Can DEM time series produced by UAV be used to quantify diffuse erosion in an agricultural watershed?

Highlights

• UAS can produce high precision DEM with replicable accuracy.

• UAS-DEMs diachronical analysis can provide erosion patterns at watershed scale.

• Elevation differences detected with UAS are confirmed by ground truth measurements.

Abstract

Erosion and deposition modelling should rely on field data. Currently these data are seldom available at large spatial scales and/or at high spatial resolution. In addition, conventional erosion monitoring approaches are labour intensive and costly. This calls for the development of new approaches for field erosion data acquisition.

As a result of rapid technological developments and low cost, unmanned aerial vehicles (UAV) have recently become an attractive means of generating high resolution digital elevation models (DEMs). The use of UAV to observe and quantify gully erosion is now widely established. However, in some agro-pedological contexts, soil erosion results from multiple processes, including sheet and rill erosion, tillage erosion and erosion due to harvest of root crops. These diffuse erosion processes often represent a particular challenge because of the limited elevation changes they induce. In this study, we propose to assess the reliability and development perspectives of UAV to locate and quantify erosion and deposition in a context of an agricultural watershed with silt loam soils and a smooth relief. Erosion and deposition rates derived from high resolution DEM time series are compared to field measurements.

The UAV technique demonstrates a high level of flexibility and can be used, for instance, after a major erosive event. It delivers a very high resolution DEM (pixel size: 6 cm) which allows us to compute high resolution runoff pathways. This could enable us to precisely locate runoff management practices such as fascines. Furthermore, the DEMs can be used diachronically to extract elevation differences before and after a strongly erosive rainfall and be validated by field measurements. While the analysis for this study was carried out over 2 years, we observed a tendency along the slope from erosion to deposition. Erosion and deposition patterns detected at the watershed scale are also promising. Nevertheless, further development in the processing workflow of UAV data is required in order to make this technique accurate and robust enough for detecting sediment movements in an agricultural watershed affected by diffuse erosion. This area of investigation holds much potential as the images processing is relatively new and expanding.

Introduction

European Union identified erosion as one of the major threats to soils, considering its negative impacts in terms of decreases in crop yields (Verity and Anderon, 1990, Souchere et al., 1998, Boardman et al., 2003, Papiernick et al., 2009), siltation of rivers, loss of surface water quality (Papy and Douyer, 1991, Boardman et al., 1994, Dosskey, 2001, Cerdan et al., 2002, Berger et al., 2006), and muddy floods (Verstraeten and Poesen, 1999, Bielders et al., 2003, Evrard et al., 2007). Large-scale quantification of erosion is urgently needed in order to assess these environmental impacts and design the required mitigation measures. Erosion modelling and forecasting also remain a major challenge (Stroosnijder, 2005). Indeed, a wide variety of models is currently available ranging from quite simple approaches (Wischmeier and Smith, 1978) to physically-based schemes (Williams, 1985). However, these models share a common requirement: they all need ground data for model calibration and validation. Hence, the quantification of erosion and deposition of soil particles is a recurring endeavour in this area of research, undertaken in order to support policies and soil conservation programs. Currently, there is a lack of field data for calibration and validation of erosion models, particularly at watershed scale. Indeed, most available erosion data are acquired at the plot scale; therefore relatively few data are available for catchments. In addition, the data are usually available only at the catchment outlet, providing no insight as to the spatial patterns for sediment transport and deposition within the watershed. The connectivity between plots within the watersheds may significantly change the sediment delivery at the outlet. Understanding the physical processes behind erosion/deposition at watershed scale calls for new data collection methods of distributed field data at large spatial scale and quite short time scale (Slattery et al., 2002, Saavedra, 2005).

Permanent changes in soil topography occur in agricultural landscapes. The repeated long-term monitoring of these changes might be one of the ways to obtain the spatial distribution of erosion and deposition. Traditionally, these repeated measurements of surface elevation (Jester and Klik, 2005) are carried out using reference stakes or with profile meters (Hudson, 1993, Sirvent et al., 1997, Casali et al., 1999, Vandekerckhove et al., 2001, Descroix and Claude, 2002, Guzha, 2004, Avni, 2005, Clarke and Rendell, 2006, De Santisteban et al., 2006, Della Seta et al., 2007, Moreno et al., 2008, Della Seta et al., 2009, Keay-Bright and Boardman, 2009, Vergari et al., 2013); by means of surveying with theodolite or terrestrial LiDAR (Belyaev et al., 2004, Haubrock et al., 2009, Nelson et al., 2009, Barneveld et al., 2013, Milenkovic et al., 2015); and terrestrial photogrammetry (Warner, 1995, Hancock and Willgoose, 2001, Rieke-Zapp and Nearing, 2005, Gessesse et al., 2010; Kaiser et al., 2014, Frankl et al., 2015). However, these methods are applicable only to small areas, whereas the most adapted scale to understand erosion and deposition processes seems to be the catchment scale. Moreover, depending on the consistency of soils, some contact methods may disturb the ground surface (Ouédraogo et al., 2014). Standard large-format historical aerial photographs are sometimes used to look at a diachronic evolution, allowing the cover of more extended areas (Aucelli et al., 2014, Gomez, 2014, Gomez et al., 2015). However, this method is only promising for the exploration of active landscapes that widely changed in time. As the flight height is more than 2000 m, the measurement precision attainable does not allow short-term monitoring or identification of subprocesses involved in diffuse erosion. Gomez (2014) can reach a horizontal accuracy of ± 2 to 6 m, as the method is also constrained by the number of pixels in each image. Moreover, permanent GCPs are needed with this approach while no features are present along the years in a predominantly agricultural watershed, leading to imprecision. Digital photogrammetry with an unmanned aerial vehicle (UAV) equipped with a handheld non-metric camera is a non-destructive alternative that could be less time consuming and cheaper than the traditional methods described above. In addition, it provides the user with continuous space coverage, and permits a high sampling density.

The combination of advanced photogrammetry and the more and more widespread small UAVs led geoscientists to reviewed the opportunities and challenges of this fast and low-cost technique to try to improve it, which shows great promise (e.g. Puech et al., 2009, Aber et al., 2010, Pierrot-Deseilligny and Clery, 2011, Zhang et al., 2011, Gruen, 2012, James and Robson, 2012, Remondino et al., 2012, Turner et al., 2012, Fonstad et al., 2013, Hugenholtz et al., 2013, Colomina and Molina, 2014, Tarolli, 2014). In general, photogrammetry requires equipment cheaper and lighter than lasergrammetry (Pierrot-Deseilligny and Clery, 2011, White et al., 2013). Various studies also compared the UAV results and those from laser surveys (e.g. Eisenbeiss and Zhang, 2006 ; Hugenholtz et al., 2012, Westoby et al., 2012, Fonstad et al., 2013, Mancini et al., 2013, Stumpf et al., 2013, White et al., 2013, Obanawa et al., 2014, Ouédraogo et al., 2014) and often demonstrated that the advances in photogrammetric technique (named SfM for “structure from motion”) can deliver data quality and resolutions that are comparable with LiDAR and classic photogrammetry (differing from modern photogrammetry by the use of manned aircraft, metric camera and low images overlap). Furthermore, it can produce point clouds with horizontal and vertical precision in the centimetre order. Aber et al. (2010) and d'Oleire-Oltmanns et al. (2012) describe the photogrammetric technique as a way to reduce the existing gap between field scale and satellite scale data collection. It can be used at different scales for various applications (James and Robson, 2014).

The SfM photogrammetric technique enables the fast reconstruction of three-dimensional scene geometry from two-dimensional pictures (Westoby et al., 2012). The multiple overlapping images are captured by a consumer grade camera moving around the scene, and algorithms detect characteristic image feature points which match between images (Verhoeven, 2011). Tie points are automatically determined between the images and aerotriangulation by bundle block adjustment result in a sparse 3D model. The generated point cloud is then translated and rotated in a specific reference system by the use of ground control points (GCPs) (Sona et al., 2014). The georeferenced camera positions can be used through dense matching of images (Fonstad et al., 2013) in order to create digital elevation products. The techniques of image processing by photogrammetry and SfM are increasingly used, but the associated algorithms are in constant evolution.

The SfM photogrammetry technique was validated for various fields (see Smith and Vericat, 2015 for a synthesis of existing validation studies). Originally applied to the study of coastal or river morphology (Hapke and Richmond, 2000, Westaway et al., 2000, Westaway et al., 2001, Mertes, 2002, Carbonneau et al., 2003, Gilvear and Bryant, 2003, Westaway et al., 2003, Carbonneau et al., 2004, Carbonneau et al., 2005, Carbonneau et al., 2006, Lejot et al., 2007, Marcus and Fonstad, 2008), photogrammetry has more recently been used in studies on landslides (Niethammer et al., 2012, Stumpf et al., 2013, Lucieer et al., 2013) and badlands or gully erosion (Giménez et al., 2009, Marzolff and Poesen, 2009, Puech et al., 2009, Marzolff et al., 2011, d'Oleire-Oltmanns et al., 2012, Peter et al., 2014). About the latter, Aber et al. (2010) concluded that small format aerial photography can be considered an advantageous alternative to field methods. The detected topographic changes due to gully erosion were validated by laboratory experiments (Rieke-Zapp and Nearing, 2005) or field measurements (Marzolff et al., 2011) including rainfall simulation (Gessesse et al., 2010, Peter et al., 2014). As these analyses often showed that the results are better for the less complex terrains, it's interesting to try the technique on our smoother terrains where different (diffuse) erosion types are predominant. Indeed, in many regions, the diffuse forms of erosion (interrill erosion, tillage erosion, harvest erosion) represent important contributions to total erosion. They represent a particular challenge as regards photogrammetry considering the small changes in soil elevation while phenomena such as coastal erosion or gully erosion induce large difference in elevation (dm or more), which are relatively easy to detect by photogrammetry. The challenge is therefore to spatialize the phenomena in landscapes where the changes in relief are less affected over the short term than in other existing studies. Hence, this research aims to investigate if these technological developments make it possible to quantify these small variations by the use of multidate imagery acquired with a small UAV.

There are two main objectives of this study. One is to establish a technique chain to obtain high quality digital elevation models (DEMs). The other is to explore whether regular drone surveys allow us to quantify the spatial and temporal distribution of erosion/deposition due to rainfall events and agricultural practices. By making a temporal analysis recommended by Tarolli (2014), we investigate whether a diachronic analysis based on drone data allows us to locate and quantify soil redistribution at the watershed scale, for silty-loam soil affected mainly by diffuse erosion.