In the first part of this study, the potential application of hyperspectral remote sensing (i.e., with information on many, very fine, contiguous spectral bands) in estimating N concentration in sugarcane leaves was examined. In the last part of this study, the ability of vegetation indices derived from multispectral data (Landsat TM and ETM+) in predicting sugarcane yield was investigated.
Plagiarism
Publications and manuscripts
Background
These stressors can affect crop function and consequently limit its growth (Baret et al., 2007; Jones and Schofield, 2008). In this context, remote sensing can play a unique role due to its ability to provide fast, synoptic, relatively inexpensive and near-real-time data over large areas (de Boer, 1993; Lillesand and Kiefer, 2001; Kumar et al., 2003; Aronoff , 2005).
Sugarcane nitrogen status
However, these methods do not meet the need for real-time and non-destructive strategies for monitoring crop N status (Zhu et al., 2008). Therefore, methods that utilize the advantages of remote sensing, especially hyperspectral remote sensing (i.e. information on many, very fine, contiguous spectral bands), are needed for diagnosing the N status of crops.
Sugarcane thrips
Complementary methods are needed that can provide repeatable, fairly accurate, unbiased means of monitoring thrips damage and infestation. Remote sensing, especially spectroscopy, offers advanced techniques that can provide such necessary thrips monitoring protocols.
Aim and objectives
To investigate the potential of spectroscopic data to detect sugarcane thrips damage at leaf level. To investigate the potential use of spectroscopic data in predicting leaf-level thrips counts, and.
Scope of the study
To explore the use of in situ spectroscopic data to estimate sugarcane leaf nitrogen concentrations,. To investigate the potential of imaging spectroscopy in predicting sugarcane leaf nitrogen concentrations using Hyperion data,.
Description of the study area
- General
- Umfolozi mill supply area
9 growing areas in South Africa, there are two types of producers in Umfolozi: large-scale and small-scale. Most of the farms in the large-scale growing sector have full or supplementary irrigation (irrigated or supplementary irrigation sectors respectively) and those farms that are cultivated under high rainfall conditions towards the coast have no type of irrigation (rain-fed growing sector).
Outline of the thesis
Application of remote sensing techniques in sugarcane (Saccharum spp. hybrid) production: a literature review. Applications of remote sensing techniques in sugarcane agriculture have been undertaken with particular emphasis on sugarcane classification, surface extent mapping, thermal age group identification, variety discrimination, crop health, nutritional status monitoring and forecasting. of yield.
Introduction
The use of remote sensing in identifying sugarcane varieties and monitoring the health, condition and nutritional status of sugarcane is discussed. The study also highlights the challenges and opportunities associated with the successful application of remote sensing in sugarcane production.
Light interaction with sugarcane canopies
18 The spectral response of sugarcane is also influenced by pigments in the leaves, such as chlorophyll a and b, carotene, xanthophyll and anthocyanins (Guyot, 1990; de Boer 1993; van der Meer et al., 2003). Canopies with high LAI reflect much more than a canopy with medium or low LAI (Simões et al., 2005; Fortes and Demattê, 2006).
Sugarcane classification and areal extent mapping
In India, Rao et al. 2002) estimated sugarcane area using Indian remote sensing satellites (IRS) and Landsat TM datasets. 22 In Australia, Everingham et al. 2007a) employed three classification procedures, namely discriminant analyses, random forest, and support vector machine to distinguish between nine sugarcane cycles (i.e., the number of times the plant regrown after harvest) using Hyperion data.
Varietal identification
Fortes and Demattê (2006) used data from Landsat ETM+ to distinguish between four sugarcane varieties (Figure 2.2) by analyzing individual spectral bands and spectral vegetation indices. 25 The use of fine spatial resolution sensors (e.g. IKONOS, QuickBird platform) for distinguishing sugarcane varieties has not yet been tested.
Monitoring sugarcane nutritional status, health, and condition
- Detection of nutrient and water deficiencies
- Disease detection
By using hyperspectral remote sensing, the sugarcane varieties can potentially be discriminated at a high level of accuracy. The results also verified that areas with the sugarcane rust disease show differences in spectral reflectance signatures and can be distinguished from non-diseased areas, at certain wavelengths.
Sugarcane yield prediction
The estimated LAI values resulted in better sugarcane yield forecasts when used as input to a process-based growth model. In addition, there is a need for testing fine-resolution sensors for sugarcane yield estimates, especially for large fields.
Overall challenges and opportunities
Furthermore, with the exception of Bappel et al. 2005), no study implements the combination of sugarcane growth simulation models with sugarcane biophysical parameters derived from remotely sensed data to provide relatively accurate estimates of sugarcane yield. Despite these shortcomings, there is no doubt that remote sensing technology will permeate many aspects of sugarcane farming in the future.
Introduction
36 barely (eg, Jørgensen et al., 2007) determining the appropriate wavelength or combination of wavelengths that characterize N deficiency. The aim of this study was to evaluate the use of spectroscopic data to predict N content in leaves in sugarcane.
Material and methods
- Leaf sample collection
- Leaf spectral measurements
- Chemical analysis
- Spectral transformation
- First-order derivative spectra
- Continuum-removed spectra
- Statistical analysis
For this task, the first-order derivative of the reflectance values in the range between 400 and 2500 nm and the continuum-removed reflectance in two areas of the visible part of the spectrum, namely R460−530 and R550−750, were used. 40 Rλ1i = first-order derivative of the reflectance or non-continuum reflectance at the ith wavelength, where there was a strong correlation (r ≥ 0.6 or r ≤ −0.60) between the first-order derivative of the reflectance or non-continuum reflectance and N concentration .
Results
- Foliar nitrogen concentration
- Discriminating age groups using spectral signature
- First-order derivative of spectra
- Continuum-removed spectra
- Validation
The results of the univariate correlation test showed that the leaf first-order derivative of reflectance was significantly correlated with N concentration in some wavelengths (p ≤ 0.05). The vegetation indices included in the models for estimating leaf N concentration for the 4-5 month sugar cane crops were generated based on the first order derivative of the reflection at wavebands 744 and 746 nm located in the visible near-infrared region (VNIR) of the electromagnetic radiation. spectrum and 2142 nm, which is in the mid-infrared.
Discussion
However, the models for the 6–7 month age group included combinations of the first-order derivative of reflectance only in the mid-infrared region. Models generated from the first-order derivatives of the reflectance showed better results compared to those developed from the reflectance subtracted continuum.
Conclusions
The aim of this study was to investigate the use of in situ spectroscopy for the estimation of sugarcane leaf nitrogen (N) concentration. These results suggest that the in situ spectroscopy has potential use in predicting sugarcane leaf N.
Introduction
They found that a vegetation index based on normalized difference vegetation index equation, namely NDVI could explain 86.9% of leaf N concentration in sugarcane crops at 6-7 months old. Thus, this study aims to explore the use of in situ spectroscopic data to estimate the N concentration of sugarcane leaves.
Material and methods
- Sampling procedure
- Leaf spectral measurements
- Chemical analysis
- Spectral transformation
- First-order derivative spectra
- Continuum-removed spectra
- Data analysis
Derivatives involve calculating the slope of the spectrum (the rate of change of reflectance with wavelength). Rλ1i = the first-order derivative of the reflectance or the reflectance removed from the continuum at the second wavelength at which there was strong correlation (r ≥ 0.6 or r ≤ ─0.60) between the first-order derivative of reflectance or reflectance removed from continuity and N concentration Rλ2i = first-order derivative of reflectance at the next i-th wavelength between 400 and 2500 nm or reflectance removed from continuity at 460 to 530 and 550 to at 750 nm.
Results
- Descriptive statistics
- Relationships between transformed spectral reflectance and leaf N concentration
- Validation
R2 values and root mean square error of prediction (RMSEP) were recorded to test the performance of the developed models. Low RMSEP values were obtained in the one-to-one relationship between predicted and measured leaf N concentrations, as well as high validation R2 values indicating the predictive ability of SR SR and SR indices.
Discussion
One-to-one relationships between measured and predicted sugarcane leaf nitrogen concentration (%) for a sample data set (n = 25) using the zero-one cross-validation method for modified vegetation indices from: (a), (b) and (c) the first-order derivative of the reflectance and (d), (e), and (f) the reflectance removed from the continuum.
Conclusions
Estimation of sugarcane leaf nitrogen concentration using normalized ratio indices generated from EO-1 Hyperion hyperspectral data. The aim of this study was to investigate the potential of imaging spectroscopy in predicting sugarcane leaf nitrogen concentration using Hyperion data.
Introduction
For the maize crop, Osborne et al. 2002) recommended that canopy reflectance at the red edge position could explain 81% of the variability in maize leaf nitrogen. For potato crop, Jain et al. 2007) showed that canopy reflectance index at 750 and 710 nm resulted in an R2 of 0.55 for predicting potato leaf nitrogen content.
Materials and methods
- Image acquisition and preprocessing
- Field data collection and chemical analysis
- Data analyses
- Random forest ensemble
- Stepwise multiple linear regression
- Validation
The wavebands were ranked according to their importance in predicting sugarcane leaf nitrogen content as determined by a random forest algorithm. The random forest algorithm was recalibrated against ntree and mtry and used to predict nitrogen concentration.
Results
- Descriptive statistics
- Optimisation of random forest regression models
- Random forest predictive models
- Stepwise linear regression model
- Validation
Optimization of random forest parameters (ntree and mtry) using RMSEP: (a) random forest model run with first-order derivative of reflectance and (b) random forest model run with vegetation indices based on NDVI. Selected vegetation indices were generated from wavelengths in the visible (478 and 518 nm), red (691 nm) and mid-infrared (2042 nm) regions of the electromagnetic spectrum.
Discussion
This may be due to differences in the characteristics of the sensors used or to the different scales under which the spectral data were obtained. Given the spatial resolution of Hyperion (30 X 30 meters), future work should explore the estimation of sugarcane N only in the large-scale growing sector, whether remote sensing data will be acquired using Hyperion or any other sensor of the same or larger pixel. the size.
Conclusions
As a first step towards remote sensing evaluation for thrips monitoring, a preliminary experiment and analysis was conducted at the leaf level using a handheld field spectroradiometer covering the 350 nm to 2500 nm range of the electromagnetic spectrum to detect damage of the sugarcane thrips. The results of the analyzes showed that there were significant differences in the spectral reflectance and derived variables used in the study at different levels of thrips damage.
Introduction
Applications of remote sensing for crop pests and diseases monitoring
In addition, Mirik et al. 2007) reported that wheat leaves infested by Russian wheat aphids had significantly lower reflectance in the near-infrared region and higher reflectance in the visible range of the spectrum compared to non-infested leaves. 2005) studied stress in wheat leaves caused by green aphid infestation using multispectral radiometry. In another study, Mirik et al. 2006a) measured wheat leaf reflectance using a hyperspectral field spectroradiometer and a digital camera, and they illustrated that damage sensitive spectral indices (DSSI1 and DSSI2), simple ratio (SR) index and normalized difference vegetation index (NDVI) strongly related to aphid damage.
Materials and Methods
- Field sampling and categorisation
- Leaf spectral measurements
- Data analysis
- One-way analysis of variance (ANOVA)
- Canonical discriminant analysis
96 database), whereas N12 is the most widespread variety in the rainfed regions of the South African sugar industry (SASRI, 2006). The coefficient of variation is the ratio of the standard deviation to the mean reflectance (Patel et al., 2001; SPSS, 2006).
Results
- One-way ANOVA
- Sensitivity analysis
- Canonical discriminant analysis
Results of one-way ANOVA (Figure 6.2) show significant differences between the classes of thrips damage in some parts of the visible (400-700 nm) region of the spectrum and especially in the so-called red edge region (690-720) nm) for both varieties tested. The third function of N19 distinguished between the medium damaged leaves and leaves on the other damage scales.
Discussion
The high levels of discrimination provided by the REI selection for both varieties reinforce the point of the importance of the red edge in discriminating between thrips damage classes. Since there is a strong nitrogen-chlorophyll relationship (Yoder and Pettigrew-Crosby, 1995; Nguyen and Lee, 2006), the significantly high reflectance of severely damaged leaves at the red edge position may be due to nitrogen deficiency. due to thrips damage.
Conclusions
A handheld FieldSpec® 3 spectroradiometer was used at leaf level in the 350–2500 nm range of the electromagnetic spectrum to develop a monitoring technique for a serious sugarcane pest (Fulmekiola serrata Kobus (Thysanoptera: Thripidae). Leaf reflectance measurements were converted to first-derivative order of reflectivity and analyzed to evaluate the assessment of pest populations.
Introduction
A shortcoming of PLS regression in spectroscopic data analysis is the identification of the spectral region(s) that have a relatively greater influence on PLS regression models (Huang et al. recommended that the PLS regression coefficients can be normalized by the mean spectral reflectance at all input wavelengths Spectral regions showing high values of normalized coefficients indicate the influence of such regions on the calibrated PLS regression models.
Applications of remote sensing for monitoring the incidence of crops pests
One disadvantage of the random forest algorithm in selecting variables from the spectroscopic data is that the selected relevant wavebands can still be autocorrelated (Strobl et al., 2008), especially with the very fine spectral resolutions produced by handheld sensors. For cotton crops, Sudbrink et al. 2003) analyzed reflectance data at leaf level as well as from an aerial image (at canopy level) to identify any association between larval densities of beet armyworm (BAW) and narrow-band NDVI values.
Materials and methods
- Leaf sample collection
- Leaf spectral measurements
- Determination of thrips numbers
- Data analysis
- Spectral transformation
- Statistical analyses
- Validation
A so-called random true regression algorithm was first used to reduce the number of variables in the spectroscopic data set, while preserving the maximum relevant information for thrips occurrence. The optimum number of spectral variables and the appropriate number of components (factors) that can be included in each PLS regression model were determined based on root mean square errors of prediction (RMSEP) values with a leave-one-out cross-validation method.
Results
In the near-infrared (NIR; 700–1300 nm) and mid-infrared (MIR nm) regions of the spectrum, positive and negative relationships were shown between the first-order derivative of reflectance and thrips numbers. The results of the validation models developed from the data set collected in summer (December 2007) using the data set collected in autumn (March 2008) as independent test data are shown in Figure 7.11.
Discussion
Nymphs + adults in age group 4–5 months Nymphs + adults in age group 6–7 months Nymphs + adults in combined age group data. PLS regression models developed from the December 2007 training data set were used for prediction: (a) age group 4–5 months, (b) age group 6–7 months, and (c) data on combined age groups.
Conclusions
Random forest regression for sugarcane yield prediction based on Landsat TM and ETM+ spectral vegetation indices. This study investigated the use of vegetation indices derived from multi-date Landsat TM and ETM+ data in predicting sugarcane yield (t ha–1) in South Africa.
Introduction
Except for eg, Singh et al. 2004), most of these studies investigated crop yield prediction at regional and district levels using vegetation indices (eg, Normalized Difference Vegetation Index: NDVI) derived from coarse-resolution sensors such as the Radiometer advanced very high resolution (AVHRR onboard NOAA) . With the exception of Jiang et al. 2004b), who used an artificial neural network model with a back-propagation algorithm, studies have applied simple and multiple linear regression models to estimate crop yield.
Rationale
There are relatively new approaches such as boosting and the random forest algorithms that can be used for both purposes, where the variable selection process is included during the development of the regression models. The random forest algorithm can be used for a dual purpose: developing nonlinear regression models and selecting variables.
Materials and Methods
- Image acquisition and preprocessing
- Spectral vegetation indices
- Field data collection
- Random forest ensemble
In each tree, the random forest algorithm uses randomness in the regression process by selecting a random subset of variables (mtry) to determine the split at each node (Breiman, 2001). VI RNIR–RRED Vegetation index: sensitive to green leaf material or photosynthetically active biomass in plant canopies.
Results
- Sugarcane yield (t ha –1 )
- Optimisation of random forest regression models
- Random forest prediction models
- Selection of variables
One-to-one relationships between measured and predicted sugarcane yields to validate the random forest prediction models developed using all spectral vegetation indices: (a) variety N19 and (b) variety NCo376. One-to-one relationships between measured and predicted sugarcane yield to validate the random forest prediction models developed using only the selected spectral vegetation indices: (a) variety N19 and (b) variety NCo376.
Discussion
The selected spectral vegetation indices for both sugarcane varieties are either sensitive to the amount of green materials (PCA1, EVI, NDVI*SR) or to other biochemical compounds (BC, BC1 and NDVI green) in the plants (Gitelson et al., 1996) ; Gong et al., 2003). Furthermore, EVI increases the sensitivity when estimating high biomass canopies such as sugarcane plantations (Gitelson et al., 1996).
Conclusions
In this study, the yield estimation of only two varieties was assessed, while many varieties are grown in the study area. On the other hand, since the average field size in the study area is large (~6.5 ha), fine resolution sensors should also be tested for predicting sugarcane yield.
Introduction
Summary of findings
The potential use of EO-1 Hyperion data in predicting sugarcane leaf N concentrations was tested (Chapter 5). A random forest ensemble was used to reduce redundancy in complex Hyperion hyperspectral data and predict sugarcane leaf N concentrations.
Conclusions
Sugarcane yield (t ha-1) was measured in test fields under three different irrigation conditions (fully irrigated, supplementally irrigated and rainfed). It was found that models for detecting a stressor or predicting yield in sugarcane vary depending on age group, cultivar, sampling season, conditions at which spectral data are collected (controlled laboratory or natural field conditions), level at which remotely sensed data are captured (leaves or canopy levels), and watering conditions.
Recommendations
Overall, this study developed predictive models for detecting sugarcane N status, thrips damage and incidence, and predicting sugarcane yield. The capability of multispectral sensors other than Landsat TM and ETM+ (eg SPOT, RapidEye, IKONOS, QuickBird, LREye, Sumbandilasat) should be tested for predicting sugarcane yield.
The practical and operational use of remote sensing techniques in sugarcane
These confounding effects can be detected using spectral features in the same region of the electromagnetic spectrum. Spectral features at the red end of the electromagnetic spectrum, for example, can be used to detect insect damage, disease infection, N deficiencies, and so on.
