Riyad Ismail

A comparison of regression tree ensembles: Predicting Sirex noctilio induced water stress in Pinus patula forests of KwaZulu-Natal, South Africa

In this study we compared the performance of regression tree ensembles using hyperspectral data. More specifically, we compared the performance of bagging, boosting and random forest to predict Sirex noctilio induced water stress in Pinus patula trees using nine spectral parameters derived from hyperspectral data. Results from the study show that the random forest ensemble achieved the best overall performance (R 2 = 0.73) and that the predictive accuracy of the ensemble was statistically different (p < 0.001) from bagging and boosting. Additionally, by using random forest as a wrapper we simplified the modeling process and identified the minimum number (n = 2) of spectral parameters that offered the best overall predictive accuracy (R 2 = 0.76). The water index and Ratio975 had the best ability to assay the water status of S. noctilio infested trees thus making it possible to remotely predict and quantify the severity of damage caused by the wasp.

Discriminating the papyrus vegetation Cyperus papyrus L. and its co-existent species using random forest and hyperspectral data resampled to HYMAP

Techniques for discriminating swamp wetland species are critical for the rapid assessment and proactive management of wetlands. In this study, we tested whether the random forest (RF) algorithm could discriminate between papyrus swamp and its co-existent species (Phragmites australis, Echinochloa pyramidalis and Thelypteris interrupta) using in situ canopy reflectance spectra. Canopy spectral measurements were taken from the species using analytical spectral devices but later resampled to Hyperspectral Mapper (HYMAP) resolution. The RF algorithm and a simple forward variable selection (FVS) technique were used to identify key wavelengths for discriminating papyrus swamp and its co-existence species. The method yielded 10 wavelengths located in the visible and short-wave infrared portions of the electromagnetic spectrum with a lowest out-of-bag (OOB) estimate error rate of 9.5% and .632+ bootstrap error of 8.95%. The use of RF as a classification algorithm resulted in overall accuracy of 90.50% and a kappa value of 0.87, with individual class accuracies ranging from 93.73% to 100%. Additionally, the results from this study indicate that the RF algorithm produces better classification results than conventional classification trees (CTs) when using all HYMAP wavelengths (n = 126) and when using wavelengths selected by the FVS technique.

Intra-and-Inter Species Biomass Prediction in a Plantation Forest: Testing the Utility of High Spatial Resolution Spaceborne Multispectral RapidEye Sensor and Advanced Machine Learning Algorithms

The quantification of aboveground biomass using remote sensing is critical for better understanding the role of forests in carbon sequestration and for informed sustainable management. Although remote sensing techniques have been proven useful in assessing forest biomass in general, more is required to investigate their capabilities in predicting intra-and-inter species biomass which are mainly characterised by non-linear relationships. In this study, we tested two machine learning algorithms, Stochastic Gradient Boosting (SGB) and Random Forest (RF) regression trees to predict intra-and-inter species biomass using high resolution RapidEye reflectance bands as well as the derived vegetation indices in a commercial plantation. The results showed that the SGB algorithm yielded the best performance for intra-and-inter species biomass prediction; using all the predictor variables as well as based on the most important selected variables. For example using the most important variables the algorithm produced an R2 of 0.80 and RMSE of 16.93 t·ha−1 for E. grandis; R2 of 0.79, RMSE of 17.27 t·ha−1 for P. taeda and R2 of 0.61, RMSE of 43.39 t·ha−1 for the combined species data sets. Comparatively, RF yielded plausible results only for E. dunii (R2 of 0.79; RMSE of 7.18 t·ha−1). We demonstrated that although the two statistical methods were able to predict biomass accurately, RF produced weaker results as compared to SGB when applied to combined species dataset. The result underscores the relevance of stochastic models in predicting biomass drawn from different species and genera using the new generation high resolution RapidEye sensor with strategically positioned bands.

Random forest regression and spectral band selection for estimating sugarcane leaf nitrogen concentration using EO-1 Hyperion hyperspectral data

Nitrogen (N) is one of the most important limiting nutrients for sugarcane production. Conventionally, sugarcane N concentration is examined using direct methods such as collecting leaf samples from the field followed by analytical assays in the laboratory. These methods do not offer real-time, quick, and non-destructive strategies for estimating sugarcane N concentration. Methods that take advantage of remote sensing, particularly hyperspectral data, can present reliable techniques for predicting sugarcane leaf N concentration. Hyperspectral data are extremely large and of high dimensionality. Many hyperspectral features are redundant due to the strong correlation between wavebands that are adjacent. Hence, the analysis of hyperspectral data is complex and needs to be simplified by selecting the most relevant spectral features. The aim of this study was to explore the potential of a random forest (RF) regression algorithm for selecting spectral features in hyperspectral data necessary for predicting sugarcane leaf N concentration. To achieve this, two Hyperion images were captured from fields of 6–7 month-old sugarcane, variety N19. The machine-learning RF algorithm was used as a feature-selection and regression method to analyse the spectral data. Stepwise multiple linear (SML) regression was also examined to predict the concentration of sugarcane leaf N after the reduction of the redundancy in hyperspectral data. The results showed that sugarcane leaf N concentration can be predicted using both non-linear RF regression (coefficient of determination, R 2 = 0.67; root mean square error of validation (RMSEV) = 0.15%; 8.44% of the mean) and SML regression models (R 2 = 0.71; RMSEV = 0.19%; 10.39% of the mean) derived from the first-order derivative of reflectance. It was concluded that the RF regression algorithm has potential for predicting sugarcane leaf N concentration using hyperspectral data.

Estimating standing biomass in papyrus Cyperus papyrus L. swamp: exploratory of in situ hyperspectral indices and random forest regression

Accurate estimates of papyrus (Cyperus papyrus) biomass are critical for an efficient papyrus swamp monitoring and management system. The objective of this study was to test the utility of random forest (RF) regression and two narrow-band vegetation indices in estimating above-ground biomass (AGB) for complex and densely vegetated swamp canopies. The normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI) were calculated from field spectrometry data and fresh AGB was measured in 82 quadrats at three different areas in the iSimangaliso Wetland Park, South Africa. NDVI was calculated from all possible band combinations of the electromagnetic spectrum (350 and 2500 nm), while EVI was calculated from possible band combinations in the blue, red, and near infrared of the spectrum. Backward feature elimination and RF regression were used as variable selection and modelling techniques to predict papyrus AGB. Results showed that the effective portions of electromagnetic spectrum for estimation AGB of papyrus swamp were located within the blue, red, red-edge, and near-infrared regions. The three best selected EVIs were computed from bands located at (i) 445, 682, and 829 nm, (ii) 497, 676, and 1091 nm, and (iii) 495, 678, and 1120 nm. These indices produced better predictive accuracies (R2 = 0.90; root mean square error of prediction (RMSEP) = 0.289 kg m−2; 7.99% of the mean) than the best selected NDVIs (R2 = 0.85; RMSEP = 0.343 kgm−2; 9.49% of the mean) that were calculated from bands located at (i) 739 and 829 nm, (ii) 739 and 814 nm, (iii) 744 and 789 nm, and (iv) 734 and 909 nm. The results of the present study demonstrate the utility of narrow-band vegetation indices (EVI and NDVI) and RF regression in estimating papyrus AGB at high density, a previously challenging task with broadband satellite sensors.

Discriminating the early stages of Sirex noctilio infestation using classification tree ensembles and shortwave infrared bands

The woodwasp Sirex noctilio is causing extensive damage to Pinus patula trees in the summer rainfall areas of South Africa. The ability to remotely detect S. noctilio infestation remains crucial for monitoring purposes and for the effective deployment of suppression activities. In this study, we evaluated whether random forest and boosting trees can accurately discriminate between healthy trees and the early stages of S. noctilio infestation using reflectance measurements in the shortwave infrared (SWIR). Three variable selection methods, namely, a filter, the random forest out-of-bag samples and a wrapper algorithm, were used to select the smallest subset of SWIR bands. The results show that random forest produces better classification results than the competing boosting trees for all three variable selection methods, even when noise is introduced into the SWIR bands and class labels. The ability of the bands centred at 1990, 2009, 2028, 2047 and 2065 nm to discriminate between healthy trees and the early stages of infestation could be explained due to the rapid physiological changes that occur as a result of the toxic mucus and a fungus that S. noctilio injects into the tree. Overall, the results are encouraging and show that there is a link between the selected SWIR bands and existing physiological knowledge, thereby improving the chances of detecting the early stages of S. noctilio infestation at a canopy or landscape level.

Forest health and vitality: the detection and monitoring of Pinus patula trees infected by Sirex noctilio using digital multispectral imagery

The Eurasian woodwasp, Sirex noctilio, causes considerable tree mortality in commercial pine plantations in southern KwaZulu-Natal, South Africa. Broad-scale visual assessments of infestation provided by forest managers are currently used to measure forest health and vitality. The effectiveness of visual assessments is questionable because they are qualitative, subjective and dependent on the skill of the surveyor. Remote sensing technology provides a synoptic view of the canopy and thus offers an alternative to the conventional methods of monitoring forest health and vitality. In this study, high resolution (0.5 × 0.5m) digital multispectral imagery (DMSI) was acquired over commercial Pinus patula trees of varying age classes, which had been ground assessed and ranked on an individual tree crown basis using a severity scale. The severity scale was based on a hierarchy of decline symptoms that are visibly apparent on the infested tree and are represented in this study as the green, red and grey stages. A series of ratio- and linear-based vegetation indices were then calculated and compared to the different crown condition classes as determined by severity scale. Of the vegetation indices derived from the high-resolution DMSI, significant differences between the pre-visual (healthy and green stages) and visual (red and grey stages) crown condition classes were obtained. Canonical variate analysis further revealed that greater discriminatory power between the different crown condition classes is obtained when using the normalised difference vegetation index (NDVI). Overall the study demonstrated the potential benefit of using high-resolution DMSI to discriminate between healthy trees and trees that were in the visual stage of infestation.

Investigating the Capability of Few Strategically Placed Worldview-2 Multispectral Bands to Discriminate Forest Species in KwaZulu-Natal, South Africa

WorldView-2 multispectral wavebands (8 wavebands; 427-908 nm spectral range; 2 m spatial resolution) were utilized to classify six commercial forest species (Eucalyptus grandis, Eucalyptus nitens, Eucalyptus smithii, Pinus patula, Pinus elliotii and Acacia mearnsii) in South Africa using the partial least squares discriminant analysis (PLS-DA) technique. Results indicate that the WorldView-2 imagery produced an overall accuracy of 85.42% and a kappa statistic value of 0.83, with individual forest species accuracies ranging between 63% and 100%. The variable importance in the projection (VIP) method was then used to identify the most important wavebands that were most effective in discriminating the forest species. Four VIP bands were ranked with thresholds greater than one and produced an overall accuracy of 84.38% and kappa value of 0.81, with individual forest species accuracies between 69% and 100%. More specifically, the VIP bands that were found to be important in the classification were the coastal blue (427 nm), blue (478 nm), green (546 nm) and red (659 nm) and confirmed the relative importance of the visible region of the electromagnetic spectrum in discriminating forest species. Overall, results indicate that multispectral information characterized by greater spatial resolution can successfully discriminate between and within forest species, thus providing an accurate framework for commercial forest species mapping.

Random Forests Unsupervised Classification: The Detection and Mapping of Solanum mauritianum Infestations in Plantation Forestry Using Hyperspectral Data

Detecting and mapping plant invaders using hyperspectral remote sensing is necessary in mitigating the extensive ecologic and economic damage these alien plants induce on our forest ecosystems. Using AISA Eagle image data, this study investigated the capability of two unsupervised classification methods for the detection and mapping of Solanum mauritianum located within commercial forestry ecosystems. The existing random forest (RF) outlier detection method when used in conjunction with Anselins Morans I produced a detection rate (DR) of 89% with a false positive rate (FPR) of 9.26%. In comparison, the newly developed methodology which is based on the decomposition of the RF proximity matrix using principal component analysis (PCA) resulted in a DR of 95% with a lower FPR (6.39%). Overall, this research has demonstrated the potential of utilizing an unsupervised and accurate RF framework for the detection and mapping of alien invasive plants.

Modeling the Potential Distribution of Pine Forests Susceptible to Sirex Noctilio Infestations in Mpumalanga, South Africa

Reducing the impact of the siricid wasp, Sirex noctilio is crucial for the future productivity and sustainability of commercial pine resources in South Africa. In this study we present a machine learning model that serves as a spatial guide and allows forest managers to focus their existing detection and monitoring efforts on key areas and proactively adopt the most appropriate course of intervention. We implemented the random forest model within a spatial framework to determine which pine forests in Mpumalanga are highly susceptible to S. noctilio infestations. Results indicate that a majority (63%) of pine forest plantations located in Mpumalanga have a high susceptibility (>70%) to S. noctilio infestation. A KHAT value of 0.84 and F measures above 0.87 indicate that the random forest model is a robust classifier that produces accurate results. Additionally, the use of the backward variable selection method enabled us to simplify the random forest modeling process and identify the minimum number of explanatory variables that offer the best discriminatory power and help in the empirical interpretation of the final random forest model. Overall, the results show that pine forests that experience stress caused by evapotranspiration and evaporation followed by rainfalls, especially during the summer months are more susceptible to S. noctilio infestations. tgis_1229