Yield mapping and prediction

Yield prediction is crucial in agricultural activities from a farm-scale to a country or even a global scale. Market management, fertilizer demand estimation, performance assessment of different cultivars, and timely import and export policies are examples of activities that demand accurate yield predictions[1]. Conventionally, farmers used their experience and historic data to predict the approximate yield based on the field's current condition [2]. However, for large-scale and economical production, a more reliable and accurate yield prediction method is required. The potential yield (the maximum yield possible) depends on various factors, such as the weather, soil properties, topography, irrigation, fertilizer management, and, more importantly, the plants' characteristics[3]. However, actual yield often falls significantly below potential yield due to one or more limiting factors at each stage [4]. The interplay of numerous factors makes yield prediction a complex problem, especially for nuts with an alternate bearing. However, these factors can be classified into three major categories: environmental factors, genotypic factors, and interaction of them [5]. The genotypic and interacting factors will be discussed in the phenotyping section.

Environmental factors include but are not limited to weather, precipitation, location of the orchard (mainly latitude and altitude), soil properties, water quantity and quality, and availability of pollinating factors. Studying environmental factors helps to find the limiting parameter and resolve it. For instance, a study on pistachio orchard revealed that the irrigation water amount, soil soluble magnesium, soil electrical conductivity (EC), leaf phosphorus, and leaf nitrogen were the main determinant factors for yield in order of their importance [6]. A similar study on almond orchards showed that light interception was highly influential factor in most orchards so that one percent increase in light interception could increase the potential yield by 57.9 lbs kernel /acre[4]. Although it might not be practical to monitor and incorporate all the affecting environmental factors for yield prediction, remote sensing methods provide a cost-effective monitoring and data collection technique covering many constraining factors[7].

Remotely sensed data for yield estimation can be classified into three categories. 1) physical characteristics of the trees such as canopy volume, height, and LAI, 2) visually identifiable features such as blossom or fruits, and 3) spectral features. In studies based on physical characteristics, usually, a drone equipped with a high-resolution RGB camera or a lidar collects data to create a digital surface models (DSM) and digital terrain models (DTM). Then physical attributes of each tree such as height, crown area and volume could be extracted [8][9]. Additionally, the light interception can be estimated from the physical characteristics of canopies [10]. Yield estimation based on blossom or fruit detection usually takes advantage of the distinct color of the flowers or the fruits to identify them, count them or estimate their percentage. Then, based on weighted average, yield can be estimated [11][12]. These methods heavily depends on image processing techniques as well as machine learning algorithms[13].

The spectral radiance that is used for a variety of purposes can also be used for yield prediction. For example the fraction of absorbed Photosynthetically Active Radiation (fPAR, 400–700 nm), a significant factor in yield, can be estimated from spectral data[14], especially VIs such as the photochemical reflectance index (PRI) [15] . In a an inclusive way, a full-range spectrum can be analyzed to find the most informative bands or a combination of several bands that could potentially model the yield [16].

Although each dataset (physical characteristics, visually identifiable features, and spectral features) can be used individually, in a comprehensive study a mixture of these data should be used to estimate the yield in an orchard. For example, Zhang et al. were used orchard physical characteristics, vegetation indices, orchard age, and weather data and used machine learning techniques to fuse data reaching almond yield prediction at the orchard level by R2 of 0.71 in California [17]. Currently, the main knowledge gap for yield prediction is a method to combine the affecting data of different resolutions and types to predict yield in a holistic picture.


[1]       S. Khaki and L. Wang, “Crop yield prediction using deep neural networks,” Frontiers in plant science, vol. 10, p. 621, 2019.

[2]       G. Ruß and A. Brenning, “Spatial variable importance assessment for yield prediction in precision agriculture,” in International Symposium on Intelligent Data Analysis, 2010, pp. 184–195.

[3]       A. Chlingaryan, S. Sukkarieh, and B. Whelan, “Machine learning approaches for crop yield prediction and nitrogen status estimation in precision agriculture: A review,” Computers and electronics in agriculture, vol. 151, pp. 61–69, 2018.

[4]       Y. Jin, B. Chen, B. D. Lampinen, and P. H. Brown, “Advancing agricultural production with machine learning analytics: yield determinants for California’s almond orchards,” Frontiers in plant science, vol. 11, p. 290, 2020.

[5]       N. Heslot, D. Akdemir, M. E. Sorrells, and J.-L. Jannink, “Integrating environmental covariates and crop modeling into the genomic selection framework to predict genotype by environment interactions,” Theoretical and applied genetics, vol. 127, no. 2, pp. 463–480, 2014.

[6]       I. Esfandiarpour-Borujeni, S. Javad Hosseinifard, H. Shirani, M. Zeinadini, and A. Asghar Besalatpour, “Identifying soil and plant nutrition factors affecting yield in irrigated mature pistachio orchards,” Communications in Soil Science and Plant Analysis, vol. 49, no. 12, pp. 1474–1490, 2018.

[7]       M. Maimaitijiang, V. Sagan, P. Sidike, S. Hartling, F. Esposito, and F. B. Fritschi, “Soybean yield prediction from UAV using multimodal data fusion and deep learning,” Remote Sensing of Environment, vol. 237, p. 111599, 2020.

[8]       J. P. Underwood, C. Hung, B. Whelan, and S. Sukkarieh, “Mapping almond orchard canopy volume, flowers, fruit and yield using lidar and vision sensors,” Computers and electronics in agriculture, vol. 130, pp. 83–96, 2016.

[9]       J. Sarron, É. Malézieux, C. A. B. Sané, and É. Faye, “Mango yield mapping at the orchard scale based on tree structure and land cover assessed by UAV,” Remote Sensing, vol. 10, no. 12, p. 1900, 2018.

[10]     H. Lee, K. C. Slatton, B. E. Roth, and W. P. Cropper, “Prediction of forest canopy light interception using three-dimensional airborne LiDAR data,” International Journal of Remote Sensing, vol. 30, no. 1, pp. 189–207, 2009.

[11]     G. Lin, Y. Tang, X. Zou, J. Li, and J. Xiong, “In-field citrus detection and localisation based on RGB-D image analysis,” Biosystems Engineering, vol. 186, pp. 34–44, 2019.

[12]     R. Horton, E. Cano, D. Bulanon, and E. Fallahi, “Peach flower monitoring using aerial multispectral imaging,” Journal of Imaging, vol. 3, no. 1, p. 2, 2017.

[13]     Y. Chen et al., “Strawberry yield prediction based on a deep neural network using high-resolution aerial orthoimages,” Remote Sensing, vol. 11, no. 13, p. 1584, 2019.

[14]     T. Majasalmi, P. Stenberg, and M. Rautiainen, “Comparison of ground and satellite-based methods for estimating stand-level fPAR in a boreal forest,” Agricultural and Forest Meteorology, vol. 232, pp. 422–432, Jan. 2017, doi: 10.1016/j.agrformet.2016.09.007.

[15]     D. Peng et al., “Assessing spectral indices to estimate the fraction of photosynthetically active radiation absorbed by the vegetation canopy,” International Journal of Remote Sensing, vol. 39, no. 22, pp. 8022–8040, 2018.

[16]     X. Ye, K. Sakai, A. Sasao, and S. Asada, “Potential of airborne hyperspectral imagery to estimate fruit yield in citrus,” Chemometrics and intelligent laboratory systems, vol. 90, no. 2, pp. 132–144, 2008.

[17]     Z. Zhang, Y. Jin, B. Chen, and P. Brown, “California almond yield prediction at the orchard level with a machine learning approach,” Frontiers in plant science, vol. 10, p. 809, 2019.