Measures of Canopy Structure from Low-Cost UAS for Monitoring Crop Nutrient Status

Abstract

Deriving crop information from remotely sensed data is an important strategy for precision agriculture. Small unmanned aerial systems (UAS) have emerged in recent years as a versatile remote sensing tool that can provide precisely-timed, fine-grained data for informing management responses to intra-field crop variability (e.g., nutrient status and pest damage). UAS sensors with high spectral resolution used to compute informative vegetation indices, however, are practically limited by high cost and data dimensionality. This research extends spectral analysis for remote crop monitoring to investigate the relationship between crop health and 3D canopy structure using low-cost UAS equipped with consumer-grade RGB cameras. We used flue-cured tobacco as a case study due to its known sensitivity to fertility variation and nutrient-specific symptomology. Fertilizer treatments were applied to induce plant health variability in a 0.5 ha field of flue-cured tobacco. Multi-view stereo images from three UAS surveys collected during crop development were processed into orthoimages used to compute a visible band spectral index and photogrammetric point clouds using Structure from Motion (SfM). Plant structural metrics were then computed from detailed high resolution canopy surface models (0.05 m resolution) interpolated from the photogrammetric point clouds. The UAS surveys were complimented by nutrient status measurements obtained from plant tissues. The relationships between foliar nitrogen (N), phosphorus (P), potassium (K), and boron (B) concentrations and the UAS-derived metrics were assessed using multiple linear regression. Symptoms of N and K deficiencies were well captured and differentiated by the structural metrics. The strongest relationship observed was between canopy shape and N foliar concentration (adj. r2 = 0.59, increasing to adj. r2 = 0.81 when combined with the spectral index). B foliar concentration was consistently better predicted by canopy structure with a maximum adj. r2 = 0.41 observed at the latest growth stage surveyed. Overall, combining information about canopy structure and spectral reflectance increased model fit for all measured nutrients compared to spectral alone. These results suggest that an important relationship exists between relative canopy shape and crop health that can be leveraged to improve the usefulness of low cost UAS for precision agriculture.

Determining Spatiotemporal Distribution of Macronutrients in a Cornfield Using Remote Sensing and a Deep Learning Model

“…In addition to the novelties, the soil NPKC macronutrient prediction accuracies have surpassed the results of the benchmark studies in [7,[9][10][11][12][13] and were comparable to the results in [8]. Several reasons could be cited which might have induced such an improvement.…”

“…The field was divided into 5 equal sections, S1-S5, from which soil samples were collected for determining the NPKC nutrient distributions. The field's water flow distribution was arranged evenly, and there was no slope on the field that would have resulted in water accumulation within the area of study; hence, the infield probable nutrient displacements were prevented after rainy days [11].…”

“…On the other hand, in the detailed works of [10], the N uptake and concentration in rice plants have been investigated via multispectral drone images using regression models, rendering moderate performance results; consequently, providing applicable insights about the variable spatiotemporal excess or lack of this nutrient in the soil and the plant biomass. Meanwhile, in [11], the remotely collected spectral reflectance distribution and canopy structural information in tobacco plants at different growth stages were utilized to investigate foliar NPK deficiencies via regression models. The study has produced significant findings regarding N predictions with an R 2 value of 0.80; however, in forecasting the P and K nutrient constituencies, the accuracy levels were moderate.…”

“…The study provided convincing results with R 2 values of about 0.92 from different indices; but, overall the study was limited in analyzing only one macronutrient at a single vegetative growth stage. One could ponder that in such studies [10][11][12][13], employing different state-of-the-art deep learning methods might have rendered superior results as well. Similarly, in [14] and [15], studies were carried out for forecasting nutrients, such as NPK, using statistical and machine learning (ML) methods.…”

Lab-on-a-Drone: remote voltammetric analysis of lead in water with real-time data transmission

“…[8][9][10][11][12] Although increasingly accessible, the cost of most commercially-available UAV place them outside the reach of most low-income communities, where the contaminants are typically discharged. Many of these systems can evaluate chlorophyll levels, 13 estimate nutrition of plants, 14 detect deforestation, 15 compute urbanization 16 or even identify cracks in concrete structures. 17 In addition, these sophisticated UAV need greater ight stability, greater autonomy and the possibility of carrying loads, use multiple rotors like the hexacopter, 8,18 leading to substantial cost increases.…”