UAV-based approach for municipal solid waste landfill monitoring and water ponding issue detection using sensor fusion

Municipal solid waste (MSW) management encompasses various challenges, including addressing issues such as greenhouse gas emissions, waste compaction, and odor control (Townsend et al. 2015). Unmanned aerial vehicles (UAVs) equipped with sensors have emerged as effective tools to monitor MSW landfills, offering a cost-effective and efficient means of conducting volumetric measurements (Hassan et al. 2022a). In the earlier stages, researchers (Mudura et al. 2014) pioneered the use of UAV-driven photogrammetry data in combination with a traditional global positioning system (GPS) to calculate landfill area and volume. Building upon this, recent studies (Filkin et al. 2022) conducted a comparative analysis between low-cost and geodetic UAVs, assessing their efficiency in estimating volumes of waste sites. Expanding the utility of UAVs, UAV-based photogrammetry techniques were employed to identify instability points within landfills (Cunha et al. 2022) over extended periods by utilizing digital elevation models. This interconnected research underscores the evolving role of UAVs in advancing MSW management practices.

In addition to landfill volume estimation, UAV-based sensing applications have also proven useful in environmental monitoring, such as landfill gas (LFG) emissions measurement (Kim et al. 2021). LFG contributes to greenhouse gas emissions (e.g., CH₄, CO₂) and landfill odor. In the United States, the yearly release of 111 million metric tons of CO₂-equivalent from landfills accounts for roughly 2% of the nation's overall greenhouse gas emissions. Methane emissions from U.S. landfills stand as the third most substantial single source, contributing to 17% of the total methane emissions (USEPA 2020). The main challenge for measuring landfill CH₄ emissions is the high spatial and temporal variations. Although several measurement methods for landfill CH₄ emissions quantification were developed, the challenges in their accuracy and applicability are yet to be addressed. For example, surface emission monitoring (SEM) is the prevalent technique to locate emission hotspots on a landfill cover routinely. However, conducting SEM is often time- and effort-intensive because each measurement requires a technician to walk the entire cover, which can take several days for the full survey. Other commonly used methods include flux chambers (Chakraborty et al. 2011), tracer gas dispersion method (Scheutz & Kjeldsen 2019), micro-meteorological method, differential absorption lidar (DiAL) method (Robinson et al. 2011), and air-borne imaging spectrometer method (Duren et al. 2019), most of which have disadvantages associated with high cost and weather dependency. Remote sensing technology has also been employed on a UAV platform (e.g., with 10-cm spatial resolution) compared to satellite platforms (e.g., with 10-m spatial resolution) to achieve a higher spatial resolution. Attempts were made in the past few years to monitor the emission of CH₄ from the surface of a landfill using UAV-based concentration mapping and quantification of CH₄ emission fluxes using various UAV platforms (e.g., fixed-wing, rotary-wing type) and different air sampling methods (Shaw et al. 2021). Integration of UAVs and sensors, such as CH₄ detectors (Emran et al. 2017) and thermal infrared (TIR) cameras (Fjelsted et al. 2019), was implemented to measure LFG emissions.

Remote sensing using multispectral (MS) and hyperspectral (HS) imaging has been used to assess landfills since the late twentieth century. Normalized difference water index (NDWI) maps were used to measure the water content of vegetation from space remotely. The initial studies included the characterization of solid waste landfills and hazardous waste sites by geological remote sensing in two ways (Vincent 1995); one was to compute the digital elevation model (DEM) and the other was to identify significant chemical compositional changes. Vegetation anomalies in terms of gas emission, leachate seeps, and soil wetness were investigated using thermal/MS scanning and infrared images (Henseleit et al. 1990). Expanding the use of multi-band image scanning as an application for agriculture monitoring, an early work (Teke et al. 2013) highlighted the importance of using multi-band imaging from MS/HS sensing for landfill cover monitoring via normalized difference vegetation indices (NDVI). Recently, it was found that these multi-band images, which were collected from an MS instrument of the Sentinel 2 satellite (Dancheva 2020), had great potential to inspect landfills and could be used to calculate the surface temperature. NDVI maps generated from MS imagery were used to determine CH₄ emission via vegetation condition detection (Daugėla et al. 2020). The use of an object-based imagery analysis processing chain was demonstrated in reconstructing landcover maps using UAV-based MS imagery data (Wyard et al. 2022).

The integrity of covering systems (e.g., daily covers, intermediate covers, final covers) is essential for the successful operation of an MSW landfill. The settlement of waste typically causes the deformation of the landfill cover system (Bleiker et al. 1995). This is especially important in areas with high precipitation (e.g., Florida, USA) or after extreme events (e.g., hurricanes), where the uneven settlement can induce surface water ponding and further accelerate the damage to the landfill cover because of the additional weight of the ponding water. The enlarged depression volume, in turn, will allow more infiltration and ultimately provide paths for LFG to escape which would be additionally harmful to the surrounding environment and community [e.g., odors (Du et al. 2023)] if they are left undetected or untreated. Therefore, there is a need to identify (potential and existing) water pond locations.

Some UAV-based measurements can provide useful information about potential or existing water ponds on top of landfills. For example, LiDAR-based DEM from UAV surveying can reveal direct landscape structures at a fine spatial scale (e.g., 10 cm × 10 cm) which relates to hydrological connectivity (Ussyshkin & Theriault 2011). However, the LiDAR-based DEM should be generated during the relatively dry season without much water on top of the land cover, because the laser pulse from the common air-borne LiDAR sensors (rather than bathymetric LiDAR for underwater depth research) can only be reflected from solid material and cannot penetrate water to measure the bottom level of an existing water pond. Geophysical researchers (Chu et al. 2013) studied topographical depression identification (TDI), which could be achieved using DEM generated from LiDAR measurements. Topographical depressions indicate areas with no lateral surface flow and calculate the steepest downward slope (relative to the surrounding area) in terms of the flow direction of each cell/pixel (via DEM) with respect to the surrounding pixels. These flow direction values were then used for the identification of flow accumulation that indicated the areas on the surface having water-storage capacity that could influence soil moisture states, hydrologic connectivity, and climate–soil–vegetation interactions (Le & Kumar 2014). TDI models used the D8 algorithm (O'Callaghan & Mark 1984) which computed flow direction by assigning each center cell to one of its eight neighbors with the steepest downward slope to identify the local elevation minima of each depression location. Local search algorithms were then implemented from these elevation minima to identify depression cells and the elevation thresholds were set to determine the level of water to calculate the water ponding area and volume (Le & Kumar 2014).

Although the TDI method can detect concave areas on a larger space with sharp contrast topographical features (e.g., ridge, valley) using geophysics, TDI is less well suited for a relatively flat landfill top surface with smaller ponds. Hence, more data were necessary to infer these potential/existing ponding locations. In addition to the topographical feature from DEM, the intensity of the reflected laser pulses collected from the raw LiDAR measurement could provide useful information about moisture. The collected LiDAR data file contains one key element of reflective intensity, which was a relative strength measurement of the return pulse compared to the originally emitted pulse. The reflective intensity was much lower from a water surface than from land (Hooshyar et al. 2015). Due to the high UAV LiDAR acquisition altitudes and incidence angles of laser pulses, the reflective intensities returning from a water surface that were collected by the receiver were often negligible. Hence, the point cloud on water surfaces was missing or at low density (Höfle et al. 2009). Remotely sensed MS or HS images from satellites or UAVs could also be used to identify open water bodies. For example, near-infrared (NIR) radiation was absorbed by water but reflected by vegetation (e.g., grass and bushes on top of landfills) and dry soil which was explored to identify ponds and lakes (Work & Gilmer 1976).

There were two common UAV-based methods to obtain the 3D shape of a solid object experimentally: one is from a LiDAR-based point cloud (Collis 1970) and the other is from an image-generated point cloud [e.g., using structure from motion method with bundle adjustment algorithm (Triggs et al. 1999; Hassan et al. 2022b)]. Although either type of point cloud could be used for 3D topographical surveying, there are differences in data features. For example, a LiDAR-based point cloud contains the reflectance intensity features and the return count/profile of laser pulses; while a photogrammetry-based point cloud contains the red-green-blue (RGB) color feature extracted from the images. Either one can be processed into a digital surface model (DSM) representing the surficial features of objects. However, when a DEM with a landcover feature is needed, a LiDAR-based point cloud is the preferred choice due to the classification capacity from laser pulse returned from different objects (e.g., tree canopy, building, ground) and LiDAR's active sensing mode. Another reason to choose LiDAR over photogrammetry is the penetration ability with dense LiDAR pulses through vegetation. LiDAR determines variable distance by targeting an object with a laser and measuring the time for the reflected light to return to the receiver. Laser pulses emitted from a LiDAR system reflected from objects both on and above the ground surface (e.g., vegetation, gas wells) causing different returns. Thus, LiDAR can be used to make digital 3D representations of areas on the earth's surface due to differences in laser return times.

In addition to topology mapping, monitoring surface emissions is an important task for landfill management. The conventional practice in the MSW industry is to conduct SEM by technicians equipped with a handheld sensor who will traverse the area of interest. The cost of such SEM may vary depending upon the required scrutiny level and the total distance of the traversal pathway. Method-21 is the first regulatory method for the detection of CH₄ that was introduced by the US Environmental Protection Agency (EPA) (Fox et al. 2019). Method-21 is applicable for detecting volatile organic compound (VOC) leaks and the detection is performed using a portal device. The types of commonly used detectors include flame ionization, infrared absorption, photoionization, and catalytic oxidation. In this study, a multiple-functional gas sensor (e.g., Soarability Sniffer4Dv2) was mounted onto the UAV platform to analyze air quality a few meters above the land cover. The air quality sensor was affixed to the UAV platform to measure CH₄, CO₂, and H₂S concentrations in real time as the UAV flew over the surface. The sensor uses CH₄ sensing module based on non-dispersive infrared (NDIR) spectroscopy, while using H₂S and CO₂ sensing modules based on electrochemical methods. One should be noted that excessive LFG emissions can infer damage on landfill cover if water ponding induces damage on landfill covers. However, such damages will only occur once water ponding has exacerbated to a certain extent and the chances are relatively low for a well-maintained landfill. Hence, the LFG concentration map is one optional feature for water ponding detection in the proposed sensing framework.

During air quality sensing, the UAV was programed to maintain a low height (e.g., 3–5 m) above the surface of the ground to collect samples of CH₄, CO₂, and H₂S just above the landfill cover. To perform the challenging task of flying a UAV only a few meters above landfills, DSM obtained from LiDAR measurement was provided to assist the route planning so that the UAV could detect nearby terrain elevation changes in its vicinity. Such context-aware air quality sensing not only enables air sampling near the ground but also makes sure no static obstacles in the surrounding environment would impose danger to UAV flights. In addition, the UAV was programmed to stop for 10 s while facing into the wind to accommodate the response time of the sensor, as specified by Method-21.

These index maps maximize the characteristics of vegetation while minimizing confounding factors like soil reflectance or atmospheric effects (Fang & Liang 2014). Various sensing purposes (e.g., vegetation sensing and moisture sensing) have been described in the literature using MS indices. The distributions of MS clues (e.g., NDWI map) may be used to infer water content in the soil of the landfill cover. To explore the moisture-related characteristics of the landfill cover, an MS imaging sensor (e.g., Micasense RedEdge MX) was utilized with the UAV platform for 2D orthomosaic mapping. The route planning for the MS imaging sensor was performed by maintaining both the side and front overlaps (between nearby images) as 75%. The flight height was set at 50 m above ground during the data collection. The MS sensor could capture images in five different bands at the wavelength of 475 nm (blue), 560 nm (green), 668 nm(red), 717 nm (red edge), and 842 nm (near infrared) with the bandwidths of 32, 27, 16, 12, and 57 nm, respectively.

One objective of this study is to identify the areas of topographical depressions on the surface of a landfill. The depressed areas on landfills are the cause of water ponding issues following heavy rainfall events. These concave areas can aggravate settlement effects on the landfill surface due to the excessive pressure resulting from the trapped water. When the permeability of the landfill cover for downward percolation or upward evapotranspiration is insufficient to effectively lower the water level of a pond, the elevated pressure resulting from stagnant water will exacerbate local settling on the surrounding surface and will further expand the depression. The characteristics of the ponds need to be studied to determine which types of measurements are optimal for robust water pond detection.

To normalize the PI for comparison, the total weight of the three weight coefficients should be fixed as 1 (i.e., ⁠). For example, (used in this study) represents using the equal weights of the three features to infer water pond locations; represents using the equal weights of the NFC feature and LRI feature while omitting the NDWI feature. Due to the limit of the extent, the optimization of the weight coefficients is not discussed in this work.

As shown in Figure 5(a), the laser reflective intensity varied but was bounded between 0 and 100% with a low value denoting little reflectance (e.g., water surface, completely soaked soil) or a high value denoting complete reflectance (e.g., rock, concrete). In addition, the LiDAR 3D point cloud (Figure 5(c)) obtained from the raw LiDAR measurement contained information on multiple laser pulse returns. To obtain the topographical information of the ground rather than trees or grasses, the classification of the ground return points (Figure 5(d)) was performed with the features of elevation and pulse return order using a LiDAR data processing software (e.g., Geocue LP360 Professional). Similar to the LiDAR-based point cloud, the one generated from the bundle adjustment algorithm is shown in Figure 5(b). DSM and DEM were generated by processing the LiDAR-based point cloud data using classification software (e.g., LP360 Professional) and GIS software (e.g., Esri ArcGIS Pro). These elevation models were used to compute the flow accumulation map on the landfill cover and the intensity reflectance map was generated from the raw LiDAR data. The topography of the ground was revealed in a LiDAR-based DEM by automatically filtering out the obstruction (e.g., trees, bushes, tall grass) that would otherwise remain in an image-based DSM. There was a gradual increase in the elevation of the landfill cover starting from the bottom corners. On the top of the studied landfill, there were three small ash pile hills stacked contributing to the highest elevation. The ground point classification process could filter the point cloud associated with non-ground points.

During concentration measurements, the UAV was programed to maintain a fixed height and fixed air sampling direction above the surface of the landfill cover. The route planning for the MS imaging task was similar to the LiDAR scanning task but used a fixed pose of the MS camera without a gimbal (Figure 2(d)). Both the side and front overlaps (between nearby images) were maintained at 75% in the MS imaging task. The flight height was set at 50 m (150 ft) above the ground during the data collection. The total approximated scanned area (and time) for MS data collection before the hurricane was 173,641 m² (31 min) and after the hurricane around 167,000 m² (50 min). As the second phase of the data processing, the NDWI map was generated using MS data in which higher NDWI implied the presence of water or moisture on the landfill's surface.

Although minor differences are observed in these air quality measurements, more measurements are needed to find the underlying reasons. However, the results obtained from the air quality survey of the landfill show that no specific hotspots in LFG emissions were detected. This means that the hurricane did not cause damage to the landfill cover leading to additional fugitive LFG emission. The small amount of local variations in concentrations can be attributed to differences in waste composition, age, and microbial activity across the landfill. This may also highlight the complexity of gas generation under variations of external factors like weather or wind. These factors can also be investigated in future studies to incorporate their effect on LFG emissions.

There are spikes of NDWI on top of the landfill, especially near the ash area. The observation indicates the presence of a ponded area which is validated and aligned with findings of the manual visual survey documented by real-time images (Figure 8(b)) captured on the landfill at the time of the survey. The preliminary results (in section 3.4) show that MS mapping is a good candidate for inferring soil moisture content by remote sensing. To gain a more comprehensive understanding, we conducted manual soil sampling tests both before and after the hurricane (Figure 8). These manual tests unequivocally demonstrate a significant increase in overall soil moisture levels in the post-hurricane survey. The NDWI maps match well with the substantial rise in soil moisture levels. The match indicates that NDWI can effectively represent the heightened moisture content in the soil post-hurricane, making the NDWI map a valuable feature for calculating PI maps in specific scenarios.

After the hurricane scenario, the water ponding areas are most prevalent in the ash area along with several other locations in the vegetated area of the landfill as can also be verified from the manual survey (Figure 8). A comparison between the vegetated area and the ash area on top of the landfill before and after the hurricane is presented in Figure 14(b)–14(e). It can be observed that PI increased after the hurricane in several regions (Figure 14(c)). The PI increase is a combined result of both moisture content (as reflected by the NDWI map and laser reflective intensity map) and the depressed areas. The topography did not change much, and the depressed area existed before the hurricane when the landfill was in dry condition (Figure 9(a) and 9(e)) and after the hurricane when the landfill was in wet condition (Figure 9(b) and 9(f)). Similar to the vegetated area, the ash area is shown for scenarios before and after the hurricane in Figure 14(d) and 14(e), respectively. The PI indicated in red observed after the stormwater event (Figure 14(e)) shows water ponding areas than before the stormwater event (Figure 14(d)) in warmer (yellow to red) color. The detection results are aligned with observations as indicated in Figure 8(b).

MSW landfill operation and management need regular surveying (e.g., topology) and monitoring (e.g., CH₄ emission) which is time-consuming and labor-intensive if using traditional methods. In addition, a large amount of stagnant water on top of a landfill represents a threat to landfill cover and landfill management, causing further issues such as cracking, LFG leaking, and water infiltration. To satisfy the practical needs in MSW landfill management, such as effective surveying and landfill cover issue detection and remediation (e.g., water ponding), the study proposes a UAV-based sensing approach and data/image processing and analysis strategy for robust water pond detection on the surfaces of landfills. In addition, the protocol of UAV route planning and data analysis for individual sensing tasks (e.g., topology, MS, air quality) have been planned and conducted. Based on the individual measurements (e.g., flow accumulation processed from DEM, reflective intensity from LiDAR pulses, MS imaging index), data fusion of the features leads to the detection of topographical depressions which are potential or existing water ponds on the surface of landfills which may cause damage to the landfill cover. A damaged cover will accelerate fugitive CH₄ emissions in the environment contributing to climate change and global warming causing adverse effects on the quality of daily life. Hence, the study constructed a PI map fused from multiple-modality sensory measurements which will help landfill operators monitor and maintain the integrity of landcover through periodic UAV-based assessment enabling authorities to take remedial actions to mitigate and prevent the release of potent LFG in the environment once needed. The authors envision the UAV sensing system as a versatile sensing platform to achieve other surveying tasks in landfill operations and management. In particular, the novel PI map using multimodal sensor fusion proposed in this pilot study can help prioritize cover maintenance in an automated and robust fashion.

A case study of the influence of Hurricane Ian on a landfill cover is presented to evaluate the proposed automatic sensing approach and pond index (PI) map detection method. DEMs are generated from the data which are then used for the calculation of laser-reflective intensity maps and water flow accumulation maps. NDWI is computed from the MS images collected before and after the hurricane. NDWI and reflective intensity maps provide moisture estimation while flow accumulation provides depression information on the surface. The resulting PI map shows a match with the in-field manual survey on top of the landfill. The current study can be extended to further facilitate MSW landfill management using landfill site information (e.g., waste filling history, gas quantity/quality) which can be requested from the landfill operators. These data can be incorporated into the existing data platform including remote sensing imagery, air quality data, and meteorological measurements at the site to feed LFG generation models. Further studies are needed to calibrate the effect of different covering materials on top of the soil (e.g., bushes, grasses, ash). In addition to moisture mapping, MS sensing can also provide useful features in detecting vegetation abnormality in a landfill. Further research could focus on refining the index and assessing its potential for application in other waste management scenarios. For example, a sensitivity study can be performed in future studies to optimize the values of the weighting parameters within the PI formula.

The study is funded by the Florida Department of Environmental Protection William W. ‘Bill’ Hinkley Center for Solid and Hazardous Waste Management with award number AWD08952 and project number P0184923. The authors appreciate the support from Mr Larry E. Ruiz at Hillsborough County Solid Waste Management Department for providing landfill space for study and Dr Dingbao Wang from the University of Central Florida for the technical support in GIS computation. In addition, the authors thank Mr Sina Shid-Moosavi, Mr Mohammad Vasef, and Mr Poyu Zhang from the University of Central Florida for their part in the data collection.

S.Z.H. and P.S. carried out the integration/evaluation of the sensing system and the development of the water ponding detection method. S.Z.H, P.S., M.G., and J.C. conducted the in-field experiment on landfills. S.Z.H, P.S., J.C., and D. R. performed data analysis and discussed the results. All the authors contributed to the writing and revision of the manuscript. P.S. conceptualized the main idea of the work and supervised the entire execution of the project.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.