Course of Action

Throughout this paper, the necessity, collection methods, and processing pipelines of remote sensing data using available platforms and sensors for various applications with a focus on nut orchards were discussed. Due to the rapidly growing acreage of nut orchards, and increasing need for food in general, new agricultural methods based on remote sensing data are inevitable. The available remote sensing platforms can be classified to satellites, manned aircraft, and UAS, each with pros and cons, making platform selection a subject-dependent task. Sensors are also mostly problem-dependent, and the requirements of the issue determine the best possible sensor. However, in many cases, a combination of sensors and platforms are combined for the best results. Remote sensing applications are diverse, including irrigation management, disease control, yield estimation, nutrition mapping, salinity control or detection, phenotyping, and other managerial applications. However, reviewing the published papers on nut crops showed that most of the papers focused on water management problems and other applications are limited or not practiced, indicating research opportunities and the gap in the knowledge.

Before analyzing data, some preprocessing steps should be taken to ensure refinement and removing unsupportive/redundant data.  Radiometric correction is an essential preprocessing step that has not been taken seriously in most studies, although it is crucial, especially for long-term temporal comparisons. Additionally, since most remote sensing data rely on electromagnetic reflectance, preserving the original reflectance values (pure pixel) is of great concern, and employing software that manipulate the original reflectance values are not recommended. Before feature extraction, noise and outlier detection and removal, which can be achieved using supervised and unsupervised methods, guarantee the accuracy and repeatability of the results. In many cases, the raw features cannot explain all the variability in response, raising the need for feature engineering and feature selection. Features can be used in linear or non-linear models. Although linear models, such as linear regression, are potent and most practiced approaches in modeling, nonlinear methods such as machine learning and artificial intelligence techniques, which tackle many input variables and capture the most relevant information from the input features, are expected to increase. A favorable analytics model or a data fusion framework would determine all essential factors, weight them based on their importance level, and use all available information for the best prediction possible.