Fethi Ahmed

The application of remote sensing techniques to sugarcane (Saccharum spp. hybrid) production: a review of the literature

Remote sensing techniques provide timely, up‐to‐date and relatively accurate information for the management of sugarcane crop. This article reviews the literature on the application of remote sensing to sugarcane agriculture and highlights the challenges and opportunities pertinent to the success of this application. The aim of the review was to provide accurate and fundamental information relating the spectral properties of sugarcane to its agronomic, health and nutritional status characteristics that would be of importance to cane farmers and farm managers. The applications of the remote sensing techniques in sugarcane agriculture have been undertaken with particular emphasis on sugarcane classification and areal extent mapping, thermal age group identification, varietal discrimination, yield prediction and crop health and nutritional status monitoring. It can be concluded that by selecting appropriate spatial and spectral resolution as well as suitable processing techniques for extracting sugarcane spectral information, remotely sensed data should find use in sugarcane agriculture in all areas of application with satisfactory results.

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 plot-level tree height and volume of Eucalyptus grandis plantations using small-footprint, discrete return lidar data

This study explores the utility of small-footprint, discrete return lidar data in deriving important forest structural attributes with the primary objective of estimating plot-level mean tree height, dominant height, and volume of Eucalyptus grandis plantations. The secondary objectives of the study were related to investigating the effect of lidar point densities (1 point/m2, 3 points/m2, and 5 points/m2) on height and volume estimates. Tree tops were located by applying local maxima (LM) filtering to canopy height surfaces created at each density level, followed by buffering using circular polygons. Maximum and mean height values of the original lidar points falling within each tree polygon were used to generate lidar mean and dominant heights. Lidar mean value was superior to the maximum lidar value approach in estimating mean plot height (R2∼0.95; RMSE∼7%), while the maximum height approach resulted in superior estimates for dominant plot height (R2 ∼0.95; RMSE∼5%). These observations were similar across all lidar point density levels. Plot-level volume was calculated using approaches based on lidar-derived height variables and stems per hectare, as well as stand age. The level of association between estimated and observed volume was relatively high (R2=0.82—0.94) with non-significant differences among estimates at high lidar point densities and field observation. Nearly all estimates, however, exhibited negative biases and RMSE ranging in the order of 20—43%. Overall, the results of the study demonstrate the potential of lidar-based approaches for forest structural assessment in commercial plantations, even though further research is required on improving stems per hectare (SPHA) estimation.

Assessment of image fusion procedures using entropy, image quality, and multispectral classification

The use of disparate data sources within a pixel level image fusion procedure has been well documented for pan-sharpening studies. The present paper explores various image fusion procedures for the fusion of multi-spectral ASTER data and a RadarSAT-1 SAR scene. The research sought to determine which fusion procedure merged the largest amount of SAR texture into the ASTER scenes, while also preserving the spectral content. An additional application based maximum likelihood classification assessment was also undertaken. Three SAR scenes were tested namely, one backscatter scene and two textural measures calculated using grey level co-occurrence matrices (GLCM). Each of these were fused to the ASTER data using the following established approaches; Brovey transformation, Intensity Hue and Saturation, Principal Component Substitution, Discrete wavelet transformation, and a modified discrete wavelet transformation using the IHS approach. Resulting data sets were assessed using qualitative and quantitative (entropy, universal image quality index, maximum likelihood classification) approaches. Results from the study indicated that while all post fusion data sets contained more information (entropy analysis), only the frequency-based fusion approaches managed to preserve the spectral quality of the original imagery. Furthermore results also indicated that the textural (mean, contrast) SAR scenes did not add any significant amount of information to the post-fusion imagery. Classification accuracy was not improved when comparing ASTER optical data and pseudo optical bands generated from the fusion analysis. Accuracies range from 68.4% for the ASTER data to well below 50% for the component substitution methods. Frequency based approaches also returned lower accuracies when compared to the unfused optical data. The present study essentially replicated (pan-sharpening) studies using the high resolution SAR scene as a pseudo panchromatic band.

Predicting forest structural attributes using ancillary data and ASTER satellite data

Abstract This study assessed the suitability of both visible and shortwave infrared ASTER data and vegetation indices for estimating forest structural attributes of Eucalyptus species in the southern KwaZulu Natal, South Africa. The empirical relationships between forest structural attributes and ASTER data were derived using stepwise multiple regression analysis; Modified Soil Adjusted Vegetation Index (MSVI) and band 3 were selected for analysis as it showed best relationships with forest structural attributes. The ancillary data such as age and site index were also included in the analysis. Although the results of this study have indicated statistically significant relationships between the forest structural attributes and the ASTER data in the plantation forests stands with adjusted R 2 -values for volume, basal area (BA), stem per hectare (SPHA), and tree height of 0.51, 0.67, 0.65, and 0.52, respectively, but these results are not suitable for operational purpose in a forest company. However, the structural forest attribute predictions were markedly improved after incorporating age and site index as predictor variable. R 2 -values for the stands increased by 42%, 20.2%, 16.8%, and 42.2% for volume, basal area, SPHA, and tree height, respectively. These results imply that ASTER satellite data alone are not applicable to forest structural attribute estimation; however, ASTER data can provide useful information if it is used in conjunction with age and site index data for forest structural attribute estimation in plantation forests.

Forest structural assessment using remote sensing technologies: an overview of the current state of the art

The Forestry and Forest Products Research Centre (CSIR), University of KwaZulu-Natal and MONDI Business Paper have recently embarked on a remote sensing cooperative. The primary focus of this cooperative is to explore the potential benefits associated with using remote sensing for forestry-related activities. A subproject within the cooperative is exploring the utility of various remote sensing technologies for forest structural assessment. This paper reports on the primary findings of a state-of-the-art review conducted by members of the cooperative and seeks to inform and contribute to the development of focused research projects. Both active and passive sensors are reviewed at varying spatial scales focusing primarily on accuracies attained. Medium-resolution studies focus on contextual forest attributes while high-resolution studies focused on location-based forest variables. Results from research consulted indicate that while remote sensing has a strong theoretical background, there are several limiting factors that need to be explored within a South African context. These include the saturation of satellite signals in mature forests, underestimation of tree heights using LiDAR data and the cost of LiDAR surveys. The review ends with recommendations for future research activities.

Individual tree detection based on variable and fixed window size local maxima filtering applied to IKONOS imagery for even-aged Eucalyptus plantation forests

Detection of individual trees remains a challenge for forest inventory efforts especially in homogeneous, even-aged plantation scenarios. Airborne imagery has mainly been used for detection of individual trees using local maxima filtering, as point spread function and signal-to-noise ratio are smaller than with satellite-borne imagery. This led to the development of a novel approach to local maxima filtering for tree detection in plantation forests in KwaZulu-Natal, South Africa, using satellite remote sensing imagery. Our approach is based on Gaussian smoothing for noise elimination and image classification, that is, natural break classification to determine the threshold for removing pixels of extremely bright and dark areas in the imagery. These pixels are assumed to belong to the background and hinder the search for tree peaks. A semivariogram technique was applied to determine variable window sizes for local maxima filtering within a plantation stand. A fixed window size for local maxima filtering was also applied using pre-determined tree spacing. Evaluation of the various approaches was based on aggregated assessment methods. The overall accuracy using a variable window size was 85%, root mean square error (RMSE) = 189 trees, whereas a fixed window size resulted in an accuracy of 80%, RMSE = 258 trees. The approach worked remarkably well in mature forest stands as compared to young forest stands. These results are encouraging for temperate–warm climate plantation forest companies, who deal with even-aged, broadleaf plantations and forest inventory practices that require assessment 1 year before harvesting.

Potential of spectroscopic data sets for sugarcane thrips Fulmekiola serrata Kobus damage detection

Sugarcane thrips was detected in South African sugarcane in 2004. Since then it has become widespread in South Africa. The South African Sugarcane Research Institute (SASRI) conducts field surveys to monitor this pest, but this is time intensive and costly. As a first step towards evaluation of remote sensing for thrips monitoring, a preliminary experiment and analysis at leaf level were conducted using a handheld field spectroradiometer covering the 350 to 2500 nm range of the electromagnetic spectrum to detect sugarcane thrips damage. Reflectance spectra of sugarcane leaves with different levels of thrips damage, from two popular varieties (N19 and N12), were measured and statistically analysed using one-way analysis of variance, sensitivity analysis and canonical discriminant analysis. The results of the analyses showed that there were significant differences in spectral reflectance and derived variables used in the study at the different levels of damage. The red edge region of the visible portion gave the highest significant differences and levels of separability among the damage classes. It is hypothesized that this might be associated with chlorophyll and nitrogen deficiencies induced by thrips.

Investigating the impact of discrete-return lidar point density on estimations of mean and dominant plot-level tree height in Eucalyptus grandis plantations

The accuracy of lidar remote sensing in characterizing three-dimensional forest structural attributes has encouraged foresters to integrate lidar approaches in routine inventories. However, lidar point density is an important consideration when assessing forest biophysical parameters, given the direct relationship between higher spatial resolution and lidar acquisition and processing costs. The aim of this study was to investigate the effect of point density on mean and dominant tree height estimates at plot level. The study was conducted in an intensively managed Eucalyptus grandis plantation. High point density (eight points/m2) discrete-return, small-footprint lidar data were used to generate point density simulations averaging 0.25, one, two, three, four, five, and six points/m2. Field surveyed plot-level mean and dominant heights were regressed against metrics derived from lidar data at each simulated point density. Stepwise regression was used to identify which lidar metrics produced the best models. Mean height was estimated at accuracy of R2 ranging between 0.93 and 0.94 while dominant height was estimated with an R2 of 0.95. Root mean square error (RMSE) was also similar at all densities for mean height (∼1.0 m) and dominant height (∼1.2 m); the relative RMSE compared to field-measured mean was constant at approximately 5%. Analysis of bias showed that the estimation of both variables did not vary with density. The results indicated that all lidar point densities resulted in reliable models. It was concluded that plot-level height can be estimated with reliable accuracy using relatively low density lidar point spacing. Additional research is required to investigate the effect of low point density on estimation of other forest biophysical attributes.

Image fusion for enhanced forest structural assessment

This research explores the potential benefits of fusing active and passive medium-resolution satellite-borne sensor data for forest structural assessment. Image fusion was applied as a means of retaining disparate data features relevant to modelling and mapping of forest structural attributes in even-aged (4–11 years) Eucalyptus plantations, located in the southern KwaZulu-Natal midlands of South Africa. Remote-sensing data used in this research included the visible and near-infrared bands of the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), as well as a fine-beam (6.25 m resolution) Radarsat-1 image. Both datasets were collected during the spring of 2006 and fused using a modified discrete wavelet transformation. Spatially referenced forest-inventory data were also collected during this time, with 122 plots enumerated in 38 plantation compartments. Empirical relationships (ordinary and multiple regression) were used to test whether fused data sources produced superior statistical models. Secondary objectives of the article included exploring the roles of texture, derived from grey-level co-occurrence matrices, and scale in terms of forest modelling at the plot and extended plot levels (Voroni diagrams). Results indicate that single bands from both the optical and Synthetic Aperture Radar (SAR) datasets were not adept at modelling basal area and merchantable timber volume with adjusted R 2 (coefficient of determination) values < 0.3. An optimized multiple-regression approach (adjusted R 2) improved results based on mean, range and standard deviation statistics when compared to single bands, but were still not suitable for operational forest applications (basal area: R 2 = 0.55, volume: R 2 = 0.59). No significant difference was found between fused and non-fused datasets; however, optical and fused datasets produced superior models when compared to SAR results. Investigations into potential benefits of using textural indices and varied scales also returned inconclusive results. Findings indicate that the spatial resolutions of both sensors are inappropriate for plantation forest assessment. The frequency of the C-band Radarsat-1 data is, for instance, unable to penetrate the canopy and interact with the woody structures below canopy, leading to weak statistical models. The lack of variability in both the optical and SAR data lead to unconvincing results in the fused imagery, where, in some cases, the adjusted R 2 results were worse than the single-dataset approach. It was concluded that future research should focus on high-spatial-resolution optical and Light Detection and Ranging (LiDAR) data and the development of automated and semi-automated forest-inventory procedures.