Estimating the number and size of irrigation center pivot systems (CPS) from remotely sensed data, using artificial intelligence (AI), is a potential information source for assessing agricultural water use. In this study, we identified two technical challenges in the neural-network-based classification: Firstly, an effective reduction of the feature space of the remote sensing data to shorten training times and increase classification accuracy is required. Secondly, the geographical transferability of the AI algorithms is a pressing issue if AI is to replace human mapping efforts one day. Therefore, we trained the semantic image segmentation algorithm U-NET on four spectral channels (U-NET SPECS) and the first three principal components (U-NET principal component analysis (PCA)) of ESA/Copernicus Sentinel-2 images on a study area in Texas, USA, and assessed the geographic transferability of the trained models to two other sites: the Duero basin, in Spain, and South Africa. U-NET SPECS outperformed U-NET PCA at all three study areas, with the highest f1-score at Texas (0.87, U-NET PCA: 0.83), and a value of 0.68 (U-NET PCA: 0.43) in South Africa. At the Duero, both models showed poor classification accuracy (f1-score U-NET PCA: 0.08; U-NET SPECS: 0.16) and segmentation quality, which was particularly evident in the incomplete representation of the center pivot geometries. In South Africa and at the Duero site, a high rate of false positive and false negative was observed, which made the model less useful, especially at the Duero test site. Thus, geographical invariance is not an inherent model property and seems to be mainly driven by the complexity of land-use pattern. We do not consider PCA a suited spectral dimensionality reduction measure in this. However, shorter training times and a more stable training process indicate promising prospects for reducing computational burdens. We therefore conclude that effective dimensionality reduction and geographic transferability are important prospects for further research towards the operational usage of deep learning algorithms, not only regarding the mapping of CPS.

Digital terrain models (DTMs) are a fundamental source of information in Earth sciences. DTM-based studies, however, can contain remarkable biases if limitations and inaccuracies in these models are disregarded. In this work, four freely available datasets, including Shuttle Radar Topography Mission C-Band Synthetic Aperture Radar (SRTM C-SAR V3 DEM), Advanced Spaceborne Thermal Emission and Reflection Radiometer Global Digital Elevation Map (ASTER GDEM V2), and two nationwide airborne light detection and ranging (LiDAR)-derived DTMs (at 5-m and 1-m spatial resolution, respectively) were analysed in three geomorphologically contrasting, small (3–5 km2) catchments located in Mediterranean landscapes under intensive human influence (Mallorca Island, Spain). Vertical accuracy as well as the influence of each dataset’s characteristics on hydrological and geomorphological modelling applicability were assessed by using ground-truth data, classic geometric and morphometric parameters, and a recently proposed index of sediment connectivity. Overall vertical accuracy—expressed as the root mean squared error (RMSE) and normalised median deviation (NMAD)—revealed the highest accuracy for the 1-m (RMSE = 1.55 m; NMAD = 0.44 m) and 5-m LiDAR DTMs (RMSE = 1.73 m; NMAD = 0.84 m). Vertical accuracy of the SRTM data was lower (RMSE = 6.98 m; NMAD = 5.27 m), but considerably higher than for the ASTER data (RMSE = 16.10 m; NMAD = 11.23 m). All datasets were affected by systematic distortions. Propagation of these errors and coarse horizontal resolution caused negative impacts on flow routing, stream network, and catchment delineation, and to a lower extent, on the distribution of slope values. These limitations should be carefully considered when applying DTMs for catchment hydrogeomorphological modelling.

No abstract