz e, Xavier Pons f a Grumets Research Group, Departament de Geografia, Universitat Autònoma de Barcelona, Departament de Geografia Office B1092, Edifici B, Universitat Autònoma de Barcelona, 08193 Bellaterra, Catalonia, Spain b Centre de Recerca Ecològica i Aplicacions Forestals (CREAF), Edifici C, Campus UAB, 08193 Bellaterra, Catalonia, Spain c Centre Tecnològic Forestal de Catalunya (CTFC), Centre de Recerca Ecològica i Aplicacions Forestals (CREAF), Ctra Sant Llorenç km 2, 24280 Solsona, Catalonia, Spain d Centre Tecnològic Forestal de Catalunya (CTFC), Centre de Recerca Ecològica i Aplicacions Forestals (CREAF), Consejo Superior de Investigaciones Científicas (CSIC), Ctra Sant Llorenç km 2, 24280 Solsona, Catalonia, Spain e Universitat Autònoma de Barcelona, Centre de Recerca Ecològica i Aplicacions Forestals (CREAF), Edifici C, Universitat Autònoma de Barcelona, 08193 Bellaterra, Catalonia, Spain f Grumets Research Group, Departament de Geografia, Universitat Autònoma de Barcelona, Departament de Geografia Office B1094, Edifici B, Universitat Autònoma de Barcelona, 08193 Bellaterra, Catalonia, Spain H I G H L I G H T S G R A P H I C A L A B S T R A C T • Fast, repeatable and systematic moni- toring at sub-decimeter spatial resolu- tion • R2 between drone imagery and field spectroradiometric measurements over 0.9 • Multispectral imagery allows monitor- ing mine wastes, restored soil and vege- tation. • A solution for easy detection of failing or successful restoration actions • Affordable equipment and protocol for rigorous and non-conflictive mine monitoring a r t i c l e i n f o a b s t r a c t Article history: Open-pit mine is still an unavoidable activity but can become unsustainable without the restoration of degraded Received 15 June 2018 sites. Monitoring the restoration after extractive activities is a legal requirement for mine companies and public Received in revised form 19 November 2018 administrations in many countries, involving financial provisions for environmental liabilities. The objective of Accepted 10 December 2018 this contribution is to present a rigorous, low-cost and easy-to-use application of Unmanned Aerial Systems Available online 13 December 2018 (UAS) for supporting opencast mining and restoration monitoring, complementing the inspections with very Keywords: high (b10 cm) spatial resolution multispectral imagery, and improving any restoration documentation with de- Mine restoration tailed land cover maps. The potential of UAS as a tool to control restoration works is presented in a calcareous Monitoring restoration quarry that has undergone different post-mining restoration actions in the last 20 years, representing 4 reclaimed Abbreviations: AGL, above ground level; ASPRS, American Society for Photogrammetry and Remote Sensing; DCSM, Digital Cast Shadow Model; DEM, Digital Elevation Model; DIM, Digital Illumination Model; DLS, Downwelling Light Sensor; DSM, Digital Surface Model; DTM, Digital Terrain Model; EVA, Ethylene-Vinyl Acetate; EXIF, Exchangeable Image File Format; FWHM, Full Width at Half Maximum; GCP, Ground Control Point; GIS, Geographical Information System; GNSS, Global Navigation Satellite System; GPS, Global Positioning System; GRE, green (spectral region); kNN, k-Nearest Neighbor; LDCM, Landsat Data Continuity Mission; lidar, Light Detection and Ranging; MSAVI, Modified Soil Adjusted Vegetation Index; NDVI, Normalized Difference Vegetation Index; NDWI, Normalized Difference Water Index; NIR, near-infrared (spectral region); RED, red (spectral region); REF, reference; REG, red edge (spectral region); ROI, Region Of Interest; RS, remote sensing; SAVI, Soil Adjusted Vegetation Index; TA, Training Area; TIN, Triangulated Irregular Network; TRA, Transit Area; UAS, Unmanned Aerial System; UAV, Unmanned Aerial Vehicle; VNIR, visible and near-infrared (spectral region); VA, Validation Area. ⁎ Corresponding author. E-mail addresses: JoanCristian.Padro@uab.cat (J.-C. Padró), V.Carabassa@creaf.uab.cat (V. Carabassa), Jaume.Balague@ctfc.cat (J. Balagué), Lluis.Brotons@ctfc.cat (L. Brotons), JoseMaria.Alcaniz@uab.cat (J.M. Alcañiz), Xavier.Pons@uab.cat (X. Pons). https://doi.org/10.1016/j.scitotenv.2018.12.156 0048-9697/© 2018 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). J.-C. Padró et al. / Science of the Total Environment 657 (2019) 1602–1614 1603 Drone stages. We used a small (b2 kg) drone equipped with a multispectral sensor, along with field spectroradiometer Unmanned Aerial System measurements that were used to radiometrically correct the UAS sensor data. Imagery was processed with pho- UAS togrammetric and Remote Sensing and Geographical Information Systems software, resulting in spectral infor- UAV remote sensing mation, vegetation and soil indices, structural information and land cover maps. Spectral data and land cover Multispectral imagery classification, which were validated through ground-truth plots, aided in the detection and quantification of mine waste dumping, bare soil and other land cover extension. Moreover, plant formations and vegetation devel- opment were evaluated, allowing a quantitative, but at the same time visual and intuitive comparison with the surrounding reference systems. The protocol resulting from this research constitutes a pipeline solution intended for the implementation by public administrations and privates companies for precisely evaluating restoration dy- namics in an expedient manner at a very affordable budget. Furthermore, the proposed solution prevents subjec- tive interpretations by providing objective data, which integrate new technologies at the service of scientists, environmental managers and decision makers. © 2018 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). 1. Introduction to 10 m) is not detailed enough to monitor slight land cover changes in relatively small areas of interest, such as some opencast mine restora- Severe ecosystem changes are the result of opencast mining activi- tions, despite some studies have been done combining Landsat imagery ties, which affect the soil, vegetation, fauna, hydrology and landforms and in-situ monitoring (e.g. Bonifazi et al., 2003). Alternatively, a much (Drake et al., 2010). The restoration of degraded soils and human- more detailed spatial and spectral resolution can be reached with air- disrupted landscapes involves a large number of priorities (Ockendon borne optical sensors, such as the Airborne Hyperspectral Scanner, et al., 2018). Among them, post-mining restoration represents the spe- which features 80 bands from visible to thermal regions, or the Compact cific challenge of creating ecological structures that become integrated Airborne Spectrographic Imager 1500, which features 288 bands be- into the surrounding environment, which starts from “point zero” tween 0.35 μm and 1050 μm, allowing sub-metric pixels at a standard (Hüttl and Weber, 2001). Mine wastes and rocky debris are common flight height of 3000 m above ground level (AGL). Light Detection and initial substrate conditions, being necessary artificially replacing soil Ranging (lidar) sensors can be effectively used to reconstruct the vege- (i.e. technosol) and introducing local vegetation by means of planting, tation structure (Listopad et al., 2015) and classify land cover surfaces in sowing and encroaching the surrounding vegetation to drive the resto- combination with optical imagery, but the cost of taking off a plane with ration process to pre-mining landscape conditions (Korjus et al., 2014). this expensive payload solely for the purpose of monitoring a mine res- Ecological engineering scientists assume that after intensive opencast toration is ultimately not cost-effective. Moreover, satellite and airborne mining, the final soil conditions are usually extremely degraded thus optical imagery is often constrained by cloud cover conditions. On the leading the restoration process to either fail or simply not begin other hand, such conditions do not pose a problem for lidar or optical (Bradshaw, 1997). Consequently, the task of monitoring reclaimed sensors onboard Unmanned Aerial Systems (UAS or drones) when fly- areas becomes a fundamental feature in any restoration process, as it ing below the cloud cover thanks to their high sensitivity even when provides crucial information to environmental managers for those in- only diffuse irradiance exists. It is important to note that in such condi- volved in assessing the success of adopted techniques (DITR, 2016). tions, reflectance is much more Lambertian than under direct sunlight, The objective of a restoration is to achieve a self-sustainable ecosystem and incoming diffuse irradiance can be measured in-situ at the time of at the overall affected landscape (Elmqvist et al., 2003). Restoration of the drone flight. A radiometric analysis of multispectral photogrammet- opencast mines is regulated in most developed countries, and expected ric imagery acquired in different illumination conditions can be found in to be also in the remaining ones. For instance, in the European Union we Markelin (2013). Then, the use of well-known remote sensing satellite can find different levels of legislation, the European (Directive 2006/21/ and aerial techniques, which can be adapted to the imagery provided EC), member states (e.g. Spanish RD975/2009, German by multispectral optical sensors onboard low-weight drones (Sanders, Bundesberggesetz 1980) and regional governments (e.g. Catalan Law 2017), can considerably facilitate the task of covering an area dimen- 12/1981), all of which require monitoring and periodically inspection sioned to the platform, in that a much higher spatial resolution can be of mine restorations. Moreover, extractive companies are required to obtained with a low-cost revisit time and is not constricted to cloud provide a financial guarantee for environmental liabilities, which are cover. With all these advantages, such a solution can prove more suit- blocked by the administration to assure that the companies comply able for the operative goal of acquiring RS data in monitoring mine ex- with the regulations ensuring the proper restoration of the mined ploitations and their restoration. areas. Altogether, not only public administrations are interested in up- Although the regulation of b25 kg UAS may vary in each country, in holding legal measures for land restoration, also private companies most of them there are no major restrictions on their use in non-urban want to recover the financial provision. areas, such as in opencast mine undergoing restorations; nevertheless Environmental characterization, mapping and monitoring they comply with the regulations on flight height limitation and observ- (Thenkabail, 2015) is one of the first and most productive applications ing the security distance to airports, as recommended by the Interna- of remote sensing (RS). The scientific use of environmental remote tional Civil Aviation Organization (ICAO, 2011). Regulations aside, the sensing data has been evolving since the 1970s, when the first Landsat proliferation of low-cost and low-weight UAS as data acquisition tools satellite was put into orbit. The Landsat Data Continuity Mission is rapidly promoting the transference of knowledge from the research (LDCM) (NASA, 2018), which has been the standard-bearer of Earth re- sphere to the productive economy, and vice versa. Many studies make sources and environmental monitoring from space, uses sensors in the use of multispectral optical sensors onboard low-weight (b25 kg) solar [0.450 μm to 2.500 μm] and thermal [8 μm to 13 μm] regions of drones and apply remote sensing techniques for wide-ranging applica- the electromagnetic spectrum. The recent Sentinel-2 mission (ESA, tions, such as the 3D reconstruction of cultural heritage sites (Eisenbeiss 2015), which is represented by the Sentinel-2A and B platforms, is im- and Zhang, 2006), identifying and evaluating climate-related diseases proving the spatial, spectral and temporal resolution of the free avail- (Yoo et al., 2017), precision agriculture (Díaz-Varela et al., 2015), topog- able satellite acquisitions. The joint synergy of those Earth observation raphy (Lucieer et al., 2014) or wildfire severity (Pla et al., 2017). Alto- programs is currently coordinated in the Global Earth Observation Sys- gether these studies focus intensely on space and time due to the tem of Systems (GEOSS). However, the Sentinel-2 spatial resolution (up nature of the UAS platforms (fixed-wing or rotary-wing), which can 1604 J.-C. Padró et al. / Science of the Total Environment 657 (2019) 1602–1614 acquire images with ultra-high spatial resolution, usually from 1 cm to (Honkavaara et al., 2012; Iqbal et al., 2018) is the numerical fitting (usu- 10 cm depending on the flight height and the sensor instantaneous ally an empirical line through reference targets) of UAS-acquired radi- field of view. The extraction activities sector is also very interested in ometry and field spectroscopy data. Additionally, there are consumer UAS remote sensing applications. When dealing with opencast mining grade cameras with a Downwelling Light Sensor (DLS) that measures cases, we find that extractive companies are interested in 3D quantifica- the total incoming irradiance at-sensor level, which stores this informa- tion of mineral resource volumes (Xu et al., 2015) or the geomorphic tion in the EXIF file of each photogram and obtaining direct information characterization (Chen et al., 2015), while the environmental commu- to calculate at-sensor reflectance. Nevertheless, as Hakala et al. (2018) nity is mainly focused on monitoring water and soil pollution point out, the challenge in this method is that the upwards looking sen- (Capolupo et al., 2015). To the best known of the authors, there are no sor has to be horizontally stabilized, which is typically not the case applications of UAS imagery specifically designed for the “standardized” when using low-cost instruments. monitoring of opencast mine restorations, a legal issue pointed out in The main objective of this research is to provide a new tool using much of the environmental legislations and that involves public admin- UAS that generates detailed, objective, exhaustive and rigorous cartog- istrations, companies, technicians and wastes time and economic re- raphy of opencast mine restorations. Introducing of UAS for acquiring sources (Minerals Council of Australia, 2015). The interaction of these imagery as a source to obtain data and produce digital maps is expected actors is susceptible to involve subjective appreciations of a given resto- to improve the diligence and objectiveness of inspections, and as such ration, so we propose a workflow for acquiring highly detailed imagery could reduce disagreements between public administrations and pri- to produce thematic maps and complement on-field evaluation with vate companies. As a secondary goal, this research focuses on how, be- objective cartographical evidences. sides the intrinsic interest in the detailed imagery for visual The current geometric theoretical basis of UAS remote sensing finds inspections, UAS imagery can provide quantitative spatial information its roots in digital photogrammetry and computer vision background about restoration failures or success in terms of monitoring the area (Hartley and Zisserman, 2003; Granshaw and Fraser, 2015). Generally, covered by vegetation, mine waste dumping or bare soil at different res- multiple overlapped photograms are acquired and a bundle block ad- toration stages. justment is performed using aero-triangulation techniques. Homolo- gous points between adjacent photograms are automatically detected 2. Materials by using computer vision techniques, which are further used as tie points to fit the whole model, by forming a mosaic of the photograms 2.1. Study area and restoration stages and reconstructing the surface with the stereo photogrammetric infor- mation. There are two georeferencing methods mostly used in UAS im- The implementation for monitoring mine restorations using UAS agery: direct georeferencing and indirect georeferencing. Direct was carried out in Catalonia, Spain, where 483 mine exploitations georeferencing is based on integrated navigation systems, which uses were documented in 2018 (GENCAT, 2018). We have developed the Global Navigation Satellite Systems (GNSS), e.g. GPS receiver as well presented workflow on an active calcareous sandstone quarry located as Inertial Measurement Unit (IMU) information to locate the camera in central Catalonia (lat. 41° 41′ 35″N lon. 1° 49′ 43″E), with representa- perspective center and estimate its orientation parameters at the acqui- tive landforms corresponding to different stages of restoration, namely sition time (Skaloud et al., 2014; Rehak and Skaloud, 2017). In fact, the from old restorations carried out N20 years ago to recent plantations. quality and accuracy of the onboard GNSS and IMU are key factors for By accessing the historic orthophotographs of 1957, 1984, 2000 and proper direct georeferencing. In contrast, indirect georeferencing of 2016 (ICGC, 2018a, 2018b), we can report that the land use of this UAS imagery is based on Ground Control Points (GCP), scene locations area before the mining activity was dominated by vineyards (1957), with well-known coordinates (Rumpler et al., 2014). GCP are com- which were abandoned and spontaneously forested by the encroach- monly measured with static post-processed GNSS methods and ment of autochthonous species (1984). This is a good example of the surrounded with a chessboard target that is visible in the UAS imagery. transition forest theory (Mather and Needle, 1998) in Mediterranean Comparatively, indirect georeferencing is more time-consuming than environments (Padró and Badia, 2017; Poyatos et al., 2003). In this direct georeferencing, and the accuracy depends on the correct mea- area, mining began in the mid-1980's, and hit its activity peak on the surement of GCP positions, the technician's skill in locating the refer- 2000's decade. The restoration stages considered were, from the present ences in the imagery, and the number of GCPs, reaching pixel size to the past, T0 (0–5 years), T1 (5–10 years), T2 (10–15 years) and T3 accuracies (Turner et al., 2014). A recent comparison of four UAS (15–20 years); additionally, a non-disturbed area was taken as refer- georeferencing methods for environmental monitoring purposes can ence (REF), and a Transit Area (TRA) was masked to avoid a bias in be found in Padró et al. (2019). the quantification (Fig. 1). The radiometric theoretical basis of optical multispectral remote The material extracted was calcareous sandstone, exploited trough sensing in the visible and near infrared regions (VNIR), i.e. between 10 m × 10 m banks. Bank-berm was the common landform model, 400 nm and 900 nm, is attained by measuring the spectral surface re- resulting in steeped slopes. After a partial landform adaptation using flectance (ρλ) in order to be comparable when imagery is acquired on mining debris, the surface was covered by new topsoil. The topsoil different dates and/or by different sensors. Imagery is converted to ρλ was a technosol, which was constructed using mine spoils and mineral at pixel level; ρλ is computed as the amount of spectral radiance (Lλ) sludge from the tailing process and was mixed with a small amount of (W·m−2·sr−1·μm−1) reflected by a surface located in a given pixel previous topsoil stored for a long time. Organic amendments were with respect to the total spectral irradiance (Iλ) (W·m−2·μm−1) re- added in order to increase soil fertility. Specifically, anaerobically ceived (Schaepmann-Strub et al., 2006). Satellite imagery is radiometri- digested sewage sludge was incorporated in a dose around 50 Mg/ha cally corrected according to one of the different existing approaches in order to increase soil organic matter, P and N contents, and improve (Kaufman and Sendra, 1988; Chavez, 1988; Pons et al., 2014; Vermote soil aggregation. Revegetation measures consisted of sowing a mixture et al., 2016, as examples), and can be validated by using field of herbaceous seeds (mainly grasses and leguminous) and planting spectroradiometric measurements as ground-truth (Padró et al., local shrubs (mainly Rosmarinus officinalis) and trees (mainly Pinus 2017). Moreover, as the incoming irradiance is difficult to approximate halepensis), at a mean density of about 400 seedlings ha−1. Vegetation by modeling, it can be estimated empirically (Abdollahnejad et al., in restored areas presented different grades of development according 2018), and as the UAS sensor is usually located in a very low AGL alti- to the age of sowing and planting activities, which range from adult tude (b120 m) the radiative transfer modeling has to be adapted or trees and dense shrub cover in T3, to a dominant herbaceous cover in even substituted by other approaches (Aasen et al., 2015). A common T0. The reference system (REF, undisturbed area) is a Pinus halepensis approach for obtaining surface reflectance values from drone data forest, with a dense shrub layer and a low herbaceous cover. J.-C. Padró et al. / Science of the Total Environment 657 (2019) 1602–1614 1605 Fig. 1. Top left: Flight planned area and study area over the official local agency orthophotomap (ICGC, 2018a, 2018b). Bottom left: Restoration stages T0 (0–5 years), T1 (5–10 years), T2 (10–15 years) and T3 (15–20 years); non-altered area taken as reference (REF); Transit Area (TRA) masked to avoid bias in the quantification. Top right: Location maps. Bottom right: Restoration landscape. The total dimension of the study area measured at 64,386 m2, which Sequoia was tested with in-situ reference targets, and finally at-sensor comprised of T0 4364 m2, T1 12,627 m2, T2 16,638 m2, T3 18,561 m2, REF reflectance was numerically fitted to surface reflectance 8206 m2 and TRA 3990 m2. The planned flight (cf. Table S1 in the Sup- spectroradiometric measurements. The platform used was a DJI Phan- plementary materials) covered an additional area to assure the com- tom III Advanced (DJI, 2018) quadcopter (cf. Table S3 in the Supplemen- plete overlapping and preventing geometric distortions of the tary materials) due to its compatibility with Android and iOS systems, marginal photograms of the block. Previous flights were performed in fitness to the proposed procedures, low-cost, easy manageability, re- mining activities to define a suitable flight altitude (90 m AGL) that bal- duced dimensions, and capacity of embedding the multispectral sensor ances the need for covering a typical opencast area and using a high spa- that we used therein. The default optical sensor was substituted by the tial resolution product (b10 cm). These recommendations are in Parrot Sequoia, which was mounted on the gimbal (Fig. 2). accordance with other experiences made in land cover classification of UAS imagery at different spatial resolutions (Ahmed et al., 2017). 2.3. Field spectroradiometer It is worth noting that the season and the hour of the UAS data acqui- sition are crucial parameters to consider. After testing our method in all The in-situ ground-truth reflectance was measured with an four seasons and at different hours of the day, we suggest that for ac- OceanOptics USB2000+ portable field spectroradiometer quiring imagery, in spring and summer are the best periods due to the (OceanOptics, 2006) (Table S4 in the Supplementary materials). leaf presence of deciduous vegetation, and during the hours of solar af- USB2000+ widely covers the VNIR spectral range of the multispectral ternoon due to the high sun position, what minimizes shadows. UAS sensor commonly used. The physical magnitude measured with the spectroradiometer is the spectral surface reflectance (ρλ), as ex- 2.2. UAS sensor and platform plained in Pons et al. (2015) and Meroni and Colombo (2009). In order to account for the variability of measurements, 100 readings The optical sensor used on this study was a Parrot Sequoia (cf. were obtained for each measurement and the overall data was statisti- Table S2 in the Supplementary materials), which is a multispectral sen- cally analyzed (mean, median, standard deviation, etc.) to evaluate its sor light enough to be embedded in b2 kg maximum take-off weight coherence, whereby we synthesized 1 nm of spectral resolution signa- platforms. This sensor has four spectral bands (green (GRE), red ture of every given surface. (RED), red-edge (REG) and near infrared (NIR)), which are specifically We used a set of 24 matt Ethylene-Vinyl Acetate (EVA) rubber disposed to improve the detection of vegetation and its features. The (foam) artificial targets of different colors (violet, blue, sky blue, green, relative spectral response function (RSRF) is not provided by the manu- yellow, orange, pink, red, brown, grey and black) in order to have sev- facturer, and we assumed the Full Width Half Maximum (FWHM) pro- eral spectral reflectance references. A lower number of targets can be vided in the specification sheet, by guessing the shape of the RSRF; other used for an empirical line correction (e.g. using two, the targets should sensors providing detailed RSRF allow a more detailed integration (e.g. be dark and bright). When more than two targets are used, they should Padró et al., 2018). A detailed analysis of the Parrot Sequoia can be found properly cover the reflectance range, but avoiding reflectance spectra in Fernández-Guisuraga et al. (2018). The Parrot Sequoia includes a DLS intersections; if it is the case, a large number of spectra is recommended located at the top of the UAS, which registers the Sun total spectral irra- to be used (Richter and Schläpfer, 2016), and thus, we used a large num- diance at-sensor level and thus facilitates the automatic determination ber of targets. The EVA foam material presented a highly Lambertian be- of the at-sensor reflectance. The performance of the DLS of the Parrot havior and was useful in different campaigns. We worked with two 1606 J.-C. Padró et al. / Science of the Total Environment 657 (2019) 1602–1614 Fig. 2. UAS materials: (a) Parrot Sequoia multispectral sensor (b) Downwelling Light Sensor (DLS) (c) Nominal FWHM bandwidth (d) DJI Phantom III platform with the optical sensor equipment. sizes of artificial targets, a group of 14 units of 50 cm × 50 cm × 2 cm of each image to register the multiple bands (Jhan et al., 2017). The ori- (namely “EVA” in Fig. 3) and a group of 10 units of 40 cm × 60 cm × entation process was optimized by filtering out erroneous points and 0.2 cm (namely “foam” in Fig. 3). Those targets were big enough to be using GCPs that were obtained from the official reference points of clearly seen in the UAS images (9 cm pixel size) with an equivalent sur- this mining activity reported in the respective exploitation labor plan face of 30.9 pixels and 5.5 pixels × 5.5 pixels around the central pixel in cartography, which were used for indirectly georeferencing the bundle the squared targets, and with an equivalent surface of 29.6 pixels and block. The validation of the geometric accuracy was achieved using the 4.4 pixels × 6.7 pixels in the rectangular targets. This always allowed cross-validation method and complying with the planimetry and altim- one “pure” pixel (surrounded by 5 × 5 pixels) at the center in the etry standards of the American Society for Photogrammetry and Remote squared targets, and, consequently, they were the preferred ones to re- Sensing (ASPRS) for digital orthoimagery (ASPRS, 2014) (cf. Table S5 in late the drone sensor values and the in-situ (spectroradiometer) values the Supplementary materials). This process generated the bundle block through fitting and testing. Rectangular targets were only used to com- adjustment with aero-triangulation, the individual images mosaicking plement the measurements since their pixels were not so symmetrically and the photogrammetric construction of a dense point cloud with X, distributed around the central pixel. Y and Z coordinates. The 3D point cloud allowed the generation of a Tri- Seven EVA targets were located in a shadow to experiment with dif- angulated Irregular Network (TIN), which was further rasterized in a fuse irradiation conditions and support the de-shadowing process, Digital Surface Model (DSM), and finally outputting a georeferenced while the remaining 17 targets were located under direct sunlight. To orthophotomosaic. Points classified as terrain were used to generate a this end, the spectral signature of each target was integrated at the Digital Elevation Model (DEM) and derive contour lines. DLS irradiance FWHM provided by the Parrot Sequoia manufacturer (Fig. 3). information stored in EXIF files was used to radiometrically correct each photogram, so as to obtain at-sensor reflectance values. The photogram- 3. Methods metric software output was composed by the orthophotomap, the DEM and the DSM, the elevation contour lines from the DEM, and the dense 3.1. Use of the photogrammetric software point cloud of each model. We used Agisoft PhotoScan v.1.4.1 (Agisoft, 2018) photogrammetric 3.2. Use of the RS&GIS software software, which is widely used in current UAS projects (Kraaijenbrink et al., 2016). The software workflow initially locates the individual We used MiraMon (Pons, 2016) RS&GIS software to radiometrically frames with the geotagged photocenters provided by the navigation correct the at-sensor drone imagery, to generate the derived spectral GNSS, then uses automatic tie points to estimate more accurately the and structural variables, to perform the digital classification to obtain external coordinates (X,Y,Z) for every frame, and the orientation angles land cover maps, and to integrate drone imagery with official mapping (ω,φ,κ). The Parrot Sequoia is a frame multilens sensor that generates agencies cartography. four non-registered raw images in DN values, but the photogrammetric Shadow reduction is the first issue to be solved, processing sepa- software uses the internal geometric information stored in the EXIF file rately those pixels irradiated with diffuse radiance (shadowed pixels) Fig. 3. Field spectroradiometric materials: (a) Field spectroradiometric measurements on targets (b) testing shadow effects (c) spectral signatures of targets, overlapped to Sequoia sensor bandwidth. The “foam” and the “EVA” panels differed in the size and, as seen in the spectral signature, in the color. J.-C. Padró et al. / Science of the Total Environment 657 (2019) 1602–1614 1607 from those pixels also illuminated by direct Sun irradiance (non- Digital Aspect Model used in the terrain morphology characterization shadowed pixels) to recover information from as much shadowed and is capable of describing water runoff flows. A very valuable product pixels as possible (Movia et al., 2016). In our approach, the detailed is the height difference between the DSM and the DEM, which result in shadow detection was obtained combining empirical and physical providing structural information ranging from the bare soil (~0 m of dif- methods (Adeline et al., 2013). In order to limit the possible histogram ference) up to the tree formations (N2 m of difference). commissions within the geometrically candidate pixels, the joint condi- tions for selecting shadowed pixels was: i) under the histogram thresh- 3.3. Classification methods and accuracy evaluation old of the first quartile in the intensity image, in the REG band and in the NIR band; ii) under the shadow of the Digital Cast Shadow Model The automatic land cover classification was carried out using k- (DCSM) or Digital Illumination Model (DIM) [self-shadows]. Once Nearest Neighbor (kNN) classification algorithm (Cover and Hart, masked, the reduction was achieved with the invariant color model 1967) implemented as a parallelized 64-bit version in the MiraMon method (Adeline et al., 2013) taking advantage of radiometric values software. kNN is particularly robust when different spectral responses taken in pairs of targets located in shadowed and non-shadowed condi- are associated to a unique informational class (Vidal-Macua et al., 2017). tions simultaneously. Non-shadowed pixels and de-shadowed pixels Ground-truth plots with a known land cover played a key role at this were tagged and, finally, a 5 pixels × 5 pixels mean filter was applied step. Delimited areas were digitized over the orthophotomap, whereby in the boundary belt of de-shadowed areas to smooth the transition and half of them were used to train the classifiers (Training Areas, TA) (cf. Fig. S7 in Supplementary materials). Afterwards, at-sensor drone and the other half was set aside to validate the thematic accuracy (Val- data obtained with DLS was radiometrically corrected by applying the idation Areas, VA) (Fig. 4). empirical line technique supported in field spectroscopy measure- Six land cover categories in the legend of thematic areas were ments, so as to improve the fitting to target references and correcting considered: possible biases. When radiometrically corrected, imagery was used for calculating Mine wastes: Outcrops of bare mineral substrates, very poor in or- spectral variables to statistically discriminate land covers at pixel reso- ganic matter (b0.5%), and without vegetation, formerly not restored lution. A large quantity of spectral indices using VNIR information are soil (mine spoils dump). This category is well characterized spec- related in the specialized literature, which focused on detecting vegeta- trally in the restored area with a b1% standard deviation in in-situ tion (Xue and Su, 2017) and soil (Viscarra Rossel et al., 2016) features. In measurements (Fig. 5). order to distinguish and quantify the area occupied by mine spoil out- crops from replaced topsoil (technosol) and the considered vegetation Topsoil: Bare soil with any organic matter content (N0.5%) but with- strata (grassland, shrubland and tree cover), multiple referenced indices out vegetation, replaced former soil coming from restoration activi- were tested according to the availability of UAS sensor bands. Calculat- ties (technosol). This category was well characterized spectrally in ing the correlation matrix between all the tested indices and the spec- the restored area although a 2% standard deviation is observed tral bands showed that indices with less redundant information throughout the measurements (Fig. 5). included: Tree cover: Tree plant formations, typically forested zones of the lat- i) Soil Adjusted Vegetation Index (SAVI) (Huete, 1988) [1] and est stages in restored areas. Not only spectral features characterized Modified Soil Adjusted Vegetation Index 2 (MSAVI2) this category, but structural information was used to consider that (Richardson and Weigand, 1977) [2], which minimize the effects all vegetation pixels with a height AGL over 2 m corresponded to of the soil background in the vegetation detection, and is widely trees. used in soil erosion, drought and desertification analyses Shrublands: Plant formations of up to 2 m AGL, mainly bushes corre- [1] SAVI = (NIR − RED) 1.5 / (NIR + RED + 0.5) sponding to medium successional stages. Several species and forma- [2] MSAVI2 = 0.5 [2 NIR + 1 − √((2 NIR + 1)2 − 8 (NIR tions were included in this category, which also featured a wide − RED))] spectral signature variety (Fig. 5). Grasslands: Herbaceous plant formations with heights of up to 1 m, ii) Normalized Difference Water Index 1 (NDWI1) (McFeeters, mainly consisting of grasses and forbs species that correspond to 1996) [3] which was designed to delimit water bodies. However, early successional stages in restored areas. Several grassland forma- in the study area context, it enhances the detection of vegetation tions were included in this category, which featured a wide spectral and soil moisture signature variety (Fig. 5). [3] NDWI1 = (GRE − NIR) / (GRE + NIR) Remaining shadows: Shadows not corrected in the shadow reduc- tion process can be considered as missing data. This category is iii) Normalized Difference Vegetation Index (NDVI) (Rouse et al., scene-dependent (terrain morphology, solar zenith angle (θ) and 1974) [4], which is the most used index to discriminate vegeta- solar azimuth angle (ϕ)). tion and monitor its evolution in multi-temporal analysis [4] NDVI = (NIR − RED) / (NIR + RED). 4. Results UAS-acquired spectral information in green, red, red-edge and near- infrared wavelength regions resulted in generating orthophotomaps of Soil and vegetation structural features were important inputs to de- 9 cm pixel size that clearly showed mine wastes, topsoil, and vegetation scribe the categories we aimed to map. Thus Digital Terrain Models structures. Locations that were inaccessible to technicians could also be (DTM) provided information that was not in the spectral bands and in- observed and registered. A systematic multispectral measurement was dices. Note that DTM are digital models that represent any variable, a done over the entire area of interest, complementing the traditional DEM is a DTM where the variable is the terrain elevation, and a DSM transecting operations. is a DTM where the variable is the surface elevation (including vegeta- Radiometric correction resulted in a good fitting to ground-truth ra- tion, buildings, etc.). To this end, the DSM was the base for calculating diometric measurements. Although the coefficient of determination be- the DIM and the DCSM used in the shadow reduction process, while tween field data and at-sensor (Sequoia) reflectance data was over 0.9 the DEM was the base to calculate the Digital Slope Model and the in all bands, bias and gain were not good enough for highly accurate 1608 J.-C. Padró et al. / Science of the Total Environment 657 (2019) 1602–1614 Fig. 4. General workflow for land cover mapping from UAS data using the kNN classifier. applications. Establishing synergy between the conventional field spec- products as contour line vector maps, which, in combination with offi- troscopy measurements and novel UAS sensors, we finally applied the cial topographic maps could complement the terrain morphology mon- empirical line correction to at-sensor (DLS) reflectance values to fit itoring. Combining the information on sun position and the DSM, the image data to in-situ observed values in the seven targets located important landscape and ecological information was retrieved, as the in direct irradiance illumination conditions. However, we previously vegetation projected shadows and sun illumination rates. Moreover, tested the accuracy of the results by performing the empirical line the differential map between the DSM and the DEM provided relevant with three spectrally representative targets, including dark and bright information about the relative heights of structures, and thus, the vege- ones, and evaluating the results with the remaining four. The obtained tation height AGL. Firstly, in the classification step, the layer of the dif- bias and gain clearly provided a better fitting to ground truth than fitting ferential map between DSM and DEM (in combination with the using the at-sensor Sequoia reflectance data. Once this was clarified, and spectral information layers) was used to statistically contribute to given that the bias and gain values obtained using three or seven refer- distinguishing high vegetation from low vegetation; in this respect it ences provided similar model results, we finally applied the empirical is important to note that a pixel with a spectral response different to line correction using all targets (seven) in order to minimize any error vegetation can have a relative height but is note classified as vegetation. and obtain the more robust model. On the other hand, shadow reduc- Secondly, in the interpretation of the results step, those pixels classified tion facilitated the process of classifying previously shadowed pixels, al- as vegetation can be crossed with the differential map between DSM though the method was not perfect and some undetected shadows and DEM to determine the vegetation height (Fig. 8). remained (Fig. 6). Finally, selected digital information was used to generate a stack of Spectral information was combined to generate additional informa- nine statistical variables (spectral and structural) to characterize each tion and detect vegetation and soil properties (i.e. vegetation activity land cover. Four variables represented the spectral bands of the Parrot (NDVI), moisture content (NDWI1)). By visualizing false color combina- Sequoia (GRE, RED, REG and NIR radiometrically corrected), three vari- tions of drone images it was intuitive to find the presence of vegetation ables represented the vegetation indices (SAVI, NDWI1 and NDVI), one due to the high response of active vegetation in the NIR band. By map- variable represented the intensity part of the color space transformation ping spectral indices, we could focus on the spatial distribution of the and the remaining variable represented the AGL difference between the vegetation presence and activity (NDVI), the moisture content DSM and the DEM. (NDWI1) and the soil effects on vegetated areas (SAVI and MSAVI2), A total of 60,402 pixels were used as Training Areas for the digital which altogether facilitated in the process of quantifying vegetated classification process, which were selected by photointerpretation on areas and non-vegetated areas at 9 cm × 9 cm pixels (Fig. 7). delimited plots visible in the drone image. The combination of training Additionally, structural information about the terrain morphology data, statistical remote sensing variables and the kNN classification was constructed from the DEM and allowed simple topographic method, produced a land cover map where the categories of interest Fig. 5. Ground-truth data: Field plots of (a) mine-wastes soil (b) replaced topsoil (c) grasslands (d) shrublands (e) mine wastes and topsoil spectral signatures. J.-C. Padró et al. / Science of the Total Environment 657 (2019) 1602–1614 1609 Fig. 6. Radiometric correction of drone data. (a) Regression resulting from the empirical line radiometric correction using three reference targets. (b) Validation, using four independent targets, of the empirical line correction performed from (a) results. (c) Regression resulting from the empirical line correction using seven targets (regression finally used for the empirical line correction). (d) Detail of the shadow reduction result in imagery and land cover classification (cf. Fig. S7 in Supplementary materials for more details). were identified, and therefore individually monitored. A total of was carried out by considering all the restoration stages together or at 11,872 pixels were used as test areas to evaluate the thematic accuracy. each individual stage scale (Table 1). The overall accuracy was 94.71% and the kappa index was 0.92. The con- The land cover area occupied in each restoration zone was moni- fusion matrix showed a good accuracy when splitting mine wastes and tored not only in absolute units, but also comparing relative changes topsoil (N97%), tree cover was mapped with good results (N96%), and in the internal distribution of each stage. The analysis of the internal var- most of the confusion was found between grass and shrub vegetation, iation of land covers distribution was found as one of the most interest- due principally to the intrinsic structural and spectral diversity of ing results of the provided information. The study area presented these categories (cf. Table S6 in the Supplementary materials). different restoration stages and in all of them, the spatial particularities The thematic land cover mapping was principally useful for quanti- were mapped, which constitute a valuable product for environmental fying diverse plant covered area and its spatial distribution trend, which managers (Fig. 9). 1610 J.-C. Padró et al. / Science of the Total Environment 657 (2019) 1602–1614 Fig. 7. Spectral information from UAS data: Top: Vegetation indices calculated with the RS&GIS software. Bottom: False color orthophoto generated from near-infrared, red and green bands. The mapping products obtained were fully integrated into the actual monitoring restored mining areas. In addition to scientific interest, in- official cartography, which locally improves both the spatial resolution troducing UAS into the periodic inspections of mine restorations could and the temporal resolution. Therefore, the visual inspection of restora- be considered as an objective and effective solution for reducing dis- tion processes is greatly improved, as it provides crucial information in agreements between public administration and private companies, the event that a regularly updated official orthoimagery would not be and at the same time, contributing to monitoring the sustainable devel- feasible. If previous cartography is available, UAS products are useful opment of extractive mining activities. in detecting anomalies in the restoration process, and thus make it pos- Validation issues regarding radiometric corrections were solved in sible to focus on zones where vegetation development remains low synergy with in-situ spectroradiometric measurements, using inexpen- (Fig. 10). sive reference panels. In this study we used a large number of targets to test different spectral responses and for validation purposes. Neverthe- 5. Discussion less, a lower number of radiometric references can be used for this ap- plication (Richter and Schläpfer, 2016). Indeed, the empirical line This medium-sizes opencast mine restoration (b10 ha) was easily regression using seven targets and using three targets is very similar covered by a low-weight rotary-wing UAS. Fast multispectral informa- (Fig. 6), but in this study we used seven references to perform a more tion acquired from the UAS sensor is particularly interesting due to robust regression. The use of the Parrot Sequoia DLS did not provide sat- the spatial resolution (9 cm) and coverage (64,386 m2), which makes isfactory reflectance values, as demonstrated in the regressions be- it possible to provide a very detailed visual inspection of the entire re- tween image data and field spectroscopy data (Fig. 6). The coefficient stored area, thus making it a highly valuable tool for controlling and of determination was good (R2 N 0.9 in all bands) but the bias was higher Fig. 8. Structural information from UAS data: Top: Digital Terrain Models accounting for vegetation, soil elevation, sun illumination and topographic contour line product. Bottom: Differential height between Digital Surface Model and Digital Elevation Model, representing vegetation height of over 2 m. J.-C. Padró et al. / Science of the Total Environment 657 (2019) 1602–1614 1611 Table 1 economically robust transitions in environmental management in ac- Land cover area (m2) mapped by soil and vegetation typologies and restoration stages. Ab- cordance with the Sustainable Development Goals (United Nations, breviations: T0 = 0–5 years, T1 = 5–10 years, T2 = 10–15 years, T3 = 15–20 years, REF = Reference area, TRA = Transit Area, ROI = Region Of Interest. 2018). The categorical treatment of land surfaces led to the quantification of Surface cover/stage T0 T1 T2 T3 REF TRA Total the area covered by the different soil and vegetation types: not only Mine wastes 829 2536 4426 3608 233 3601 15,233 within the overall restoration range, but also specifically within the dif- Topsoil 28 240 241 603 130 51 1293 ferent restoration stages. Detecting and distinguishing topsoil and mine Tree cover 862 4700 6470 4932 5480 42 22,485 waste cover was carried out and yielded good results. This is important Shrubland 833 1895 2363 2977 1273 28 9370 Grassland 1802 3206 2962 6387 1039 265 15,661 when considering that topsoil and mine waste cover pose real concerns Remaining shadows 10 51 175 54 51 4 345 for identifying and interpreting restoration failures and risks (Espigares Total ROI 4364 12,627 16,638 18,561 8206 3990 64,386 et al., 2011). Indeed, on the one hand, bare soil cover could be an indica- tor of sowings and planting failures, and/or of poor soil quality, which increase the risk of erosion (Van der Knijff et al., 2000; Espigares et al., than 4% in visible bands and reached 7% in the red-edge band. These biases 2011). On the other hand, mine waste dumping or mine waste used as have been corrected using the empirical line correction by fitting the image soil substitute without a proper organic and/or mineral amendment, to the reflectance of reference targets (Padró et al., 2018). Nevertheless, heavily limits vegetation establishment and encroachment, supposing previous works that used more accurate DLS instruments have demon- an unacceptable situation in mine restoration (Ram et al., 2006; strated that the use of upwards looking irradiance sensors is a good system Jordán et al., 2008). Therefore, an expedient and precise quantification for converting images to at-sensor reflectance (Hakala et al., 2018; Markelin of these types of land cover is a useful tool for estimating erosion risks et al., 2018), but the challenge in the aforementioned method is ensuring or soil quality limitations. Regarding monitoring vegetation and soil at that the upwards looking sensor are consistently horizontally leveled. stages, grasslands were the dominant cover in T0 but some areas cov- Shadow reduction allowed the recovering of information from pixels that ered by mine wastes were still detected. Therefore, corrective measures would be lost in the land cover mapping, although a relatively few number are still imperative for improving vegetation encroachment (Jordán of pixels remained classified as shadows due to conservative masking deci- et al., 2008). In the restored areas from five and ten years ago (T1) tree sions, as pointed out in previous studies (Movia et al., 2016; Adeline et al., cover was relatively dominant in the landscape despite the same pro- 2013). portion of mine waste cover was detected. Restored areas from 10 and Band calculation using near infrared and visible information facili- 15 years ago (T2) occupied the most extensive tree cover area of the tates the quantification of vegetation activity in intensity maps, which four restoration scenarios considered, but also mine waste outcrops point out areas with high vegetation development and areas with veg- were the second land cover in this stage, which could be attributed to etation cover deficits. Using photogrammetric DSM and DEM provided failing restoration processes in problematic areas (steeped slopes, slope the essential basis creating derived terrain models that contributed in N 90%). The late stage restoration scenario (T3), in which restoration mea- obtaining valuable information by computing the sun illumination re- sures were carried out N15 years ago, showed a land cover distribution ceived in every pixel, the detailed terrain slope, or the vegetation height. far away from the reference landscape (REF) and the more recent stage It is worthy noting that spatial resolution of 9 cm is a fine systematic considered, and thus demonstrating the need of revising the classical meth- landscape sampling, which is enough to establish vegetation height odologies used for the restoration in this area. A plausible explanation of classes reflecting the dynamics of the vegetation cover from the initial this result can be attributed to less exigent need for restoration regulations phases to the more advanced stages of the restoration. Airborne lidar 20 years ago, whereby the usual practice at the time was the direct affores- data could typically offer spatial resolutions and samplings below 1 m, tation over the mine wastes with Aleppo pine (P. halepensis), without soil and rarely below 10 cm at low-cost. However, lidar sensors are cur- replacement, or over a very unfertile substrate. rently being tested in combination with UAS platforms that could A fixed-wing platform could be more suitable for much more exten- allow ultra-high spatial sampling. Perhaps in the future, the presented sive restorations and perhaps further improvements could be achieved protocol could be complemented with lidar sensors. However, for using multispectral sensors with more bands (e.g. MicaSense RedEdge) many, the high expense of using and maintaining lidar sensors can or hyperspectral sensors with narrower bands to enhance the spectral prove ultimately non-cost effective (Manfreda et al., 2018). As a more information. Regarding image geometry, low-cost single-frequency viable alternative, photogrammetric solutions, such as those presented GPS receivers do not allow direct georeferencing. As such, geometric ac- in this paper, offer more than reasonable answers to opencast mine res- curacy relies on the accuracy of delimitation stacks used as GCPs, which toration monitoring. Although the use of low-cost UAS imagery for mining companies are required to maintain by law, as well as the monitoring purposes is plenty operative in developed countries, we technician's skill to locate them in the imagery. A heavier multicopter point to developing countries, where often there are higher economi- typically offers more stability, while a fixed-wing platform typically of- cally constrains. In this land cover degradation and restoration study, fers much more autonomy and coverage, and a double-frequency we found a solution that can effectively contribute towards integrating carrier-phase GNSS provides much more direct georeferencing accu- scientific methods and technologies, while working towards racy, and a lidar sensor could provide structural information (Listopad et al., 2015), but this material is much more expensive. Ultimately, our goal was to design a protocol that is low-cost and capable of offering ac- ceptable results. The material is useful in different campaigns and costs altogether under 6000 €, which included the platform, the UAS sensor, the field spectroradiometer and the radiometric reference targets. 6. Conclusions Monitoring opencast mining restoration with UAS imagery has been proven operational and useful, as it provides expedient and accurate spatial and thematic information. Resulting products can systematically and recur- rently sample the restoration area of interest at sub-decimeter pixel size, which improves airborne imagery spatial resolution and allows monitoring Fig. 9. Land cover relative distribution within each restoration stage. in inaccessible, restricted or dangerous areas. In addition to the intrinsic 1612 J.-C. Padró et al. / Science of the Total Environment 657 (2019) 1602–1614 Fig. 10. Top: False color orthophotomap and land cover classification result integrated with official cartography. Bottom: Land cover classification of the study area. interest of the imagery for visual inspections, photogrammetric and RS&GIS Acknowledgments software treatment provided additional spatial information about vegeta- tion development and restoration failure or success, plant cover and height, This work was supported by the European Union as part of the mine waste dumping, bare soil cover, and geomorphic terrain features. ECOPOTENTIAL Project (H2020 641762-2 EC), by an Innovation grant With the relatively low-cost material used in this study i.e. low- from the Spanish Ministry of Science and Innovation (ACAPI, weight platform, multispectral sensor, low-cost radiometric reference CGL2015-69888-P (MINECO/FEDER-ERDF, EU)), and by the Catalan panels and field spectroradiometer, it is possible to generate rigorous Government (Department of Territory and Sustainability) under the cartographical documentation as orthophotomaps, Digital Elevation project “Research and innovation in the monitoring and process of ex- Models, Digital Surface Models, vegetation indices and thematic land tractive activities restoration”. GRUMETS Research Group is partially cover maps fully integrated to previous cartography and official data- supported by the Catalan Government under Grant (SGR2017-1690). bases. The inclusion of new unmanned aerial technologies gives us a Joan-Cristian Padró is a recipient of the FI-DGR scholarship grant glimpse of new horizons for monitoring restoration with inexpensive (2016B_00410). Xavier Pons is a recipient of the ICREA Academia Excel- materials and intensive samplings. lence in Research Grant (2016–2020). Not only did automatic land cover mapping allow us to dispense with the manual photointerpretation, it also provided accuracies of Author contributions over 94%, although individual focusing on problematic areas could also be improved manually, if needed. The overall workflow was highly au- C.P. and V.C. conceived and designed the experiments. J.B., V.C. and tomatized with batch processes. Nevertheless, it is convenient for the C.P., performed the field data experiments. C.P. and V.C. analyzed the specialized technician to verify the correct execution and help in impor- data. C.P., V.C. and X.P. wrote the paper. L.B., J.A. and X.P. supervised tant decisions, such as digitizing of the training and testing areas. More- the paper and the overall experiments. over, compared to what would have been carried out in a conventional inspection of the restored areas, the fieldwork time in our case study Conflicts of interest was reduced considerably, which altogether included geometric stacks maintenance, field spectroradiometric measurements, and UAS flight. The authors declare that there are no conflicts of interest. The To summarize, the time used for obtaining ground-truth data and the founding sponsors had no role in the design of the study; in the collec- UAS flight, 2 h is sufficient for inspecting a restored area of 10 ha, tion, analyses, or interpretation of data; in the writing of the manuscript, while data processing can be fully done by one person in two working or in the decision to publish the results. days. It is worth noting that while UAS can help and support the moni- toring of restoration activities, it should not by any means replace in- Appendix A. Supplementary data situ inspections or technician supervision. In addition to the interest of monitoring older restorations with an Supplementary data to this article can be found online at https://doi. unprecedented detail, current UAS remote sensing products will cer- org/10.1016/j.scitotenv.2018.12.156. tainly become a valuable precedent for future opencast mine monitor- ing studies all around the world, as it provides improved time-series References information. Considering the potential, the obtained results and the ob- jective information produced, the proposed protocol can be a suitable Aasen, H., Burkart, A., Bolten, A., Bareth, G., 2015. Generating 3D hyperspectral informa- tion with lightweight UAV snapshot cameras for vegetation monitoring: from camera tool for administrations, mining and consulting companies as well as calibration to quality assurance. ISPRS J. Photogramm. Remote Sens. 108, 245–259. for environmental scientists. https://doi.org/10.1016/j.isprsjprs.2015.08.002. J.-C. Padró et al. / Science of the Total Environment 657 (2019) 1602–1614 1613 Abdollahnejad, A., Panagiotidis, D., Surový, P., Ulbrichová, I., 2018. UAV capability to de- Huete, A.R., 1988. A soil-adjusted vegetation index (SAVI). Remote Sens. Environ. 25 (3), tect and interpret solar radiation as a potential replacement method to hemispherical 259–309. https://doi.org/10.1016/0034-4257(88)90106-X. photography. J. Remote Sens. 10 (3), 423–440. https://doi.org/10.3390/rs10030423. Hüttl, R.F., Weber, E., 2001. Forest ecosystem development in post-mining landscapes: a Adeline, K.R.M., Chen, M., Briottet, X., Pang, S.K., Paparoditis, N., 2013. Shadow detection case study of the Lusatian lignite district. Naturwissenschaften 88 (8), 322–329. in very high spatial resolution aerial images: a comparative study. ISPRS https://doi.org/10.1007/s001140100241. J. Photogramm. Remote Sens. 80, 21–38. https://doi.org/10.1016/j. ICAO (International Civil Aviation Organization), 2011. “Unmanned Aircraft Systems isprsjprs.2013.02.003. (UAS)”. CIR328 AN/190. Agisoft LLC, 2018. PhotoScan user manual, professional edition, version 1.4. http://www. ICGC (Institut Cartogràfic i Geològic de Catalunya), 2018a. Full 274–99 de l'Ortofoto de agisoft.com/pdf/photoscan-pro_1_4_en.pdf, Accessed date: 1 November 2018. Catalunya 1:2500 (OF–25C) (1:2500 Catalonia orthophoto Tile 274–99). . URL. Ahmed, O.S., Shemrock, A., Chabot, D., Dillon, C., Williams, G., Wasson, R., Franklin, S.E., http://www.icc.cat/vissir3/, Accessed date: 1 November 2018. 2017. Hierarchical land cover and vegetation classification using multispectral data ICGC (Institut Cartogràfic i Geològic de Catalunya), 2018b. VISSIR, Visor de avançat de acquired from an unmanned aerial vehicle. Int. J. Remote Sens. 38, 2037–2052. geoinformació de Catalunya (VISSIR, Advanced geoinformation viewer of Catalonia). https://doi.org/10.1080/01431161.2017.1294781. . URL. http://www.icc.cat/vissir3/, Accessed date: 1 November 2018. ASPRS (American Society for Photogrammetry and Remote Sensing), 2014. ASPRS Posi- Iqbal, F., Lucieer, A., Barry, K., 2018. Simplified radiometric calibration for UAS-mounted tional Accuracy Standards for Digital Geospatial Data. Edition 1, version 1.0.0. URL:. multispectral sensor. Eur. J. Remote Sens. 51 (1), 301–313. https://doi.org/10.1080/ http://www.asprs.org/a/society/divisions/pad/Accuracy/ASPRS_Positional_Accuracy_ 22797254.2018.1432293. Standards_for_Digital_Geospatial_Data_Edition1_V1_FinalDraftForPublication.docx, Jhan, J.P., Rau, J.Y., Haala, N., Cramer, M., 2017. Investigation of parallax issues for multi- Accessed date: 1 November 2018. lens multispectral camera band co-registration. Int. Arch. Photogramm. Remote. Bonifazi, G., Cutaia, L., Massacci, P., Roselli, I., 2003. Monitoring of abandoned quarries by Sens. Spat. Inf. Sci. XLII-2/W6, 157–163. https://doi.org/10.5194/isprs-archives-XLII- remote sensing and in situ surveying. Ecol. Model. 170, 213–218. https://doi.org/ 2-W6-157-2017. 10.1016/S0304-3800(03)00228-X. Jordán, M.M., Pina, S., García-Orenes, F., Almendro-Candel, M.B., García-Sánchez, E., 2008. Bradshaw, A., 1997. Restoration of mined lands - using natural processes. Ecol. Eng. 8, Environmental risk evaluation of the use of mine spoils and treated sewage sludge in 255–269. https://doi.org/10.1016/S0925-8574(97)00022-0. the ecological restoration of limestone quarries. Environ. Geol. 55, 453–462. https:// Capolupo, A., Pindozzi, S., Okello, C., Fiorentino, N., Boccia, L., 2015. Photogrammetry for doi.org/10.1007/s00254-007-0991-4. environmental monitoring: the use of drones and hydrological models for detection Kaufman, Y.J., Sendra, C., 1988. Algorithm for automatic atmospheric corrections to visible of soil contaminated by copper. Sci. Total Environ. 514, 298–306. https://doi.org/ and near-IR satellite imagery. Int. J. Remote Sens. 9, 1357–1381. https://doi.org/ 10.1016/j.scitotenv.2015.01.109. 10.1080/01431168808954942. Chavez Jr., P.S., 1988. An improved dark-object subtraction technique for atmospheric Korjus, H., Laarmann, D., Sims, A., Paluots, T., Kangur, A., 2014. Assessment of novel forest scattering correction of multispectral data. Remote Sens. Environ. 24, 459–479. ecosystems on post-mining restoration site in Aidu, Estonia. In: Polgár, A., Bazsó, T., https://doi.org/10.1016/0034-4257(88)90019-3. Nagy, G., Gálos, B. (Eds.), Local and Regional Challenges of Climate Change Adaptation Chen, J., Li, K., Chang, K.-J., Sofia, G., Tarolli, P., 2015. Open-pit mining geomorphic feature and Green Technologies. University of West Hungary Press, Sopron, pp. 35–44 characterization. Int. J. Appl. Earth Obs. Geoinf. 42, 76–86. https://doi.org/10.1016/j. https://doi.org/10.13140/2.1.4557.6640. jag.2015.05.001. Kraaijenbrink, P., Meijer, S.W., Shea, J.M., Pellicciotti, F., De Jong, S.M., Immerzeel, W.W., Cover, T., Hart, P., 1967. Nearest neighbor pattern classification. IEEE Trans. Inf. Theory 13 2016. Seasonal surface velocities of a Himalayan glacier derived by automated corre- (1), 21–27. https://doi.org/10.1109/TIT.1967.1053964. lation of unmanned aerial vehicle imagery. Ann. Glaciol. 57 (71), 103–113. https:// Díaz-Varela, R.A., De La Rosa, R., León, L., Zarco-Tejada, P.J., 2015. High-resolution airborne doi.org/10.3189/2016AoG71A072. UAV imagery to assess olive tree crown parameters using 3D photo reconstruction: Listopad, C., Masters, R., Drake, J., Weishampel, J., Branquinho, C., 2015. Structural diver- application in breeding trials. Remote Sens. 7 (4), 4213–4232. https://doi.org/ sity indices based on airborne LiDAR as ecological indicators for managing highly dy- 10.3390/rs70404213. namic landscapes. Ecol. Indic. 57, 268–279. https://doi.org/10.1016/j. DITR (Department of Industry, Tourism and Resources), 2016. Mine Rehabilitation. ecolind.2015.04.017. Australian Government 0-642-72481-4, p. 77 URL:. https://archive.industry.gov.au/ Lucieer, A., Turner, D., King, D.H., Robinson, S.A., 2014. Using an Unmanned Aerial Vehicle resource/Documents/LPSDP/LPSDP-MineRehabilitationHandbook.pdf, Accessed (UAV) to capture micro-topography of Antarctic moss beds. Int. J. Appl. Earth Obs. date: 1 November 2018. Geoinf. 27, 53–62. https://doi.org/10.1016/j.jag.2013.05.011. DJI, 2018. Phantom 3 Professional. URL. https://www.dji.com/es/phantom-3-pro/info, Manfreda, S., McCabe, M., Miller, P., Lucas, R., Pajuelo Madrigal, V., Mallinis, G., Ben-Dor, E., Accessed date: 1 November 2018. Helman, D., Estes, L., Ciraolo, G., Müllerová, J., Tauro, F., de Lima, M.I., de Lima, J., Mal- Drake, J.A., Greene, R.S.B., Macdonald, B.C.T., Field, J.B., Pearson, G.L., 2010. A review of tese, A., Francés, F., Caylor, K., Kohv, M., Perks, M., Tóth, B., 2018. On the use of Un- landscape rehabilitation in ecosystem engineering for mine closure. In: Fourie, A., manned Aerial Systems for environmental monitoring. Remote Sens. 10, 641–669. Tibett, M., Wiertz, J., Perth, W.A. (Eds.), Proceedings of International Conference on https://doi.org/10.3390/rs10040641. Mine Closure. Australian Center for Geomechanics, pp. 241–249. Markelin, L., 2013. Radiometric Calibration, Validation and Correction of Multispectral Eisenbeiss, H., Zhang, L., 2006. Comparison of DSMs generated from mini UAV imagery Photogrammetric Imagery. , pp. 1–86 URL:. https://aaltodoc.aalto.fi/bitstream/han- and terrestrial laser scanner in a cultural heritage application. ISPRS Commission V dle/123456789/10948/isbn9789517112956.pdf?sequence=1&isAllowed=y (Doc- Symposium ‘Image Engineering and Vision Metrology’, IAPRS Volume XXXVI, Part toral dissertation for the degree of Doctor of Science in Technology, Aalto 5, Dresden 25–27 Sep. 2006, pp. 90–96. University School of Engineering, Espoo, Finland, September 2013). Elmqvist, T., Folke, C., Nystrom, M., Peterson, G., Bengtsson, J., Walker, B., Norberg, J., 2003. Markelin, L., Suomalainen, J., Hakala, T., Oliveira, R., Viljanen, N., Näsi, R., Scott, B., Response diversity, ecosystem change and resilience. Front. Ecol. Environ. 1, Theocharous, T., Greenwell, C., Fox, N., Honkavaara, E., 2018. Methodology for direct 488–494. https://doi.org/10.1890/1540-9295(2003)001[0488:RDECAR]2.0.CO;2. reflectance measurement from a drone: system description, radiometric calibration ESA (European Space Agency), 2015. Sentinel-2A user handbook. Rev. 2. URL:. https:// and latest results. ISPRS - International Archives of the Photogrammetry, Remote sentinels.copernicus.eu/web/sentinel/user-guides/document-library/-/asset_pub- Sensing and Spatial Information Sciences vol. XLII-1, pp. 283–288. https://doi.org/ lisher/xlslt4309D5h/content/sentinel-2-user-handbook, Accessed date: 1 November 10.5194/isprs-archives-XLII-1-283-2018. 2018. Mather, A.S., Needle, C.L., 1998. The forest transition: a theoretical basis. Area 30, Espigares, T., Moreno-de las Heras, M., Nicolau, J.M., 2011. Performance of vegetation in 117–124. https://doi.org/10.1111/j.1475-4762.1998.tb00055.x. reclaimed slopes affected by soil erosion. Restor. Ecol. 19, 35–44. https://doi.org/ McFeeters, S.K., 1996. The use of the normalized difference water index (NDWI) in the de- 10.1111/j.1526-100X.2009.00546.x. lineation of open water features. Int. J. Remote Sens. 17, 1425–1432. https://doi.org/ Fernández-Guisuraga, J.M., Sanz-Ablanedo, E., Suárez-Seoane, S., Calvo, L., 2018. Using un- 10.1080/01431169608948714. manned aerial vehicles in postfire vegetation survey campaigns through large and Meroni, M., Colombo, R., 2009. 3S: a novel program for field spectroscopy. Comput. heterogeneous areas: opportunities and challenges. Sensors 18 (2), 586. https://doi. Geosci. 35, 1491–1496. https://doi.org/10.1016/j.cageo.2009.01.005 URL:. https:// org/10.3390/s18020586. www.sciencedirect.com/science/article/pii/S0098300409000922, Accessed date: 1 GENCAT (Generalitat de Catalunya), 2018. Orígens i evolució de la mineria (Origins and November 2018. evolution of mining). . Available online:. http://www.gencat.cat/icaen/ Minerals Council of Australia, 2015. In: Minerals Council of Australia (Ed.), Mine Rehabil- recorregutenergia/unitat14/bloc3_unitat14_pag13.html, Accessed date: 1 November itation in the Australian Minerals Industry , p. 48 URL:. http://www.minerals.org.au/ 2018. file_upload/files/reports/Mine_rehabilitation_in_the_Australian_minerals_industry_ Granshaw, S.I., Fraser, C.S., 2015. Editorial: computer vision and photogrammetry: inter- FINAL.pdf, Accessed date: 1 November 2018. action or introspection? Photogramm. Rec. 30, 3–7. https://doi.org/10.1111/ Movia, A., Beinat, A., Crosilla, F., 2016. Shadow detection and removal in RGB VHR images phor.12092. for land use unsupervised classification. ISPRS J. Photogramm. Remote Sens. 119, Hakala, T., Markelin, L., Honkavaara, E., Scott, B., Theocharous, T., Nevalainen, O., Näsi, R., 485–495. https://doi.org/10.1016/j.isprsjprs.2016.05.004. Suomalainen, J., Viljanen, N., Greenwell, C., Fox, N., 2018. Direct reflectance measure- NASA (National Aeronautics and Space Administration), 2018. Landsat Science. ments from drones: sensor absolute radiometric calibration and system tests for for- URL:. https://landsat.gsfc.nasa.gov/about/history/, Accessed date: 1 November est reflectance characterization. Sensors 18, 1417. https://doi.org/10.3390/ 2018. s18051417. OceanOptics, 2006. USB200+ Data Sheet. OceanOptics, Dunedin, FL, p. 38 URL:. https:// Hartley, R., Zisserman, A., 2003. Multiple View Geometry in Computer Vision. Cambridge oceanoptics.com/wp-content/uploads/OEM-Data-Sheet-USB2000-.pdf, Accessed University Press 0521540518. date: 1 November 2018. Honkavaara, E., Kaivosoja, J., Mäkynen, J., Pellikka, I., Pesonen, L., Saari, H., Salo, H., Hakala, Ockendon, N., Thomas, D., Cortina, J., Adams, W., Aykroyd, T., Barov, B., Boitani, L., Bonn, A., T., Marklelin, L., Rosnell, T., 2012. Hyperspectral reflectance signatures and point Branquinho, C., Brombacher, M., Burrell, C., Carver, S., Crick, H., Duguy, B., Everett, S., clouds for precision agriculture by light weight UAV imaging system. ISPRS Annals Fokkens, B., Fuller, R.J., Gibbons, D.W., Gokhelashvili, R., Sutherland, W., 2018. One of Photogrammetry, Remote Sensing and Spatial Information Sciences vol. I-7, hundred priority questions for landscape restoration in Europe. Biol. Conserv. 221. pp. 353–358. https://doi.org/10.5194/isprsannals-I-7-353-2012. https://doi.org/10.1016/j.biocon.2018.03.002. 1614 J.-C. Padró et al. / Science of the Total Environment 657 (2019) 1602–1614 Padró, J.-C., Badia, A., 2017. L'incendi de 1998 a la Catalunya Central. Anàlisi dels canvis en pp. 309–317 URL:. https://ntrs.nasa.gov/archive/nasa/casi.ntrs.nasa.gov/ les cobertes del sòl (1956-2009) a diferents escales, d'incendi i de finca (The wildfire 19740022614.pdf, Accessed date: 1 November 2018. of 1998 in Central Catalonia. Analysis of land cover changes (1956–2009) at different Rumpler, M., Daftry, S., Tscharf, A., Prettenthaler, R., Hoppe, C., Mayer, G., Bischof, H., 2014. scales: at the level of fire and property). Treballs de la Societat Catalana de Geografia Automated end-to-end workflow for precise and geo-accurate reconstructions using 83, 79–108. https://doi.org/10.2436/20.3002.01.125. fiducial markers. ISPRS Technical Commission III Symposium, 5–7 September 2014, Padró, J.-C., Pons, X., Aragonés, D., Díaz-Delgado, R., García, D., Bustamante, J., Pesquer, Ll., Zurich, Switzerland, pp. 135–142 https://doi.org/10.5194/isprsannals-II-3-135-2014. Domingo-Marimon, C., González-Guerrero, O., Cristóbal, J., Doktor, D., Lange, M., Sanders, A., 2017. Mapping the distribution of understorey Rhododendron ponticum using 2017. Radiometric correction of simultaneously acquired Landsat-7/Landsat-8 and low-tech multispectral UAV derived Imagery. In: Díaz-Delgado, R., Lucas, R., Hurford, Sentinel-2A imagery using pseudoinvariant areas (PIA): contributing to the Landsat C. (Eds.), The Roles of Remote Sensing in Nature Conservation. A Practical Guide and time series legacy. Remote Sens. 9 (12), 1319. https://doi.org/10.3390/rs9121319. Case Studies. Springer International Publishing AG2017, Springer, Cham, Switzerland. Padró, J.-C., Muñoz, F.-J., Avila, L.-A., Pesquer, L., Pons, X., 2018. Radiometric correction of ISBN: 978-3-319-64332-8, pp. 167–181. Landsat-8 and Sentinel-2A scenes using drone imagery in synergy with field Schaepmann-Strub, G., Schaepmann, M.E., Painterd, T.H., Dangelb, S., Martonchik, J.V., spectroradiometry. Remote Sens. 10 (11), 1687. https://doi.org/10.3390/rs10111687. 2006. Reflectance quantities in optical remote sensing – definitions and case studies. Padró, J.-C., Muñoz, F.-J., Planas, J., Pons, X., 2019. Comparison of four UAV georeferencing Remote Sens. Environ. 103, 27–42. https://doi.org/10.1016/j.rse.2006.03.002. methods for environmental monitoring purposes focusing on the combined use with Skaloud, J., Rehak, M., Lichti, D., 2014. Mapping with MAV: experimental study on the airborne and satellite remote sensing platforms. Int. J. Appl. Earth Obs. Geoinf. 75, contribution of absolute and relative aerial position control. ISPRS - International Ar- 130–140. https://doi.org/10.1016/j.jag.2018.10.018. chives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Pla, M., Duarte, A., Brotons, L., 2017. Potencial de las imágenes UAV como datos de verdad Volume XL-3/W1, 2014. EuroCOW 2014, the European Calibration and Orientation terreno para la clasificación de la severidad de quema de imágenes Landsat: Workshop, 12–14 Feb. 2014, Castelldefels, Spain https://doi.org/10.5194/ aproximaciones a un producto útil para la gestión post incendio (Potential of UAV im- isprsarchives-XL-3-W1-123-2014. ages as ground-truth data for the burning severity classification of Landsat imagery: Thenkabail, P.S., 2015. Land Resources Monitoring, Modeling, and Mapping With Remote approximations to a useful product for post-fire management). Revista de Sensing. CRC Press, Boca Raton, FL, USA 9781482217957. teledetección 49, 91–102. https://doi.org/10.4995/raet.2017.7140. Turner, D., Lucieer, A., Malenovský, Z., King, D.H., Robinson, S.H., 2014. Spatial co- Pons, X., 2016. MiraMon. Sistema d'Informació Geogràfica i software de Teledetecció. registration of ultra-high resolution visible, multispectral and thermal images ac- Versió 8.01b. Centre de Recerca Ecològica i Aplicacions Forestals, CREAF, Bellaterra quired with a micro-UAV over Antarctic moss beds. Remote Sens. 6, 4003–4024. 84-931323-4-9 URL:. http://www.creaf.uab.cat/miramon/, Accessed date: 1 Novem- https://doi.org/10.3390/rs6054003. ber 2018. United Nations, 2018. Sustainable development goals. URL. https://www.un.org/ Pons, X., Pesquer, L., Cristóbal, J., González-Guerrero, O., 2014. Automatic and improved sustainabledevelopment/sustainable-development-goals/, Accessed date: 1 Novem- radiometric correction of Landsat imagery using reference values from MODIS surface ber 2018. reflectance images. Int. J. Appl. Earth Obs. Geoinf. 33, 243–254. https://doi.org/ Van der Knijff, J.M., Jones, R.J.A., Montanarella, L., 2000. Soil Erosion Risk Assessment in 10.1016/j.jag.2014.06.002. Europe. European Soil Bureau, Joint Research Center, Ispra (34 pp.). Pons, X., Moré, G., Cea, C., Serra, P., Mira, M., González, O., Zabala, A., Pérez, F., Tardà, A., Vermote, E., Justice, C., Claverie, M., Franch, B., 2016. Preliminary analysis of the perfor- Alamús, R., Palà, P., Lopez, A., 2015. A protocol for optimum usage of handheld mance of the Landsat8/OLI land surface reflectance product. Remote Sens. Environ. spectroradiometers in airborne and satellite applications: the Ocean Optics 185, 46–56. https://doi.org/10.1016/j.rse.2016.04.008. USB2000+ case. Proceedings of the 9th EARSeL SIG Imaging Spectroscopy Workshop, Vidal-Macua, J.J., Zabala, A., Ninyerola, M., Pons, X., 2017. Developing spatially and the- Luxembourg (Luxembourg) , pp. 126–127 URL:. https://www.upf.edu/documents/ matically detailed backdated maps for land cover studies. Int. J. Digital Earth 10 (2), 3254890/0/program_short.pdf/bc40e341-f2a5-4b87-a89c-c8a84fe6fb8b, Accessed 175–206. https://doi.org/10.1080/17538947.2016.1213320. date: 1 November 2018. Viscarra Rossel, R.A., Behrens, T., Ben-Dor, E., Brown, D.J., Demattê, J.A.M., Shepherd, K.D., Poyatos, R., Latron, J., Llorens, P., 2003. Land use and land cover change after agricultural Shi, Z., Stenberg, B., Stevens, A., Adamchuk, V., Aïchi, H., Barthès, B.G., Bartholomeus, abandonment. Mt. Res. Dev. 23, 362–368. https://doi.org/10.1659/0276-4741(2003) H.M., Bayer, A.D., Bernoux, M., Böttcher, K., Brodský, L., Du, C.W., Chappell, A., Fouad, 023[0362:LUALCC]2.0.CO;2. Y., Genot, V., Gomez, C., Grunwald, S., Gubler, A., Guerrero, C., Hedley, C.B., Knadel, M., Ram, L.C., Srivasta, N.K., Tripathi, R.C., Jha, S.K., Sinha, A.K., Singh, G., Manoharan, V., 2006. Morrás, H.J.M., Nocita, M., Ramirez-Lopez, L., Roudier, P., Rufasto Campos, E.M., Management of mine spoils for crop productivity with lignite fly ash and biological Sanborn, P., Sellitto, V.M., Sudduth, K.A., Rawlins, B.G., Walter, C., Winowiecki, L.A., amendments. J. Environ. Manag. 79, 173–187. https://doi.org/10.1016/j. Hong, S.Y., Ji, W., 2016. A global spectral library to characterize the world's soil. jenvman.2005.06.008. Earth Sci. Rev. 155, 198–230. https://doi.org/10.1016/j.earscirev.2016.01.012. Rehak, M., Skaloud, J., 2017. Time synchronization of consumer cameras on Micro Aerial Xu, Z.-H., Wu, L.-X., Chen, S.-J., Wang, Z., 2015. Method of engineering volume monitoring Vehicles. ISPRS J. Photogramm. Remote Sens. 123, 114–123. https://doi.org/ and calculation for open-pit mine from UAV images. J. Northeast. Univ. Nat. Sci. 37 10.1016/j.isprsjprs.2016.11.009. (1), 84–88. https://doi.org/10.3969/j.issn.1005-3026.2016.01.018. Richardson, A.J., Weigand, C.L., 1977. Distinguishing vegetation from soil background in- Xue, J., Su, B., 2017. Significant remote sensing vegetation indices: a review of develop- formation. PE&RS 43, 1541–1552 URL:. https://www.asprs.org/wp-content/uploads/ ments and applications. J. Sens., 1353691 https://doi.org/10.1155/2017/1353691 pers/1977journal/dec/1977_dec_1541-1552.pdf, Accessed date: 1 November 2018. (17). Richter, R., Schläpfer, D., 2016. Atmospheric/Topographic Correction for Satellite Imagery Yoo, H.-T., Lee, H., Chi, S., Hwang, B.-G., Kim, J., 2017. A preliminary study on disaster (ATCOR-2/3 User Guide, Version 9.0.2, March 2016). URL:. http://www.rese.ch/pdf/ waste detection and volume estimation based on 3D spatial information. Comput. atcor3_manual.pdf, Accessed date: 1 November 2018. Civ. Eng., 428–435 https://doi.org/10.1061/9780784480823.051. Rouse Jr., J.W., Haas, R., Schell, J., Deering, D., 1974. Monitoring vegetation systems in the Great Plains with ERTS. Third ERTS Symposium. NASA Special Publication 351 ,