Computers and Electronics in Agriculture 213 (2023) 108246
Contents lists available at ScienceDirect 
Computers and Electronics in Agriculture 
journal homepage: www.elsevier.com/locate/compag 
Differentiating nutritional and water statuses in Hass avocado plantations 
through a temporal analysis of vegetation indices computed from aerial 
RGB images 
Itamar Salazar-Reque a, Daniel Arteaga a, Fabiola Mendoza b, Maria Elena Rojas b, Jonell Soto b, 
Samuel Huaman a, Guillermo Kemper a,* 
a INICTEL-UNI, Universidad Nacional de Ingeniería, Lima, Peru 
b Instituto Nacional de Innovación Agraria, Lima, Peru   
A R T I C L E  I N F O   A B S T R A C T   
Keywords: Maximizing crop production efficiently and sustainably through plant health monitoring is key for global food 
Hass avocado security. Monitoring large areas with remote sensing technologies such as unmanned aerial vehicles (UAVs) with 
Aerial RGB images sensors deals with time and money issues; however, the usage of advanced sensors such as hyperspectral, 
Vegetation Indices multispectral and thermal cameras limit their usage among all the stakeholders. In this study we explore different 
Nutrient Status Monitoring 
Water Status Monitoring vegetation indices (VIs) extracted from aerial RGB images acquired in different flights to differentiate the 
nutritional and water statuses of Hass avocado plantations. We used an image processing workflow consisting of 
image selection through a convolutional neural network (CNN) model, tree crown segmentation, color correction 
and feature extraction to automate the computation of VIs from RGB images. To compare the performance of VIs 
in the differentiation of nutritional and water statuses, we proposed a comparison metric called Mean Distance 
between Vegetation Indices (MDVI), analyzed the evolution of the extracted features, and studied their re-
lationships with gold standard Normalized Difference Vegetation Index (NDVI) measurements. Since the 
extracted features from each group vary from flight to flight due to multiple factors such as the light intensity of 
each season and the phenological stage of the plant, the proposed comparison metric leverages the differences 
between the features extracted from each group, thus reducing these temporal effects. We found that Modified 
Green Red Vegetation Index (MGRVI) allows a better differentiation of nutritional and water statuses. 
Furthermore, the correlation coefficients of this VI in the three statuses and NDVI for nitrogen group range 
between 0.63 and 0.85, indicating a positive strong relationship. The results of this work show that MGRVI has a 
potential to be used as a correlation variable in studies that only use RGB sensors in order to monitor the 
nutritional and water status of crops.   
1. Introduction preventing diseases, and optimizing the use of agrichemicals, fertilizers, 
and water. By leveraging precision agriculture techniques, farmers can 
With the current socio-economic and climate change scenarios, maximize crop production while minimizing costs and environmental 
ensuring global food security has become a major concern. Projections impact, thereby ensuring long-term sustainability. 
to 2050 indicate a significant increase in global food demand by 35% to One crop that has garnered considerable attention due to the nutri-
56% between 2010 and 2050 (van Dijk et al., 2021), placing the onus on tional and commercial significance, not only of the pulp, but also of the 
the agriculture sector to enhance crop production in an efficient and seeds and peel, is the avocado (Persea Americana). For this reason, some 
sustainable manner. Monitoring plant health through nutrient and water studies propose the utilization of seeds and peels (Nyakang’i et al., 2023; 
status is a viable approach to optimize yields while mitigating envi- Bayomy et al., 2023) based on their abundance of bioactive compounds, 
ronmental impacts. This monitoring provides valuable insights that can including unsaturated fatty acids, vitamins C, B, and E, dietary fiber, 
be utilized for diagnosing existing nutrient deficiencies, preemptively lutein, phenolic compounds, as well as various pigments such as 
* Corresponding author. 
E-mail address: guillermo.kemper@gmail.com (G. Kemper).  
https://doi.org/10.1016/j.compag.2023.108246 
Received 12 August 2022; Received in revised form 30 August 2023; Accepted 9 September 2023   
Available online 22 September 2023
0168-1699/© 2023 Elsevier B.V. All rights reserved.
I. Salazar-Reque et al.                                                                                                                                                              C  o  m   p  u t e  r s  a  n  d   E  l e c  t r o  n i c  s  i n   A  g  r i c u  l t u  r e  213 (2023) 108246
chlorophylls, anthocyanins, and carotenoids (Kosińska et al., 2012; Lu intake for plants is closely related with the amount of water in the soil; 
et al., 2005; Saavedra et al., 2017; Wang et al., 2010; Corrales-García therefore, during drought, crops are negatively affected due to water 
et al., 2019; Ong et al., 2022). Consequently, the global production and and nitrogen constraints (Plett et al., 2020). In order to determine and 
consumption of avocados have witnessed a substantial surge in recent control the nutrients and water in plants, different monitoring proced-
years. This surge has not only contributed to increased interest in avo- ures have been employed. 
cado as a functional ingredient in foods but has also led to diverse Silber et al. (2018) assessed the seasonal nutritional requirement of 
commercial applications, ranging from frozen products and ice cream to Hass avocado trees grown in lysimeters. Through this in-situ monitoring 
avocado oil, guacamole, and even cosmetic products (Colombo and of NPK macronutrients, they determined that continuous nutrient 
Papetti, 2019). application during the year, and the modification of fertilization rate 
Precision agriculture approaches that enable effective monitoring of together with the nutrient combination is required for guaranteeing 
nutrient and water status, especially in high-value crops like avocados, optimal yield. Moreover, they conclude that fruit analyses should be 
hold significant promise to cope with the escalating global food demand. used for scheduling fertilization instead of leaf analyses. Gaona et al. 
Although the advantages provided by precision agriculture, time and (2020) also employed an in-situ monitoring procedure to determine the 
money are limitations when monitoring large areas. Current plant effect of two doses of nitrogen and potassium in the initial growth phase 
analysis in laboratory is not suitable for this scenario because it involves of Hass avocado. They observed significant differences in the greenery 
sample collection, sample preparation, laboratory analysis and posterior index and nutrient concentration variables. 
interpretation. Emerging technologies have been applying to face these In-situ monitoring allows accurate evaluation of the nutritional and 
limitations. Roper et al. (Roper et al., 2021) discuss some approaches of water status of plants; nevertheless, it is time consuming and normally 
in vivo plant sensors for plant health monitoring. Remote sensing requires the completion of specific steps to interpret the data (Dezordi 
technologies are also used for the same purpose. UAVs and imaging et al., 2016). Remote sensing technologies cope with this drawback. 
sensors have been widely employed as a remote sensing system to Monitoring of the nutritional and water status with remote sensing 
compute different VIs, which are then used to diagnose nutrient prob- systems allows to cover medium and large areas in less time and data 
lems (Garza et al., 2020; Janoušek et al., 2021; Noguera et al., 2021) or processing is done autonomously through computer programs. 
predict nutrient status (Prado Osco et al., 2019; Zha et al., 2020; Osco Satellites are very popular remote sensing monitoring systems that 
et al., 2020). have been used for many precise agriculture applications such as 
In most of the studies, the calculation of VIs involves the usage of at pesticide usage monitoring (Adan et al., 2021), pest severity assessment 
least one spectral band other than those of the visible spectrum (the true (Salgadoe et al., 2018), yield mapping (Robson et al., 2017), and 
color bands red, green and blue), such as the near-infrared (NIR) or red nutritional plant status monitoring (Sharifi, 2020; Sims et al., 2013). 
edge (RE) bands. This imply the usage of multispectral cameras over Sims et al. (2013) presented a synopsis of studies using satellite hyper-
UAVs for image acquisition, which can produce a limitation for stake- spectral imagery for foliar nutrition assessment which collectively 
holders that only possess RGB cameras. To make more extensible the demonstrate that several important nutrients can be accurately mapped 
monitoring of nutritional and water status through UAVs and imaging at critical ages for the management of plantation forests. However, some 
sensors, identifying some VIs extracted from aerial RGB images that are spatial and radiometric characteristics limit the practical usage. Sharifi 
useful for nutritional and water status assessment is required. (2020) used satellite imagery to evaluate the use of VIs for crop nutrition 
This work aims to explore different VIs extracted from aerial RGB mapping. They found that the two best VIs were TCARI (Transformed 
images that allow to differentiate trees with adequate and inadequate Chlorophyll Absorption in Reflectance Index) (Haboudane et al., 2002) 
nutritional and water status. The study was developed with Hass avo- and MCARI (Modified Chlorophyll Absorption in Reflectance Index) 
cado trees. Because of the advantages of fertigation (Incrocci et al., (Daughtry et al., 2000). Moreover, they conclude that the use of NIR and 
2017), it was used to induce deficiencies and excesses of nitrogen, RE multispectral bands in mid-season led to better results than VIs 
phosphorus and potassium (NPK) macronutrients, and water de- calculated at the end of the season or that do not include the RE band. 
ficiencies. During the study period, a UAV was employed to acquire With the development of UAVs and imagery sensors in the recent 
aerial RGB images of trees with adequate nutritional status (control years, there are more studies in the monitoring of nutritional and water 
groups) and trees with an induced inadequate nutritional and water status of plants. A review article by Jayme Arnal (Barbedo, 2019) pre-
status. Different VIs were calculated for each group and flight, which sented 30 and 31 works dealing with the monitoring of nutrient and 
then were used to compute the comparison features. We proposed a water status in crops respectively. On one hand, multispectral cameras 
comparison metric called Mean Distance between Vegetation Indices were the sensors used in most of the studies for monitoring of nutritional 
(MDVI) that allows better differentiation among statuses for different status. On the other hand, most of the studies aiming to monitor water 
groups and VIs. In addition, a temporal analysis of these features and a status employed thermal cameras. 
correlation study with NDVI (Rouse et al., 1974) measurements were Regarding the studies with multispectral images, Lu et al. (2019) 
performed. Our results show that MGRVI (Bendig et al., 2015) per- acquire images at seven view zenith angles for three critical growth 
formed better in differentiating nutritional and water statuses. stages of winter wheat and computed four VIs as input variables: Visible 
Therefore, in this study, we propose a novel approach to differentiate Atmospherically Resistant Index (VARI) (Gitelson et al., 2002), Red edge 
the nutritional and water statuses in Hass avocado plantations through a Chlorophyll Index (CIred-edge) (Gitelson et al., 2003; Gitelson, 2005; 
temporal analysis of VIs computed from aerial RGB images. By Gitelson et al., 2006), Green band Chlorophyll Index (CIgreen) (Gitelson 
employing remote sensing techniques and advanced image analysis al- et al., 2003; Gitelson, 2005; Gitelson et al., 2006), and Modified 
gorithms, we aim to develop a non-invasive and cost-effective method Normalized Difference Vegetation Index with a blue band (mNDblue) 
for assessing the nutrient and water status of Hass avocado trees. Such (Jay et al., 2017). They used leaf nitrogen concentration (LNC), plant 
an approach has the potential to empower farmers with valuable in- nitrogen concentration (PNC), leaf nitrogen accumulation (LNA), and 
formation to make informed decisions regarding nutrient and water plant nitrogen accumulation (PNA) as target variables to be estimated by 
management, leading to improved yields and sustainable avocado a linear regression model. When using multiple view zenith angles, they 
production. used a two-variable regression model. The best result with single-angle 
images was obtained with CIgreen for LNC from a view zenith angle of 
2. Related work − 60◦. Zhang et al. (2016) monitored heading rice growing based on 
UAV multispectral images by using linear regression model. They used 
One of the main limitations for the increase in crop productivity is GNDVI (Green Normalized Difference Vegetation Index) (Buschmann 
the availability of water and nutrients like nitrogen. Macronutrients and Nagel, 1993) as input variable, and SPAD values (chlorophyll) and 
2
I. Salazar-Reque et al.                                                                                                                                                              C  o  m   p  u t e  r s  a  n  d   E  l e c  t r o  n i c  s  i n   A  g  r i c u  l t u  r e  213 (2023) 108246
the nitrogen content as target variables, finding a higher correlation method the accuracies were 0.75, 0.8 and 0.94 respectively. 
between GNDVI and SPAD values. Another usage of multispectral im- RGB images acquired via UAVs have been also used in monitoring 
ages to compute VIs for estimating nutritional status was presented by plant health – nutrient and water – status. Garza et al. (2020) correlates 
Chungcharoen et al. (2022). In this study, they computed 34 VIs and 10 the triangular greenness index (TGI) (Hunt et al., 2011) with field 
features based on average and standard deviation values to be used as measurements such as mineral nutrients and foot rot disease severity to 
input variables for analyses against 10 target nutritional status variables distinguish trees with disease. They found that TGI was different 
(encompassing NPK). They employed four machine learning techniques depending on disease or in combination of diseases and also explained 
(Random Forest, Support Vector Regression, Partial Least Square which factors were involved in these differences. Qiu et al. (Qiu et al., 
Regression, and Artificial Neural Network) finding that models for 2021) calculated different VIs from RGB images acquired through a UAV 
chlorophyll, nitrogen and calcium predictions were acceptable for and showed they are correlated with nitrogen nutrition index (NNI) for 
screening. The screening model based on Support Vector Regression for rice at different growth stages. Ballesteros et al. (2018) estimated crop 
nitrogen obtain a coefficient of determination ranging from 0.665 to biomass (which involves pest and weed status, soil quality, water stress, 
0.718. Similarly, Guerra-Hernández et al. (2021) acquired multispectral yield prediction, among others) from high-resolution RGB images ob-
images with a UAV to monitor health status in priority riparian forest. tained with a UAV. They extracted green canopy cover, crop height and 
They considered 34 remote sensing variables such as VIs, texture fea- canopy volume for predictor variables, finding strong correlation be-
tures from NDVI and digital surface model, among others. The objective tween canopy volume and dry leaf biomass and dry bulb biomass. A 
was to classify trees in four categories: asymptomatic, dead, and defo- method for monitoring health status of Eucalyptus pellita in a large-scale 
liation above and below 50% threshold. The classification using Random area through the usage of RGB images acquired with a UAV was pro-
Forest with four, three (asymptomatic, defoliated, dead) and two (alive, posed by Megat Mohamed Nazir et al., (2021). They generated an ortho- 
dead) classes yielded overall accuracies of 0.67, 0.72 and 0.91 respec- image and computed VARI-green (Gitelson et al., 2002) to classify four 
tively. By using three logistic models with leave-out cross validation levels of pest and disease: dead, severely infected, mildly infected and 
Fig. 1. Geographic location from the study area. a) Lima location (brown colored) related to Peru. b) La Molina location (brown colored) related to Lima. c) National 
Institute of Agrarian Innovation (INIA) and study area (pink box). d) Mosaicked image from study area in false color. Fill-colored rectangles group trees with some 
common features. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) 
3
I. Salazar-Reque et al.                                                                                                                                                              C  o  m   p  u t e  r s  a  n  d   E  l e c  t r o  n i c  s  i n   A  g  r i c u  l t u  r e  213 (2023) 108246
healthy trees. The VARI-green indices were verified by manual ground- N-7 has excess of nitrogen; P-19 has deficiency of phosphorus and K-12 
true values and by comparison with NDVI, showing a correlation of 0.73. is control of potassium (i.e., potassium in optimal conditions). To induce 
Ge et al. (2021) estimated PNC of rice using a fusion of VIs and color these statuses, macronutrients were provided in the amounts shown in 
moments from UAV-based RGB images. They used Partial Least Square Table 1 in three different dates: May 10th, September 6th and November 
Regression and Random Forest models and demonstrated that the fusion 15th. 
of VIs and color moments as input variables improve the accuracy of 
PNC estimation compared to VIs or color moments only. 3.2.2. Water variation experiment 
The last three rows in the study area were used to study the effect of 
3. Methodology different levels of water in the trees. We varied the level of water while 
keeping macronutrients in optimal conditions. Each row had a different 
3.1. Area of study level of water: 50% (black), 75% (gray) and control (sky blue) (see 
Fig. 1d and Table 1). Water pressure through the pipes was controlled to 
This study was conducted over a hectare of Hass Avocado at the induce the three different water statuses. 
National Institute of Agrarian Innovation located (INIA) in La Molina, 
Lima, Peru (see Fig. 1a-c). We analyzed 123 trees that were distributed 3.3. NDVI measurements 
in seven rows and twenty-one columns (pink box in Fig. 1d). The sep-
aration between trees was five meters approximately as show in Fig. 2. Normalized Difference Vegetation Index (NDVI) has been used 
Latitude and longitude coordinates from study area vertices are: widely in the literature to analyze crops. We captured in-field NDVI 
(− 12.074588, − 76.947483), (− 12.074266, − 76.946131), measurements using the GreenSeeker Handheld Crop Sensor. Measure-
(− 12.074826, − 76.946021) and (− 12.075139, − 76.947359). Trees ments were taken approximately 60 cm above the top of the tree with 
were six meter tall and three years old on average. The soil was loan with the help of a step ladder. 
an average pH of 7.55 and an electrical conductivity of 2.06 mS cm− 1 
(measured according to UNE 77308:2001 using a mixture of soil and 
water in the ratio 2 to 1). The organic matter value was 0.55. Soil 
phosphorus (P) and potassium (K) values were 18.7 ppm and 355 ppm Table 1 
respectively. These values were obtained from a laboratory analysis of Amount of macronutrients and water provided to trees per group and statuses.  
soil samples at different layers made in January 2020. 
Group (G) Status (s) N P K Water 
N Excess 120 kg ha− 1 60 kg ha− 1 80 kg ha− 1 100% 
3.2. Experimental design Control 80 kg ha− 1 60 kg ha− 1 80 kg ha− 1 100% 
Deficiency 0 kg ha− 1 60 kg ha− 1 60 kg ha− 1 100% 
3.2.1. Macronutrient variation experiment P Excess 80 kg ha− 1 100 kg ha− 1 80 kg ha− 1 100% 
The first three rows in the study area were used to study the effect of Control 80 kg ha− 1 60 kg ha− 1 80 kg ha− 1 100% 
different levels of macronutrients, either nitrogen (N), phosphorus (P) or Deficiency 80 kg ha− 1 0 kg ha− 1 60 kg ha− 1 100% 
− 1 − 1 − 1 
potassium (K) in the trees. We provided nutrients using fertigation as it K Excess 80 kg ha 60 kg ha 120 kg ha 100% 
Control 80 kg ha− 1 60 kg ha− 1 80 kg ha− 1 100% 
allowed us to vary the amount of macronutrients in a practical way. We Deficiency 80 kg ha− 1 60 kg ha− 1 0 kg ha− 1 100% 
varied the level of one macronutrient while keeping the others and Water Control 80 kg ha− 1 60 kg ha− 1 80 kg ha− 1 100% 
water in optimal conditions. Trees in each row were subdivided in three 75 80 kg ha− 1 60 kg ha− 1 80 kg ha− 1 75% 
statuses according to the level of the nutrient: excess (red), control (blue 50 80 kg ha− 1 60 kg ha− 1 80 kg ha− 1 50%  
sky) and deficiency (yellow) (see Fig. 1d). For example, a tree located in 
Fig. 2. Sequential images acquired from a single tree using UAV. Note that before any image processing we need to select the best acquired image (number 3 in 
this case). 
4
I. Salazar-Reque et al.                                                                                                                                                              C  o  m   p  u t e  r s  a  n  d   E  l e c  t r o  n i c  s  i n   A  g  r i c u  l t u  r e  213 (2023) 108246
3.4. Image acquisition First, we have to clean our image set per flight and tag them according to 
the group and status they belong to by performing image selection 
Images from the top of the trees were acquired using a Phantom Pro through a CNN. Next, we need to perform a proper crown segmentation 
V2 UAV from DJI (https://www.dji.com/phantom-4-pro-v2) equipped of the trees. Finally, leveraging these segmentations several features per 
with a SENTERA Double 4 K multispectral camera. This camera has two group, status and flight are extracted for comparison. These steps were 
lenses: one of them acquires RGB images and the other, NIR and RE performed through Python scripts. A summary of this procedure is 
bands together. In this study we are interested in the use of RGB images. shown in Fig. 3. 
The UAV flights followed a single grid path starting in tree N-1 and 
finishing in tree C-21 as shown in Fig. 1d. To do so, coordinates of each 3.5.1. Image selection based on CNN 
row extreme were set in the flight planning software considering a As mentioned before, our flight configuration allowed us to complete 
height of 7 m above the ground. UAV was configured to have an average the acquisition in a few minutes but with the compromise of having 
speed of 2 m/s and images were acquired every 1 s. Although, this unnecessary images (see Fig. 2). Thus, we need to clean our image set 
configuration implies to have more images than trees, we guarantee to and tag each image according to the group and status the tree belongs to. 
complete the desired path rapidly and to have at least one image of a To do so, we perform the following steps which are based on our pre-
complete tree with the least number of images per flight. A sample of vious work on tree image selection (Salazar-Reque et al., 2021): 
continuous images acquired under this configuration is shown in Fig. 2, 
note that as there are many images per tree we will need to select the A manual selection of “best images” per tree was made for one flight. 
best. The procedure to deal with image selection is described in section This is done only once and will be used as a reference in future 
3.5. flights. Manual selection means human selection and “best images” 
We conducted 21 flights over a period of seven months. Each flight stand for those images in which all or the most of the tree crown 
acquired an average of 700 georeferenced images (some of them will be appears, and in which it is as centered as possible. 
discarded). Dates, number of images as well as some flight conditions are An automatic pre-selection of images per tree was made based on 
reported in Table 2. proximity using geolocation information from images from step 1. 
We generate binary masks per each image from step 2 using a UNET 
(Ronneberger et al., 2015) like architecture trained for coarse crown 
3.5. Image processing segmentation of trees. 
Binary images from step 3 were used to discard images that do not 
For each flight we have to follow a procedure composed of three meet the criteria of a good acquisition (e.g., tree centered in the 
steps in order to generate a proper feature to compare between statuses. image). 
We select the best image available per tree by comparing binary 
Table 2 masks from images from step 4 with binary masks from the images of 
UAV flights information.  reference from step 1. As each tree is associated with a group, the 
Date DOY # of Flight conditions 
(t) images 
March 29 88 835 RH: 56, T: 24, WS: 2 
April 14 104 820 RH: 67, T: 22, WS: 3, LCC: 6, MCC: 0, 
HCC: 82 
April 28 118 760 T: 27, WS: 2.5, LCC: 1, MCC: 1, HCC: 41 
May 7 127 702 T: 25.5, WS: 3, LCC: 3, MCC: 13, HCC: 0 
May 12 132 805 T: 24, WS: 2, LCC: 13, MCC: 5, HCC: 20 
May 19 139 730 RH: 81, T: 23.5, WS: 3, LCC: 0, MCC: 0, 
HCC: 0 
May 26 146 750 RH: 78.5, T: 24, WS: 1, LCC: 13, MCC: 8, 
HCC: 0 
Jun 02 153 835 WS: 2, LCC: 59, MCC: 0, HCC: 7 
Jun 11 162 820 RH: 62, T: 19, WS: 3, LCC: 17, MCC: 3, 
HCC: 0 
Jun 16 167 795 RH: 73, T: 27, WS: 3, LCC: 2, MCC: 0, 
HCC: 2 
Jun 23 174 889 T: 18, WS: 3, LCC: 12, MCC: 0, HCC: 19 
July 02 183 803 RH: 54, T: 19.5, WS: 2, LCC: 43, MCC: 0, 
HCC: 1 
July 09 190 786 RH: 53, T: 22, WS: 2, LCC: 4, MCC: 0, 
HCC: 0 
July 14 195 708 – 
July 23 204 750 RH: 62, T: 20, WS: 3, LCC: 21, MCC: 0, 
HCC: 0 
August 05 217 770 RH: 76, T: 20.5, WS: 3, LCC: 55, MCC: 0, 
HCC: 0 
August 13 225 802 RH: 60, T: 20.5, WS: 3, LCC: 5, MCC: 1, 
HCC: 10 
August 19 231 815 – 
September 258 752 RH: 60, T: 19, WS: 3, LCC: 5, MCC: 1, 
15 HCC: 2 
September 267 798 RH: 62, T: 21, WS: 2, LCC: 10, MCC: 0, 
24 HCC: 6 
October 15 288 770 – 
October 24 302 821 RH: 62, T: 23, WS: 2, LCC: 62, MCC: 4, 
HCC: 4 
Note: RH: relative humidity, T: ambient temperature in Celsius degrees, WS: 
wind speed, LCC: percentage of low cloud cover, MCC: percentage of medium 
cloud cover, HCC: percentage of high cloud cover. Fig. 3. Image processing flowchart.  
5
I. Salazar-Reque et al.                                                                                                                                                              C  o  m   p  u t e  r s  a  n  d   E  l e c  t r o  n i c  s  i n   A  g  r i c u  l t u  r e  213 (2023) 108246
final outcome is one image per tree labeled according to their cor- 3.6. Comparison metric 
responding group. 
To compare the differentiation between nutritional and water sta-
If we repeat this process in every flight, we ended up with a set of tuses that each VI indicates, we proposed the Mean Distance between 
images ordered by tree and date. Vegetation Indices (MDVI) as a metric (see Equation 3). 
1∑
3.5.2. Crown segmentation MDVIVI(s1, s2) = |zs1 (t) − zs2 (t)| (3) 
Previous binary masks were coarse crown segmentations that were T t
useful for tree image comparison with reference images from another In Equation 3, z(t) is the value of VI at time t. The value of VI at time t 
flight. However, when viewed in detail, these binary masks also include for a certain status s is denoted as zs(t), so the distance between VIs for 
shadows or poorly lit areas of the trees that can generate artifacts in the statuses s1 and s2 at time t is the absolute value of the difference between 
features we are intended to generate. To reduce these artifacts, we zs1 (t) and zs2 (t). Finally, to compute the mean value of this distance, the 
eliminate these dark areas by means of the Otsu’s method (Otsu, 1979) absolute value of the difference (|zs1 (t) − zs2 (t)|) for each time t is sum-
on each of the color components of the RGB images clipped to the med together and divided by the total number of time instants T. This 
rectangle with the maximum area that is inside the segmented tree. allows comparisons between studies with different temporal resolutions. 
Otsu’s method maximizes the means and minimizes the variance of the MDVI basically computes the area of the difference between the tem-
distribution of pixel values by assuming that there are two groups poral evolution (signals) of VIs for the compared statuses per time 
(leaves and dark areas in our case). Thus, we reduce the influence of instant, and therefore this is a measurement of how similar or different 
dark areas in the final features. the temporal evolution of VIs for each status are. 
3.5.3. Feature extraction 4. Results 
Using our previous crown segmentations, we can create a set of Np 
square patches per group G in a given time t. Then, for each of these 4.1. Comparison of VIs 
patches we compute its density function h(i)G (x; t) for a random variable X 
where i = 1, ...,Np. Our objective here is to generate a signal zG(t) that Taking into consideration the groups, denoted by G, Equation 3 can 
represents the temporal evolution of group G. We compute this signal as be formulated as follow: 
the expected value over X as follows: 
G 1∑MDVI (s , s G GVI 1 2) = |z (t) − z (t)| (3) zG(t) = E[fG(x; t)] (1)  T
s1 s2
t
G
where fG(x; t) represents the probability density function of a group G in For each VI in Table 3, we computed a value of MDVIVI(s1, s2) where 
a given time t. This is computed as the average of the previous density G represented the group and was either {N, P, K or W} and s1 and s2 
functions h(i) x t . represented the statuses and were chosen from {Excess, Control or G ( ; ) Deficiency} for macronutrients or from {50, 75 or Control} for water 
1 ∑Np (W). This computation also involves to calculate zG(i) s (t) for each DOY (see fG(x; t) = hG (x; t) (2) N Table 2). It is important to mention that MDVI computation for a group p i=1
involves multiple trees, so zGs (t) is the average value of VI for all the trees 
The random variable X is defined by the pixel-wise combination belonging to group G at time t for a certain status s. 
between red, green and blue channels according to the VI of interest. The The MDVI values for each combination of groups, statuses and VIs 
VIs explored in this work are the same used in (Qiu et al., 2021) which are summarized in the plots shown in Figs. 4 and 5. Better differentiation 
we summarized in Table 3. between statuses is achieved for nitrogen because most MDVI values are 
higher for this group (MDVINVI > MDVI
P N K
VI and MDVIVI > MDVIVI) as 
shown by red lines being in the outermost part of the radar plot 
Table 3 compared to blue and green lines. Similar MDVI values (but usually 
Vegetation indexes explored.  lower than nitrogen) were obtained in the phosphorus group. For po-
tassium, the MDVI values for any VIs were always lower than for ni-
Vegetation Index X References 
trogen and phosphorus. In general, this shows that it is harder to 
Red Green Ratio Index (RGRI) r/g (Gamon and Surfus, distinguish between groups for this macronutrient, especially when a 
1999) 
Blue Green Ratio Index (BGRI) b g (Yue et al., 2018) deficiency status is involved. /
Green Leaf Algorithm (GLA) (2g − r − b)/(2g + r + (Louhaichi et al., When analyzing only the results for nitrogen and phosphorus groups, 
b) 2001) we note that higher values of MDVI were always given by MGRVI. This 
Green Leaf Index (GLI) (2g − r + b)/(2g + r + (Qiu et al., 2021) indicates that this VI better differentiates between statuses for these 
b) macronutrients. Not surprisingly, the Excess vs Deficiency statuses 
Green Red Vegetation Index (g − r)/(g + r) (Tucker, 1979) 
(GRVI) comparison was easy to differentiate than the others as can be noted by 
Modified Green Red Vegetation (g2 − r2)/(g2 + r2) (Bendig et al., the higher MDVI values reported. On the other hand, the Deficiency and 
Index (MGRVI) 2015) Control statuses comparison had the smallest values indicating that lack 
Excess Green minus Excess Red (2g − r − b) − (1.4r − g) (Meyer and Neto, of nitrogen or phosphorus macronutrients are difficult to distinguish 
(ExGR) 2008) 
Excess Red Vegetation Index (ExR) 1.4r − g (Meyer et al., 1998) from controls. 
Excess Blue Vegetation Index 1.4b − g (Mao et al., 2003) A similar analysis can be done for the water experiment results. 
(ExB) Again, MGRVI has the higher values of MDVI (see Figs. 4 and 5) but only 
Excess Green Vegetation Index 2g − r − b (Woebbecke et al., for 50% vs CONTROL and 50% vs 75% comparisons. In the case of of 
(ExG) 1995) 75% vs CONTROL comparison we can expect that these statuses to 
Visible Atmospherically Resistant (g − r)/(g + r + b) (Gitelson et al., 
Index (VARI) [] 2002) behave very similar as MDVI values for this comparison are very low for 
Red Green Blue Vegetation Index (g2 − br2)/(g2 + br2) (Bendig et al., all the VIs. 
(RGBVI) 2015) 
Note: All operations are pixel-wise operations. 
6
I. Salazar-Reque et al.                                                                                                                                                              C  o  m   p  u t e  r s  a  n  d   E  l e c  t r o  n i c  s  i n   A  g  r i c u  l t u  r e  213 (2023) 108246
Fig. 4. MDVI values for each combination of groups, statuses and VIs. First row: Comparison between excess and control statuses (left), comparison between 
deficiency and control statuses (middle) and comparison between excess and deficiency statuses (right). Second row: Comparison between water50 and control 
statuses (left), comparison between water75 and control statuses (middle) and comparison between water50 and water75 statuses (right). 
Fig. 5. Bar plots with MDVI values for multiple VIs. Each row represents a group, and each column represents a pair of statuses being compared.  
7
I. Salazar-Reque et al.                                                                                                                                                              C  o  m   p  u t e  r s  a  n  d   E  l e c  t r o  n i c  s  i n   A  g  r i c u  l t u  r e  213 (2023) 108246
4.2. Temporal analysis of VIs for nitrogen group Table 4 
Correlation between NDVI and RGB-based vegetation indices for different ni-
As the MDVI values were higher for nitrogen group, we decided to trogen statuses.  
analyze the temporal evolution of NDVI compared with three VIs that VI Deficiency Control Excess 
allow good (MGRVI), normal (GLA) and poor (ExB) differentiation GLA 0.6585 0.7921 0.4775 
(according to MDVI) for this group (see Fig. 6). This selection is due to MGRVI 0.8086 0.8488 0.6289 
we wanted to perform a posterior correlation study with NDVI, so the ExB − 0.2863 − 0.4269 − 0.2909  
possibility of unexpected correlations had to be considered. For the 
computation of the distributions, we have used 50 bins in the range of 
[-0.2, 0.8]. We used this range because VIs varied between those values. VIs align with previous studies in the field. Nitrogen has demonstrated 
NDVI values were computed from multispectral images acquired in the the highest differentiating power among the macronutrients analyzed, 
same flight conditions with a Sentera Double 4 K Sensor Multispectral which is consistent with the strongest influence this macronutrient has 
camera (https://sentera.com/products/fieldcapture/sensors/double-4 on tree growth and development as reported by Jiaying et al. (2022) and 
k/). Silber et al. (2022). 
In Figs. 4 and 5, we presented radar and bar plots that demonstrate 
the differentiation between macronutrient statuses based on MDVI 
4.3. Correlation with NDVI measurements values for each VI. Notably, MGRVI, BGRI, RGRI, and VARI consistently 
outperformed other indices in differentiating between the nitrogen and 
The results of correlation study (see Table 4) demonstrate a positive phosphorus groups, particularly for nitrogen. These findings are 
relationship between MGRVI for nitrogen group and NDVI due to the consistent with the results reported in previous studies (Qiu et al., 2021; 
correlation coefficients in the three statuses range from 0.63 to 0.85. Megat Mohamed Nazir et al., 2021). 
These values were computed considering the fruit and harvesting period Interestingly, the worst-performing indices (ExB, ExR, and ExG) were 
(approximately from DOY 163 to DOY 259). those computed using linear operations of color components, while best 
VIs always involved a non-linear operation (a division). This might be 
5. Discussion due to the fact that images acquired by UAV are prone to light variations 
and division between color components are more robust to this kind of 
The purpose of this study was to investigate the relationship between artifacts than linear combination, as discussed by (Barbedo, 2017) in the 
NPK macronutrients status and VIs as well as water status and VIs. In this context of disease segmentation over leaves. 
discussion, we will compare our results with findings from other au- As anticipated, the MDVI values were highest when comparing the 
thors, providing a more comprehensive analysis of the obtained results excess and deficiency groups, demonstrating their strong discriminatory 
and their implications. Our findings regarding macronutrient status and 
Fig. 6. Comparison of temporal evolution of NDVI, MDVIN , MDVINMGRVI GLA, and MDVI
N
ExB for trees under different nitrogen statuses.  
8
I. Salazar-Reque et al.                                                                                                                                                              C  o  m   p  u t e  r s  a  n  d   E  l e c  t r o  n i c  s  i n   A  g  r i c u  l t u  r e  213 (2023) 108246
capability in identifying extreme nutrient statuses. In contrast, the Acknowledgement 
comparison between the deficiency and control groups resulted in lower 
values, indicating the difficulty in distinguishing nutrient deficiencies To the World Bank and “PROCIENCIA” (formerly FONDECYT), an 
from the normal range. This finding poses a challenge in practical ap- initiative of CONCYTEC, for the funds allocated to the project under the 
plications where the detection of deficiencies is of significant interest. contract N◦ 097-2018-FONDECYT-BM-IADT-AV. 
Notably, during the fruit production and harvest period (DOY 163 to 
DOY 259), the density distributions for the deficiency status in nitrogen References 
and phosphorus groups shifted to higher values compared to the control 
and excess statuses. This observation can be explained by the heavy Adan, M., Abdel-Rahman, E.M., Gachoki, S., Muriithi, B.W., Lattorff, H.M.G., Kerubo, V., 
utilization of available nutrients in the crop during this period. Thus, the Landmann, T., Mohamed, S.A., Tonnang, H.E., Dubois, T., 2021. Use of earth 
observation satellite data to guide the implementation of integrated pest and 
increased differentiation during this period further emphasizes the pollinator management (IPPM) technologies in an avocado production system. 
effectiveness of VIs in capturing nutrient variations. Remote Sens. Appl.: Soc. Environ. 23, 100566. 
The strong correlation between MGRVI for nitrogen group and NDVI Ballesteros, R., Ortega, J.F., Hernandez, D., Moreno, M.A., 2018. Onion biomass 
supports the idea that MGRVI serves as a suitable indicator of vegetation monitoring using uav-based rgb imaging. Precis. Agric. 19, 840–857. Barbedo, J.G.A., 2019. A review on the use of unmanned aerial vehicles and imaging 
health and biomass estimation. Bendig et al. (2015), which focuses on sensors for monitoring and assessing plant stresses. Drones 3 (2), 40. 
biomass prediction in a summer barley experiment, also highlights the J. G. A. Barbedo, A new automatic method for disease symptom segmentation in digital 
potential of visible band vegetation indices, including MGRVI, for photographs of plant leaves 147 (2) (2017) 349–364. Bayomy, H.M., Alamri, E.S., Rozan, M.A., 2023. Effect of roasting hass avocado kernels 
biomass estimation. This concurrence with a similar study reinforces the on nutritional value and volatile compounds. Processes 11 (2), 377. 
notion that MGRVI, along with other visible band indices, holds promise Bendig, J., Yu, K., Aasen, H., Bolten, A., Bennertz, S., Broscheit, J., Gnyp, M.L., 
for various applications in vegetation monitoring and assessment. Bareth, G., 2015. Combining UAV-based plant height from crop surface models, visible, and near infrared vegetation indices for biomass monitoring in barley. Int. J. 
Overall, our study aligns with the findings of Ge et al. (2021) and Qiu Appl. Earth Obs. Geoinf. 39, 79–87. 
et al. (2021) by highlighting the importance of VIs derived from UAV- Buschmann, C., Nagel, E., 1993. In vivo spectroscopy and internal optics of leaves as 
based RGB images in assessing macronutrient status. The correlation basis for remote sensing of vegetation. Int. J. Remote Sens. 14 (4), 711–722. Chungcharoen, T., Donis-Gonzalez, I., Phetpan, K., Udompetaikul, V., Sirisomboon, P., 
between MGRVI and nutrient status observed in our study reinforces the Suwalak, R., 2022. Machine learning-based prediction of nutritional status in oil 
feasibility of using UAV-based remote sensing systems for cost-effective palm leaves using proximal multispectral images. Comput. Electron. Agric. 198, 
and efficient monitoring of crop growth traits and nutrient parameters. 107019. 
Colombo, R., Papetti, A., 2019. Avocado (Persea americana mill.) by-products and their 
impact: from bioactive compounds to biomass energy and sorbent material for 
6. Conclusion removing contaminants. a review. Int. J. Food Sci. Technol. 54 (4), 943–951. 
Corrales-García, J.E., del Rosario García-Mateos, M., Martínez-López, E., Barrientos- 
Priego, A.F., Ybarra-Moncada, M.C., Ibarra-Estrada, E., Méndez-Zúñiga, S.M., 
We showed that a temporal analysis of VIs computed from aerial RGB Becerra-Morales, D., 2019. Anthocyanin and oil contents, fatty acids profiles and 
images can be used to differentiate between macronutrient and water antioxidant activity of mexican landrace avocado fruits. Plant Foods Hum. Nutr. 74, 
statuses for Hass avocado trees. Moreover, we proposed a comparison 210–215. 
metric called MDVI that allows better differentiation among statuses for Daughtry, C.S., Walthall, C., Kim, M., De Colstoun, E.B., Mc- Murtrey Iii, J., 2000. Estimating corn leaf chlorophyll concentration from leaf and canopy reflectance. 
different groups and VIs. Over eleven VIs, MGRVI has been found to be Remote Sens. Environ. 74 (2), 229–239. 
the best on this purpose based on a comparison metric that measures the Dezordi, L.R., Aquino, L.A.d., Aquino, R.F.B.d.A., Clemente, J.M., Assunção, N.S., 2016. 
difference of temporal signals computed from the VIs. MGRVI better Diagnostic methods to assess the nutritional status of the carrot crop. Rev. Bras. Ciênc. Solo 40. 
differentiates between statuses for nitrogen and phosphorus groups in a Gamon, J., Surfus, J., 1999. Assessing leaf pigment content and activity with a 
period of time that coincides with the fruit production and harvesting reflectometer. New Phytol. 143 (1), 105–117. 
period of the Hass avocado trees. In addition, MGRVI shows a positive P. Gaona, L. Vásquez, C. Morales, W. Viera, P. Viteri, A. Sotomayor, L. Medina, P. Mejía, Y. Cartagena, Efecto de dos niveles de nitrogeno y potasio aplicados por fertirriego 
strong relationship with NDVI. Future research will explore the use of en las variables de crecimiento y concentración de macro y micronutrientes en 
MGRVI in studies using only RGB images as variable to correlate with plantas de aguacate (persea americana mill.) var. hass, in: ECUADOR ES CALIDAD, 
macronutrient and water status obtained from laboratory analysis. Vol. 7, 2020, 315 pp. 41–48. https://doi.org/10.36331/revista.v7i2.114. 
Garza, B.N., Ancona, V., Enciso, J., Perotto-Baldivieso, H.L., Kunta, M., Simpson, C., 
2020. Quantifying citrus tree health using true color UAV images. Remote Sens. 
CRediT authorship contribution statement (Basel) 12 (1), 170. 
Ge, H., Xiang, H., Ma, F., Li, Z., Qiu, Z., Tan, Z., Du, C., 2021. Estimating plant nitrogen 
concentration of rice through fusing vegetation indices and color moments derived 
Itamar Salazar-Reque: Conceptualization, Methodology, Software, from UAV-RGB images. Remote Sens. (Basel) 13 (9), 1620. 
Formal analysis, Investigation, Data curation, Writing – original draft, Gitelson, A.A., Kaufman, Y.J., Stark, R., Rundquist, D., 2002. Novel algorithms for 
Writing – review & editing, Visualization. Daniel Arteaga: Conceptu- remote estimation of vegetation fraction. Remote Sens. Environ. 80 (1), 76–87. 
alization, Methodology, Investigation, Data curation, Writing – original Gitelson, A.A., Gritz, Y., Merzlyak, M.N., 2003. Relationships between leaf chlorophyll content and spectral reflectance and algorithms for non-destructive chlorophyll 
draft, Writing – review & editing, Visualization. Fabiola Mendoza: assessment in higher plant leaves. J. Plant Physiol. 160 (3), 271–282. 
Conceptualization, Investigation, Data curation, Validation. Maria Gitelson, A.A., Vi ̃na, A., Ciganda, V., Rundquist, D.C., Arkebauer, T.J., 2005. Remote 
Elena Rojas: Conceptualization, Investigation, Resources, Supervision. estimation of canopy chlorophyll content in crops. Geophys. Res. Lett. 32 (8). Gitelson, A.A., Keydan, G.P., Merzlyak, M.N., 2006. Three-band model for noninvasive 
Jonell Soto: Investigation, Resources, Supervision, Funding acquisition. estimation of chlorophyll, carotenoids, and anthocyanin contents in higher plant 
Samuel Huaman: Conceptualization, Methodology, Resources, Super- leaves. Geophys. Res. Lett. 33 (11). 
vision. Guillermo Kemper: Conceptualization, Investigation, Re- Guerra-Hernández, J., Díaz-Varela, R.A., Ávarez-González, J.G., Rodríguez-González, P. M., 2021. Assessing a novel modelling approach with high resolution UAV imagery 
sources, Supervision, Project administration, Funding acquisition. for monitoring health status in priority riparian forests. Forest Ecosystems 8 (1), 
1–21. 
Declaration of Competing Interest Haboudane, D., Miller, J.R., Tremblay, N., Zarco-Tejada, P.J., Dextraze, L., 2002. Integrated narrow-band vegetation indices for prediction of crop chlorophyll content 
for application to precision agriculture. Remote Sens. Environ. 81 (2–3), 416–426. 
The authors declare that they have no known competing financial Hunt Jr, E.R., Daughtry, C., Eitel, J.U., Long, D.S., 2011. Remote sensing leaf chlorophyll 
interests or personal relationships that could have appeared to influence content using a visible band index. Agron. J. 103 (4), 1090–1099. 
Incrocci, L., Massa, D., Pardossi, A., 2017. New trends in the fertigation management of 
the work reported in this paper. irrigated vegetable crops. Horticulturae 3 (2), 37. 
Janoušek, J., Jambor, V., Marcoň, P., Dohnal, P., Synková, H., Fiala, P., 2021. Using 
Data availability UAV-based photogrammetry to obtain correlation between the vegetation indices 
and chemical analysis of agricultural crops. Remote Sens. (Basel) 13 (10), 1878. 
Jay, S., Gorretta, N., Morel, J., Maupas, F., Bendoula, R., Rabatel, G., Dutartre, D., 
Data will be made available on request. Comar, A., Baret, F., 2017. Estimating leaf chlorophyll content in sugar beet canopies 
9
I. Salazar-Reque et al.                                                                                                                                                              C  o  m   p  u t e  r s  a  n  d   E  l e c  t r o  n i c  s  i n   A  g  r i c u  l t u  r e  213 (2023) 108246
using millimeter- to centimeter-scale reflectance imagery. Remote Sens. Environ. Qiu, Z., Ma, F., Li, Z., Xu, X., Ge, H., Du, C., 2021. Estimation of nitrogen nutrition index 
198, 173–186. in rice from UAV RGB images coupled with machine learning algorithms. Comput. 
Jiaying, M., Tingting, C., Jie, L., Weimeng, F., Baohua, F., Guangyan, L., Hubo, L., Electron. Agric. 189, 106421 https://doi.org/10.1016/j.compag.2021.106421. 
Juncai, L., Zhihai, W., Longxing, T., Guanfu, F., 2022. Functions of nitrogen, Robson, A., Rahman, M.M., Muir, J., 2017. Using worldview satellite imagery to map 
phosphorus and potassium in energy status and their influences on rice growth and yield in avocado (persea americana): A case study in bundaberg, australia. Remote 
development. Rice Sci. 29 (2), 166–178. Sens. (Basel) 9 (12), 1223. 
Kosińska, A., Karamać, M., Estrella, I., Hernández, T., Bartolomé, B., Dykes, G.A., 2012. Ronneberger, O., Fischer, P., Brox, T., 2015. U-net: Convolutional networks for 
Phenolic compound profiles and antioxidant capacity of persea americana mill - biomedical image segmentation. In: Navab, N., Hornegger, J., Wells, W.M., 
peels and seeds of two varieties. J. Agric. Food Chem. 60 (18), 4613–4619. PMID: Frangi, A.F. (Eds.), Medical Image Computing and Computer-Assisted Intervention – 
22494370.  MICCAI 2015. Springer International Publishing, Cham, pp. 234–241. 
Louhaichi, M., Borman, M.M., Johnson, D.E., 2001. Spatially located platform and aerial Roper, J.M., Garcia, J.F., Tsutsui, H., 2021. Emerging technologies for monitoring plant 
photography for documentation of grazing impacts on wheat. Geocarto Int. 16 (1), health in vivo. ACS Omega 6 (8), 5101–5107. 
65–70. Rouse, J.W., Haas, R.H., Schell, J.A., Deering, D.W., et al., 1974. Monitoring vegetation 
Lu, Q.-Y., Arteaga, J.R., Zhang, Q., Huerta, S., Go, V.L.W., Heber, D., 2005. Inhibition of systems in the great plains with erts. NASA Spec. Publ 351 (1), 309. 
prostate cancer cell growth by an avocado extract: role of lipid-soluble bioactive Saavedra, J., Córdova, A., Navarro, R., Díaz-Calderón, P., Fuentealba, C., Astudillo- 
substances. J. Nutr. Biochem. 16 (1), 23–30. Castro, C., Toledo, L., Enrione, J., Galvez, L., 2017. Industrial avocado waste: 
Lu, N., Wang, W., Zhang, Q., Li, D., Yao, X., Tian, Y., Zhu, Y., Cao, W., Baret, F., Liu, S., Functional compounds preservation by convective drying process. J. Food Eng. 198, 
et al., 2019. Estimation of nitrogen nutrition status in winter wheat from unmanned 81–90. 
aerial vehicle based multi-angular multispectral imagery. Front. Plant Sci. 10, 1601. Salazar-Reque, I., Arteaga, D., Huaman, K.G., Huamán Bustamante, S., 2021. A CNN- 
Mao, W., Wang, Y., Wang, Y., 2003. Real-time detection of between-row weeds using based algorithm for selecting tree-of-interest images acquired by UAV. In: 2021 IEEE 
machine vision. In: 2003 ASAE Annual MeetIng. American Society of Agricultural International Conference on Machine Learning and Applied Network Technologies 
and Biological Engineer, p. 1. (ICMLANT), pp. 1–6. https://doi.org/10.1109/ICMLANT53170.2021.9690556. 
Megat Mohamed Nazir, M.N., Terhem, R., Norhisham, A.R., Mohd Razali, S., Meder, R., Salgadoe, A.S.A., Robson, A.J., Lamb, D.W., Dann, E.K., Searle, C., 2018. Quantifying the 
2021. Early monitoring of health status of plantation-grown eucalyptus pellita at severity of phytophthora root rot disease in avocado trees using image analysis. 
large spatial scale via visible spectrum imaging of canopy foliage using unmanned Remote Sens. (Basel) 10 (2), 226. 
aerial vehicles. Forests 12 (10), 1393. Sharifi, A., 2020. Remotely sensed vegetation indices for crop nutrition mapping. J. Sci. 
Meyer, G., Mehta, T., Kocher, M., Mortensen, D., Samal, A., 1998. Textural imaging and Food Agric. 100 (14), 5191–5196. 
discriminant analysis for distinguishingweeds for spot spraying. Trans. ASAE 41 (4), Silber, A., Naor, A., Cohen, H., Bar-Noy, Y., Yechieli, N., Levi, M., Noy, M., Peres, M., 
1189–1197. Duari, D., Narkis, K., et al., 2018. Avocado fertilization: Matching the periodic 
Meyer, G.E., Neto, J.C., 2008. Verification of color vegetation indices for automated crop demand for nutrients. Sci. Hortic. 241, 231–240. 
imaging applications. Comput. Electron. Agric. 63 (2), 282–293. Silber, A., Goldberg, T., Shapira, O., Hochberg, U., 2022. Nitrogen uptake and 
Noguera, M., Aquino, A., Ponce, J.M., Cordeiro, A., Silvestre, J., Arias-Calderón, R., da macronutrients distribution in mango (Mangifera indica L. cv. Keitt) trees. Plant 
Encarnação Marcelo, M., Jordão, P., Andújar, J.M., 2021. Nutritional status Physiol. Biochem. 181, 23–32. 
assessment of olive crops by means of the analysis and modelling of multispectral Sims, N.C., Culvenor, D., Newnham, G., Coops, N.C., Hopmans, P., 2013. Towards the 
images taken with UAVs. Biosyst. Eng. 211, 1–18. operational use of satellite hyperspectral image data for mapping nutrient status and 
Nyakang’i, C.O., Ebere, R., Marete, E., Arimi, J.M., 2023. Avocado production in kenya fertilizer requirements in australian plantation forests. IEEE J. Sel. Top. Appl. Earth 
in relation to the world, avocado by-products (seeds and peels) functionality and Obs. Remote Sens. 6 (2), 320–328. 
utilization in food products. Appl. Food Res., 100275 Tucker, C.J., 1979. Red and photographic infrared linear combinations for monitoring 
Ong, E.S., Low, J., Tan, J.C.W., Foo, S.Y., Leo, C.H., 2022. Valorization of avocado seeds vegetation. Remote Sens. Environ. 8 (2), 127–150. 
with antioxidant capacity using pressurized hot water extraction. Sci. Rep. 12 (1), van Dijk, M., Morley, T., Rau, M.L., Saghai, Y., 2021. A meta-analysis of projected global 
13036. food demand and population at risk of hunger for the period 2010–2050. Nature 
Osco, L.P., Junior, J.M., Ramos, A.P.M., Furuya, D.E.G., Santana, D.C., Teodoro, L.P.R., Food 2 (7), 494–501. 
Gonçalves, W.N., Baio, F.H.R., Pistori, H., C. A. d. S. Junior,, et al., 2020. Leaf Wang, W., Bostic, T.R., Gu, L., 2010. Antioxidant capacities, procyanidins and pigments 
nitrogen concentration and plant height prediction for maize using UAV-based in avocados of different strains and cultivars. Food Chem. 122 (4), 1193–1198. 
multispectral imagery and machine learning techniques. Remote Sens. (Basel) 12 Woebbecke, D.M., Meyer, G.E., Von Bargen, K., Mortensen, D.A., 1995. Color indices for 
(19), 3237. weed identification under various soil, residue, and lighting conditions. Trans. ASAE 
Otsu, N., 1979. A threshold selection method from gray-level histograms. IEEE Trans. 38 (1), 259–269. 
Syst. Man Cybern. 9 (1), 62–66. Yue, J., Feng, H., Jin, X., Yuan, H., Li, Z., Zhou, C., Yang, G., Tian, Q., 2018. 
Plett, D.C., Ranathunge, K., Melino, V.J., Kuya, N., Uga, Y., Kronzucker, H.J., 2020. The A comparison of crop parameters estimation using images from uav-mounted 
intersection of nitrogen nutrition and water use in plants: new paths toward snapshot hyperspectral sensor and high-definition digital camera. Remote Sens. 
improved crop productivity. J. Exp. Bot. 71 (15), 4452–4468. (Basel) 10 (7), 1138. 
Prado Osco, L., Marques Ramos, A.P., Roberto Pereira, D., Akemi Saito Moriya, É., Zha, H., Miao, Y., Wang, T., Li, Y., Sun, J.W., Feng, Z., Kusnierek, K., 2020. Improving 
Nobuhiro Imai, N., Takashi Matsubara, E., Estrabis, N., de Souza, M., Marcato unmanned aerial vehicle remote sensing-based rice nitrogen nutrition index 
Junior, J., Gonçalves, W.N., et al., 2019. Predicting canopy nitrogen content in prediction with machine learning. Remote Sens. (Basel) 12 (2), 215. 
citrus-trees using random forest algorithm associated to spectral vegetation indices Zhang, Y., Su, Z., Shen, W., Jia, R., Luan, J., 2016. Remote monitoring of heading rice 
from UAV-imagery. Remote Sens. (Basel) 11 (24), 2925. growing and nitrogen content based on UAV images. International Journal of Smart 
Home 10 (7), 103–114. 
10