In precision agriculture (PA) practices, the accurate delineation of management zones (MZs), with each zone having similar characteristics, is essential for map-based variable rate application of farming inputs. However, there is no consensus on an optimal clustering algorithm and the input data format. In this paper, we evaluated the performances of five clustering algorithms including k-means, fuzzy C-means (FCM), hierarchical, mean shift, and density-based spatial clustering of applications with noise (DBSCAN) in different scenarios and assessed the impacts of input data format and feature selection on MZ delineation quality. We used key soil fertility attributes (moisture content (MC), organic carbon (OC), calcium (Ca), cation exchange capacity (CEC), exchangeable potassium (K), magnesium (Mg), sodium (Na), exchangeable phosphorous (P), and pH) collected with an online visible and near-infrared (vis-NIR) spectrometer along with Sentinel2 and yield data of five commercial fields in Belgium. We demonstrated that k-means is the optimal clustering method for MZ delineation, and the input data should be normalized (range normalization). Feature selection was also shown to be positively effective. Furthermore, we proposed an algorithm based on DBSCAN for smoothing the MZs maps to allow smooth actuating during variable rate application by agricultural machinery. Finally, the whole process of MZ delineation was integrated in a clustering and smoothing pipeline (CaSP), which automatically performs the following steps sequentially: (1) range normalization, (2) feature selection based on cross-correlation analysis, (3) k-means clustering, and (4) smoothing. It is recommended to adopt the developed platform for automatic MZ delineation for variable rate applications of farming inputs.

Thermal and hyperspectral proximal disease sensing are valuable tools towards increasing pesticide use efficiency. However, some practical aspects of the implementation of these sensors remain poorly understood. We studied an optimal measurement setup combining both sensors for disease detection in leek and potato. This was achieved by optimising the signal-to-noise ratio (SNR) based on the height of measurement above the crop canopy, off-zenith camera angle and exposure time (ET) of the sensor. Our results indicated a clear increase in SNR with increasing ET for potato. Taking into account practical constraints, the suggested setup for a hyperspectral sensor in our experiment involves (for both leek and potato) an off-zenith angle of 17°, height of 30 cm above crop canopy and ET of 1 ms, which differs from the optimal setup of the same sensor for wheat. Artificial light proved important to counteract the effect of cloud cover on hyperspectral measurements. The interference of these lamps with thermal measurements was minimal for a young leek crop but increased in older leek and after long exposure. These results indicate the importance of optimising the setup before measurements, for each type of crop.

Soil surveys with line-scanning platforms appear to have great advantages over the traditional methods used to collect soil information for the development of field-scale soil mapping and applications. These carry VNIR (visible and near infrared) spectrometers and have been used in recent years extensively for the assessment of soil fertility at the field scale, and the delineation of site-specific management zones (MZ). A challenging feature of VNIR applications in precision agriculture (PA) is the massiveness of the derived datasets that contain point predictions of soil properties, and the interpolation techniques involved in incorporating these data into site-specific management plans. In this study, fixed-rank kriging (FRK) geostatistical interpolation, which is a flexible, non-stationary spatial interpolation method especially suited to handling huge datasets, was applied to massive VNIR soil scanner data for the production of useful, smooth interpolated maps, appropriate for the delineation of site-specific MZ maps. Moreover, auxiliary Sentinel-2 data-based biophysical parameters NDVI (normalized difference vegetation index) and fAPAR (fraction of photosynthetically active radiation absorbed by the canopy) were included as covariates to improve the filtering performance of the interpolator and the ability to generate uniform patterns of spatial variation from which it is easier to receive a meaningful interpretation in PA applications. Results from the VNIR prediction dataset obtained from a pivot-irrigated field in Albacete, southeastern Spain, during 2019, have shown that FRK variants outperform ordinary kriging in terms of filtering capacity, by doubling the noise removal metrics while keeping the computation cost reasonably low. Such features, along with the capacity to handle a large volume of spatial information, nominate the method as ideal for PA applications with massive proximal and remote sensing datasets.