The Ph.D. study is based on three years’ field experiments in Denmark from 2017-2019. The main aim was to explore the feasibility of using unmanned aerial vehicles (UAV) and Sentinel-2 images to monitor potato growth (net primary productivity, NPP), nitrogen as well as drought status and provide agricultural management support. The experiments were conducted during the growth season (May-September) of potatoes with a combination of different nitrogen (N) and irrigation treatments.
The intercepted photosynthetically active radiation (Ipar) could be accurately estimated from three scales of remote sensing (RS) data: Rapidscan, UAV and Sentinel-2 data from ground, airborne and spaceborne scales, respectively. The Ipar was proportionally related to NPP via the optimal radiation use efficiency (ranging from 4.19 to 4.98 g MJ-1 depending on the different scales) and the relationship was significantly affected by environmental constraints from maximum temperature and cloudiness index. The successful estimation of daily NPP has a large potential for use in guiding agro-environmental management decisions by supplying basic growth information.
Regarding the N status assessment, parametric analyses (PR) with five different type regressions were compared but were out-performed by the non-parametric random forest regression (RFR) algorithm, which turned out to be a robust method to link the reflectance bands from three scales RS data and the N status parameters, such as plant N uptake (PNU), plant N concentration (PNC) and N nutrition index (NNI). The estimated NNI map could supply the N status information needed for site-specific split application of N in potatoes. Thus, the nitrogen requirements (NR) were calculated based on the estimated PNU, NNI, and expected nitrogen use efficiency, and shown to be able to avoid overdosing of N when pre-seasonal animal manure was applied. However, the RFR has drawbacks as it could not extrapolate outside the range of the training data and does not provide an explicit model formulation. For creating a more representative and applicable model, enlarging the training datasets is necessary.
For assessing drought stress of potato in the field, a two-layer (canopy and soil) energy balance model based on Priestley-Taylor approximation equation (TSEB-PT) was run by using land surface temperature (LST) data from thermal images, estimated fraction of cover from basic RGB images, estimated leaf are index (LAI) from multispectral images, as well as several meteorological data. The modelled transpiration were in good agreement with 10 minutes interval measurements of sap flow during the season. The crop water stress index (CWSI) map derived from LST data could provide straightforward information of the drought stress condition. Finally, the recommended irrigation amount was calculated from the difference between potential and modelled actual evapotranspiration divided by the irrigation water use efficiency (IWUE).
The combination of the N status assessment and drought stress detection supplied useful information not only for detecting crop status but also for discriminating a possible “false signal”, such as estimated N deficiency that in reality derives from drought stress. Thereby, precision agriculture might be conducted more accurate and efficient.
Overall, this study demonstrated the feasibility of using UAV and Sentinel-2 images for assessing crop status with respect to N deficiency and drought stress. Future studies should focus on reducing the underestimation of the fraction of Ipar from Sentinel-2 images and increasing the accuracy of predicting N status also from Sentinel-2 images. The main reason for this was the relatively low spatial resolution compared to the experimental fields and that the soil parts could not be removed from image pixels. Spectral un-mixing could be helpful. Furthermore, a combination of Sentinel-2 multispectral data and Sentinel-3 thermal data should be tested as well.