This study explores a method to classify seven tropical rainforest tree species from full-range (400–2,500 nm) hyperspectral data acquired at tissue (leaf and bark), pixel and crown scales using laboratory and airborne sensors. Metrics that respond to vegetation chemistry and structure were derived using narrowband indices, derivative- and absorption-based techniques, and spectral mixture analysis. We then used the Random Forests tree-based classifier to discriminate species with minimally-correlated, importance-ranked metrics. At all scales, best overall accuracies were achieved with metrics derived from all four techniques and that targeted chemical and structural properties across the visible to shortwave infrared spectrum (400–2500 nm). For tissue spectra, overall accuracies were 86.8% for leaves, 74.2% for bark, and 84.9% for leaves plus bark. Variation in tissue metrics was best explained by an axis of red absorption related to photosynthetic leaves and an axis distinguishing bark water and other chemical absorption features. Overall accuracies for individual tree crowns were 71.5% for pixel spectra, 70.6% crown-mean spectra, and 87.4% for a pixel-majority technique. At pixel and crown scales, tree structure and phenology at the time of image acquisition were important factors that determined species spectral separability.

In this study, a 1-D Convolutional Neural Network (CNN) architecture was developed, trained and utilized to classify single (summer) and three seasons (spring, summer, fall) of hyperspectral imagery over the San Francisco Bay Area, California for the year 2015. For comparison, the Random Forests (RF) and Support Vector Machine (SVM) classifiers were trained and tested with the same data. In order to support space-based hyperspectral applications, all analyses were performed with simulated Hyperspectral Infrared Imager (HyspIRI) imagery. Three-season data improved classifier overall accuracy by 2.0% (SVM), 1.9% (CNN) to 3.5% (RF) over single-season data. The three-season CNN provided an overall classification accuracy of 89.9%, which was comparable to overall accuracy of 89.5% for SVM. Both three-season CNN and SVM outperformed RF by over 7% overall accuracy. Analysis and visualization of the inner products for the CNN provided insight to distinctive features within the spectral-temporal domain. A method for CNN kernel tuning was presented to assess the importance of learned features. We concluded that CNN is a promising candidate for hyperspectral remote sensing applications because of the high classification accuracy and interpretability of its inner products.

Web-based applications that integrate geospatial information, or the geoweb, offer exciting opportunities for remote sensing science. One such application is a Web‑based system for automating the collection of reference data for producing and verifying the accuracy of land-use/land-cover (LULC) maps derived from satellite imagery. Here we describe the capabilities and technical components of the Virtual Interpretation of Earth Web-Interface Tool (VIEW-IT), a collaborative browser-based tool for “crowdsourcing” interpretation of reference data from high resolution imagery. The principal component of VIEW-IT is the Google Earth plug-in, which allows users to visually estimate percent cover of seven basic LULC classes within a sample grid. The current system provides a 250 m square sample to match the resolution of MODIS satellite data, although other scales could be easily accommodated. Using VIEW-IT, a team of 23 student and 7 expert interpreters collected over 46,000 reference samples across Latin America and the Caribbean. Samples covered all biomes, avoided spatial autocorrelation, and spanned years 2000 to 2010. By embedding Google Earth within a Web-based application with an intuitive user interface, basic interpretation criteria, distributed Internet access, server-side storage, and automated error-checking, VIEW-IT provides a time and cost efficient means of collecting a large dataset of samples across space and time. When matched with predictor variables from satellite imagery, these data can provide robust mapping algorithm calibration and accuracy assessment. This development is particularly important for regional to global scale LULC mapping efforts, which have traditionally relied on sparse sampling of medium resolution imagery and products for reference data. Our ultimate goal is to make VIEW-IT available to all users to promote rigorous, global land-change monitoring.

The Global Ecosystem Dynamics Investigation (GEDI) lidar began data acquisition from the International Space Station in March 2019 and is expected to make over 10 billion measurements of canopy structure and topography over two years. Previously, airborne lidar data with limited spatial coverage have been used to examine relationships between forest canopy structure and faunal diversity, most commonly bird species. GEDI’s latitudinal coverage will permit these types of analyses at larger spatial extents, over the majority of the Earth’s forests, and most importantly in areas where canopy structure is complex and/or poorly understood. In this regional study, we examined the impact that GEDI-derived Canopy Structure variables have on the performance of bird species distribution models (SDMs) in Sonoma County, California. We simulated GEDI waveforms for a two-year period and then interpolated derived Canopy Structure variables to three grid sizes of analysis. In addition to these variables, we also included Phenology, Climate, and other Auxiliary variables to predict the probability of occurrence of 25 common bird species. We used a weighted average ensemble of seven individual machine learning models to make predictions for each species and calculated variable importance. We found that Canopy Structure variables were, on average at our finest resolution of 250 m, the second most important group (32.5%) of predictor variables after Climate variables (35.3%). Canopy Structure variables were most important for predicting probability of occurrence of birds associated with Conifer forest habitat. Regarding spatial analysis scale, we found that finer-scale models more frequently performed better than coarser-scale models, and the importance of Canopy Structure variables was greater at finer spatial resolutions. Overall, GEDI Canopy Structure variables improved SDM performance for at least one spatial resolution for 19 of 25 species and thus show promise for improving models of bird species occurrence and mapping potential habitat.

Land change in the Greater Antilles differs markedly among countries because of varying socioeconomic histories and global influences. We assessed land change between 2001 and 2010 in municipalities (second administrative units) of Cuba, Dominican Republic, Haiti, Jamaica, and Puerto Rico. Our analysis used annual land-use/land-cover maps derived from MODIS satellite imagery to model linear change in woody vegetation, mixed-woody/plantations and agriculture/herbaceous vegetation. Using this approach, we focused on municipalities with significant change (p ≤ 0.05). Between 2001 and 2010, the Greater Antilles gained 801 km² of woody vegetation. This increase was mainly due to the return of woody vegetation in Cuba, and smaller increases in Puerto Rico and the Dominican Republic. Despite relatively similar environments, the factors associated with these changes varied greatly between countries. In Puerto Rico, Dominican Republic, and Jamaica, agriculture declined while mixed-woody vegetation increased, mostly in montane regions. In contrast, Cuba experienced an extensive decline in sugarcane plantations, which resulted in the spread of an invasive woody shrub species and the increase in woody vegetation in areas of high agricultural value. In Haiti, the growing population, fuelwood consumption, and increase in agriculture contributed to woody vegetation loss; however, woody vegetation loss was accompanied with a significant increase in the mixed woody and plantations class. Most regional analyses often treated the Greater Antilles as a homogeneous unit; our results suggest that historical and socio-economic differences among countries are crucial for understanding the variation in present day land change dynamics.