Tree Species Classification in Mixed Deciduous Forests Using Very High Spatial Resolution Satellite Imagery and Machine Learning Methods

Capacity and capability of remote sensing to inform invasive plant species management in the Pacific Islands region

The Pacific Islands region is home to several of the world's biodiversity hotspots, yet its unique flora and fauna are under threat because of biological invasions. These invasions are likely to proliferate as human activity increases and large-scale natural disturbances unfold, exacerbated by climate change. Remote sensing data and techniques provide a feasible method to map and monitor invasive plant species and inform invasive plant species management across the Pacific Islands region. We used case studies taken from literature retrieved from Google Scholar, 3 regional agencies' digital libraries, and 2 online catalogs on invasive plant species management to examine the uptake and challenges faced in the implementation of remote sensing technology in the Pacific region. We synthesized remote sensing techniques and outlined their potential to detect and map invasive plant species based on species phenology, structural characteristics, and image texture algorithms. The application of remote sensing methods to detect invasive plant species was heavily reliant on species ecology, extent of invasion, and available geospatial and remotely sensed image data. However, current mechanisms that support invasive plant species management, including policy frameworks and geospatial data infrastructure, operated in isolation, leading to duplication of efforts and creating unsustainable solutions for the region. For remote sensing to support invasive plant species management in the region, key stakeholders including conservation managers, researchers, and practitioners; funding agencies; and regional organizations must invest, where possible, in the broader geospatial and environmental sector, integrate, and streamline policies and improve capacity and technology access.

Use of Multi-Date and Multi-Spectral UAS Imagery to Classify Dominant Tree Species in the Wet Miombo Woodlands of Zambia

Accurate maps of tree species distributions are necessary for the sustainable management of forests with desired ecological functions. However, image classification methods to produce species distribution maps for supporting sustainable forest management are still lacking in the Miombo woodland ecoregion. This study used multi-date multispectral Unmanned Aerial Systems (UAS) imagery collected at key phenological stages (leaf maturity, transition to senescence, and leaf flushing) to classify five dominant canopy species of the wet Miombo woodlands in the Copperbelt Province of Zambia. Object-based image analysis (OBIA) with a random forest algorithm was used on single date, multi-date, and multi-feature UAS imagery for classifying the dominant canopy tree species of the wet Miombo woodlands. It was found that classification accuracy varies both with dates and features used. For example, the August image yielded the best single date overall accuracy (OA, 80.12%, 0.68 kappa), compared to October (73.25% OA, 0.59 kappa) and May (76.64% OA, 0.63 kappa). The use of a three-date image combination improved the classification accuracy to 84.25% OA and 0.72 kappa. After adding spectral indices to multi-date image combination, the accuracy was further improved to 87.07% and 0.83 kappa. The results highlight the potential of using multispectral UAS imagery and phenology in mapping individual tree species in the Miombo ecoregion. It also provides guidance for future studies using multispectral UAS for sustainable management of Miombo tree species.

Forest tree species distribution for Europe 2000-2020: mapping potential and realized distributions using spatiotemporal machine learning

This article describes a data-driven framework based on spatiotemporal machine learning to produce distribution maps for 16 tree species (Abies alba Mill., Castanea sativa Mill., Corylus avellana L., Fagus sylvatica L., Olea europaea L., Picea abies L. H. Karst., Pinus halepensis Mill., Pinus nigra J. F. Arnold, Pinus pinea L., Pinus sylvestris L., Prunus avium L., Quercus cerris L., Quercus ilex L., Quercus robur L., Quercus suber L. and Salix caprea L.) at high spatial resolution (30 m). Tree occurrence data for a total of three million of points was used to train different algorithms: random forest, gradient-boosted trees, generalized linear models, k-nearest neighbors, CART and an artificial neural network. A stack of 305 coarse and high resolution covariates representing spectral reflectance, different biophysical conditions and biotic competition was used as predictors for realized distributions, while potential distribution was modelled with environmental predictors only. Logloss and computing time were used to select the three best algorithms to tune and train an ensemble model based on stacking with a logistic regressor as a meta-learner. An ensemble model was trained for each species: probability and model uncertainty maps of realized distribution were produced for each species using a time window of 4 years for a total of six distribution maps per species, while for potential distributions only one map per species was produced. Results of spatial cross validation show that the ensemble model consistently outperformed or performed as good as the best individual model in both potential and realized distribution tasks, with potential distribution models achieving higher predictive performances (TSS = 0.898, R2 logloss = 0.857) than realized distribution ones on average (TSS = 0.874, R2 logloss = 0.839). Ensemble models for Q. suber achieved the best performances in both potential (TSS = 0.968, R2 logloss = 0.952) and realized (TSS = 0.959, R2 logloss = 0.949) distribution, while P. sylvestris (TSS = 0.731, 0.785, R2 logloss = 0.585, 0.670, respectively, for potential and realized distribution) and P. nigra (TSS = 0.658, 0.686, R2 logloss = 0.623, 0.664) achieved the worst. Importance of predictor variables differed across species and models, with the green band for summer and the Normalized Difference Vegetation Index (NDVI) for fall for realized distribution and the diffuse irradiation and precipitation of the driest quarter (BIO17) being the most frequent and important for potential distribution. On average, fine-resolution models outperformed coarse resolution models (250 m) for realized distribution (TSS = +6.5%, R2 logloss = +7.5%). The framework shows how combining continuous and consistent Earth Observation time series data with state of the art machine learning can be used to derive dynamic distribution maps. The produced predictions can be used to quantify temporal trends of potential forest degradation and species composition change.

Individual Tree Species Classification Based on Convolutional Neural Networks and Multitemporal High-Resolution Remote Sensing Images

The classification of individual tree species (ITS) is beneficial to forest management and protection. Previous studies in ITS classification that are primarily based on airborne LiDAR and aerial photographs have achieved the highest classification accuracies. However, because of the complex and high cost of data acquisition, it is difficult to apply ITS classification in the classification of large-area forests. High-resolution, satellite remote sensing data have abundant sources and significant application potential in ITS classification. Based on Worldview-3 and Google Earth images, convolutional neural network (CNN) models were employed to improve the classification accuracy of ITS by fully utilizing the feature information contained in different seasonal images. Among the three CNN models, DenseNet yielded better performances than ResNet and GoogLeNet. It offered an OA of 75.1% for seven tree species using only the WorldView-3 image and an OA of 78.1% using the combinations of WorldView-3 and autumn Google Earth images. The results indicated that Google Earth images with suitable temporal detail could be employed as auxiliary data to improve the classification accuracy.

A Deep Fusion uNet for Mapping Forests at Tree Species Levels with Multi-Temporal High Spatial Resolution Satellite Imagery

It is critical to acquire the information of forest type at the tree species level due to its strong links with various quantitative and qualitative indicators in forest inventories. The efficiency of deep-learning classification models for high spatial resolution (HSR) remote sensing image has been demonstrated with the ongoing development of artificial intelligence technology. However, due to limited statistical separability and complicated circumstances, completely automatic and highly accurate forest type mapping at the tree species level remains a challenge. To deal with the problem, a novel deep fusion uNet model was developed to improve the performance of forest classification refined at the dominant tree species level by combining the beneficial phenological characteristics of the multi-temporal imagery and the powerful features of the deep uNet model. The proposed model was built on a two-branch deep fusion architecture with the deep Res-uNet model functioning as its backbone. Quantitative assessments of China’s Gaofen-2 (GF-2) HSR satellite data revealed that the suggested model delivered a competitive performance in the Wangyedian forest farm, with an overall classification accuracy (OA) of 93.30% and a Kappa coefficient of 0.9229. The studies also yielded good results in the mapping of plantation species such as the Chinese pine and the Larix principis.

Methodology for the Definition of Durum Wheat Yield Homogeneous Zones by Using Satellite Spectral Indices

One of the main questions facing precision agriculture is the evaluation of different algorithms for the delineation of homogeneous management zones. In the present study, a new approach based on the use of time series of satellite imagery, collected during two consecutive growing seasons, was proposed. Texture analysis performed using the Gray-Level Co-Occurrence Matrix (GLCM) was used to integrate and correct the sum of the vegetation indices maps (NDVI and MCARI2) and define the homogenous productivity zones on ten durum wheat fields in southern Italy. The homogenous zones identified through the method that integrates the GLCM indices with the spectral indices studied showed a greater accuracy (0.18–0.22 Mg ha−1 for ∑NDVIs + GLCM and 0.05–0.49 Mg ha−1 for ∑MCARI2s + GLCM) with respect to the methods that considered only the sum of the indices. Best results were also obtained with respect to the homogeneous zones derived by using yield maps of the previous year or vegetation indices acquired in a single day. Therefore, the survey methods based on the data collected over the entire study period provided the best results in terms of estimated yield; the addition of clustering analysis performed with the GLCM method allowed to further improve the accuracy of the estimate and better define homogeneous productivity zones of durum wheat fields.

An Evaluation of Pixel- and Object-Based Tree Species Classification in Mixed Deciduous Forests Using Pansharpened Very High Spatial Resolution Satellite Imagery

Quality tree species information gathering is the basis for making proper decisions in forest management. By applying new technologies and remote sensing methods, very high resolution (VHR) satellite imagery can give sufficient spatial detail to achieve accurate species-level classification. In this study, the influence of pansharpening of the WorldView-3 (WV-3) satellite imagery on classification results of three main tree species (Quercus robur L., Carpinus betulus L., and Alnus glutinosa (L.) Geartn.) has been evaluated. In order to increase tree species classification accuracy, three different pansharpening algorithms (Bayes, RCS, and LMVM) have been conducted. The LMVM algorithm proved the most effective pansharpening technique. The pixel- and object-based classification were applied to three pansharpened imageries using a random forest (RF) algorithm. The results showed a very high overall accuracy (OA) for LMVM pansharpened imagery: 92% and 96% for tree species classification based on pixel- and object-based approach, respectively. As expected, the object-based exceeded the pixel-based approach (OA increased by 4%). The influence of fusion on classification results was analyzed as well. Overall classification accuracy was improved by the spatial resolution of pansharpened images (OA increased by 7% for pixel-based approach). Also, regardless of pixel- or object-based classification approaches, the influence of the use of pansharpening is highly beneficial to classifying complex, natural, and mixed deciduous forest areas.

Mapping Floods in Lowland Forest Using Sentinel-1 and Sentinel-2 Data and an Object-Based Approach

The impact of floods on forests is immediate, so it is necessary to quickly define the boundaries of flooded areas. Determining the extent of flooding in situ has shortcomings due to the possible limited spatial and temporal resolutions of data and the cost of data collection. Therefore, this research focused on flood mapping using geospatial data and remote sensing. The research area is located in the central part of the Republic of Croatia, an environmentally diverse area of lowland forests of the Sava River and its tributaries. Flood mapping was performed by merging Sentinel-1 (S1) and Sentinel-2 (S2) mission data and applying object-based image analysis (OBIA). For this purpose, synthetic aperture radar (SAR) data (GRD processing level) were primarily used during the flood period due to the possibility of all-day imaging in all weather conditions and flood detection under the density of canopy. The pre-flood S2 imagery, a summer acquisition, was used as a source of additional spectral data. Geographical information system (GIS) layers—a multisource forest inventory, habitat map, and flood hazard map—were used as additional sources of information in assessing the accuracy of and interpreting the obtained results. The spectral signature, geometric and textural features, and vegetation indices were applied in the OBIA process. The result of the work was a developed methodological framework with a high accuracy and speed of production. The overall accuracy of the classification is 94.94%. Based on the conducted research, the usefulness of the C band of the S1 in flood mapping in lowland forests in the leaf-off season was determined. The paper presents previous research and describes the SAR parameters and characteristics of floodplain forest with a significant impact on the accuracy of classification.