Introduction

Potato (Solanum tuberosum L.) growth monitoring and tuber yield prediction are of utmost importance for effective management of crops and planning of farm activities, such as harvesting, storage, distribution, and marketing logistics (Stone and Meinke 2005; Van der Velde and Nisini 2019). Moreover, monitoring potato crop development during the growing season allows for an adaptive management of fertilizers, irrigation, and pests and diseases (van Evert et al. 2012; Cucho-Padin et al. 2020; Gold et al. 2020).

Crop growth monitoring and yield prediction through ground-based observations and destructive sampling during the growing season tend to be costly, time-consuming, and prone to errors (Basso and Liu 2019; Tiedeman et al. 2022). As such, process-based crop models have been developed to simplify growth and yield prediction processes. Well-calibrated crop models can provide reliable estimates of potato growth and tuber yield before harvest (Raymundo et al. 2014). However, calibration for local conditions is necessary to accommodate the spatial variability in soil and weather conditions, crop variety, and management practices, which require a large amount of data input (Boote et al. 1996). These requirements can result in simulation uncertainties if input data are inaccurate or incomplete (Hoogenboom et al. 2019).

Recently, remote sensing technology has been adopted to monitor crop growth and predict yield during the season, and the spectral reflectance of green plants has been related to crop growth variables such as leaf area index (LAI), canopy cover, biomass, leaf chlorophyll content (LCC), and yield (Haboudane et al. 2002; Al-Gaadi et al. 2016; Tenreiro et al. 2021). Various vegetation indices (VIs), such as the normalized difference vegetation index (NDVI), weighted difference vegetation index (WDVI), enhanced vegetation index (EVI), red-edge inflection point (REIP), and ratio vegetation index (RVI) have been derived from remote sensor observations and used as proxies for plant growth and productivity (Prasad et al. 2006; Herrmann et al. 2011; Xue and Su 2017). Although remote sensing has been widely used in crop monitoring, there is a lack of literature regarding the appropriate phenological growth stage for acquiring remote sensing imagery and suitable VIs for optimal potato growth monitoring and tuber yield prediction.

This paper presents a systematic literature review of published original research on the application of remote sensing in potato growth monitoring and yield prediction. The objective of this study was to address the gaps in the existing literature by answering the following research questions:

  • What are the most suitable remote sensing techniques for potato growth monitoring and tuber yield prediction?

  • What are the most widely used vegetation indices to monitor potato growth and predict yield?

  • What is the most appropriate phenological stage for accurate potato yield prediction?

The remainder of this paper is organized as follows. Sect. "Overview of Remote Sensing Applications for Potato" provides an overview of the remote sensing applications for potato growth monitoring and yield prediction. Sect. "Research Methodology" describes the methodology used in the systematic review. Sect. "Results and Discussion" presents the results and discussion, and Sect. "Conclusions" presents the conclusions.

Overview of Remote Sensing Applications for Potato

Remote sensing information can be retrieved using various sensors, such as cameras, video recorders, multispectral and hyperspectral scanners, mounted on satellites, airplanes, unmanned aerial vehicles (UAVs), and ground-based platforms. The application of remote sensing for potato crop monitoring can be categorized into three main areas: growth monitoring, vegetation condition status monitoring, and tuber yield prediction.

Growth Monitoring

Remote sensing is based on acquiring electromagnetic wave reflectance from vegetation surfaces. The reflected light from vegetation surfaces depends on factors such as plant type, growth stage, water content and intrinsic tissue factors. Reflectance within the ultraviolet (10 – 380 nm), the visible region (450 – 750 nm) and the near infrared wave band (850 – 1100 nm), are prominent in agricultural applications (Xue and Su 2017). The change in reflectance at different wavelengths during the different crop growth stages is used to derive VIs that are related to canopy state variables, such as aboveground biomass (AGB), LAI, ground canopy cover, plant height, and vigour assessment (Delegido et al. 2008). The NDVI is the most popular VI related directly or indirectly to growth variables through regression or machine learning (ML) models (Peng et al. 2021a; Tenreiro et al. 2021).

Vegetation Condition Status

Remote sensing has been employed to estimate the leaf nitrogen (N) and LCC of potato crops (Clevers and Kooistra 2012; Kooistra and Clevers 2016). Plant health and vigour are used to assess the plant vegetation condition status. Healthy and vigorous plants usually exhibit rapid emergence, early ground coverage, and high concentrations of leaf N and LCC (Ter Steege et al. 2005; Basu and Groot 2023). Furthermore, leaf N and LCC can serve as indicators of plant nutritional status, photosynthetic rate, and biomass production (Clevers and Kooistra 2012; Kooistra and Clevers 2016). Therefore, in-season measurement of leaf N and LCC using remote sensing can be of great assistance in timely N management and optimising N use efficiency by the crop (van Evert et al. 2012).

Remote sensing has been used to monitor and assess the incidence and severity of diseases in potatoes (Couture et al. 2018; Duarte-Carvajalino et al. 2018; Gold et al. 2020). According to Polder et al. (2019), there is a significant difference between the reflectance of healthy and diseased potato leaves. Plant pathogens (fungi, bacteria, and viruses) attack the epidermal and mesophyll cells of leaves, which affects the biophysical and biochemical properties of crop vegetation (Couture et al. 2018). Disease infection therefore influences the spectral reflectance of vegetation, which in turn affects spectral metrics, such as spectral distance and VIs (Griffel et al. 2018). Simultaneously, spectral reflectance is affected by factors such as water and nutrient stress, natural plant senescence, variability of canopy structure, and spectral resolution of the sensor, which limits the accuracy of disease assessment using remote sensing (Franceschini et al. 2019). Despite these limitations, several studies reported successful disease assessment using high spectral resolution imagery and spectral reflectance classification techniques, including parametric and non-parametric modelling techniques, as well as classification methods, such as quadratic discriminant analysis, support vector machine (SVM), and classification trees (Duarte-Carvajalino et al. 2018; Franceschini et al. 2019). Griffel et al. (2018) and Couture et al. (2018) used support vector machine classification methods and found that potato plants infected with potato Virus Y (Potyviridae PVY) had significantly lower reflectance values between 700 – 1300 nm wavelengths than healthy plants. Other foliar diseases including potato early blight (Alternaria solani) (Van De Vijver et al. 2020) and late blight (Phytophthora infestans) (Franceschini et al. 2017a, b; Gold et al. 2020; Hou et al. 2022) have been monitored using remotely sensed data.

Remote sensing has been used to assess plant water status and water stress in crops (Gerhards et al. 2016). This can be achieved through the estimation of leaf water content using indicators such as water potential, relative water content, equivalent water thickness of leaves and canopy temperature (Ahmad et al. 2021). Moreover, remote sensing can provide information for estimating crop water requirements through the determination of crop evapotranspiration (ET) (Jayanthi et al. 2007; Campos et al. 2017; Pôças et al. 2020). Knowledge of crop ET facilitates irrigation scheduling. Crop ET can be estimated through remote sensing using two approaches. The first approach involves using thermal bands as inputs to the surface energy balance algorithm for land (SEBAL), mapping evapotranspiration at high resolution using internalized calibration (METRIC), surface energy balance index (SEBI), and surface energy balance system (SEBS) (Bastiaanssen et al. 1998; Allen et al. 2005; Aryalekshmi et al. 2021). The required energy fluxes (net radiation Rn, soil heat flux, and sensible heat H) are derived from satellite images (Allen et al. 2011; Irmak et al. 2011). Successful applications of energy balance models for estimating ET in various crops have been reported (Bastiaanssen et al. 2005; Tasumi et al. 2005; Tasumi and Allen 2007; Bashir et al. 2008; Kumar et al. 2020). Although this approach provides a precise measure of crop ET, it requires extensive data input and is limited by the availability of satellites equipped with thermal sensors (Glenn et al. 2010).

The second method entails estimating crop coefficients (the single crop coefficient, Kc, and the basal crop coefficient, Kcb) based on VIs (VI approach) (Jayanthi et al. 2007; Mukiibi et al. 2023). The VI-based Kc or Kcb values are then multiplied by the reference evapotranspiration to determine ET. Relationships between VIs and crop coefficients have been established for various crops using linear and nonlinear equations (Choudhury et al. 1994; Duchemin et al. 2006; Campos et al. 2017). The VI approach is relatively simple and requires fewer computations than the surface energy balance models (Glenn et al. 2010). However, reductions in ET due to stomatal closure, particularly during periods of water deficit, cannot be detected using the VI approach (González-Dugo and Mateos 2008).

Yield Prediction

Remote sensing techniques have been employed to forecast and predict crop yields using linear and non-linear models. The most common models for remote sensing applications are empirical relationships between VIs and final yield (Lobell 2013). However, empirical models require calibration using ground measured data for accurate yield predictions. A major limitation of empirical models is that they are site-specific and may not provide accurate yield predictions in different locations and years (Basso et al. 2013). Recent advancements in data processing techniques using ML algorithms have led to the development of more precise yield prediction models that incorporate VIs and factors that affect crop yields, such as fertilizer application rates (nitrogen, phosphorus, and potassium), irrigation, soil properties, weather parameters, and crop management information (plant population) (Abrougui et al. 2019; Abbas et al. 2020; Muruganantham et al. 2022). Additionally, these factors have a significant influence on the spectral reflectance of the crop canopy and can be used to explain the spatial variability in crop yields.

Remotely sensed information can also be incorporated into a crop simulation model during calibration, or to adjust the initial conditions of the model (Doraiswamy et al. 2003; Pinter et al. 2003; Awad 2019). The integration of remotely sensed data into process-based crop models can provide more accurate results than using process-based crop models alone (Launay and Guerif 2005; Dente et al. 2008; Jin et al. 2018). For instance, canopy state variables and soil moisture can be integrated into potato models to enhance the yield prediction accuracy (Dente et al. 2008; Jin et al. 2016; 2018; Zhou et al. 2017b).

Research Methodology

This study followed the systematic review guidelines outlined by Kitchenham and Charters (2007). Additionally, the study followed the preferred reporting items for systematic reviews and meta-analyses (PRISMA) (Liberati et al. 2009). The PRISMA framework describes the flow of information through the different phases of a systematic review, and includes article identification, screening, eligibility, and data analysis (Fig. 1).

Fig. 1
figure 1

The PRISMA flow diagram indicating the number of articles in each phase of the selection process

Full size image

Article Search Procedure

A literature search was performed on the internet using the following databases: Google Scholar, ScienceDirect, Scopus, Web of Science, IEEE Explorer, MDPI, Taylor and Francis, and SpringerLink. More studies were sourced by scanning the reference lists of articles obtained from the databases. Databases were selected based on the comprehensiveness of archiving and accessibility. This research focused on potato yield prediction using remote sensing technologies; therefore, the search strategy in the databases included keywords and Boolean operators “AND” and “OR” with the main search string as “potato yield prediction” AND “remote sensing”. As a consequence, less emphasis was put on vegetation condition status. To include any other relevant articles in our study, the search string was modified by adding the following keywords: “potato yield prediction” OR “potato yield forecasting” OR “potato yield estimation” AND “remote sensing” OR “satellite imagery.” Minor adjustments were made to the search string to suit each database (Table 1). Original research studies conducted between 2000 and 2022 were used for this study, as the use of remote sensing technologies in agriculture gained momentum with the launch of MODIS and Landsat-7 satellite sensors in 1999 (Kasampalis et al. 2018; Khanal et al. 2020). Since then, various remote sensing platforms, including satellites, manned and UAVs, have been developed and used to collect vegetation spectral data for different precision agricultural applications.

Table 1 Search string and number of papers retrieved from each database
Full size table

Article Selection Criteria

The search query in each database returned several inapplicable records, most of which were out of the scope (Table 1). To be included, studies had to meet the minimum criteria of including remote sensing applications for potato growth and yield prediction. Publication titles, abstracts, and keywords were used to select articles for further analysis. Articles were excluded using the following criteria:

  • Articles not focused on potato.

  • Studies not including remote sensing data or remote sensing applications not related to growth monitoring or yield prediction.

  • Publications without access to the full-text version, review articles, and articles in a language other than English.

Data Analysis

The number of articles published per annum and the number of articles per remote sensing application were calculated. Information that answered the research questions was extracted from the publications. A bibliometric analysis was conducted using VOSviewer software (www.vosviewer.com, van Eck and Waltman 2010) to identify authors’ keywords appearing in three or more publications. This allowed the visualization of the network between the most dominant themes in remote sensing applications for potato research.

Results and Discussion

Descriptive Statistics of Selected Publications

The number of publications on remote sensing application in potato increased rapidly in recent years (Fig. 2). After applying the inclusion criteria, 79 articles were selected for further analysis. Remote sensing was mainly used for potato yield prediction (37% of the total studies), followed by leaf N status estimation (21% of the total studies) (Fig. 3). The most frequently used keywords related to growth and yield prediction were potato yield, tuber yield, LAI, plant height, AGB, and phenology (Fig. 4). Other common keywords included remote sensing, machine learning, precision agriculture, random forest, climate change, and UAV.

Fig. 2
figure 2

Number of publications from 2000 to 2022 included in the review

Full size image
Fig. 3
figure 3

Distribution of growth monitoring and yield prediction aspects from selected articles

Full size image
Fig. 4
figure 4

Overlay visualization of co-occurrence of author keywords from the Web of Science (a) and Scopus (b) databases

Full size image

The use of remote sensing technology to monitor crop growth and predict yield is rooted in the premise that remotely sensed features, mainly spectral reflectance, VIs and canopy texture, can serve as proxies for plant growth variables, such as LAI, AGB, and fraction of photosynthetically active radiation (fPAR) (Lobell 2013; Zhou et al. 2018). These plant growth variables play a vital role in biophysical processes, such as light interception, photosynthetic activity, and biomass accumulation, which directly affect final crop yield (Mulla 2013; Vannoppen and Gobin 2022). Remote sensing enables the real-time observation of vegetation condition, which is influenced by both crop genetics and management practices (Zhou et al. 2018). As VIs are linked to the primary productivity of crops, potato yield prediction using various models based on VIs has been explored (Table 2).

Table 2 Summary of the remote sensing types and models used for potato yield prediction and model accuracy
Full size table

Potato Yield Prediction

Remote Sensing Techniques for Potato Yield Prediction

Optical satellite systems were the most widely used remote sensing platforms for predicting yield (Table 2). Sentinel-2 was the most popular satellite platform, while other platforms adopted by researchers included Landsat satellite series 5–8, Planet Scope, National Oceanic and Atmospheric Administration (NOAA) equipped with an Advanced Very High-Resolution Radiometer (AVHRR) sensor, TERRA, and Aqua satellites equipped with moderate-resolution imaging spectroradiometer (MODIS) sensors (Table 2). Optical satellite platforms are equipped with high-resolution multispectral sensors that capture fine-scale details of crop vegetation over large areas, facilitating potato yield prediction at both the field and regional scales (Salvador et al. 2020). However, the main limitation of optical satellite platforms is the trade-off between acquiring images with sufficiently high spatial and temporal resolution to obtain multiple cloud-free images during the growing season (Lobell et al. 2007; Mulla 2013; Khabbazan et al. 2019). For example, Sentinel-2 satellites have a temporal resolution of 5 days and a relatively high spatial resolution (10 or 20 m/pixel for land applications) (Herrmann et al. 2011; Sun et al. 2022). Landsat satellites have a relatively high spatial resolution of 30 m/pixel, but a low temporal resolution of 16 days. The TERRA satellite has a high temporal resolution of 1 – 2 days but a low spatial resolution of 250 – 1000 m/pixel. The Planet Scope satellite provides images with the highest spatial resolution (3 m/pixel) and highest temporal resolution (daily). However, these images are costly ($218 per 100 km2), limiting the use of Planet Scope images for agricultural monitoring (Sun et al. 2022).

Owing to the limitations of optical remote sensing, microwave (radar) sensors, which can penetrate clouds with high spatial and temporal resolution and are independent of light conditions, have been investigated for Earth observations (Bouman and van Kasteren 1990; Clevers and Van Leeuwen 1996; Steele-Dunne et al. 2017). Microwave remote sensing uses synthetic aperture radar (SAR) sensors that emit low-frequency microwave pulses (1 – 10 GHz) towards the Earth’s surface (Steele-Dunne et al. 2017; Khabbazan et al. 2019). The pulses are scattered upon interaction with different surfaces (vegetation and soil) and sent back to the receiver, which is also known as radar backscattering (Moran et al. 2002). Radar backscattering from a vegetation surface is mainly influenced by canopy size, ridge orientation, architecture of individual plants, crop type, growth stage, water content of the plant parts, and roughness of the vegetation canopy (Bouman and van Kasteren 1990). Radar backscattering by the soil surface is mainly influenced by soil water content and surface roughness. Therefore, radar remote sensing has potential for agricultural applications, particularly for crop monitoring, classification, and soil/vegetation moisture estimation (Moran et al. 2002; Steele-Dunne et al. 2017). With the launch of the Sentinel-1 satellites (Sentinel-1A and Sentinel-1B), high temporal (6 – 12 days), and high spatial (2.3 – 13.9 m) SAR data can be freely accessed in different parts of the world (Mercier et al. 2020). The potential of Sentinel-1 SAR data for monitoring phenological development of various crops including maize, potato, sugar beet, winter wheat, rapeseed and rye grass was evaluated by Khabbazan et al. (2019) and Mercier et al. (2020). Radar backscattering coefficients can be used to estimate crop growth variables such as crop height (Abdikan et al. 2018; Arslan et al. 2022), crop biomass (Ndikumana et al. 2018), LAI (Clevers and Van Leeuwen 1996; Hirooka et al. 2015), and crop yield (Clevers and Van Leeuwen 1996).

Besides optical satellites, ground-based systems were the second most widely used platforms, including the Field Scout NDVI meter, handheld hyperspectrometers, infrared cameras, red, green, and blue (RGB) digital cameras, GreenSeeker, and Holland Scientific Crop Circle™ sensors (Table 2). Ground-based remote sensing devices provide benefits such as high spatial resolution, real-time data, cost-effectiveness for small research plots, control over measurement conditions, and access to ground-truth measurements, which can be used to validate and improve the accuracy of potato yield prediction (Sun et al. 2022).

The Most Used Vegetation Indices for Potato Yield Prediction

A wide range of VIs was used to predict potato yield (Table 3). NDVI was the most widely used index (Tables 2 and 3). Other commonly used indices included GNDVI, normalized difference red-edge (NDRE), SAVI, EVI, red-edge chlorophyll index (CI red-edge) and RVI (Table 3).

Table 3 Most widely used vegetation indices for estimating potato yield, aboveground biomass, leaf area index, and canopy chlorophyll
Full size table

The accuracy of final tuber yield prediction was evaluated based on the coefficient of determination (R2) between VI and observed yields. The ranges of R2 values reported for the six most commonly used VIs are shown in Fig. 5. Positive associations between NDVI and final potato yield had R2 values ranging from 0.23 – 0.84 (median of 0.67). Relating NDVI (median R2 = 0.67) with final potato yield provided a higher median R2 than other comparable indices, such as GNDVI (median R2 = 0.58 and NDRE (median R2 = 0.61), despite the fact that NDVI is influenced by soil background reflectance and tends to saturate at LAI values greater than 3 (Gitelson 2004).

Fig. 5
figure 5

The coefficient of determination (R2) for vegetation indices used in final tuber yield prediction (NDVI is normalized difference vegetation index, GNDVI is green NDVI, NDRE is Normalized difference red-edge, SAVI is soil-adjusted vegetation index, EVI is enhanced vegetation index, and CIred-edge is red-edge chlorophyll index)

Full size image

A few studies used NDRE to predict yield and reported R2 values between 0.12 – 0.85 (median of 0.61) (Fig. 5). For instance, Luo et al. (2020) observed a strong association between NDRE and tuber yield (R2 = 0.85) during the starch accumulation stage (80 – 100 DAP). The R2 for the association between GNDVI and tuber yield ranged between 0.26 – 0.75 (median of 0.58). This indicates that the spectral absorbance of the green portion of the electromagnetic spectrum can be used to predict the final tuber yield. This is supported by Mhango et al. (2021), who reported that the absorbance of the green portion of the magnetic spectrum is significantly associated with the tuber yield. The EVI and CIred-edge also showed good associations with the tuber yield (R2 = 0.4 – 0.87). Luo et al. (2020) estimated potato yield using six VIs and identified CIred-edge as the best index for potato yield estimation at any growth stage of the crop, with R2 greater than 0.70.

Preferred Growth Stage for Potato Yield Prediction Using Remote Sensing

The results in Fig. 6 show variability in the R2 values obtained for the relationship between tuber yield and VIs as the number of days after planting increased from 35 – 95 DAP. The highest median R2 values were observed between 36 – 55 DAP. These findings suggest that the optimal growth stage for obtaining VIs to predict tuber yield often occurs approximately during maximum ground cover and tuber initiation.

Fig. 6
figure 6

Box plot comparing the coefficients of determination (R2) between tuber yield and all vegetation indices combined on different days after planting

Full size image

Dry matter accumulation in potato is a function of intercepted photosynthetically active radiation, which exponentially increases to a maximum at 100% ground cover (Haverkort 2018). For most potato varieties, maximum ground cover is attained between 60 – 80 DAP, which also coincides with the peak spectral reflectance and peak VI values reported for potato (Mhango et al. 2022). The reflectance in the NIR region may be inconsistent in the early stages of potato because of the influence of soil background reflectance (Morier et al. 2015). Additionally, the biophysical and biochemical composition of the leaves, which are responsible for light absorption and reflection, are immature during the early stages of the crop, resulting in low absorption of red light and low NIR reflectance (Gómez et al. 2021). Therefore, VIs obtained during the early stages (0 – 40 DAP) of the crop may provide less reliable yield predictions.

Potato Aboveground Biomass Monitoring

Remote Sensing Techniques for Aboveground Biomass Monitoring

Aboveground biomass (AGB) is an agronomic variable widely used to assess crop growth and development, physiological conditions, light use efficiency, and the effects of agricultural management practices, as well as to estimate crop yields (Poorter et al. 2012; Gnyp et al. 2014; Liu et al. 2022a). AGB is also important for determining the nitrogen nutrition index (NNI), which is used to monitor crop N status (van Evert et al. 2012; Jin et al. 2021; Sun et al. 2022). Conventional methods for estimating AGB involve destructive plant sampling, which is time-consuming, labour-intensive, and often unable to detect spatial and temporal variability (Gnyp et al. 2014). Alternative methods for estimating AGB include process-based crop models; however, these models often require numerous inputs that are not readily available for accurate AGB estimation (Craufurd et al. 2013; Wan et al. 2021).

Remote sensing technology provides real-time and non-destructive AGB measurements (Table 4), which are based on the hypothesis that the spectral reflectance of vegetation at specific wavelengths is strongly associated with LAI, canopy coverage, fPAR, and CCC, which are related to crop biomass production (Weiss et al. 2020; Shu et al. 2023). Changes in the spectral reflectance of vegetation at different growth stages are mainly influenced by the biochemical properties of the leaf, such as leaf pigments, water content, and dry matter content, and canopy structure properties, such as LAI and leaf inclination (He et al. 2021; Jin et al. 2021).

Table 4 Comparison of types of remote sensing, models, and main input features for aboveground potato biomass monitoring
Full size table

Unmanned aerial vehicles (UAVs) were the most commonly used platforms for the in-season estimation of potato AGB (Table 4). UAVs allow frequent and timely monitoring, are easy to operate, and usually provide images with higher spatial and temporal resolution than satellite imagery (Liu et al. 2022b). UAVs can fly at low altitudes to capture images with an extremely high spatial resolution of up to 1 cm/pixel. These images contain vital information regarding the spatial distribution of plants, soil coverage, and canopy structure, which are essential factors for estimating the AGB (Turner et al. 2012; Yu et al. 2016). UAVs are usually equipped with digital, multispectral, and hyperspectral sensors that capture images of varying properties. In particular, hyperspectral sensors have narrow spectral bands that allow visualization of spatial variability in the canopy structure and are highly sensitive to reflected light across the bands (Liu et al. 2022a). Hyperspectral sensors provide shortwave (1100 – 2500 nm) plant spectra, which are strongly associated with leaf water content, lignin, cellulose, and starch content (Gnyp et al. 2014; Marshall and Thenkabail 2015). These leaf traits are directly related to fresh and dry leaf weight. However, the use of hyperspectral sensors is limited because of the large number of spectral bands, high cost, and image processing difficulties (Jin et al. 2021; Mao et al. 2021). Moreover, because UAVs operate at low altitudes, images typically exhibit a small field of view (Turner et al. 2012). Consequently, many images must be captured to adequately cover the required area of interest. The processing of a large number of images necessitates the use of techniques, such as mosaicking, geometric correction, and ortho-rectification, which can be particularly challenging (Turner et al. 2012; Zhang and Kovacs 2012).

Some studies employed ground-based sensors, specifically handheld field spectroradiometers, for potato AGB estimation (Table 4). Ground-based sensors provide high temporal and spectral resolution images for extracting canopy traits, such as plant height and texture features, which are related to AGB (Liu et al. 2022b, c). However, ground-based sensors are unable to capture large-scale images that illustrate the spatial distribution of crop growth (Zhu et al. 2019). Furthermore, image acquisition using ground-based sensors often results in damage to crops and is labour intensive, which complicates the crop growth monitoring process (Liu et al. 2022b).

Few studies employed satellite platforms to estimate the AGB of potatoes (Table 4). The limited application of satellite remote sensing could be due to the low spatial and temporal resolutions of most freely available satellites. The long revisit cycles of most satellites make it difficult to obtain crop AGB data for the desired growth stages (Wan et al. 2021).

Aboveground biomass of potato has been predicted with varying accuracy (R2 = 0.20 – 0.90) using different models (Table 4). Estimation of crop AGB through remote sensing is mainly performed using physical and statistical models. Physically based models for estimating AGB involve the use of radiative transfer models (RTMs), such as the PROSAIL model (Duan et al. 2014). The physical models use canopy reflectance at 400 – 2500 nm wavelengths as input (Duan et al. 2014; Weiss et al. 2020). The inversion of RTMs using canopy reflectance values from remote sensing allows the estimation of plant variables such as chlorophyll content, dry matter content, LAI, carotenoid content, and leaf equivalent water thickness (Wan et al. 2021), which are used to calculate AGB (Duan et al. 2014). Statistical models are empirical regression equations relating ground measured AGB and remotely sensed spectral features such as VIs, canopy texture variables, geometric variables, canopy height, and fractional vegetation cover (Luo et al. 2022). Statistical models employed for AGB estimation included multiple regression, partial least squares regression (PLSR) and principal component analysis (PCA) (Table 4). Recently ML techniques such as Random Forest, artificial neural network (ANN) and SVM that incorporate multiple VIs, canopy texture features and crop height have been explored for AGB estimation (Zhu et al. 2019; Li et al. 2020c; Liu et al. 2022c).

Vegetation Indices for Aboveground Biomass Monitoring

The most commonly used VIs included NDVI, optimized SAVI (OSAVI), re-normalized difference vegetation index (RDVI), and modified simple ratio (MSR) (Table 3). Most studies associated NDVI with the AGB of potato, with relationships showing varying strengths. However, NDVI is affected by reflectance from the soil surfaces, especially at low canopy cover, and has low sensitivity to AGB at high canopy cover (Xue and Su 2017). This affects the accuracy of AGB estimation using NDVI. Therefore, alternative VIs such as RDVI, SAVI, OSAVI, TSAVI, CIred-edge, MSR among others, with the capability to distinguish between crop vegetation and soil reflectance, have been evaluated for accurate AGB estimation. Although most VIs are positively correlated with AGB (Pei et al. 2019), they tend to saturate at high vegetation cover, making them inefficient for estimating AGB at different crop growth stages (Liu et al. 2022e). Therefore, techniques, including the use of plant height extracted from digital surface models (DSMs) (Roth and Streit 2018), microwave SAR backscatter (Gao et al. 2013; Hosseini et al. 2019), light detection and ranging (LiDAR) (Li et al. 2015; ten Harkel et al. 2020) and canopy texture features from ultrahigh ground-resolution RGB digital images have been investigated (Yue et al. 2018; Liu et al. 2022b).

Plant height is used to characterize vertical crop growth; therefore, it can be used to estimate other crop variables, including AGB (Yue et al. 2018; Lu et al. 2019). Crop AGB has been estimated using plant height extracted from DSMs and digital terrain models (Bendig et al. 2015; Mao et al. 2021). DSMs can be generated from RGB ultrahigh-ground resolution digital images using structure-from-motion algorithms (Aasen et al. 2015). Combining plant height and VIs in statistical models improves the accuracy of AGB estimation (Bendig et al. 2015; Yue et al. 2018; Lu et al. 2019), as it enables the use of both spectral and structural information (Lu et al. 2019; Niu et al. 2019). Liu et al. (2022b) found that the use of crop height and canopy texture features results in more accurate estimation of potato AGB (R2 = 0.73 – 0.78) than using texture features alone (R2 = 0.59 – 0.73).

Microwave SAR is sensitive to biophysical crop variables including LAI, biomass, and canopy height (Jin et al. 2015; Ndikumana et al. 2018). Synthetic aperture radar backscatter is advantageous over optical remote sensing because it can penetrate crop canopies and circumvent the premature saturation of AGB estimates, particularly during periods of full canopy coverage (Gao et al. 2013; Liu et al. 2022b). Research has demonstrated that the backscattering coefficients from SAR can accurately estimate the AGB in various crops, including wheat (Jin et al. 2015; Han et al. 2019), maize (Gao et al. 2013; Hosseini et al. 2019), rice (Ndikumana et al. 2018), and soybean (Mandal et al. 2019).

LiDAR, or laser-scanning remote sensing, involves the transmission of electromagnetic pulses (laser pulses) with a specific penetration ability to interact with vegetation and ground surfaces (Poley and McDermid 2020; Sun et al. 2022). The interaction between laser pulses and ground objects is used to characterize crop canopy structural and biophysical variables, such as volume, height, density, LAI, stem diameter, and coverage, which are useful for estimating the AGB (Jimenez-Berni et al. 2018; Poley and McDermid 2020; ten Harkel et al. 2020). For example, Li et al. (2015) demonstrated that the canopy height and LAI acquired by LiDAR correlated strongly with aboveground and belowground maize biomass. Jimenez-Berni et al. (2018) accurately (R2 > 0.90) estimated the canopy height, ground cover, and biomass of wheat using a LiDAR-based 3-dimensional profile index (3DPI) model. Similarly, ten Harkel et al. (2020) evaluated the potential of a 3DPI model based on LiDAR point clouds to estimate winter wheat, sugar beet, and potato biomass. The 3DPI estimated biomass well for sugar beet (R2 = 0.68) and winter wheat (R2 = 0.82), but poorly estimated potato biomass (R2 = 0.24). Additional research is needed to assess the potential of LiDAR technology for potato AGB estimation.

Canopy texture features such as gray scale variance, homogeneity, contrast, dissimilarity, entropy and second moment obtained from ultrahigh-ground-resolution RGB digital images provide useful information for estimating crop variables, such as LAI, plant density, chlorophyll content, nitrogen content, and AGB (Yue et al. 2018; Poley and McDermid 2020; Zhai et al. 2023). Canopy texture features have been used to accurately estimate crop AGB for rice (Zheng et al. 2019), winter wheat (Yue et al. 2019) and potato (Liu et al. 2022c, b). Studies by Li et al. (2020b) and Liu et al. (2022c) revealed that combining multiple VIs, texture features, plant height, and canopy cover improved the accuracy of potato AGB estimation.

The strength of the relationship between VIs and AGB increases from emergence to tuber bulking stages and declines from tuber bulking until crop maturity (Liu et al. 2022d). VIs are strongly associated with AGB at the tuber bulking stage (R2 = 0.62 – 0.92) (Pei et al. 2019; Yang et al. 2021; Liu et al. 2022a; d). This trend can be related to the potato growth cycle, where maximum vegetative growth and full canopy cover are attained during the tuber initiation and bulking growth stages (Haverkort 2018). Therefore, the spectral information obtained between the tuber initiation and bulking stages reflected AGB relatively well (Pei et al. 2019; Liu et al. 2022d). During the later growth stages, leaves start to senesce, resulting in a reduction in canopy cover; hence, crop canopy is a poorer predictor of AGB in this stage.

Potato Leaf Area Index

Leaf area index, defined as the total one-sided area of leaf tissue per unit ground surface area, is a good indicator of vegetation status, photosynthetic rate, biomass accumulation, and evapotranspiration (Wan et al. 2021). Timely estimation of potato LAI using remote sensing has been investigated by various researchers (Table 5). Unmanned aerial vehicles equipped with hyperspectral sensors and satellites equipped with multispectral sensors were the most commonly used remote sensing platforms for potato LAI monitoring (Table 5). However, no single remote sensing platform provides images with high spectral, spatial, and temporal resolutions at a low cost. Therefore, the use of different optical sensors on various platforms has been explored for estimating potato LAI (Franceschini et al. 2017a). Ground-based sensors, such as multispectral cameras (Roosjen et al. 2018) and Cropscan multispectral radiometers (Clevers et al. 2017; Franceschini et al. 2017a), were employed to obtain ground-truth reflectance data for calibrating the UAV hyperspectral and satellite multispectral data. Gevaert et al. (2014, 2015), investigated the utility of continuous surface reflectance (spectral-temporal response surfaces, STRSs) derived from the integration of Formosat-2 multispectral satellite imagery and UAV hyperspectral imagery. The findings indicated that STRSs obtained from both satellite multispectral and UAV hyperspectral data can be employed to derive VIs for estimating potato LAI and chlorophyll content, with R2 ranging from 0.1 – 0.84. However, studies on image fusion techniques utilizing satellite and UAV data to estimate potato LAI remain limited.

Table 5 Comparison of remote sensing type, models and main input features for potato leaf area index estimation
Full size table

Leaf area index can be estimated from remote sensing using three approaches: biophysical processors, RTMs, and statistical models (Wan et al. 2021). The LAI can be retrieved from the biophysical processor tool of the Sentinel Application Platform (SNAP) created by the European Space Agency (Mourad et al. 2020). The SNAP software includes neural network algorithms trained to process biophysical traits such as LAI and LCC from Sentinel-2 imagery. Additionally, Mourad et al. (2020) indicated that the SNAP biophysical processor tool underestimated the LAI for various crops, including potato (R2 = 0.27), compared to VI-based statistical models (R2 = 0.51 – 0.63). However, additional research is needed to fully assess the dependability of the SNAP biophysical processor in estimating the LAI of crops.

The PROSAIL model, which integrates the PROSPECT and SAIL models, is the most commonly used RTM for simulating crop canopy reflectance as a function of LAI, leaf angle distribution function, chlorophyll content, dry matter content, carotenoid content, leaf equivalent water thickness, canopy reflectance background, sensor viewing angle, sun zenith, and azimuth angles (Roosjen et al. 2018; Wan et al. 2021). To retrieve LAI using RTMs, model inversion techniques such as iterative optimization, look-up tables, and neural networks must be applied (Duan et al. 2014; Verrelst et al. 2015). A specific set of model parameter values that result in simulated canopy reflectance similar to remote sensing canopy reflectance is utilized to estimate the LAI. Duan et al. (2014) assessed the capacity of the PROSAIL model to estimate LAI for maize, potato, and sunflower using UAV hyperspectral data. The results indicated that the model accurately simulated LAI for all three crops, with root mean square error ranging from 0.55 – 0.60 m2 m−2. However, differences were observed between the model-estimated and ground measured LAI values, which were attributed to the inability of the model to account for the shading effect of row crops. The shading effect can lead to an enhancement or reduction in crop canopy reflectance, depending on the UAV flight direction (Duan et al. 2014).

Statistical models are based on established empirical relationships between VIs and ground-measured LAI using devices such as a Plant Canopy Analyzer (LAI-2000, LI-COR), AccuPAR LP-80 ceptometer, and LI-3100C leaf area meter (Clevers et al. 2017). These empirical relationships can take various forms, including linear, exponential, logarithmic polynomial, and inverse exponential expressions (Ray et al. 2006). Haboudane et al. (2004) and Ray et al. (2006) found that exponential relationships between LAI of various crops and VIs had the highest R2 values compared to linear, power, and logarithmic relationships. This may be because indices based on the ratio or product of red and NIR wavelengths are highly sensitive to chlorophyll content and tend to saturate at LAI greater than 3 m2 m−2 (Haboudane et al. 2004). Although statistical models are simple to compute, they require extensive calibration and validation using ground truth data (Mourad et al. 2020). Moreover, these models are limited to a specific set of conditions, such as crop type, crop management, weather, and soil conditions, under which they are created, which restricts their use in different conditions. Despite these limitations, statistical models are the most frequently used for estimating the potato LAI, with the relationship between LAI and VIs showing varying R2 ranging from 0.52 – 0.95 (Table 5).

The accuracy (indicated by R2) of LAI estimations, regardless of the model used, ranged between 0.40 – 0.99, which suggests that remote sensing data can be used to estimate the LAI of potato with considerable accuracy (Table 5). The most popular indices for potato LAI estimation included NDVI, SAVI, WDVI, red-edge NDVI, red-edge inflection point (REIP) and GNDVI (Table 3). Reflectance in the green, red and NIR regions is directly related to the LAI, greenness, and canopy cover (Herrmann et al. 2011). However, indices based on NIR and red bands, such as NDVI and RVI, tended to be less sensitive to LAI greater than 2 m2 m−2, whereas indices combining green and red-edge bands were highly sensitive to LAI above 4 m2 m−2 (Gitelson 2004; Herrmann et al. 2011). In addition, reflectance in the NIR region tended to be affected by soil reflectance, particularly during the early crop stages (Morier et al. 2015). To eliminate the effects of the soil background, VIs such as the WDVI, wide dynamic range vegetation index (WDRVI), SAVI, and OSAVI were created with a soil reflectance correction factor in the NIR region, which improved their sensitivity to LAI (Huete 1988; Clevers 1989; Haboudane et al. 2002; Gitelson 2004). Although VIs with minimal interference from soil background reflectance are highly sensitive to LAI, there is currently no specific VI for LAI estimation. Furthermore, the dependence of spectral reflectance in the visible and NIR regions on both LAI and chlorophyll content makes it challenging to find a VI that is not influenced by the chlorophyll content. Haboudane et al. (2004), proposed the MCARI2 and the modified triangular vegetation index 2 (MTVI2), which are highly sensitive to changes in LAI and have low sensitivity to chlorophyll content. Therefore, future research should investigate the nature of the relationship between MCARI2, MTVI2, and potato LAI.

Few studies evaluated the relationship between the VIs and LAI of potato at different growth stages. Clevers et al. (2017) evaluated the temporal pattern of WDVI as an estimator of LAI of potato during the season. The R2 values between the WDVI and LAI of potato at 52, 85, 95, and 122 DAP were 0.71, 0.92, 0.87, and 0.92, respectively (Clevers et al. 2017). More research is needed to determine the most suitable time to estimate the LAI using VIs.

Estimating Potato Leaf and Canopy Chlorophyll Content

Leaf chlorophyll content (LCC) and canopy chlorophyll content (CCC) serve as indicators of crop physiological status, health, nutritional status, and productivity (Evans 1989; Elarab et al. 2015). Leaf chlorophyll content is closely related to leaf N, and can therefore be used as an indicator for N nutrition in leaves to guide fertilizer application in potatoes (Evans 1989; Clevers et al. 1994; van Evert et al. 2012). Therefore, timely and accurate estimation of chlorophyll content can aid in the implementation of crop management interventions for potatoes, leading to improved crop growth and optimized yield. Traditional methods for estimating chlorophyll content involve laboratory techniques such as liquid chromatography, atomic absorption, and spectrophotometry (Li et al. 2020b). These methods require destructive leaf sampling, which is time-consuming and laborious. Traditional methods are limited to small areas, necessitating the use of remote sensing techniques for real-time chlorophyll estimation over relatively large areas (Gao et al. 2021). Research has shown that reflectance in specific portions of the spectrum between the visible and NIR regions is sensitive to chlorophyll content. Vegetation indices based on reflectance in the green, red-edge and NIR bands are strongly correlated with LCC and CCC (Gitelson and Merzlyak 1996; Borhan et al. 2017).

The remote sensing platforms, sensors, and models used for estimating LCC and CCC of potato are listed in Table 6. UAVs equipped with multispectral and hyperspectral sensors, and ground-based (handheld) devices such as the Field spectrometer, Cropscan multispectral radiometers, charged-coupled devices (CCD) digital RGB cameras and the soil plant analysis development (SPAD) chlorophyll meter were the most used platforms, whereas a few studies used satellite multispectral platforms for potato LCC and CCC estimation (Table 6).

Table 6 Comparison of remote sensing models and main input features used for potato leaf and canopy chlorophyll estimation
Full size table

Ground-based sensors are quick, easy to operate and provide LCC and CCC estimates that can be used for ground truthing UAV and satellite-based estimations. Various regression techniques, including linear, exponential, logarithmic, multiple linear regression, and PLSR, as well as non-linear regression methods, such as ANN and Random Forest, have been employed to establish LCC and CCC prediction models. Relationships between the ground-measured chlorophyll content of potato with SPAD chlorophyll meter readings (Uddling et al. 2007), CCD digital image features, such as mean gray values, spectral luminosity, and mean brightness ratios (Yadav et al. 2010; Gupta et al. 2013; Borhan et al. 2017), and VIs derived from Cropscan radiometer reflectance measurements (Kooistra and Clevers 2016) have been established. Borhan et al. (2017) used simple linear and multiple linear regressions to show that the mean gray values of CCD digital images at 550 nm and 700 nm exhibited strong correlations with the SPAD chlorophyll meter readings of potato. Gupta et al. (2013) applied linear regression and ANN to relate the mean brightness ratios of CCD RGB images of leaves with the chlorophyll content of micropropagated potato plants. The ANN model demonstrated a higher accuracy in predicting the LCC (R2 = 0.82) than simple linear regression (R2 = 0.59). Kooistra and Clevers (2016) found that VIs based on Cropscan reflectance values of potato crops were linearly related to SPAD chlorophyll meter readings with an R2 ranging from 0.43 – 0.64. In the same study, the PROSAIL model estimated chlorophyll content well (R2 > 0.72) (Kooistra and Clevers 2016). Although some studies have shown promising results for the estimation of potato chlorophyll content using SPAD chlorophyll meter and CCD digital cameras, environmental factors such as nutrient deficiency, leaf disease infection, and the inherent light scattering properties of leaves can impact the accuracy of these estimates (Borhan et al. 2017). Additionally, the non-homogeneous distribution of chlorophyll within the leaves can affect the estimated chlorophyll concentration (Borhan et al. 2017).

The use of UAVs' spectral information was employed to derive VIs for establishing potato chlorophyll content estimation models (Table 6). Yin et al. (2022) implemented twelve VIs derived from UAV multispectral images as inputs for ML algorithms, including Random Forest, Support Vector Regression (SVR), PLSR, and Ridge Regression, to predict potato chlorophyll content. The findings indicated that the Random Forest model demonstrated the highest accuracy in predicting chlorophyll content, with a R2 value of 0.76, followed by the SVR model, with a R2 value of 0.74. Li et al. (2020b) reported that VIs derived from UAV hyperspectral data showed a strong correlation with potato chlorophyll content at all growth stages, with R2 values ranging from 0.53 – 0.77. Additionally, the original spectrum, fractional differential spectra, spectral position, and spectral area parameters of potato reflectance were used to establish potato chlorophyll prediction models. The study conducted by Li et al. (2021a) established linear and nonlinear relationships between fractional differentiation of the canopy spectrum and chlorophyll content in potato plants at various growth stages. The R2 for the relationship between different fractional differentiation spectra orders and chlorophyll content ranged from 0.65 – 0.85 (Li et al. 2021a). In a separate study, Li et al. (2020b) utilized ML algorithms in conjunction with UAV hyperspectral VIs and spectral characteristics based on spectral position and spectral area to estimate the chlorophyll content in potato plants at different growth stages. The stepwise regression model demonstrated the highest prediction accuracy (R2 ranging from 0.52 – 0.78) at all growth stages.

Most studies used CIred-edge, CIgreen, NDVI, and OSAVI to estimate the LCC and CCC of potato (Table 3). In addition, indices such as OSAVI (TCI/OSAVI), TCARI/OSAVI, DVI, and RVI have been strongly associated with the LCC and CCC of potatoes (Kooistra and Clevers 2016). Chlorophyll estimation requires VIs that are highly sensitive to chlorophyll concentration, resistant to variations in LAI, and unaffected by background soil reflectance (Clevers et al. 2017). These indices combine the bands of minimum chlorophyll absorption (550 nm and 700 nm) and maximum chlorophyll absorption (670 nm) (Haboudane et al. 2002). This idea led to the generation of ratio indices such as TCARI/OSAVI and MCARI/OSAVI (Gitelson and Merzlyak 1996; Haboudane et al. 2002). In addition, reflectance in the green- and red-edge regions was found to be highly sensitive to chlorophyll content (Gitelson et al. 2003). Therefore, VIs based on red-edge properties, such as red-edge position (REP), CIred-edge, CIgreen, CVI, and MERIS terrestrial chlorophyll index (MTCI), have been recommended for chlorophyll content estimation (Gitelson et al. 2003; Clevers et al. 2017).

As expected, chlorophyll content showed the strongest correlation with spectral features during the vegetative growth stage, with an R2 value of 0.85. This relationship was followed by tuber initiation (R2 = 0.70), tuber bulking stage (R2 = 0.69), and maturation stages (R2 = 0.54) (Li et al. 2020b). This finding was consistent with the use of fractional differentiation spectra orders, which produced an R2 of 0.85 during the vegetative stage, followed by an R2 of 0.79 at tuber initiation, an R2 of 0.71 at tuber bulking, and an R2 of 0.72 at starch maturation (Li et al. 2021a). Leaf greenness, leaf N content, and chlorophyll concentration are highest during the vegetative growth stage, and they gradually decline as the season progresses until crop senescence, when leaf colour changes from dark green to yellow (Borhan et al. 2017; Clevers et al. 2017; Li et al. 2020b). It is important to note that this decline in leaf greenness is a natural part of crop growth and development. Changes in the structure and biochemical components of leaves during the growing season affect spectral reflectance. Consequently, the vegetative growth period, including tuber bulking stage, is the most appropriate time for estimating chlorophyll content of potato.

Estimating Potato Leaf Nitrogen Status

Ground-based platforms were most commonly used to estimate the N status of potato leaves (Table 7). These included sensors such as Cropscan, Rapidscan, NIR analyzers, shortwave-infrared (SWIR) cameras, and SPAD meters. A possible reason for the high usage of ground-based sensors could be their high spatial resolution and the opportunities they offer for accurate ground-truth calibration. Moreover, handheld sensors are portable, facilitate measurements in small research fields, and have low operational costs compared with UAVs and satellites. The R2 values for all models combined for estimating leaf/petiole N concentrations ranged from 0.52 – 0.95, suggesting that remote sensing has the potential to estimate the N status of potato with considerable accuracy (Table 7).

Table 7 Comparison of the remote sensing models used for estimating the N concentration of potato leaves and their main input features
Full size table

The most commonly used indices were NDVI, MCARI, TCARI, and TCARI/OSAVI ratio (Table 3). Cohen et al. (2010) evaluated the relationship between TCARI and potato leaf N levels, and they found that TCARI is strongly associated with leaf N% and petiole NO3-N only at the tuber bulking stage (R2 = 0.80 and 0.76). Goffart et al. (2022) showed that the best linear and Random Forest models for estimating shoot N concentration and N uptake of potato were those combining the TCARI/OSAVI, WDVI, CIred-edge, and RVI indices. Jain et al. (2007) showed that the ratio of reflectance at red-edge bands 750 nm and 710 nm (RRE,750/710) is strongly associated with leaf N content of potato between 40 – 60 DAP. Peng et al. (2021a) suggested that the RVI, NDRE, and transformed chlorophyll index (TCI) were the most suitable indices for estimating the NNI of potato. Based on the above analysis, there is substantial variation in opinions regarding the most suitable VIs for estimating potato N status; however, red-edge-based indices appear to be the most suitable. This suggests that more research is required to confirm the suitability of red-edge indices for potato N status estimation.

Conclusions

  • Satellite images were the most widely used source of remote sensing data for potato yield prediction, whereas UAVs and handheld sensors were most widely used for potato growth monitoring.

  • A combination of regression analysis and ML models was used to generate prediction models for all aspects of potato growth and yield, with VIs and spectral bands as the main features. The most common VIs for the yield prediction models were the NDVI, CIred-edge, OSAVI, and CIgreen.

  • Strong associations between tuber yield and NDVI, EVI, NDRE, and CIred-edge, suggested that these indices are most appropriate for estimating tuber yield. A strong association was also reported between potato LAI and NDVI, OSAVI, WDVI, and GNDVI.

  • The chlorophyll content of potato correlated well with CIgreen, CIred-edge, TCI/OSAVI, and TCARI/OSAVI. However, combining multiple VIs with original canopy spectrum measurements using PLSR resulted in improved chlorophyll content estimation. Indices calculated from red-edge bands, such as the CIred-edge, NDRE, and red-edge 740/720 indices, were strongly associated with leaf N concentration.

  • Most studies reported that VIs have a weak association with potato AGB because they tend to saturate at high vegetation cover. Accurate AGB estimation can be achieved by combining multiple VIs, plant height, canopy cover, and the original canopy spectrum in ML models.

  • Vegetation indices acquired during maximum ground cover and tuber initiation, (approximately 36 – 55 DAP) were strongly associated with potato growth variables and final tuber yield.

  • The findings of this systematic review should be helpful in informing researchers, agronomists and farmers about the most suitable VIs and the most appropriate time for using remote sensing for potato growth monitoring and tuber yield prediction.