The performance of foliage mass and crown radius models in forming the input of a forest reflectance model: A test on forest growth sample plots and Landsat 7 ETM+ images

Several published foliage mass and crown radius regression models were tested on the preparation of the input for the reflectance model of Kuusk and Nilson [Kuusk, A. and Nilson, T. (2000), A directional multispectral forest reflectance model. Remote Sensing of Environment, 72(2):244–252.] for 246 forest growth sample plots in Estonia. In each test, foliage mass and crown radius for trees in the sample plots were predicted with a particular pair of allometric regression models. The forest reflectance model was then run using the estimated foliage mass and crown radius values. Reflectance factors were simulated and compared with the reflectance values obtained from three atmospherically corrected Landsat 7 Enhanced Thematic Mapper (ETM+) scenes. The statistics of linear regression between the simulated and measured reflectance factors were used to assess the performance of foliage and crown radius models. The hypothesis was that the best allometric regression models should provide the best fit in reflectance. The strongest correlation between the simulated and measured reflectance factors was found in the short-wave infrared band (ETM + 5) for all the images. The highest R² = 0.71 was observed in Picea abies dominated stands. No excellent combination of foliage mass and crown radius functions was found, but the ranking based on determination coefficients showed that some linear crown radius models are not applicable to our data. Processing of raster images, reflectance measurement for small sample plots, usage of tree-species-specific fixed parameters (specific leaf area, etc.), and the ignored influence of phenology introduced additional variation into the relationships between simulated and measured reflectance factors. Further studies are needed, but these preliminary results demonstrate that the proposed method could serve as an effective way of testing the performance of foliage mass and canopy cover regressions.     

Use of high-resolution satellite imagery in an integrated model to predict the distribution of shade coffee tree hybrid zones

In New Caledonia (21°S, 165°E), shade-grown coffee plantations were abandoned for economic reasons in the middle of the 20th century. Coffee species (Coffea arabica, C. canephora and C. liberica) were introduced from Africa in the late 19th century, they survived in the wild and spontaneously cross-hybridized. Coffee species were originally planted in native forest in association with leguminous trees (mostly introduced species) to improve their growth. Thus the canopy cover over rustic shade coffee plantations is heterogeneous with a majority of large crowns, attributed to leguminous trees. The aim of this study was to identify suitable areas for coffee inter-specific hybridization in New Caledonia using field based environmental parameters and remotely sensed predictors. Due to the complex structure of tropical vegetation, remote sensing imagery needs to be spatially accurate and to have the appropriate bands for monitoring vegetation cover. Quickbird panchromatic (black and white) imagery at 0.6 to 0.7 m spatial resolutions and multispectral imagery at 2.4 m spatial resolution were pansharpened and used for this study. The two most suitable remotely sensed indicators, canopy heterogeneity and tree crown size, were acquired by the sequential use of tree crown detection (neural network), image processing (such as textural analysis) and classification. All models were supervised and trained on learning data determined by human expertise. The final model has two remotely sensed indicators and three physical parameters based on the Digital Elevation Model: elevation, slope and water flow accumulation. Using these five predictive variables as inputs, two modelling methods, a decision tree and a neural network, were implemented. The decision tree, which showed 96.9% accuracy on the test set, revealed the involvement of ecological parameters in the hybridization of Coffea species. We showed that hybrid zones could be characterized by combinations of modalities, underlining the complexity of the environment concerned. For instance, forest heterogeneity and large crown size, steep slopes (> 53.5%) and elevation between 194 and 429 m asl, are favourable factors for Coffea inter-specific hybridization. The application of the neural network on the whole area gave a predictive map that distinguished the most suitable areas by means of a nonlinear continuous indicator. The map provides a confidence level for each area. The most favourable areas were geographically localized, providing a clue for the detection and conservation of favourable areas for Coffea species neo-diversity.     

Assessing the utility of airborne hyperspectral and LiDAR data for species distribution mapping in the coastal Pacific Northwest, Canada

To effectively manage forested ecosystems an accurate characterization of species distribution is required. In this study we assess the utility of hyperspectral Airborne Imaging Spectrometer for Applications (AISA) imagery and small footprint discrete return Light Detection and Ranging (LiDAR) data for mapping 11 tree species in and around the Gulf Islands National Park Reserve, in coastal South-western Canada. Using hyperspectral imagery yielded producer's and user's accuracies for most species ranging from > 52-95.4 and > 63-87.8%, respectively. For species dominated by definable growth stages, pixel-level fusion of hyperspectral imagery with LiDAR-derived height and volumetric canopy profile data increased both producer's (+ 5.1-11.6%) and user's (+ 8.4-18.8%) accuracies. McNemar's tests confirmed that improvements in overall accuracies associated with the inclusion of LiDAR-derived structural information were statistically significant (p < 0.05). This methodology establishes a specific framework for mapping key species with greater detail and accuracy then is possible using conventional approaches (i.e., aerial photograph interpretation), or either technology on its own. Furthermore, in the study area, acquisition and processing costs were lower than a conventional aerial photograph interpretation campaign, making hyperspectral/LiDAR fusion a viable replacement technology.     

Remote sensing of photosynthetic light-use efficiency across two forested biomes: Spatial scaling

Eddy covariance (EC) measurements have greatly advanced our knowledge of carbon exchange in terrestrial ecosystems. However, appropriate techniques are required to upscale these spatially discrete findings globally. Satellite remote sensing provides unique opportunities in this respect, but remote sensing of the photosynthetic light-use efficiency (ε), one of the key components of Gross Primary Production, is challenging. Some progress has been made in recent years using the photochemical reflectance index, a narrow waveband index centered at 531 and 570 nm. The high sensitivity of this index to various extraneous effects such as canopy structure, and the view observer geometry has so far prevented its use at landscape and global scales. One critical aspect of upscaling PRI is the development of generic algorithms to account for structural differences in vegetation. Building on previous work, this study compares the differences in the PRI: ? relationship between a coastal Douglas-fir forest located on Vancouver Island, British Columbia, and a mature Aspen stand located in central Saskatchewan, Canada. Using continuous, tower-based observations acquired from an automated multi-angular spectro-radiometer (AMSPEC II) installed at each site, we demonstrate that PRI can be used to measure ? throughout the vegetation season at the DF-49 stand (r² = 0.91, p < 0.00) as well as the deciduous site (r² = 0.88, p < 0.00). It is further shown that this PRI signal can be also observed from space at both sites using daily observations from the Moderate Resolution Imaging Spectro-radiometer (MODIS) and a multi-angular implementation of atmospheric correction (MAIAC) (r² = 0.54 DF-49; r² = 0.63 SOA; p < 0.00). By implementing a simple hillshade model derived from airborne light detection and ranging (LiDAR) to approximate canopy shadow fractions (α_s), it is further demonstrated that the differences observed in the relationship between PRI and ε at DF-49 and SOA can be attributed largely to differences in α_s. The findings of this study suggest that algorithms used to separate physiological from extraneous effects in PRI reflectance may be more broadly applicable and portable across these two climatically and structurally different biome types, when the differences in canopy structure are known.     

Effects of different sensors, flying altitudes, and pulse repetition frequencies on forest canopy metrics and biophysical stand properties derived from small-footprint airborne laser data

Canopy height distributions were created from small-footprint airborne laser scanner (ALS) data collected over 40 field sample plots with size 1000 m² located in mature conifer forest. ALS data were collected with two different instruments, i.e., the ALTM 1233 and ALTM 3100 laser scanners (Optech Inc.). The ALTM 1233 data were acquired at a flying altitude of 1200 m and a pulse repetition frequency (PRF) of 33 kHz. Three different acquisitions were carried out with ALTM 3100, i.e., (1) a flying altitude of 1100 m and a PRF of 50 kHz, (2) a flying altitude of 1100 m and a PRF of 100 kHz, and (3) a flying altitude of 2000 m and a PRF of 50 kHz. Height percentiles, mean and maximum height values, coefficients of variation of the heights, and canopy density at different height intervals above the ground were derived from the four different ALS datasets and for single + first and last echoes of the ALS data separately. The ALS-derived height- and density variables were assessed in pair-wise comparisons to evaluate the effects of (a) instrument, (b) flying altitude, and (c) PRF. A systematic shift in height values of up to 0.3 m between sensors when the first echoes were compared was demonstrated. Also the density-related variables differed significantly between the two instruments. Comparisons of flying altitudes and PRFs revealed upwards shifted canopy height distributions for the highest flying altitude (2000 m) and the lowest PRF (50 kHz). The distribution of echoes on different echo categories, i.e., single and multiple (first and last) echoes, differed significantly between acquisitions. The proportion of multiple echoes decreased with increasing flying altitude and PRF. Different echo categories have different properties since it is likely that single echoes tend to occur in the densest parts of the tree crowns, i.e., near the apex where the concentration of biological matter is highest and distance to the ground is largest. To assess the influence of instrument, flying altitude, and PRF on biophysical properties derived from ALS data, regression analysis was carried out to relate ALS-derived metrics to mean tree height (h_L) and timber volume (V). Cross validation revealed only minor differences in precision for the different ALS acquisitions, but systematic differences between acquisitions of up to 2.5% for h_L and 10.7% for V were found when comparing data from different acquisitions.     

The influence of epiphylls on remote sensing of humid forests

Epiphylls - lichens, fungi, liverworts, etc. infesting leaf surfaces - are found throughout humid forests of the world. It is well understood that epiphylls inhibit light interception by host plants, but their effect on remote sensing of colonized forests has not been examined. Incorporating leaf-level spectra from Terra Firme (primary forest) and Amazonian Caatinga (woodlands/forest growing on nutrient-deficient sandy soils), we used the GeoSAIL model to propagate leaf-level measurements to the canopy level and determine their effect on commonly used vegetation indices. In Caatinga, moderate infestations (50% leaf area epiphyll cover), lowered simulated Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) values by 6.1% and 20.4%, respectively, largely due to near infrared dampening. Heavy infestation (100% cover) simulations exhibited decreases 1.5-2 times greater than those of moderate infestations. For Terra Firme, which are generally less affected by epiphylls, moderate (20% leaf area) and heavy infestations (40%) lowered EVI by 4.4% (S.D. 0.8%) and 8.1% (S.D. 1.5%), respectively. Near infrared and green reflectance were most affected at the canopy level, showing mean decreases of 10.6% (S.D. 2.25%) and 9.5% (S.D. 3.49%), respectively, in heavy Terra Firme infestations. Time series of Moderate Resolution Imaging Spectrometer (MODIS) data corroborated the modeling results, suggesting a degree of coupling between epiphyll cover and the EVI and NDVI. These results suggest that, without explicit consideration of the presence of epiphylls, remote sensing-based methodologies may underestimate leaf area index, biomass and productivity in humid forests.     

Utility of an image-based canopy reflectance modeling tool for remote estimation of LAI and leaf chlorophyll content at the field scale

This paper presents a physically-based approach for estimating critical variables describing land surface vegetation canopies, relying on remotely sensed data that can be acquired from operational satellite sensors. The REGularized canopy reFLECtance (REGFLEC) modeling tool couples leaf optics (PROSPECT), canopy reflectance (ACRM), and atmospheric radiative transfer (6SV1) model components, facilitating the direct use of at-sensor radiances in green, red and near-infrared wavelengths for the inverse retrieval of leaf chlorophyll content (C_ab) and total one-sided leaf area per unit ground area (LAI). The inversion of the canopy reflectance model is constrained by assuming limited variability of leaf structure, vegetation clumping, and leaf inclination angle within a given crop field and by exploiting the added radiometric information content of pixels belonging to the same field. A look-up-table with a suite of pre-computed spectral reflectance relationships, each a function of canopy characteristics, soil background effects and external conditions, is accessed for fast pixel-wise biophysical parameter retrievals. Using 1 m resolution aircraft and 10 m resolution SPOT-5 imagery, REGFLEC effectuated robust biophysical parameter retrievals for a corn field characterized by a wide range in leaf chlorophyll levels and intermixed green and senescent leaf material. Validation against in-situ observations yielded relative root-mean-square deviations (RMSD) on the order of 10% for the 1 m resolution LAI (RMSD = 0.25) and C_ab (RMSD = 4.4 μg cm^− 2) estimates, due in part to an efficient correction for background influences. LAI and C_ab retrieval accuracies at the SPOT 10 m resolution were characterized by relative RMSDs of 13% (0.3) and 17% (7.1 μg cm^− 2), respectively, and the overall intra-field pattern in LAI and C_ab was well established at this resolution. The developed method has utility in agricultural fields characterized by widely varying distributions of model variables and holds promise as a valuable operational tool for precision crop management. Work is currently in progress to extend REGFLEC to regional scales.     

Development of a pit filling algorithm for LiDAR canopy height models

LiDAR canopy height models (CHMs) can exhibit unnatural looking holes or pits, i.e., pixels with a much lower digital number than their immediate neighbors. These artifacts may be caused by a combination of factors, from data acquisition to post-processing, that not only result in a noisy appearance to the CHM but may also limit semi-automated tree-crown delineation and lead to errors in biomass estimates. We present a highly effective semi-automated pit filling algorithm that interactively detects data pits based on a simple user-defined threshold, and then fills them with a value derived from their neighborhood. We briefly describe this algorithm and its graphical user interface, and show its result in a LiDAR CHM populated with data pits. This method can be rapidly applied to any CHM with minimal user interaction. Visualization confirms that our method effectively and quickly removes data pits.     

Optimization of Geoscience Laser Altimeter System waveform metrics to support vegetation measurements

The Geoscience Laser Altimeter System (GLAS) has collected over 250 million measurements of vegetation height over forests globally. Accurate vegetation heights can be determined using waveform metrics that include vertical extent and extent of the waveform's trailing and leading edges. All three indices are highly dependent upon the signal strength, background noise and signal-to-noise ratio of the waveform, as the background noise contribution to the waveforms has to be removed before their calculation. Over the last six years, GLAS has collected data during thirteen observation periods using illumination from three different lasers. The power levels of these lasers have changed over time, resulting in variable signal power and noise characteristics. Atmospheric conditions vary continuously, also influencing signal power and noise.To minimize these effects, we optimized a noise coefficient which could be constant or vary according to observation period or noise metric. This parameter is used with the mean and standard deviation of the background noise to determine a noise level threshold that is removed from each waveform. An optimization analysis was used with a global dataset of waveforms that are near-coincident with waveforms from other observation periods; the goal of the optimization was to minimize the difference in vertical extent between spatially overlapping GLAS observations. Optimizations based on absolute difference in height led to situations in which the total extent was minimized as well; further optimizations reduced a normalized difference in height extent. The simplest optimizations were based on a constant value to be applied to all observations; noise coefficients of 2.7, 3.2, 3.4 and 4.0 were determined for datasets consisting of global forests, global vegetation, forest in the legal Amazon basin and boreal forests respectively. Optimizations based on the power level or the signal-to-noise ratio of waveforms best minimized differences in waveform extent, decreasing the percent root mean squared height difference by 25-54% over the constant value approach. Further development of methods to ensure temporal consistency of waveform indices will be necessary to support long-term satellite lidar missions and will result in more accurate and precise estimates of canopy height.     

Optimal modalities for radiative transfer-neural network estimation of canopy biophysical characteristics: Evaluation over an agricultural area with CHRIS/PROBA observations

Neural networks trained over radiative transfer simulations constitute the basis of several operational algorithms to estimate canopy biophysical variables from satellite reflectance measurements. However, only little attention was paid to the training process which has a major impact on retrieval performances. This study focused on the several modalities of the training process within neural network estimation of LAI, FCOVER and FAPAR biophysical variables. Performances were evaluated over both actual experimental observations and model simulations. The SAIL and PROSPECT radiative transfer models were used here to simulate the training and the synthetic test datasets. Measurements of LAI, FCOVER and FAPAR were achieved over the Barrax (Spain) agricultural site for a range of crop types concurrently to CHRIS/PROBA satellite image acquisition. Results showed that the spectral band selection was specific to LAI, FCOVER and FAPAR variables. The optimal band set provided significantly improved performances for LAI, while only small differences were observed for the other variables. Gaussian distributions of the radiative transfer model input variables performed better than uniform distributions for which no prior information was exploited. Including moderate uncertainties in the reflectance simulations used in the training process improved the flexibility of the neural network in cases where simulations departed slightly from observations. Simple neural network architecture with a single hidden layer of five tangent sigmoid transfer functions was performing as good as more complex architectures if the training dataset was larger than ten times the number of coefficients to tune. Small sensitivity of performances was observed depending on the way the solution was selected when several networks were trained in parallel. Finally, comparison with a NDVI based approach showed the generally better retrieval accuracy of neural networks.