The optimization of crop nitrogen fertilization to accurately predict and match the nitrogen (N) supply to the crop N demand is the subject of intense research due to the environmental and economic impact of N fertilization. Excess N could seep into the water supplies around the field and cause unnecessary spending by the farmer. The drawbacks of N deficiency on crops include poor plant growth, ultimately reducing the final yield potential. The objective of this study is to use Unmanned Aerial Vehicle (UAV) multispectral imagery to predict canopy nitrogen weight (g/m²) of corn fields in south-west Ontario, Canada. Simple/multiple linear regression, Random Forests, and support vector regression (SVR) were established to predict the canopy nitrogen weight from individual multispectral bands and associated vegetation indices (VI). Random Forests using the current techniques/methodologies performed the best out of all the models tested on the validation set with an R² of 0.85 and Root Mean Square Error (RMSE) of 4.52 g/m². Adding more spectral variables into the model provided a marginal improvement in the accuracy, while extending the overall processing time. Random Forests provided marginally better results than SVR, but the concepts and analysis are much easier to interpret on Random Forests. Both machine learning models provided a much better accuracy than linear regression. The best model was then applied to the UAV images acquired at different dates for producing maps that show the spatial variation of canopy nitrogen weight within each field at that date.

Mapping wetlands with high spatial and thematic accuracy is crucial for the management and monitoring of these important ecosystems. Wetland maps in New Brunswick (NB) have traditionally been produced by the visual interpretation of aerial photographs. In this study, we used an alternative method to produce a wetland map for southern New Brunswick, Canada, by classifying a combination of Landsat 8 OLI, ALOS-1 PALSAR, Sentinel-1, and LiDAR-derived topographic metrics with the Random Forests (RF) classifier. The images were acquired in three seasons (spring, summer, and fall) with different water levels and during leaf-off/on periods. The resulting map has eleven wetland classes (open bog, shrub bog, treed bog, open fen, shrub fen, freshwater marsh, coastal marsh, shrub marsh, shrub wetland, forested wetland, and aquatic bed) plus various non-wetland classes. We achieved an overall accuracy classification of 97.67%. We compared 951 in-situ validation sites to the classified image and both the 2106 and 2019 reference maps available through Service New Brunswick. Both reference maps were produced by photo-interpretation of RGB-NIR digital aerial photographs, but the 2019 NB reference also included information from LiDAR-derived surface and ecological metrics. Of these 951 sites, 94.95% were correctly identified on the classified image, while only 63.30% and 80.02% of these sites were correctly identified on the 2016 and 2019 NB reference maps, respectively. If only the 489 wetland validation sites were considered, 96.93% of the sites were correctly identified as a wetland on the classified image, while only 58.69% and 62.17% of the sites were correctly identified as a wetland on the 2016 and 2019 NB reference maps, respectively.

Potato late blight, caused by Phytophthora infestans, is a major disease worldwide that has a significant economic impact on potato crops, and remote sensing might help to detect the disease in early stages. This study aims to determine changes induced by potato late blight in two parameters of the red and red-edge spectral regions: the red-well point (RWP) and the red-edge point (REP) as a function of the number of days post-inoculation (DPI) at the leaf and canopy levels. The RWP or REP variations were modelled using linear or exponential regression models as a function of the DPI. A Support Vector Machine (SVM) algorithm was used to classify healthy and infected leaves or plants using either the RWP or REP wavelength as well as the reflectances at 668, 705, 717 and 740 nm. Higher variations in the RWP and REP wavelengths were observed for the infected leaves compared to healthy leaves. The linear and exponential models resulted in higher adjusted R² for the infected case than for the healthy case. The SVM classifier applied to the reflectance of the red and red-edge bands of the Micasense^® Dual-X camera was able to sort healthy and infected cases with both the leaf and canopy measurements, reaching an overall classification accuracy of 89.33% at 3 DPI when symptoms were visible for the first time with the leaf measurements and of 89.06% at 5 DPI, i.e., two days after the symptoms became apparent, with the canopy measurements. The study shows that RWP and REP at leaf and canopy levels allow detecting potato late blight, but these parameters are less efficient to sort healthy and infected leaves or plants than the reflectance at 668, 705, 717 and 740 nm. Future research should consider larger samples, other cultivars and the test of unmanned aerial vehicle (UAV) imagery for field-based detection.

Optimizing nitrogen (N) fertilization in crop production by in-season measurements of crop N status may improve fertilizer N use efficiency. Hyperspectral measurements may be used to assess crop N status indirectly by estimating leaf and canopy chlorophyll content. This study evaluated the ability of the PROSAIL canopy-level reflectance model to predict leaf chlorophyll content of spring wheat (Triticum aestivum L.) during the growth stages between pre-tillering (Zadoks Growth Stage (ZGS 15)) to booting (ZGS50). Spring wheat was grown under different N fertility rates (0–200 kg N ha⁻¹) in 2002. Canopy reflectance, leaf chlorophyll content, N content and leaf area index (LAI) values were measured. There was a weakly significant trend for the PROSAIL model to over-estimate LAI and under-estimate leaf chlorophyll content. To compensate for this interdependency by the model, a canopy chlorophyll content parameter (the product of leaf chlorophyll content and LAI) was calculated. The estimation accuracy for canopy chlorophyll content was generally low earlier in the growing season. This failure of the PROSAIL model to estimate leaf and canopy variables could be attributed to model sensitivity to canopy architecture. Earlier in the growing season, full canopy closure was not yet achieved, resulting in a non-homogenous canopy and strong soil background interference. The canopy chlorophyll content parameter was predicted more accurately than leaf chlorophyll content alone at booting (ZGS 45). A strong relationship between canopy chlorophyll content and canopy N content at ZGS 45 indicates that the PROSAIL model may be used as a tool to predict wheat N status from canopy reflectance measurements at booting or later.

Management of nitrogen (N) fertilizers is an important agricultural practice and field of research to minimize environmental impacts and the cost of production. To apply N fertilizer at the right rate, time, and place depends on the crop type, desired yield, and field conditions. The objective of this study is to use Unmanned Aerial Vehicle (UAV) multispectral imagery, vegetation indices (VI), crop height, field topographic metrics, and soil properties to predict canopy nitrogen weight (g/m²) of a corn field in southwestern Ontario, Canada. Random Forests (RF) and support vector regression (SVR) models were evaluated for canopy nitrogen weight prediction from 29 variables. RF consistently had better performance than SVR, and the top-performing validation model was RF using 15 selected height, spectral, and topographic variables with an R² of 0.73 and Root Mean Square Error (RMSE) of 2.21 g/m². Of the model’s 15 variables, crop height was the most important predictor, followed by 10 VIs, three MicaSense band reflectance mosaics (blue, red, and green), and topographic profile curvature. The model information can be used to improve field nitrogen prediction, leading to more effective and efficient N fertilizer management.