An accurate and computationally efficient algorithm for ground peak identification in large footprint waveform LiDAR data

Large footprint waveform LiDAR sensors have been widely used for numerous airborne studies. Ground peak identification in a large footprint waveform is a significant bottleneck in exploring full usage of the waveform datasets. In the current study, an accurate and computationally efficient algorithm was developed for ground peak identification, called Filtering and Clustering Algorithm (FICA). The method was evaluated on Land, Vegetation, and Ice Sensor (LVIS) waveform datasets acquired over Central NY. FICA incorporates a set of multi-scale second derivative filters and a k-means clustering algorithm in order to avoid detecting false ground peaks. FICA was tested in five different land cover types (deciduous trees, coniferous trees, shrub, grass and developed area) and showed more accurate results when compared to existing algorithms. More specifically, compared with Gaussian decomposition, the RMSE ground peak identification by FICA was 2.82 m (5.29 m for GD) in deciduous plots, 3.25 m (4.57 m for GD) in coniferous plots, 2.63 m (2.83 m for GD) in shrub plots, 0.82 m (0.93 m for GD) in grass plots, and 0.70 m (0.51 m for GD) in plots of developed areas. FICA performance was also relatively consistent under various slope and canopy coverage (CC) conditions. In addition, FICA showed better computational efficiency compared to existing methods. FICA’s major computational and accuracy advantage is a result of the adopted multi-scale signal processing procedures that concentrate on local portions of the signal as opposed to the Gaussian decomposition that uses a curve-fitting strategy applied in the entire signal. The FICA algorithm is a good candidate for large-scale implementation on future space-borne waveform LiDAR sensors.     

Stochastic gradient boosting classification trees for forest fuel types mapping through airborne laser scanning and IRS LISS-III imagery

This paper presents an application of Airborne Laser Scanning (ALS) data in conjunction with an IRS LISS-III image for mapping forest fuel types. For two study areas of 165 km² and 487 km² in Sicily (Italy), 16,761 plots of size 30-m × 30-m were distributed using a tessellation-based stratified sampling scheme. ALS metrics and spectral signatures from IRS extracted for each plot were used as predictors to classify forest fuel types observed and identified by photointerpretation and fieldwork. Following use of traditional parametric methods that produced unsatisfactory results, three non-parametric classification approaches were tested: (i) classification and regression tree (CART), (ii) the CART bagging method called Random Forests, and (iii) the CART bagging/boosting stochastic gradient boosting (SGB) approach. This contribution summarizes previous experiences using ALS data for estimating forest variables useful for fire management in general and for fuel type mapping, in particular. It summarizes characteristics of classification and regression trees, presents the pre-processing operation, the classification algorithms, and the achieved results. The results demonstrated superiority of the SGB method with overall accuracy of 84%. The most relevant ALS metric was canopy cover, defined as the percent of non-ground returns. Other relevant metrics included the spectral information from IRS and several other ALS metrics such as percentiles of the height distribution, the mean height of all returns, and the number of returns.     

Evaluation of forest fire on Madeira Island using Sentinel-2A MSI imagery

A forest fire started on August 8th, 2016 in several places on Madeira Island causing damage and casualties. As of August 10th the local media had reported the death of three people, over 200 people injured, over 950 habitants evacuated, and 50 houses damaged. This study presents the preliminary results of the assessment of several spectral indices to evaluate the burn severity of Madeira fires during August 2016. These spectral indices were calculated using the new European satellite Sentinel-2A launched in June 2015. The study confirmed the advantages of several spectral indices such as Normalized Difference Vegetation Index (NDVI), Green Normalized Difference Vegetation Index (GNDVI), Normalized Burn Ratio (NBR) and Normalized Difference Vegetation Index (NDVIreXn) using red-edge spectral bands to assess the post-fire conditions. Results showed high correlation between NDVI, GNDVI, NBR and NDVIre1n spectral indices and the analysis performed by Copernicus Emergency Management Service (EMSR175), considered as the reference truth. Regarding the red-edge spectral indices, the NDVIre1n (using band B5, 705 nm) presented better results compared with B6 (740 nm) and B7 (783 nm) bands. These preliminary results allow us to assume that Sentinel-2 will be a valuable tool for post-fire monitoring. In the future, the two twin Sentinel-2 satellites will offer global coverage of the Madeira Archipelago every five days, therefore allowing the simultaneous study of the evolution of the burnt area and reforestation information with high spatial (up to 10 m) and temporal resolution (5 days).     

Upscaling plot-scale soil respiration in winter wheat and summer maize rotation croplands in Julu County,North China

Soil respiration (R_s) data from 45 plots were used to estimate the spatial patterns of R_s during the peak growing seasons of winter wheat and summer maize in Julu County, North China, by combining satellite remote sensing data, field-measured data, and a support vector regression (SVR) model. The observed R_s values were well reproduced by the model at the plot scale, with a root-mean-square error (RMSE) of 0.31 μmol CO₂ m⁻² s⁻¹ and a coefficient of determination (R²) of 0.73. No significant difference was detected between the prediction accuracy of the SVR model for winter wheat and summer maize. With forcing from satellite remote sensing data and gridded soil property data, we used the SVR model to predict the spatial distributions of R_s during the peak growing seasons of winter wheat and summer maize rotation croplands in Julu County. The SVR model captured the spatial variations of R_s at the county scale. The satellite-derived enhanced vegetation index was found to be the most important input used to predict R_s. Removal of this variable caused an RMSE increase from 0.31 μmol CO₂ m⁻² s⁻¹ to 0.42 μmol CO₂ m⁻² s⁻¹. Soil properties such as soil organic carbon (SOC) content and soil bulk density (SBD) were the second most important factors. Their removal led to an RMSE increase from 0.31 μmol CO₂ m⁻² s⁻¹ to 0.37 μmol CO₂ m⁻² s⁻¹. The SVR model performed better than multiple regression in predicting spatial variations of R_s in winter wheat and summer maize rotation croplands, as shown by the comparison of the R² and RMSE values of the two algorithms. The spatial patterns of R_s are better captured using the SVR model than performing multiple regression, particularly for the relatively high and relatively low R_s values at the center and northeast study areas. Therefore, SVR shows promise for predicting spatial variations of R_s values on the basis of remotely sensed data and gridded soil property data at the county scale.     

Above ground biomass estimation in an African tropical forest with lidar and hyperspectral data

The estimation of above ground biomass in forests is critical for carbon cycle modeling and climate change mitigation programs. Small footprint lidar provides accurate biomass estimates, but its application in tropical forests has been limited, particularly in Africa. Hyperspectral data record canopy spectral information that is potentially related to forest biomass. To assess lidar ability to retrieve biomass in an African forest and the usefulness of including hyperspectral information, we modeled biomass using small footprint lidar metrics as well as airborne hyperspectral bands and derived vegetation indexes. Partial Least Square Regression (PLSR) was adopted to cope with multiple inputs and multicollinearity issues; the Variable of Importance in the Projection was calculated to evaluate importance of individual predictors for biomass. Our findings showed that the integration of hyperspectral bands (R² = 0.70) improved the model based on lidar alone (R² = 0.64), this encouraging result call for additional research to clarify the possible role of hyperspectral data in tropical regions. Replacing the hyperspectral bands with vegetation indexes resulted in a smaller improvement (R² = 0.67). Hyperspectral bands had limited predictive power (R² = 0.36) when used alone. This analysis proves the efficiency of using PLSR with small-footprint lidar and high resolution hyperspectral data in tropical forests for biomass estimation. Results also suggest that high quality ground truth data is crucial for lidar-based AGB estimates in tropical African forests, especially if airborne lidar is used as an intermediate step of upscaling field-measured AGB to a larger area.     

Classifying individual tree species under leaf-off and leaf-on conditions using airborne lidar

In this paper, a methodology for individual tree-based species classification using high sampling density and small footprint lidar data is clarified, corrected and improved. For this purpose, a well-defined directed graph (digraph) is introduced and it plays a fundamental role in the approach. It is argued that there exists one and only one such unique digraph that describes all four pure events and resulting disjoint sets of laser points associated with a single tree in data from a two-return lidar system. However, the digraph is extendable so that it fits an n-return lidar system (n > 2) with higher logical resolution. Furthermore, a mathematical notation for different types of groupings of the laser points is defined, and a new terminology for various types of individual tree-based concepts defined by the digraph is proposed. A novel calibration technique for estimating individual tree heights is evaluated. The approach replaces the unreliable maximum single laser point height of each tree with a more reliable prediction based on shape characteristics of a marginal height distribution of the whole first-return point cloud of each tree. The result shows a reduction of the RMSE of the tree heights of about 20% (stddev = 1.1 m reduced to stddev = 0.92 m). The method improves the species classification accuracy markedly, but it could also be used for reducing the sampling density at the time of data acquisition. Using the calibrated tree heights, a scale-invariant rescaled space for the universal set of points for each tree is defined, in which all individual tree-based geometric measurements are conducted. With the corrected and improved classification methodology the total accuracy raises from 60% to 64% for classifying three leaf-off individual tree deciduous species (N = 200 each) in West Virginia, USA: oaks (Quercus spp.), red maple (Acer rubrum), and yellow poplar (Liriodendron tuliperifera).     

Augmenting forest inventory attributes with geometric optical modelling in support of regional susceptibility assessments to bark beetle infestations

Assessment of the susceptibility of forests to mountain pine beetle (Dendroctonus ponderosae Hopkins) infestation is based upon an understanding of the characteristics that predispose the stands to attack. These assessments are typically derived from conventional forest inventory data; however, this information often represents only managed forest areas. It does not cover areas such as forest parks or conservation regions and is often not regularly updated resulting in an inability to assess forest susceptibility. To address these shortcomings, we demonstrate how a geometric optical model (GOM) can be applied to Landsat-5 Thematic Mapper (TM) imagery (30 m spatial resolution) to estimate stand-level susceptibility to mountain pine beetle attack. Spectral mixture analysis was used to determine the proportion of sunlit canopy and background, and shadow of each Landsat pixel enabling per pixel estimates of attributes required for model inversion. Stand structural attributes were then derived from inversion of the geometric optical model and used as basis for susceptibility mapping. Mean stand density estimated by the geometric optical model was 2753 (standard deviation ± 308) stems per hectare and mean horizontal crown radius was 2.09 (standard deviation ± 0.11) metres. When compared to equivalent forest inventory attributes, model predictions of stems per hectare and crown radius were shown to be reasonably estimated using a Kruskal–Wallis ANOVA (p < 0.001). These predictions were then used to create a large area map that provided an assessment of the forest area susceptible to mountain pine beetle damage.     

Fire danger assessment in Iran based on geospatial information

Fire danger assessment is a vital issue to alleviate the impacts of wildland fires. In this study, a fire danger assessment system is proposed, which extensively uses geographical databases to characterize the spatial variations of fire danger conditions in Iran. This assessment requires three steps: (i) generation of the required input variables, (ii) methods to integrate those variables for creating synthetic indices and (iii) validation of those indices versus fire occurrence data. This fire danger model is based on previous works but adapted to Iranian conditions. It includes an estimation of the fire ignition potential (both considering human and climatic factors) and fire propagation potential. The former was generated from a logistic regression approach based on a wide range of input variables. The fire propagation probability was estimated from the Flammap fire behavior model. A first stage for validation of our fire danger system was based on comparing the estimated danger values to actual fire occurrence, based on satellite detected active fires and burned areas. The logistic regression model for fire ignition probability estimated 72.7% of true ignitions. Detected hotspots occurred more frequently in areas with higher fire ignition probability (average value: 0.65) than non hotspots (average value: 0.4). Propagation probability showed higher values for areas with higher proportion of burned area (r = 0.68, p < 0.001).     

3-D voxel-based solid modeling of a broad-leaved tree for accurate volume estimation using portable scanning lidar

We developed a method to produce a 3-D voxel-based solid model of a tree based on portable scanning lidar data for accurate estimation of the volume of the woody material. First, we obtained lidar measurements with a high laser pulse density from several measurement positions around the target, a Japanese zelkova tree. Next, we converted lidar-derived point-cloud data for the target into voxels. The voxel size was 0.5 cm × 0.5 cm × 0.5 cm. Then, we used differences in the spatial distribution of voxels to separate the stem and large branches (diameter > 1 cm) from small branches (diameter ⩽ 1 cm). We classified the voxels into sets corresponding to the stem and to each large branch and then interpolated voxels to fill out their surfaces and their interiors. We then merged the stem and large branches with the small branches. The resultant solid model of the entire tree was composed of consecutive voxels that filled the outer surface and the interior of the stem and large branches, and a cloud of voxels equivalent to small branches that were discretely scattered in mainly the upper part of the target. Using this model, we estimated the woody material volume by counting the number of voxels in each part and multiplying the number of voxels by the unit voxel volume (0.13 cm³). The percentage error of the volume of the stem and part of a large branch was 0.5%. The estimation error of a certain part of the small branches was 34.0%.     

Accuracy assessment of airborne photogrammetrically derived high-resolution digital elevation models in a high mountain environment

High-resolution digital elevation models (DEMs) generated by airborne remote sensing are frequently used to analyze landform structures (monotemporal) and geomorphological processes (multitemporal) in remote areas or areas of extreme terrain. In order to assess and quantify such structures and processes it is necessary to know the absolute accuracy of the available DEMs. This study assesses the absolute vertical accuracy of DEMs generated by the High Resolution Stereo Camera-Airborne (HRSC-A), the Leica Airborne Digital Sensors 40/80 (ADS40 and ADS80) and the analogue camera system RC30. The study area is located in the Turtmann valley, Valais, Switzerland, a glacially and periglacially formed hanging valley stretching from 2400 m to 3300 m a.s.l. The photogrammetrically derived DEMs are evaluated against geodetic field measurements and an airborne laser scan (ALS). Traditional and robust global and local accuracy measurements are used to describe the vertical quality of the DEMs, which show a non Gaussian distribution of errors. The results show that all four sensor systems produce DEMs with similar accuracy despite their different setups and generations. The ADS40 and ADS80 (both with a ground sampling distance of 0.50 m) generate the most accurate DEMs in complex high mountain areas with a RMSE of 0.8 m and NMAD of 0.6 m They also show the highest accuracy relating to flying height (0.14‰). The pushbroom scanning system HRSC-A produces a RMSE of 1.03 m and a NMAD of 0.83 m (0.21‰ accuracy of the flying height and 10 times the ground sampling distance). The analogue camera system RC30 produces DEMs with a vertical accuracy of 1.30 m RMSE and 0.83 m NMAD (0.17‰ accuracy of the flying height and two times the ground sampling distance). It is also shown that the performance of the DEMs strongly depends on the inclination of the terrain. The RMSE of areas up to an inclination <40° is better than 1 m. In more inclined areas the error and outlier occurrence increase for all DEMs. This study shows the level of detail to which airborne stereoscopically derived DEMs can reliably be used in high mountain environments. All four sensor systems perform similarly in flat terrain.     

Retrieving aboveground biomass of wetland Phragmites australis (common reed) using a combination of airborne discrete-return LiDAR and hyperspectral data

Wetland biomass is essential for monitoring the stability and productivity of wetland ecosystems. Conventional field methods to measure or estimate wetland biomass are accurate and reliable, but expensive, time consuming and labor intensive. This research explored the potential for estimating wetland reed biomass using a combination of airborne discrete-return Light Detection and Ranging (LiDAR) and hyperspectral data. To derive the optimal predictor variables of reed biomass, a range of LiDAR and hyperspectral metrics at different spatial scales were regressed against the field-observed biomasses. The results showed that the LiDAR-derived H_p99 (99th percentile of the LiDAR height) and hyperspectral-calculated modified soil-adjusted vegetation index (MSAVI) were the best metrics for estimating reed biomass using the single regression model. Although the LiDAR data yielded a higher estimation accuracy compared to the hyperspectral data, the combination of LiDAR and hyperspectral data produced a more accurate prediction model for reed biomass (R² = 0.648, RMSE = 167.546 g/m², RMSE_r = 20.71%) than LiDAR data alone. Thus, combining LiDAR data with hyperspectral data has a great potential for improving the accuracy of aboveground biomass estimation.     

PTrees: A point-based approach to forest tree extraction from lidar data

This paper introduces PTrees, a multi-scale dynamic point cloud segmentation dedicated to forest tree extraction from lidar point clouds. The method process the point data using the raw elevation values (Z) and compute height (H = Z − ground elevation) during post-processing using an innovative procedure allowing to preserve the geometry of crown points. Multiple segmentations are done at different scales. Segmentation criteria are then applied to dynamically select the best set of apices from the tree segments extracted at the various scales. The selected set of apices is then used to generate a final segmentation. PTrees has been tested in 3 different forest types, allowing to detect 82% of the trees with under 10% of false detection rate. Future development will integrate crown profile estimation during the segmentation process in order to both maximize the detection of suppressed trees and minimize false detections.     

Retrieval of leaf water content spanning the visible to thermal infrared spectra

The objective of this study was to investigate the entire spectra (from visible to the thermal infrared; 0.390–14.0 μm) to retrieve leaf water content in a consistent manner. Narrow-band spectral indices (calculated from all possible two band combinations) and a partial least square regression (PLSR) were used to assess the strength of each spectral region. The coefficient of determination (R²) and root mean square error (RMSE) were used to report the prediction accuracy of spectral indices and PLSR models. In the visible-near infrared and shortwave infrared (VNIR–SWIR), the most accurate spectral index yielded R² of 0.89 and RMSE of 7.60%, whereas in the mid infrared (MIR) the highest R² was 0.93 and RMSE of 5.97%. Leaf water content was poorly predicted using two-band indices developed from the thermal infrared (R² = 0.33). The most accurate PLSR model resulted from MIR reflectance spectra (R² = 0.96, RMSE = 4.74% and RMSE cross validation RMSE_CV = 6.17%) followed by VNIR–SWIR reflectance spectra (R² = 0.91, RMSE = 6.90% and RMSE_CV = 7.32%). Using thermal infrared (TIR) spectra, the PLSR model yielded a moderate retrieval accuracy (R² = 0.67, RMSE = 13.27% and RMSE_CV = 16.39%). This study demonstrated that the mid infrared (MIR) and shortwave infrared (SWIR) domains were the most sensitive spectral region for the retrieval of leaf water content.     

Detecting Sirex noctilio grey-attacked and lightning-struck pine trees using airborne hyperspectral data,random forest and support vector machines classifiers

The visual progression of sirex (Sirex noctilio) infestation symptoms has been categorized into three distinct infestation phases, namely the green, red and grey stages. The grey stage is the final stage which leads to almost complete defoliation resulting in dead standing trees or snags. Dead standing pine trees however, could also be due to the lightning damage. Hence, the objective of the present study was to distinguish amongst healthy, sirex grey-attacked and lightning-damaged pine trees using AISA Eagle hyperspectral data, random forest (RF) and support vector machines (SVM) classifiers. Our study also presents an opportunity to look at the possibility of separating amongst the previously mentioned pine trees damage classes and other landscape classes on the study area. The results of the present study revealed the robustness of the two machine learning classifiers with an overall accuracy of 74.50% (total disagreement = 26%) for RF and 73.50% (total disagreement = 27%) for SVM using all the remaining AISA Eagle spectral bands after removing the noisy ones. When the most useful spectral bands as measured by RF were exploited, the overall accuracy was considerably improved; 78% (total disagreement = 22%) for RF and 76.50% (total disagreement = 24%) for SVM. There was no significant difference between the performances of the two classifiers as demonstrated by the results of McNemar’s test (chi-squared; χ² = 0.14, and 0.03 when all the remaining ASIA Eagle wavebands, after removing the noisy ones and the most important wavebands were used, respectively). This study concludes that AISA Eagle data classified using RF and SVM algorithms provide relatively accurate information that is important to the forest industry for making informed decision regarding pine plantations health protocols.     

Modeling Mediterranean forest structure using airborne laser scanning data

The conservation of biological diversity is recognized as a fundamental component of sustainable development, and forests contribute greatly to its preservation. Structural complexity increases the potential biological diversity of a forest by creating multiple niches that can host a wide variety of species. To facilitate greater understanding of the contributions of forest structure to forest biological diversity, we modeled relationships between 14 forest structure variables and airborne laser scanning (ALS) data for two Italian study areas representing two common Mediterranean forests, conifer plantations and coppice oaks subjected to irregular intervals of unplanned and non-standard silvicultural interventions. The objectives were twofold: (i) to compare model prediction accuracies when using two types of ALS metrics, echo-based metrics and canopy height model (CHM)-based metrics, and (ii) to construct inferences in the form of confidence intervals for large area structural complexity parameters.Our results showed that the effects of the two study areas on accuracies were greater than the effects of the two types of ALS metrics. In particular, accuracies were less for the more complex study area in terms of species composition and forest structure. However, accuracies achieved using the echo-based metrics were only slightly greater than when using the CHM-based metrics, thus demonstrating that both options yield reliable and comparable results. Accuracies were greatest for dominant height (Hd) (R² = 0.91; RMSE% = 8.2%) and mean height weighted by basal area (R² = 0.83; RMSE% = 10.5%) when using the echo-based metrics, 99th percentile of the echo height distribution and interquantile distance. For the forested area, the generalized regression (GREG) estimate of mean Hd was similar to the simple random sampling (SRS) estimate, 15.5 m for GREG and 16.2 m SRS. Further, the GREG estimator with standard error of 0.10 m was considerable more precise than the SRS estimator with standard error of 0.69 m.     

Two new soil moisture indices based on the NIR-red triangle space of Landsat-8 data

In this study, the NIR-red spectral space of Landsat-8 images, which is manifested by a triangle shape, is deployed for developing two new Soil Moisture (SM) indices. First, ten parameters consisting of six distances and four angles were extracted using the position of a random pixel in this triangle. Then, some correlation assessments were made to derive those parameters that were useful for SM estimation, which were five parameters. To build a soil moisture index, all combinations of these five parameters, which were in total 31 different regression equations, were considered, and the best model was named the Triangle Soil Moisture Index (TSMI). The TSMI consists of three parameters. It showed a RMSE of 0.08 and correlation coefficient (R) of 0.67. Since the TSMI does not consider vegetation interface in SM estimation, the Modified TSMI (MTSMI), which takes into account the fraction of soil cover in each pixel, beside those parameters which were used in the TSMI, was developed (MTSMI: RMSE = 0.07, R = 0.74). The results of the TSMI and MTSMI were compared with each other, and with another soil moisture index (SMMRS introduced by Zhan et al. (2007)). It was concluded that the TSMI and MTSMI provide similar results for bare soil or sparsely vegetated surfaces. However, the MTSMI demonstrated a much better performance in densely vegetated surfaces. The accuracy of both the TSMI and MTSMI were significantly higher than the SMMRS. Moreover, the TSMI and MTSMI were validated by comparison with field measured SM data at five different depths. The results showed that satellite estimated SM by these two indices was more correlated with in situ data at 5 cm soil depth compared to other depths. Also, to show the high applicability of the proposed approach for SM estimation, we selected another set of field SM data collected in Australia. The results proved the effectiveness of the method in different study areas.     

Modeling biophysical properties of broad-leaved stands in the hyrcanian forests of Iran using fused airborne laser scanner data and ultraCam-D images

Inventories of mixed broad-leaved forests of Iran mainly rely on terrestrial measurements. Due to rapid changes and disturbances and great complexity of the silvicultural systems of these multilayer forests, frequent repetition of conventional ground-based plot surveys is often cost prohibitive. Airborne laser scanning (ALS) and multispectral data offer an alternative or supplement to conventional inventories in the Hyrcanian forests of Iran. In this study, the capability of a combination of ALS and UltraCam-D data to model stand volume, tree density, and basal area using random forest (RF) algorithm was evaluated. Systematic sampling was applied to collect field plot data on a 150 m × 200 m sampling grid within a 1100 ha study area located at 36°38′- 36°42′N and 54°24′–54°25′E. A total of 308 circular plots (0.1 ha) were measured for calculation of stand volume, tree density, and basal area per hectare. For each plot, a set of variables was extracted from both ALS and multispectral data. The RF algorithm was used for modeling of the biophysical properties using ALS and UltraCam-D data separately and combined. The results showed that combining the ALS data and UltraCam-D images provided a slight increase in prediction accuracy compared to separate modeling. The RMSE as percentage of the mean, the mean difference between observed and predicted values, and standard deviation of the differences using a combination of ALS data and UltraCam-D images in an independent validation at 0.1-ha plot level were 31.7%, 1.1%, and 84 m³ ha⁻¹ for stand volume; 27.2%, 0.86%, and 6.5 m² ha⁻¹ for basal area, and 35.8%, −4.6%, and 77.9 n ha⁻¹ for tree density, respectively. Based on the results, we conclude that fusion of ALS and UltraCam-D data may be useful for modeling of stand volume, basal area, and tree density and thus gain insights into structural characteristics in the complex Hyrcanian forests.     

A sun–crown–sensor model and adapted C-correction logic for topographic correction of high resolution forest imagery

Canopy shadowing mediated by topography is an important source of radiometric distortion on remote sensing images of rugged terrain. Topographic correction based on the sun–canopy–sensor (SCS) model significantly improved over those based on the sun–terrain–sensor (STS) model for surfaces with high forest canopy cover, because the SCS model considers and preserves the geotropic nature of trees. The SCS model accounts for sub-pixel canopy shadowing effects and normalizes the sunlit canopy area within a pixel. However, it does not account for mutual shadowing between neighboring pixels. Pixel-to-pixel shadowing is especially apparent for fine resolution satellite images in which individual tree crowns are resolved. This paper proposes a new topographic correction model: the sun–crown–sensor (SCnS) model based on high-resolution satellite imagery (IKONOS) and high-precision LiDAR digital elevation model. An improvement on the C-correction logic with a radiance partitioning method to address the effects of diffuse irradiance is also introduced (SCnS + C). In addition, we incorporate a weighting variable, based on pixel shadow fraction, on the direct and diffuse radiance portions to enhance the retrieval of at-sensor radiance and reflectance of highly shadowed tree pixels and form another variety of SCnS model (SCnS + W). Model evaluation with IKONOS test data showed that the new SCnS model outperformed the STS and SCS models in quantifying the correlation between terrain-regulated illumination factor and at-sensor radiance. Our adapted C-correction logic based on the sun–crown–sensor geometry and radiance partitioning better represented the general additive effects of diffuse radiation than C parameters derived from the STS or SCS models. The weighting factor W_t also significantly enhanced correction results by reducing within-class standard deviation and balancing the mean pixel radiance between sunlit and shaded slopes. We analyzed these improvements with model comparison on the red and near infrared bands. The advantages of SCnS + C and SCnS + W on both bands are expected to facilitate forest classification and change detection applications.     

Comparison of UAV and WorldView-2 imagery for mapping leaf area index of mangrove forest

Unmanned Aerial Vehicle (UAV) remote sensing has opened the door to new sources of data to effectively characterize vegetation metrics at very high spatial resolution and at flexible revisit frequencies. Successful estimation of the leaf area index (LAI) in precision agriculture with a UAV image has been reported in several studies. However, in most forests, the challenges associated with the interference from a complex background and a variety of vegetation species have hindered research using UAV images. To the best of our knowledge, very few studies have mapped the forest LAI with a UAV image. In addition, the drawbacks and advantages of estimating the forest LAI with UAV and satellite images at high spatial resolution remain a knowledge gap in existing literature. Therefore, this paper aims to map LAI in a mangrove forest with a complex background and a variety of vegetation species using a UAV image and compare it with a WorldView-2 image (WV2).In this study, three representative NDVIs, average NDVI (AvNDVI), vegetated specific NDVI (VsNDVI), and scaled NDVI (ScNDVI), were acquired with UAV and WV2 to predict the plot level (10 × 10 m) LAI. The results showed that AvNDVI achieved the highest accuracy for WV2 (R² = 0.778, RMSE = 0.424), whereas ScNDVI obtained the optimal accuracy for UAV (R² = 0.817, RMSE = 0.423). In addition, an overall comparison results of the WV2 and UAV derived LAIs indicated that UAV obtained a better accuracy than WV2 in the plots that were covered with homogeneous mangrove species or in the low LAI plots, which was because UAV can effectively eliminate the influence from the background and the vegetation species owing to its high spatial resolution. However, WV2 obtained a slightly higher accuracy than UAV in the plots covered with a variety of mangrove species, which was because the UAV sensor provides a negative spectral response function(SRF) than WV2 in terms of the mangrove LAI estimation.     

Winter wheat yield forecasting in Ukraine based on Earth observation,meteorological data and biophysical models

Ukraine is one of the most developed agriculture countries and one of the biggest crop producers in the world. Timely and accurate crop yield forecasts for Ukraine at regional level become a key element in providing support to policy makers in food security. In this paper, feasibility and relative efficiency of using moderate resolution satellite data to winter wheat forecasting in Ukraine at oblast level is assessed. Oblast is a sub-national administrative unit that corresponds to the NUTS2 level of the Nomenclature of Territorial Units for Statistics (NUTS) of the European Union. NDVI values were derived from the MODIS sensor at the 250 m spatial resolution. For each oblast NDVI values were averaged for a cropland map (Rainfed croplands class) derived from the ESA GlobCover map, and were used as predictors in the regression models. Using a leave-one-out cross-validation procedure, the best time for making reliable yield forecasts in terms of root mean square error was identified. For most oblasts, NDVI values taken in April–May provided the minimum RMSE value when comparing to the official statistics, thus enabling forecasts 2–3 months prior to harvest. The NDVI-based approach was compared to the following approaches: empirical model based on meteorological observations (with forecasts in April–May that provide minimum RMSE value) and WOFOST crop growth simulation model implemented in the CGMS system (with forecasts in June that provide minimum RMSE value). All three approaches were run to produce winter wheat yield forecasts for independent datasets for 2010 and 2011, i.e. on data that were not used within model calibration process. The most accurate predictions for 2010 were achieved using the CGMS system with the RMSE value of 0.3 t ha⁻¹ in June and 0.4 t ha⁻¹ in April, while performance of three approaches for 2011 was almost the same (0.5–0.6 t ha⁻¹ in April). Both NDVI-based approach and CGMS system overestimated winter wheat yield comparing to official statistics in 2010, and underestimated it in 2011. Therefore, we can conclude that performance of empirical NDVI-based regression model was similar to meteorological and CGMS models when producing winter wheat yield forecasts at oblast level in Ukraine 2–3 months prior to harvest, while providing minimum requirements to input datasets.