Monitoring Yield and Quality of Forages and Grassland in the View of Precision Agriculture Applications—A Review

Abstract

The potential of precision agriculture (PA) in forage and grassland management should be more extensively exploited to meet the increasing global food demand on a sustainable basis. Monitoring biomass yield and quality traits directly impacts the fertilization and irrigation practises and frequency of utilization (cuts) in grasslands. Therefore, the main goal of the review is to examine the techniques for using PA applications to monitor productivity and quality in forage and grasslands. To achieve this, the authors discuss several monitoring technologies for biomass and plant stand characteristics (including quality) that make it possible to adopt digital farming in forages and grassland management. The review provides an overview about mass flow and impact sensors, moisture sensors, remote sensing-based approaches, near-infrared (NIR) spectroscopy, and mapping field heterogeneity and promotes decision support systems (DSSs) in this field. At a small scale, advanced sensors such as optical, thermal, and radar sensors mountable on drones; LiDAR (Light Detection and Ranging); and hyperspectral imaging techniques can be used for assessing plant and soil characteristics. At a larger scale, we discuss coupling of remote sensing with weather data (synergistic grassland yield modelling), Sentinel-2 data with radiative transfer modelling (RTM), Sentinel-1 backscatter, and Catboost–machine learning methods for digital mapping in terms of precision harvesting and site-specific farming decisions. It is known that the delineation of sward heterogeneity is more difficult in mixed grasslands due to spectral similarity among species. Thanks to Diversity-Interactions models, jointly assessing various species interactions under mixed grasslands is allowed. Further, understanding such complex sward heterogeneity might be feasible by integrating spectral un-mixing techniques such as the super-pixel segmentation technique, multi-level fusion procedure, and combined NIR spectroscopy with neural network models. This review offers a digital option for enhancing yield monitoring systems and implementing PA applications in forages and grassland management. The authors recommend a future research direction for the inclusion of costs and economic returns of digital technologies for precision grasslands and fodder production.

1. Introduction

Forage and grassland crops provide a wide range of ecosystem services (soil, water, carbon) through producing food and feed, and show impacts on agricultural systems and the environment. Mainly forages and grassland communities involve interest in maximizing the utilization intensity and overall profits for advanced farming at the sustainable dimension. Farm management practises that sustain variable rate nitrogen (N) applications can increase N use efficiency of biomass and fodder crops without reducing economic yield [1,2]. Sustainable livestock production is dependent on animal feed and farm productivity levels, allowing the maximum utilization of daily fresh dry matter (DM) contents per animal [3]. The maximum utilization of fresh DM has become the cheapest feeding source for livestock production [4].

For the sustainable production and maximum utilization of forages and grassland species, it is very important to employ precision technologies in monitoring biomass yield and quality components that directly impact fertilization and irrigation practises [5,6,7]. However, the application of digital farming in forages and grassland management is highly dependent on monitoring systems and the procedure involved [8,9,10]. Additionally to this, monitoring technologies should not only be low-cost but also equipped with an easy installation procedure [11,12].

Traditional methods are labour-intensive and high-cost, and with low accuracy [11], grouped into direct and indirect methods [12]. The direct methods consist of harvesting by hand or machine, visual estimation, and a subjective approach as ocular estimates [13,14], recommended for vegetation cover over smaller area [11]. Manual methods are not applicable on large areas [12], and evaluating their performance on a regular basis is not very common [15]. For quality measurements, mostly laboratory-based methods are used that are time-consuming such as grinding, oven drying, and subsequent laboratory analyses [16]. The measurements by these systems could be improved by comparing with an indirect method [15,17]. Such indirect methods are based on calibration procedures like linear or multiple regressions [18,19]. In this system, yield monitors are used on self-propelled forage harvesters (SPFHs) to measure yield and moisture data of alfalfa, grass species, and corn silage [20,21]. However, these machines require precise calibration [22]. Addition to this, calibration must be performed based on different field conditions, and crop type, at least once a day or if changes occurred during harvesting. In this case, Rayburn et al. [18] used a rising plate meter to estimate forage and grass biomass through pre-calibrated regression models.

Currently, applications of digital monitoring technologies are becoming popular for grassland canopy vegetation; however, these systems are highly dependent on the procedure used, ease of implementation, and expertise [9,10]. For example, Seefeldt and Booth [23] used digital photographs for grassland canopy vegetation. However, this digital monitoring system has shown inaccuracy, much like traditional methods, and low spatial resolution and is time-consuming; thus, it is not considered feasible for large-scale canopy vegetation monitoring. Moreover, it does not consider multiple scattering between vegetation and soil, therefore creating uncertainties if applied on large and heterogeneous areas under mixed grassland [24]. Physical modelling approaches based on RTM have potential to change the physical relations between spectral values and grassland canopy cover [25]. These modelling-based measurement approaches have advantages in terms of transferability among different study areas; however, they have limited invert accuracy [26]. Furthermore, a large number of datasets are necessary to run complex RTM, especially over larger area [27]. Regression-based approaches rely on empirical relationships between plant spectral reflectance and canopy cover [28]. Many regression methods have been applied successfully in this field, such as linear/non-linear regression, random forest, back-propagation neural networks, and support vector machines [29,30]. Recently, regression-based approaches are being broadly employed for retrieving remotely sensed vegetation indices because of their high accuracy and transferability [31].

This review went further and analyzed the developments in digital monitoring systems, offering cost-effective solutions for challenges associated with regression-based approaches in mixed grasslands. For example, Un-manned Aerial Vehicle (UAV) or drone technology can estimate grassland canopy cover over on-demand high-throughput imagery (millimetre–centimetre resolution), using a similar approach like digital photograph-based classification [32,33,34]. Some of the other latest digital studies emphasizing high-resolution biomass monitoring and leaf area index (LAI) estimation for grassland-specific PA applications are the integration of Sentinel-2 and RTM [35]; combined Sentinel-1 (S1) SAR (Synthetic Aperture Radar) and daily weather data, through a machine learning-based model (Catboost) for grassland cut detection [36]; integration of MODIS (Moderate-Resolution Imaging Spectroradiometer) NDVI (Normalized Difference Vegetation Index) and weather information, through a synergistic grassland yield model, for determining the start of the growing season and phenological variables [37]; and integration of drone imagery and satellite data for grassland mapping [38].

Besides the advantages of these digital monitoring systems, precision technologies are comparatively less employed in grassland than forages [9]. Schellberg and Hejcman [39] noticed that the low adoption of digital farming in grassland might be due to high species diversity caused by spectral similarities among species, especially in mixed grassland. Currently, these issues are being resolved through digital models in mixed grasslands as highlighted via Diversity-Interactions mixed models, determining effects of various species interactions on agricultural ecosystems [40,41]. The main aim of the review is to highlight the advanced monitoring systems, available sensors, potential remote sensing technologies, and NIR spectroscopy applications in precision forages and grassland measurements. The review provides a wealth of case studies for precision forages and grassland management, a critical analysis on the applications of these technologies, an in-depth discussion of potential challenges, and a vision of future research directions. This review serves as a valuable source that will help researchers and decision makers to employ precise forages and grassland monitoring systems and to adopt digital farming in this field.

2. Review Search Methodology

The review includes scientific papers published from 2000 to 2024. It focused on the current monitoring technologies, including remote sensing methods, for forages and grassland measurements in terms of precision farming applications. In order to guarantee a thorough and objective evaluation, the articles were screened and chosen by following the individual sections in the planned review. The selection of relevant scientific papers was performed using Scopus, Springer, Wiley journal, ScienceDirect, and Google Scholar search engines. The selected papers within the search domain comprised life sciences, remote sensing, agriculture engineering, PA, agronomy, environmental science, biological sciences, and livestock production. The review focused on in-depth literature reviews, within the scope of updated forages and grassland monitoring systems in terms of digital agriculture applications, published from the years 2010 to 2024. The included articles were filtered by query strings. The keywords to search methodology are “PA”, “precision forage management”, “sensor used in agricultural machines”, “biomass sensors”, “grassland monitoring systems”, “satellite and proximal sensing”, “simulation modelling in precision forages and grassland agriculture”, “precision harvest”, “delineating sward heterogeneity”, and “DSS”. First, we analyze the monitoring systems of biomass and quality measurements, providing the advancements in remote sensing technologies, helping researchers to adopt digital options in forage and grassland management. Thereafter, the review discusses the challenges in delineating sward heterogeneity and their possible solutions in relation to the application of DSS in precision forages and grassland agriculture. The conclusion provides digital solutions to forage and grassland communities that will have positive impact on increasing their farm productivity on a sustainable basis while protecting the environment.

3. Current Monitoring Technologies

Among monitoring systems, remote sensing and UAV technology applications have been widely employed in monitoring biomass and quality traits [42,43,44,45,46,47,48,49]. LiDAR, terrestrial laser scanning, and airborne-based LiDAR are equally important in this field [50]. These robust technologies can provide accurate data concerning plant horizontal and vertical structure. Théau et al. [48] used UAV-based multispectral data for estimating fresh and dry biomass of grasslands. Some yield and quality traits of alfalfa and grasslands were estimated by canopy reflectance and biophysical traits [51]. Forage and grassland canopy vegetation were measured by Sentinel-2 data [52]. The DM and N contents of grasses were estimated by hyperspectral sensors [53]. However, aerial photogrammetry can also be used successfully in vegetation monitoring [54]. Some studies have employed the photogrammetry technique for pasture management [55,56].

Besides sensing technologies, many different sensors, load cells, and capacitance-controlled oscillators are employed on harvesting machines for measuring mass flow and impact force of the material [57,58]. In these studies, pull-type forage harvesters were used as compared to the latest SPFHs. Shinners et al. [59] installed sensors successfully on SPFHs. The John Deere moisture sensor, also known as HarvestLabTM (HL), works with the company’s GreenStarTM global positioning system (GPS) to measure the mass flow rate. However, the sensors that measure on-the-go biomass, such as a pendulum sensor, have shown more accurate results [60]. A forage-throughput sensor and large square baler can be used to measure the mass flow rate and moisture data of herbage crops [58,61,62,63].

Some harvesting machines and tools used in measurement systems are as follows: (i) the constantly weighing method using windrower machines [64], (ii) measuring the mass flow rate of grass and alfalfa using curved plates on a mowing machine [65]. Biomass yield of grass can also be measured by quantifying the torque and pressure of windrower devices or mowing machines [66]. New Holland’s bale-weighing system, if joined with GPS, has potential to develop forage maps. Waggon-based silage yield maps were produced by Lee et al. [57]. Larger and conditioner mowers can be used for monitoring hay and silage, whereas windrowers could be used in grass species. Demmel et al. [64] used a windrowing device to estimate yield of forage and hay crops. Furthermore, the application of a feed roller on a silage chopper, a harvesting machine, was used for picking up forage material from the ground, cutting them into small pieces and transferring them into a storage container for further processing. In this system, the amount of spring force was measured when crop material passed through the feed roller [59].

The DM contents of a silage crop were measured by sampling and the weighing method [67].

NIR spectroscopy is also considered as quick, non-invasive, accurate, and high-throughput technology for monitoring biomass and forage quality of switchgrass [68], and Napier grass [69]. This technology has some limitations such as the complex calibration, selection of spectra, type of available tissue, and external weather conditions. To solve these issues, understanding about the sample pre-processing and chemometrics techniques is required [70]. Akins et al. [71] measured the DM contents of forages using near-infrared spectroscopy (NIR) technology. Field spectrometers were used to estimate biomass cover and quality traits [72,73,74]. Currently, advanced monitoring systems such as the integration of satellite remote sensing and UAV technology with machine learning algorithms (such as random forest, convolutional neural network, and neural networks) have been widely used in predictive modelling for biomass crops and grassland management [30,75,76,77]. In this context, the integration of optical and SAR remote sensing data with machine learning models (such as random forest, and gradient boosting regression tree (GBRT) model) was used successfully to measure grass height with accuracy [78]. The DM yield of temperate grassland species was estimated broadly through the integration of UAV data with machine learning algorithms such as random forest, support vector machine, and partial least squares regression models [79].

3.1. Mass Flow Measurement Sensors

In mass flow measurement, weight and flow speed of the material are the key components. Basically, mass flow is measured based on the combination of weight and speed of the crop material during harvesting. Some examples are

• A laboratory-based dielectric sensor [80] in forage crops.
• Capacitance sensors with artificial neural network modelling, an Adaptive Neuro-Fuzzy Inference System, and multiple regression techniques in paddy rice [81].
• A parallel plate capacitance-throughput sensor in forage crops [82].
• Kormann [83] measured the mass flow based on the X-ray absorption technique.
• Huenink [62] used a large square baler that uses feed-roll displacement (flow speed) and bale weight to measure the mass flow rate.
• Shinners et al. [59] measured the bale weight while fitting the load cells with large square balers. They found that the dynamic bale weight method had the highest prediction accuracy (R² = 0.99) as compared to SPFH and self-propelled windrower machines.
• Shinners et al. [84] used a self-propelled forage windrower machine on forage cutting equipment, with a prediction accuracy of R² = 0.83–0.90.

3.2. Volume Flow Measurement Sensors

The volume flow rate can be measured by estimating the flowing speed of the material through sensors. It depends on the thickness of the material and growing conditions. However, mass density is important to be measured in volumetric measurement. Volume flow must be linked with the amount of plant material being harvested based on the collective integration of flowing speed, moisture content, speed, and width of the harvester (forage harvester, Green-star John Deere system, or other harvesters). These parameters influence the resulting maps of a harvested crop.

• The volume and mass flow rate of miscanthus and switchgrass were measured by a commercial auger [85].
• A linear potentiometer (Shinners et al., 2003), a vertical displacement transducer [58], and fixed load cells with springswere installed on forage harvesters, to measure volumetric flow of biomass materials [61].

3.3. Impact Sensors Installed on Harvesting Machines

There are different impact sensors available that can be used to measure crop quantity and quality contents. These sensors consist of compact plates, which are located in the windrowing apparatus of the harvesting machine (i.e., forage harvester). The efficiency of such yield measurement systems depends on the type of sensor used [86]. Many forage and grass species have different phonological and morphological characteristics (such as green-chopped, wilted hay or straw; perennial rye grass; etc.), which require a multi-sensor approach to assess their heterogeneity at the field scale. Therefore, future sensors should be developed based on multiple information that can be used over a wide range of plant diversity. Some examples are

• Kumhála et al. [87] used torque sensors and curved impact plates on a mowing machine for the measurements of forage yields. They found very good coefficients of determination (R² = 0.95) between the conditioner’s power, impact force, and material flow rate.
• Torque sensors and impact-type yield sensors were used to estimate mass flow and spatial maps of grass species [87].
• Kumhála and Prosek [65] measured the mass flow of grass and alfalfa crops through a curved impact plate mounted on a mowing machine.
• Savoie et al. [58] used a curved-type impact plate to determine the mass flow rate of hay material with 5% error, based on the moisture data through prediction modelling.

3.4. Other Available Sensors for Monitoring Biomass and Moisture Data

Several sensors can be employed successfully on forage harvesters to measure biomass traits and moisture contents [88]: a linear potentiometer (measures feed-roll displacement/volume flow), a capacitance sensor (measures moisture and mass flow rate) and NIR spectrometer (measures crop spectral properties and quality traits, as well as DM of hay and corn silage), a torque sensor (determines torque and shaft speed), load cells (measure weight of materials, crop flow, and feed-roll displacement), a strain gauge (assesses weight of crop and crop flow), and curved impact plates (measure crop impact force). These sensors must be calibrated for each plant species separately and at the optimum moisture level.

Recently, Digman and Shinners [89] used NIR sensors (also known as North American moisture calibration sensors) on John Deere SPFHs to measure the moisture and DM yield contents. Vomax stated that microwave sensing was more accurate for moisture measurement as compared to NIR sensing. However, NIR has shown optimal results in monitoring DM contents under field conditions [89] and laboratory studies [71]. Forage quality such as protein, starch, and fibre contents can be assessed by NIR sensors, in view of improving feed quality in field conditions [90].

NIR sensors installed on yield monitors can be used to measure DM contents of silage maize as well as mixed species (alfalfa and grasses) [20].

Digman and Shinners [91] used three sensors (capacitance, NIR, and microwave sensors) for moisture data.

Mechanical and electronic pasture sensors are used to measure pasture biomass [11]. Some examples of sensors that can be used to make moisture measurements are presented in

Table 1. Available sensors used for moisture measurements.

Measurement Systems	Conductivity Sensor	Dielectric Sensor	Capacitance Sensor	Microwave Sensor				NIR Sensor		Nuclear resonance Sensors	Nuclear radiation Sensors
				Transmission	Reflection	Resonator	Time area reflection	Transmission	Reflection
influencing factors	Performance
material temperature	M	M	M	M	M	M	M	L	L	VH	VH
material density	M	L	L	M	M	M	M	L	VH	H	M
layer thickness	M	M	M	M	VH	M	M	L	VH	H	M
corn size	M	M	M	M	M	M	M	M	H	M	M
sensor pressure	M	M	M	VH	VH	VH	VH	H	H	M	VH
material similarity	L	L	M	H	L	H	H	H	H	H	M
electrolyte content	L	M	H	H	H	H	H	VH	VH	VH	VH
material colour	VH	VH	VH	VH	VH	VH	VH	M	M	VH	VH
extraneous light	VH	VH	VH	VH	VH	VH	VH	L	L	VH	VH
biomass material	H	M	M	L	L	L	L	M	M	M	M
radiation	M	M	M	L	L	M	M	H	H	M	H
material velocity	H	H	H	M	M	VL	VL	H	H	M	H

Performance based on ranking “low (L), medium (M), high (H) and very high (VH)” on moisture measurement systems [88].

.

3.5. Remote Sensing Technology Applications for Monitoring Biomass Yield and Quality Traits

Recent advancements in remote sensing technologies have significantly enhanced our ability to estimate yield and quality dynamics in forages and grasslands (

Table 2. Remote sensing technologies used for monitoring biomass and quality traits in terms of precision forages and grassland management.

Methods/Sensors	Applications	Advantages	Limitations and Solutions	References
Optical remote sensing technology	Uses optical and hyperspectral images to calculate vegetation indices like NDVI for biomass estimation	Large coverage area, repeatable data collection	Lower spatial resolution as compared to drones, weather dependency. Solution: combination of multi-sensors (e.g., microwave and optical sensors)	[44,97,98]
Combined Sentinel-1 (S1) SAR and daily weather data, using a machine learning-based model (Catboost)	Detection of precision grassland cuts, for making fertilization and irrigation decisions	Enable cut detection over entire calendar year, eliminates missing and false cuts	Persistent cloud cover may affect results Solution: readings should be taken under clear sky conditions	[36]
Sentinel-1 backscatter time series	To determine the precise cut events such as harvesting frequency or utilization efficiency of grasslands by livestock	Determine the limiting factors that influence the cut detection	Factors like size, shape, and slope of plants may affect the detection of mowing events. Solution: requires dense in situ measurements to understand the behaviour of C-band backscatter of plants	[99]
Sentinel-2 data based on RTM of vegetation	Determination of biophysical variables, such as LAI calculated by advanced algorithm techniques, for biomass and LAI of grasslands	Observed lower root-mean-square error than direct measurements	Results might be affected under adverse climatic conditions. Solution: clear sky readings, coupling of cloud-free remote sensing with field spectrometer data	[35]
Sentinel-2 time series	Grassland cut detection for highest fodder utilization and production	Higher resolution over multispectral bands, high accuracy, frequent revisiting time	Can be affected by cloud cover. Solution: aggregation of Sentinel-2 and ground data with process-based algorithms	[100,101]
RGB (Red, Green, Blue) resolution sensors mountable on drone	Captures high-resolution colour images to estimate vegetation cover	High spatial resolution, flexible data acquisition	Limited battery life. Solution: further improvements in navigation and the miniaturization of these technologies	[102]
Multispectral cameras	Captures data in several spectral bands for biomass acquisition	High image quality, ease of use	Ambient conditions, requires ideal camera settings Solution: requires multiple measurements [78]	[103,104]
LiDAR (UAV/airborne biomass mapping)	Aid in assessing vegetation height, canopy structure, biomass, and LAI estimation	Detailed 3D data (resolution: 532–1064 nm), low cost, high accuracy and speed	Climatic factors such as fog, rain, snow, complex image pre-processing, segmentation, extraction, detection, and image quality assessment. Solution: specialized software and training	[105,106]
Hyperspectral imaging	Captures detailed spectral information (0.4–2.5 μm) for assessing biomass and quality traits	Accurate, time saving, broad spectral bands, produces large datasets, environmentally safe, high signal-to-noise ratio	Difficult for detecting different items within one image [107], ambient conditions. Solution: analyzing hyperspectral data with deep learning approaches [108]	[109,110,111]
Machine learning/Artificial Intelligence (AI)	Uses algorithms to analyze remote sensing data and predict biomass	Can improve accuracy and predictive capabilities	Requires large datasets and computational resources	[112,113]
Synthetic Aperture Radar (SAR) sensors	Microwave radar to estimate biomass by measuring surface roughness and vegetation structure	Effective under all-weather conditions, works day/night, penetrates canopy	Recommended for high aboveground biomass species [114]	[115,116]
Ground-based laser scanning	Laser pulses to measure vegetation structure and density	Rapid, providing dense spatial data of plant species	High cost, sensitive to extreme weather, low area coverage. Solution: Airborne Laser Scanning [117]	[118]
AVHRR (Advanced Very-High-Resolution Radiometer) and MODIS sensors	Plant health and vegetation assessment	Estimating the spatial variability of vegetation cover and plant health	Limited resolution, complex calibration, difficulties in assessing consistent time series data [119]. Solution: coupling of AVHRR and MODIS data [120]	[121,122]
Integrated remote sensing data (MODIS NDVI) and weather information through synergistic approach for grassland yield modelling	Detection of start of the grassland growing seasons for phenological observations, and biomass estimation of grasslands	Daily revisiting frequency, cover large area, time saving	This method gives high NDVI values that may fall outside of the growing season. Solution: requires future research for further validation of this technique	[37]
Small-sat technology	Equipped with sensors that can determine vegetation canopy cover, soil moisture, growth stages, and weed and pest detection	Cost effectiveness, high resolution, frequent revisit times	Limited sensor capacity, short lifespan, less coverage. Solution: digital smart technology powered by the IoT, AI, and machine learning techniques [123]	[124,125]

). These cutting-edge technologies, including satellite imagery, UAV-based sensors, and ground-based optical sensors, offer unprecedented efficiency in monitoring and utilizing (cuts of) forages and grasslands [92,93]. Remote sensing provides valuable insights into the spatial and temporal estimation of biomass, which is crucial for optimizing grazing management and ensuring sustainable land use [45,94]. As remote sensing approaches have received great attention in the advancements of PA and sustainable development, therefore, its role is becoming more and more critical in digital agriculture [95,96].

Optical remote sensing data are not only influenced by external weather, and diseases and insect pest attacks, but also by crop growth responses, the crop type, and the manner of farming [44]). Therefore, one should quantify the uncertainty in satellite sensing data produced under diversified climatic conditions and other factors [126], in order to avoid biasedness in the spatial data produced [127,128]. In this context, NIR technology is playing a promising role, which could be installed on drones, SPFHs, or combine harvesters [129]. This sensing system enables us to monitor biomass and quality traits [130]. Currently, researchers are developing new sensors using blue, green, and NIR spectra. For example, Fan et al. [131] changed the setting of a visible NIR camera for measuring LAI of Italian ryegrass.

Integrated cloud-free high-resolution data with machine learning modelling equipped us with the precise estimation of several spatial parameters that could be useful for grassland management [10,132,133]. These technologies are not only being used in understanding grassland dynamics but also in decision-making processes for sustainable development [42,134].

In the end, UAVs or LiDAR equipped with high-resolution cameras such as multispectral, hyperspectral, and thermal sensors evaluate crop and soil health and insect pest infestation in real time. The data produced from such systems and analyzed through AI and machine leaning models offer us a programme to make innovative farming decisions for maximizing resource use efficiency and advanced farm management practises. These benefits may include those such as saving labour costs, improved farm productivity, and reduced environmental impacts from irrigation, fertilization, and pesticide applications. There are some successful examples of these technologies (UAVs, LiDAR) that should be adopted in resource-constrained contexts such as integrated UAVs with AI and machine learning used for variable rate fertilization and automated crop health assessment in real time, based on data-driven farming decision making that improves resource use efficiency [135]. UAV data are analyzed by Deep Neural Networks to make decisions through resource-constrained devices [136]. The integration of UAV hyperspectral data with machine learning algorithms enables effectively eliminating noise issues, providing spatio-temporal maps of water bodies with high prediction accuracy (R² of 0.91) [137]. This method may improve farm conservation practises for agro-ecosystems. Besides this technology, infrared thermal imaging has also shown promising results in agricultural water monitoring and irrigation scheduling [138].

4. Fodder Quality Aspects Using NIR Spectroscopy

Most conventional methods for the measurement of herbage quality require plant samples from the field and pre-processing for laboratory analyses. The recent advancements in NIR have reduced the pre-processing analysis by means of the in situ NIR technique. This technique has not required the external chemical input as compared to traditional methods such as wet chemistry. The advanced NIR is a rapid method reducing laboratory workload for the chemical analysis of herbage crops [139].

It is well known that an NIR technique requires precise calibration with the individual species. Parrini et al. [140] used 105 samples of pastures, and demonstrated that NIR technology can actually measure the chemical composition of pasture species. However, laboratory-based measurements are required for robust calibration in view of restraining the error in actual data, providing a source of reference for accuracy in final results [141]. It is important to include an adequate number of samples in calibration [142]. During spectra collection, it is recommended that fresh samples should be weighed at regular intervals. In general, fresh samples are not always available; so, spectra are mostly taken from dried samples, and calibration performed is mostly based on already processed samples. Furthermore, special attention is required for drying and grinding the samples, because water is a strong absorbent to NIR wavelengths, and particle size influences the whole spectrum. The NIR spectrum must be obtained in uniform conditions through including appropriate sample quantity [143]. All these premises of NIR calibration are important for a quality analysis [140].

Thereafter, calibration should be performed by introducing the spectro-chemical modelling technique that builds an association between the spectral region and compositional trait. In this context, Decruyenaere et al. [141] used the spectro-chemical model for the prediction of chemical traits such as crude protein (CP) and DM contents. In general, high R² and low error parameters in prediction indicate the correct use of the calibration procedure [16,144,145].

4.1. Assessment of Quality Traits

Recently, forage and grass quality analyses by means of NIR have become popular in the agricultural industry [146]. The improvement in the development of the NIR technique is achieved through coupling of NIR with machine learning for detecting quality traits [147]. The main quality parameters are as follows:

4.1.1. Moisture Contents

Harvesting plant material at proper moisture is important for feed conservation and can increase the nutrient quality of the animals’ feed. Therefore, accuracy in moisture measurement is very crucial. In this context, moisture contents were measured by NIR spectroscopy in corn stover silage [148] and forage soybean [149]. Jin et al. [150] determined the leaf water contents of Miscanthus grass species by integrating visible and NIR spectrometry with multivariate modelling techniques.

4.1.2. Organic Dry Matter Contents

The DM contents can be analyzed by the NIR technique [151]. Fernandes [152] measured the DM of fresh and processed samples, with high prediction accuracy (R²) from 0.87 to 0.89. Castro et al. [153] measured the organic matter contents of dried and ground samples with R² = 0.92. Higher correlations showed the similarity between both spectral treatments and principal component analysis techniques. By comparing the pre-processing of samples for NIR calibration, the overall performance was better when processed with dried and grounded samples than fresh samples [152].

4.1.3. Protein Contents

The CP often calculated as N × 6.25 and total nitrogen are widely measured by the NIR method with very high prediction accuracy (R²), due to strong absorption bonds among –N-H in the NIR spectrum [143]. This fact is confirmed by considerably good association among plant species (R² = 0.60 to 0.99) [145,152,154]. The NIR calibration curves were developed for each chemical trait using a group of samples, and predictive equations were developed. Finally, validation was performed using external samples, and a correlation coefficient over 0.99 was evaluated for the used samples except those that showed the accuracy of the predictive equations [145].

4.1.4. Fibre Contents

The neutral detergent fibre (NDF), acid detergent fibre (ADF), and acid detergent lignin (ADL) are some of the other most expedient quality traits of forages measured by NIR technology. Many past studies developed robust calibration equations using a large number of external samples for NIR calibration, which showed R² ranging from 0.62 to 0.99 for NDF, 0.75 to 0.97 for ADF, and 0.86 to 0.94 for ADL [145,152,154,155]. These methods are highly dependent on spectra used, calibration equations, and environmental conditions [155]. Hence, the robustness in their prediction accuracy was achieved based on a large number of datasets. Additionally to this, Molano et al. [145] recommended further research to improve the accuracy of analytical methods for ADF determination.

4.1.5. Fat/Lipid Contents

The determination of lipids in plants is not very common with NIR technology, due to their low concentration in forage species. Additionally, fats do not show quite specific reflection or absorption bands in NIR. However, the experiments conducted so far have been heterogeneous, as Ullmann et al. [154] analyzed 245 samples of forage species and obtained prediction accuracy with R² = 0.94 and standard error of calibration (SEC = 2.17), whereas Fernandes [152] used 115 samples of forages and achieved medium-level prediction accuracy (R² = 0.51 and SEC = 0.60), which may be due to high similarities between databases used that should be treated as a single model for each species.

4.2. Limitations, Potential Problems, and Room for Improvement of NIR Spectroscopy

Besides technological advantages, NIR has some limitations and potential problems such as that standard error in the laboratory samples can cause error in final prediction accuracy, and other biological related measurements [156]. Other challenges and potential problems might be related to spectral measurements, regression models, and machine learning approaches, which can pose problems such as overfitting [157], pattern recognition problems [158], or a lack of understanding of choosing the right samples and processing steps. To solve these problems, sample pre-processing and prediction techniques need to be upgraded that can adapt the pattern; otherwise, the robust model must be taught each year again from a scratch version. Machine leaning-based neural network issues could be resolved through using ‘back propagation neural network’, ‘particle swarm optimization’, and ‘generalized regression neural network’ methods [158].

The necessity of measuring moisture content of samples poses challenges in a quality analysis, especially while assessing protein and sugar content. For solid samples, procedures such as lyophilization or drying are recommended for solving this issue. For liquid-based samples, the transflection spectrum method is advised [159].

Temperature plays an important role in prediction models with NIR spectrometry. It modifies the location as well as intensity of the NIR spectral bands, affecting the predictive accuracy of calibration models. This prevailing problem could be solved through the global temperature compensation technique. Global models use spectra over a full temperature range that gives better predictive results than local models [160].

Hong and Chia [161] published a review on prevailing challenges and their possible solutions of NIR measurements.

Future studies are recommended to discuss the impacts of samples on light scattering that can predict NIR measurements with high accuracy [162]. Besides this, NIR spectroscopy concerning temperature equilibrium should be conducted based on robust calibrated models or recently developed techniques like a ‘light emitting diode’ for the ease of a rapid NIR analysis. Machine learning-based neural network models can become a next-level solution in this field.

5. Precision Forages and Grassland Management

Precision and digital technologies are becoming popular for biomass and quality estimation and management in the site-specific dimension, aiming to increase the resource use efficiency for sustainable development [9,16]. Currently, this technology has been employed over a wide range of agri-food industries [163,164]. In contrast to PA technology, traditional estimation methods are often time-consuming and equipped with low accuracy [16]. For maximum benefits, low inputs and accurate biomass-monitoring methods like the estiGrass3D+ model could be a great opportunity in this field [165].

Most grasslands are semi-natural, which evolved under livestock grazing and regular mowing events. These semi-natural grasslands are highly influenced by diverse plant species, frequency of harvest intensity, and fertilization practises. Concerning this, Ineichen et al. [7] used forage samples, such as yield, botanical composition, and concentrations of net energy, utilizable crude protein, and different phenolic fractions, from three long-term fertilization experiments over a 2-year duration. This study concluded that combining the fertilization of phosphorus and potassium can offer an increment in forage productivity and enhance forage quality under mountain grasslands. Further, most pastures are located on hilly areas [166], resulting in variability in the excrement of grazed livestock and finally soil spatial nutrient availability. This can cause inaccuracy in biomass measurement, as demonstrated by Schori [167]. Borra-Serrano et al. [168] suggested to include weather data like Growth Degree Days (GDDs), as the GDD difference between harvesting and measurement dates can improve herbage biomass estimations. Additionally, the random forest algorithm is the most promising approach that could be used for monitoring grass and forage biomass [169]. Generally, the random forest method is considered more precise for biomass prediction as compared to linear models [170].

Furthermore, sensor technology, crop-modelling-based DSS, and robotics are recommended precision technologies for forage and grassland monitoring and management [39,59,171,172,173].

Most farmers are busy, having varied levels of skills and knowledge of PA technologies [9]. However, precision farming in forages and grassland is still low due to its complexity and high cost [174,175,176]. To solve these issues, the training and education of farmers by expert researchers may increase the awareness of positive impacts of digital farming on enhancing farm productivity, overall feed quality, and societal and environmental benefits [10].

5.1. Delineating Sward Heterogeneity

Delineating agricultural farm heterogeneity is crucial in precision farming [10]. It is well known that spatio-temporal heterogeneity always occurs in field conditions, due to internal and external factors that influence crop behaviours throughout the growing season [10]. These influencing factors can be related to soil entities, human intervention, biotic agents, external weather conditions, species interaction effects, previous cropping history at a particular field, and the manner of farming practises [10,40]. For understanding such complex heterogeneity influenced by various factors as mentioned above, it is recommended to cultivate crop species over a single-type or defined crop mixture, whose intrinsic behaviour could be assumed as a reference source in terms of influencing exertion occurring by these factors.

Thus, it is important to determine each influencing factor that affects sward heterogeneity over forages and grassland cultivation. In general, heterogeneity is considered higher under intensive grazing, which increases uncertainty in biomass measurement and management [177]. Previous research suggested that the integration of plant species helped in maintaining field biodiversity [178]. However, delineating sward heterogeneity has been quite difficult under mixed grasslands due to spectral similarities among species, or the limitation of remote indices like NDVI. These challenges can be resolved through integrating spectral un-mixing techniques or using hyperspectral imaging. Examples are the super-pixel segmentation technique used by Sun et al. [179]; hyperspectral un-mixing [180]; and un-mixing a Sentinel-2 median composite using a multi-level fusion procedure [181]. Authors also suggested to employ the “diversity-interactions mixed model” for evaluating growth and productivity of different species in mixed grasslands. These interaction models use species proportions and their interactions as predictors using a regression equation to determine biodiversity and ecosystem function relationships. However, it can be difficult to model numerous interactions over various species; also, interactions can be temporally variable, and may be dependent on spatial planting or other farming information. McDonnell et al. [40] developed a new Diversity-Interactions model through including weed invasion data from 16 species as random effects and within a plot species planting pattern over three years. The random effects were determined based on pairwise species within plots, rather than the more common plot level in mixed models. Brophy et al. [41] developed the Diversity-Interactions model using a data fusion from grassland and microbial experiment based on fixed coefficients and random effects to determine the interaction effects of species diversity on an agricultural ecosystem. This model offers improved standard errors for testing fixed coefficients and includes lack-of-fit tests for diversity effects. Therefore, these Diversity-Interactions models limit the complexities concerning how species interactions contribute in species-rich grassland ecosystems.

Sward height of herbage crops was measured by a rising plate meter for pasture management [16]. Barthram et al. [182] determined the sward height (variability over 30–70% of total 100%) in perennial ryegrass (Lolium perenne), because of selective grazing by sheep. Rook et al. [183] stated that continuous grazing may raise problems in managing field heterogeneity. To overcome this problem, mowing can be used to homogenize swards. Such events can be detected using remote sensing technologies in grassland farming such as integrating SAR and Sentinel-1 data [184]. This NIR technology has been widely used for monitoring biomass, quality traits, and species identification [132,185]. Xue and Su [186] produced high-quality vegetation indices during different growth stages of crop plants, but image quality might be affected by the presence of clouds and fog. Such issues could be resolved through coupling of remote sensing data with simulation modelling for developing DSS in forages and grassland agriculture [187,188]. Further, determining sward height with UAVs is highly recommended, using digital terrain model and machine learning approaches [165].

Global Navigation Satellite System (known as GNSS) technology allows for monitoring structural health of plant species [189]. Mistele and Schmidhalter [190] developed a spectral sensor with specific configuration to assess the N status of various plant species. Precision yield trends (genetic and non-genetic traits) were determined in forage maize and perennial ryegrass [191]. Li et al. [192] developed variable rate irrigation zones based on apparent soil electrical conductivity data for improving forage yields of Sudangrass (Sorghum sudanense) and alfalfa (Medicago sativa). Bailey et al. [193] collected plant samples manually for producing spatial maps of DM and nutrient contents of grass species. Spatial maps of phosphorus and potassium could be considered as a reference for evaluating nutrient status of grass species [194,195]. These sensors detect the reflectance properties of plants, which can help in managing mineral nutrients in a site-specific way. Schellberg and Hejcman [39] used the NIR technique to determine the application rates of slurry as an organic fertilizer.

5.2. Decision Support Systems

Currently, DSSs are becoming popular for PA applications in forage and grassland management. A broad spectrum of DSS has depicted many advantages within the domain of nutrient, insect pest, and livestock management [196]. Many successful studies are available that can be employed to promote digital farming in grasslands. For example, Herrmann et al. [197] integrated various factors such as N fertilizers, the defoliation rate, the type of grass species, and meteorological data to estimate biomass and CP of herbage species. AgroMANAGER is a web-based application that is supportive in keeping farm records, and helps in monitoring farms online [198]. Different European countries are offering access to open-source DSS tools such as MesParcelles in France, AgrarCommander in Austria, NMP Online in Ireland, MarkOnline in Denmark [199], and Web Module Düngung in Germany. The European project GrassQ was developed for precision grassland measurement and management, using proximal and satellite data [200].

Some recent studies demonstrated that remote sensing data integrated into DSS workflows could be used for tasks like grazing rotation or fertilization scheduling in grassland farming. Concerning this, integrating Sentinel-1 (S1), SAR, and optical remote sensing (Landsat-8 and Sentinel-2) data through multiple linear regression, support vector machine, and random forest was used to determine LAI and aboveground biomass. The results demonstrated that combining S1, Landsat-8, and S2 has a potential to monitor a grazing tallgrass prairie to offer timely and accurate data for precision grassland management [201]. Thereafter, Klingler et al. [35] developed an optimal grassland growth model as a support system based on the integration of Sentinel-2 data with RTM. The study concluded that seasonal changes showed a significant effect on canopy LAI and morphology; therefore, attention should be given during grassland management, i.e., fertilization or other decision making. Watzig et al. [100] developed a support system for improving Austrian fodder management and production systems, in terms of detecting accurate timing of grassland cuts (post-season, in-season) using Copernicus Sentinel-2 imagery. Cuts were detected based on threshold differences between a fitted NDVI curve (using Whittaker smoother) and actual NDVI observations. Moreover, false cuts were eliminated, using a gradient boosting algorithm that may be introduced due to missed cloud cover. Thereafter, Dujakovic et al. [36] went further and integrated Sentinel-1 (S1) SAR and daily weather data using a machine learning-based model (Catboost) for determining grassland cut events over an entire year and it eliminates the need for fixed growing season start/end dates. This method provides cost-effective grassland monitoring that serves as a DSS to determine mowing events influenced by fertilization inputs.

Bochtis et al. [202] developed a DSS for soil-sensitive fields to support farmers for minimizing damages caused by large agricultural machines. Delaby et al. [203] developed a DSS for precision pasture management. Navarro-Hellin et al. [204] recommended the ‘smart irrigation decision support system’. This support system is considered more accurate and efficient for developing smart irrigation plans with the minimum amount of water usage compared to traditional irrigation planning. Thereafter, Lopez-Riquelme et al. [205] developed a decision support PA application based on the FIWARE cloud, demonstrating a beneficial service in smart irrigation. Some studies on delineating field productivity use online crop and weather databases for site-specific farming in grassland management [206,207].

Besides these systems, Asher and Brosh [208] introduced a new monitoring technology, called the herd management system, which can be used to measure energy balance and forage quality for livestock production. This monitoring system is efficient for improving land management decisions on animal grazing. Subhashree et al. [209] developed a web tool including a forage economics calculator that supports forage management (alfalfa, corn). Their economic analysis includes the net return, break-even ratio, pay-back period, and final economic returns. This multi-device-based web tool can be installed on an automatic bale picker, recommended for large field areas.

Recently, advanced machine learning algorithms (such as random forest and convolutional neural networks) are being employed in predictive modelling, which help with specific decisions for grassland management [8,10,188]. Examples are the Ecological Succession Model [210], Agricultural Land Management Alternative with Numerical Assessment Criteria model [8], Simulation of Production and Utilization of Rangelands model [211], and Pasture Growth Simulation using grass measurements and weather data [212]. These models have capacity to combine various factors over complex grassland species for canopy biomass and aligned robust mappings. Xie et al. [213] employed Landsat remote sensing data to develop prediction models like artificial neural network and multiple linear regression models. Enough research has been performed on grasslands; however, the limited number of available remote indices and the support from long-term experimental data remain as potential challenges [30].

Current Challenges in the Adoption of DSS and Their Possible Solutions

The price of an agricultural commodity can be affected by factors such as gross productivity, logistics and inventory costs, demand and supply, etc. Some changes in single or more factors can cause uncertainty. Therefore, it is recommended that DSS must pay attention to prevailing uncertainty because of dynamic factors. Further to this, re-planning is a critical challenge in the development of DSS. This is because unexpected issues may arise after some time, like mechanical failures of a harvesting machine and abrupt weather changes. These issues may affect decisions on agricultural tasks such as precision harvesting and spraying, or precision nutrient or smart irrigation operations [214]. Therefore, DSS should be adjusted based on smart strategies or a new solution shall be developed for further decisions for farmers and stakeholders to complete the assigned tasks.

The application of digital DSS by smallholder farmers depends on three factors: (i) people (knowledge and skills), (ii) developed technology (ease of use and affordable price), (iii) rate of efficiency (high performance, high profits). High costs of digital DSS are the main challenge in the adoption of such high-tech methods for smallholder farmers due to the current financial crisis. To solve this issue, Oliveira-Jr et al. [215] developed an Internet of Things (IoT)-based digital system for small-scale farms. This support system is easy to use and has a low price, working on the principle ‘do-it-yourself’. The DSS can be employed on a smartphone.

Some researchers are trying to develop DSS that can solve agricultural problems autonomously. However, currently available DSSs are not yet developed with such an intelligent level because of complex agriculture and limited computation time. Further, these systems may provide inaccurate, even wrong, suggestions to farmers. So, experienced researchers are needed for validating the feasibility and correcting the mistakes of developed decision supports [216].

Further, implementation costs and potential economic returns of DSS have not been fully investigated yet in the published literature. Authors recommended the future research to further investigate the digital farming applications in terms of implementation costs and economic returns that will open new opportunities in precision grasslands and fodder research [9].

6. Conclusions and Future Research Directions

This review offers practical advice for enhancing biomass yield, and implementing quality monitoring systems and digital farming techniques in precision forages and grassland management. In this context, the authors provide current monitoring technologies and methods, including mainly mass flow sensors, volume flow sensors, impact sensors, moisture sensors, and remote sensing and NIR technology, field heterogeneity, and digital DSS, for assessing biomass, fodder, and grass quality traits, providing some clues for utilization intensity of fodder crops for livestock production.

The authors suggest that integrating multiple sensors’ data, produced by hyperspectral and thermal sensors, on-the-go sensors, torque sensors, radar sensors, NIR sensors, and optical sensors, could be a potential solution in evaluating biomass and quality traits. These types of data could be used in the decision-making process during the growing season, especially on the time of precision harvesting (cuts). It is required to search sensors for all types of plant species and mixed stands. In this context, the available digital sensors must be upgraded, and equipped with multiple data information (crop species, soil characteristics, weather information), especially for mixed grassland species.

Among plant sensing technologies, hyperspectral–UAV imaging and field spectroscopy can contribute accurately in monitoring yield and quality dynamics; however, substantial expertise is required in image processing and analyses to detect relevant crop traits. Most advanced remote sensing technologies should be employed in monitoring biomass and quality traits: the integration of remote sensing and weather data through a synergistic approach such as grassland yield modelling by combining Sentinel-2 with RTM, Sentinel-1 backscatter, and Catboost–machine learning techniques.

The NIR high-throughput technology offers advanced solutions in monitoring forage quality (such as moisture, DM contents, protein, fibre, lipids, and metabolizable energy contents); however, care should be taken during spectra collection and calibration for tissue type or temperature conditions. To solve these NIR calibration and validation issues, machine learning-based neural network models or pre-calibrated regression models are suggested.

It is quite challenging to delineate sward heterogeneity in mixed grasslands, which might occur due to spectral similarities among species or the limitation of remote indices like NDVI. To solve these problems in mixed grasslands, specific solutions are recommended like the application of mentioned Diversity-Interactions models, and integration of un-mixing spectral techniques, or hyperspectral imaging, such as the super-pixel segmentation technique, and multi-level fusion procedure.

For sustainable grassland management and production, integrating a multifunctional approach will be supportive in managing trade-offs for regulating socioeconomics and policy measures across grassland communities around the world, including implementation costs, impacts, and overall benefits of managing ecosystem services and agricultural systems. Further, there is need to include digital technologies for data acquisition or augmentation in diverse environments, such as Electronic Identification Systems, RFID, Soil and Plant Sensors, Smart Phones, and High-Throughput Proximal/Remote Sensing platforms, Diversity-Interactions models, machine learning and AI-based algorithms, and Automatic Application Controllers.

The authors recommend future studies on the advancement of the development of sensor technology by increasing their adaptability, dependability, and ease of implementation for improving biomass yield and quality in mixed grasslands. In addition, economic analyses in digital farming are also missing in the current literature; research on this subject may help farmers to adopt advanced farming in grasslands and fodder production.