Tropical forests are a major component of the global carbon cycle and home to two-thirds of terrestrial species. Upper-canopy trees store the majority of forest carbon and can be vulnerable to drought events and storms. Monitoring their growth and mortality is essential to understanding forest resilience to climate change, but in the context of forest carbon storage, large trees are underrepresented in traditional field surveys, so estimates are poorly constrained. Aerial photographs provide spectral and textural information to discriminate between tree crowns in diverse, complex tropical canopies, potentially opening the door to landscape monitoring of large trees. Here we describe a new deep convolutional neural network method, Detectree2, which builds on the Mask R-CNN computer vision framework to recognize the irregular edges of individual tree crowns from airborne RGB imagery. We trained and evaluated this model with 3797 manually delineated tree crowns at three sites in Malaysian Borneo and one site in French Guiana. As an example application, we combined the delineations with repeat lidar surveys (taken between 3 and 6 years apart) of the four sites to estimate the growth and mortality of upper-canopy trees. Detectree2 delineated 65 000 upper-canopy trees across 14 km2 of aerial images. The skill of the automatic method in delineating unseen test trees was good (F1 score = 0.64) and for the tallest category of trees was excellent (F1score = 0.74). As predicted from previous field studies, we found that growth rate declined with tree height and tall trees had higher mortality rates than intermediate-size trees. Our approach demonstrates that deep learning methods can automatically segment trees in widely accessible RGB imagery. This tool (provided as an open-source Python package) has many potential applications in forest ecology and conservation, from estimating carbon stocks to monitoring forest phenology and restoration.
Burn-area products from remote sensing provide the backbone for research in fire ecology, management, and modelling. Landsat imagery could be used to create an accurate burn-area map time series at ecologically relevant spatial resolutions. However, the low temporal resolution of Landsat has limited its development in wet tropical and subtropical regions due to high cloud cover and rapid burn-area revegetation. Here, we describe a 34-year Landsat-based burn-area product for wet, subtropical Hong Kong. We overcame technical obstacles by adopting a new LTS fire burn-area detection pipeline that (1) Automatically uniformized Landsat scenes by weighted histogram matching; (2) Estimated pixel resemblance to burn areas based on a random forest model trained on the number of days between the fire event and the date of burn-area detection; (3) Iteratively merged features created by thresholding burn-area resemblance to generate burn-area polygons with detection dates; and (4) Estimated the burn severity of burn-area pixels using a time-series compatible approach. When validated with government fire records, we found that the LTS fire product carried a low area of omission (11%) compared with existing burn-area products, such as GABAM (49%), MCD64A1 (72%), and FireCCI51 (96%) while effectively controlling commission errors. Temporally, the LTS fire pipeline dated 76.9% of burn-area polygons within two months of the actual fire event. The product represents the first Landsat-based burn-area product in wet tropical and subtropical Asia that covers the entire time series. We believe that burn-area products generated from algorithms like LTS fire will effectively bridge the gap between remote sensing and field-based studies on wet tropical and subtropical fire ecology.
Climate change is causing glaciers to retreat across much of the Himalaya, leading to a rapid shift of the vegetation cover to higher altitudes. However, the rate of vegetation shift with respect to glacier retreat, climate change, and topographic parameters is not empirically quantified. Using remote sensing measurements, we estimate (a) the rate of glacier-ice mass loss, (b) the upward vegeta- tion line shift rate, (c) regional greening trends, and (d) a relationship between the factors influencing the greenness of the landscape and vegetation change in the Himalaya. We find that the glacier mass loss rate is 10.9 ± 1.2 Gt/yr and the mean vegetation line shifts upward in altitude by 7–28 ± 1.5 m/yr. Considering the land use/land cover change pattern, the grassland area is found to be expanding the most, particularly in the de-glaciated regions. The vegetation change is found to be controlled by soil moisture and slope of the area.
Bandyopadhyay (2022), Environmental Monitoring and Assessment
Canopy gaps are openings in the forest canopy resulting from branch fall and tree mortality events. The geographical distribution of large canopy gaps may reflect underlying variation in mortality and growth processes. However, a lack of data at the appropriate scale has limited our ability to study this relationship until now.
We detected canopy gaps using a unique LiDAR data set consisting of 650 transects randomly distributed across 2500 km2 of the Brazilian Amazon. We characterized the size distribution of canopy gaps using a power-law and we explore the variation in the exponent, α. We evaluated how the α varies across the Amazon, in response to disturbance by humans and natural environmental processes that influence tree mortality rates.
We observed that South-eastern forests contained a higher proportion of large gaps than North-western, which is consistent with recent work showing greater tree mortality rates in the Southeast than the Northwest. Regions characterised by strong wind gust speeds, frequent lightning and greater water shortage also had a high proportion of large gaps, indicating that geographical variation in α is a reflection of underlying disturbance processes. Forests on fertile soils were also found to contain a high proportion of large gaps, in part because trees grow tall on these sites and create large gaps when they fall; thus canopy gap analysis picked up differences in growth as well as mortality processes. Finally, we found that human modified forests had a higher proportion of large gaps than intact forests, as we would expect given that these forests have been disturbed.
Synthesis: The proportion of large gaps in the forest canopy varied substantially over the Brazilian Amazon. We have shown that the trends can be explained by geographic variation in disturbance and growth. The frequency of extreme weather events is predicted to increase under climate change, and changes could lead to greater forest disturbance, which should be detectable as an increased proportion of large gaps in intact forests.
Logged forests cover four million square kilometres of the tropics, capturing carbon more rapidly than temperate forests and harbouring rich biodiversity. Restoring these forests is essential to help avoid the worst impacts of climate change. Yet monitoring tropical forest recovery is challenging. We track the abundance of early-successional species in a forest restoration concession in Indonesia. If the species are carefully chosen, they can be used as an indicator of restoration progress. We present SLIC-UAV, a new pipeline for processing Unoccupied Aerial Vehicle (UAV) imagery using simple linear iterative clustering (SLIC)to map early-successional species in tropical forests. The pipeline comprises: (a) a field verified approach for manually labelling species; (b) automatic segmentation of imagery into ‘superpixels’ and (c) machine learning classification of species based on both spectral and textural features. Creating superpixels massively reduces the dataset’s dimensionality and enables the use of textural features, which improve classification accuracy. In addition, this approach is flexible with regards to the spatial distribution of training data. This allowed us to be flexible in the field and collect high-quality training data with the help of local experts. The accuracy ranged from from 74.3\% for a four-species classification task to 91.7\% when focusing only on the key early-succesional species. We then extended these models across 100 hectares of forest, mapping species dominance and forest condition across the entire restoration project.
Tropical forests are subject to diverse deforestation pressures while their conservation is essential to achieve global climate goals. Predicting the location of deforestation is challenging due to the complexity of the natural and human systems involved but accurate and timely forecasts could enable effective planning and on-the-ground enforcement practices to curb deforestation rates. New computer vision technologies based on deep learning can be applied to the increasing volume of Earth observation data to generate novel insights and make predictions with unprecedented accuracy.
Here, we demonstrate the ability of deep convolutional neural networks (CNNs) to learn spatiotemporal patterns of deforestation from a limited set of freely available global data layers, including multispectral satellite imagery, the Hansen maps of annual forest change (2001–2020) and the ALOS PALSAR digital surface model, to forecast deforestation (2021). We designed four model architectures, based on 2D CNNs, 3D CNNs, and Convolutional Long Short-Term Memory (ConvLSTM) Recurrent Neural Networks (RNNs), to produce spatial maps that indicate the risk to each forested pixel (~30 m) in the landscape of becoming deforested within the next year. They were trained and tested on data from two ~80,000 km2 tropical forest regions in the Southern Peruvian Amazon.
The networks could predict the location of future forest loss to a high degree of accuracy (F1 = 0.58–0.71). Our best performing model (3D CNN) had the highest pixel-wise accuracy (F1 = 0.71) when validated on 2020 forest loss (2014–2019 training). Visual interpretation of the mapped forecasts indicated that the network could automatically discern the drivers of forest loss from the input data. For example, pixels around new access routes (e.g. roads) were assigned high risk, whereas this was not the case for recent, concentrated natural loss events (e.g. remote landslides).
Convolutional neural networks can harness limited time-series data to predict near-future deforestation patterns, an important step in harnessing the growing volume of satellite remote sensing data to curb global deforestation. The modelling framework can be readily applied to any tropical forest location and used by governments and conservation organisations to prevent deforestation and plan protected areas.
Reducing Emissions from Deforestation and forest Degradation (REDD+) projects aim to contribute to climate change mitigation by protecting and enhancing carbon stocks in tropical forests, but there are no systematic global evaluations of their impact. Using a new data set for tropical humid forests, we used a standardised evaluation approach to quantify the performance of a representative sample of 40 voluntary REDD+ certified under the Verified Carbon Standard, located in nine countries. In the first five years of implementation, deforestation within project areas was reduced by 47% (95% CI = 24–68%) compared with matched counterfactual pixels, while degradation rates were 58% lower (95% CI = 49–63%). Reductions were small in absolute terms but greater in sites located in high deforestation settings, and did not appear to be substantially undermined by leakage activities in forested areas within 10-km of project boundaries. At COP26 the international community renewed its commitment to tackling tropical deforestation as a nature-based solution to climate change. Our results indicate that incentivising forest conservation through voluntary site-based projects can slow tropical deforestation; they also highlight the particular importance of targeting financing to areas at greater risk of deforestation.
NASA’s Global Ecosystem Dynamics Investigation (GEDI) is collecting spaceborne full waveform lidar data with a primary science goal of producing accurate estimates of forest aboveground biomass density (AGBD). This paper presents the development of the models used to create GEDI’s footprint-level (~25 m) AGBD (GEDI04_A) product, including a description of the datasets used and the procedure for final model selection. The data used to fit our models are from a compilation of globally distributed spatially and temporally coincident field and airborne lidar datasets, whereby we simulated GEDI-like waveforms from airborne lidar to build a calibration database. We used this database to expand the geographic extent of past waveform lidar studies, and divided the globe into four broad strata by Plant Functional Type (PFT) and six geographic regions. GEDI’s waveform-to-biomass models take the form of parametric Ordinary Least Squares (OLS) models with simulated Relative Height (RH) metrics as predictor variables. From an exhaustive set of candidate models, we selected the best input predictor variables, and data transformations for each geographic stratum in the GEDI domain to produce a set of comprehensive predictive footprint-level models. We found that model selection frequently favored combinations of RH metrics at the 98th, 90th, 50th, and 10th height above ground-level percentiles (RH98, RH90, RH50, and RH10, respectively), but that inclusion of lower RH metrics (e.g. RH10) did not markedly improve model performance. Second, forced inclusion of RH98 in all models was important and did not degrade model performance, and the best performing models were parsimonious, typically having only 1-3 predictors. Third, stratification by geographic domain (PFT, geographic region) improved model performance in comparison to global models without stratification. Fourth, for the vast majority of strata, the best performing models were fit using square root transformation of field AGBD and/or height metrics. There was considerable variability in model performance across geographic strata, and areas with sparse training data and/or high AGBD values had the poorest performance. These models are used to produce global predictions of AGBD, but will be improved in the future as more and better training data become available.
Laser scanning technology has enabled to study three-dimensional (3D) structures in forests. For example, airborne laser scanning (ALS) point cloud has been applied to detect individual trees and segment tree crowns. However, the accuracy of such approach remains a challenge because of the intersected crowns and complicated understories. We developed a metabolic theory-based algorithm for individual tree detection and crown segmentation from ALS data. The algorithm is composed of two parts, of which one is an unscaled transporting distance-based top-to-bottom detection approach, and the other is a scaled transporting distance-based segmentation approach. The unscaled transporting distance for detection is the absolute distance from tree root to crown and then to ALS point based on the tree structure models, whereas the scaled transporting distance for segmentation is the unscaled distance that is scaled by an initial tree height obtained during detection. This is based on a basic metabolic theory that vascular plants tend to minimize the material transporting distance from root to leaves. Hence, seven types of materials transporting distance models were built based on monopodial branching structure or crown-centered structure. The performance of the proposed approach was then further examined and compared with two typical canopy height model-based approaches and one typical point cloud-based approach, taking forest in Oxfordshire, UK, as a case study. The results showed that our approach can reach a recall of 1.00, a precision of 0.96, and an F-score of 0.98 and can reach to much higher accuracy for tree height (R2 = 0.8045) than the comparison approaches (R2 < 0.2) in the study plot. One of the main reasons that led to such low accuracy of comparison approaches is much overestimation of understory height with a mean error that is 2.9 times higher than that of our approach on average. Furthermore, ALS point-to-point level accuracy assessment shows 9.7% more ALS points were truly assigned in our approach than that of comparison approaches. It is noticed that the algorithm presented is not sensitive to the two key parameters: p (a percentage determining the threshold of unscaled transporting distances) ranging from 32.0% to 34.0% and λ (the proportion between the assumed crown center height and tree height) ranging from 0.70 to 0.90 based on our data set. Such high accuracy of our approach can greatly improve detections of individual tree and crown segmentation, especially in delineating understories in complex-structured forest.
We have all seen trees swaying in the wind, but did you know that tree motion can teach us about ecology? Researchers have monitored tree motion for different purposes, from assessing wind damage risk to monitoring drought stress. Our new paper brings all this data together to study the differences between types of trees and test whether previous results generalize across a range of data sets.
We computed a set of descriptive features from the tree motion time-series data and found that the conifers clustered into a relatively tight group. This shows that trees in conifer forests move in similar ways in response to wind, presumably due to their relatively simple shapes. Conversely, there was a wide variation in the motion of open-grown trees and broadleaf trees. Data availability followed the opposite pattern, with most of the high quality (long-term, well calibrated) data collected in conifer forests. Long-term tree motion data for open grown trees and broadleaf trees were particularly rare, so future research should fill this data gap to help us understand their risk of wind damage.
Measuring the motion of a 65 m tall Coastal Redwood tree in the wind. Photo by Anthony Ambrose.
Previous studies have suggested that additional damping or amplification mechanisms become important at high wind speeds. If true, this would be a key factor in modelling wind damage, which is a driver of forest disturbance (the largest in Europe). We calculated the slope of the power spectrum for each hour of tree motion data and found that it plateaued between medium and high wind speeds. This suggests that no substantial changes in tree motion characteristics occurs across this range. Unfortunately (from a scientific point of view) none of the data sets in this study captured high enough wind speeds to cause tree damage, so we cannot rule out additional effects at higher wind speeds. However, we found no evidence for these effects in the available data.
As you can see from the large author list, this was a collaborative project involving many researchers who made their data openly available and contributed to discussions about the science. All the data we collated for this project are freely available online https://zenodo.org/record/4915883#.YO6phOhKhnJ and we encourage others to download and use it.