A 3D perspective on the effects of topography and wind on forest height and dynamics

Lead Research Organisation: University of Cambridge
Department Name: Plant Sciences

Abstract

Strong winds can unleash huge destructive power on forests. For example, Britain's Great Storm of 1987 blew over about 15 million trees. Most research into wind damage occurs in temperate plantations, because of their social and economic value, but much less is known about the impact of wind on tropical rainforest. We recently used survey forests in Sabah (a Malaysian state on the island of Borneo) using airborne laser scanning, as part of a NERC-funded programme into human-modified tropical forests. We spotted the tallest tree ever found in the tropics in the laser scans: an 89-m tall individual in the Dipterocarpaceae family of trees, which are renowned for producing giants. This discovery raises questions about the mechanical limits of height in tropical forests and the reason why Sabah's lowland forests grow so much taller than those of Amazonia and Africa. One idea is that trees are particularly tall in sheltered hillsides in Sabah because they seldom experience strong winds. Another idea is that Dipterocarps have evolved a unique set of mechanical properties that enable them to withstand strong wind better than other tropical trees. Both of these theories need critical evaluation.

The project addresses these ideas, by studying wind flow and tree responses in 40 square kilometres of old-growth hill forest situated in Sabah, where NERC's airborne research facility has already mapped ground elevation, canopy height and foliar nutrient concentrations by laser scanning (ALS) and imaging spectroscopy. The tallest forests in the tropics juxtapose with stunted heath forests in these hills, making the landscape ideal for evaluating the role of wind in determining forest height and dynamics. We will compare wind flows in the immensely tall forests of Borneo with those found in much shorter forests of South America (French Guiana) and the hurricane-impacted Caribbean (Puerto Rico), including an analysis of how wind is funneled through hilly landscapes resulting in strong wind forces on forests in exposed sites. We will also quantify the susceptibility of individual trees to windthrow by a novel approach involving 3D reconstruction of hundreds of rainforest trees using terrestrial laser scanning and simulation modelling of stresses within their trunks when exposed to wind gusts. Combining the wind flow maps with the tree susceptibility will allow us to produce maps showing where we think trees are most at risk of windthrow in each tropical region. The key question is whether sites predicted to be at high risk and found to be wind damaged. To test this, we will use the latest advances in repeat-survey ALS to track growth and mortality of about 200,000 tall trees (>30 m). Measuring tree demography at this unprecedented scale will provide a unique and powerful data set with which to test whether wind is a major determinant of tree death and canopy height. Finally, we will put this work in a global context. Wood density, modulus of elasticity and tree architecture of Bornean trees will be compared with values from French Guianan and pan-tropical datasets, to evaluate whether the exceptional tall trees are mechanically distinct. Simulations will be run using wind regimes from other tropical regions to assess whether a lack of extreme wind events may have contributed to the development of such tall forests in Borneo. ALS-measured canopy height is closely related to the aboveground carbon density, so these analyses will provide a wall-to-wall perspective on carbon dynamics in the context of climate change.

Planned Impact

A primary beneficiary will be Sabah Forestry Department (SFD), which is committed to protecting and restoring the state's remaining areas of high conservation value forest. They have taken a keen interest in the discovery of very tall trees in the state, using airborne laser scanning, and are already promoting visits to the sites for ecotourism. We will produce maps of tall tree groves in collaboration with SFD to aid conservation planning. Borneo has lost much of its lowland forests to agriculture in recent decades, but the state of Sabah is at the forefront of conservation and restoration efforts in the region. Knowing the location of very large trees would be useful for conservation planning in the region. We will run our tree segmentation algorithms over the entire laser-scanned areas of Sabah, to locate all trees over 60 m tall. These tall-tree maps will be used by SFD and presented to conservation practitioners at the 2020 Heart of Borneo conference, a key regional meeting that brings together government officials, researchers, conservationists and agribusiness leaders and provides an important opportunity to engage with stakeholders.

We plan to produce short films which will benefit SFD by providing them material for use in their education centres and will benefit the wider public via social media. Large trees are among the longest-lived and largest organisms on the planet, and as such are a source of inspiration, often appearing in artwork, films as well as news outlets. For example, our discovery of the largest trees ever found in the tropics have attracted over 100,000 views on short videos we produced, and was selected as a top ten news story (news.mongabay.com/2016/12/top-10-happy-environmental-stories-of-2016/ ). We plan to capitalise on this public interest in tall trees to communicate the wider scientific story, producing films that discuss (a) the loss of forests in SE Asia and recent initiatives aimed at protection of remaining old-growth forests and restoration of secondary forests; (b) the science of tall trees; (c) the science of airborne and terrestrial laser scanning in the context of forest science. These films will bring a sense of adventure and wonder to the ongoing efforts to conserve the world's dwindling natural resources. We will also present our work live at science festivals to inspire young people in the application of science to address pressing environmental issues.

The PI is part of the Cambridge Conservation Initiative, working in the same building as major conservation organisations that manage reserves in SE Asia. These organisations have expressed interest in our work because they have identified a need to monitor reserves more effectively and recognise the value of new remote sensing tools. We will hold a workshop with these partners to explain what we are doing and foster collaboration.

Publications

10 25 50
publication icon
Shenkin A (2019) The World's Tallest Tropical Tree in Three Dimensions in Frontiers in Forests and Global Change

 
Title Map of the probability of giant trees occurrence (> 70 m) in the Brazilian Amazon 
Description The probability of giant trees occurrence (> 70m) based on environmental conditions. The observations higher than 70 m were filtered out and used to adjust an envelope model based on maximum entropy. In its optimization routine, the algorithm tracked how much the model gain was improved when small changes were made to each coefficient value associated with a particular variable. The resulting map of predicted occurrence of the tallest trees in the Amazon from the MaxEnt model shows that the probability of maximum tree height occurrence is highest in the northeastern Amazon (Fig. 6), more specifically in the Roraima and Guianan Lowlands. We considered 18 environmental variables: (1) fraction of absorbed photosynthetically active radiation (FAPAR; in %); (2) elevation above sea level (Elevation; in m); (3) the component of the horizontal wind towards east, i.e. zonal velocity (u-speed ; in m s -1); (4) the component of the horizontal wind towards north, i.e. meridional velocity (v-speed ; in m s -1); (5) the number of days not affected by cloud cover (clear days; in days yr -1); (6) the number of days with precipitation above 20 mm (days > 20mm; in days yr -1 ); (7) the number of months with precipitation below 100 mm (months < 100mm; in months yr -1 ) ; (8) lightning frequency (flashes rate); (9) annual precipitation (in mm); (10) potential evapotranspiration (in mm); (11) coefficient of variation of precipitation (precipitation seasonality; in %); (12) amount of precipitation on the wettest month (precip. wettest; in mm); (13) amount of precipitation on the driest month (precip. driest; in mm); (14) mean annual temperature (in °C); (15) standard deviation of temperature (temp. seasonality; in °C); (16) annual maximum temperature (in °C); (17) soil clay content (in %); and (18) soil water content (in %). 
Type Of Material Database/Collection of data 
Year Produced 2020 
Provided To Others? Yes  
URL https://zenodo.org/record/4037101
 
Title Map of the probability of giant trees occurrence (> 70 m) in the Brazilian Amazon 
Description The probability of giant trees occurrence (> 70m) based on environmental conditions. The observations higher than 70 m were filtered out and used to adjust an envelope model based on maximum entropy. In its optimization routine, the algorithm tracked how much the model gain was improved when small changes were made to each coefficient value associated with a particular variable. The resulting map of predicted occurrence of the tallest trees in the Amazon from the MaxEnt model shows that the probability of maximum tree height occurrence is highest in the northeastern Amazon (Fig. 6), more specifically in the Roraima and Guianan Lowlands. We considered 18 environmental variables: (1) fraction of absorbed photosynthetically active radiation (FAPAR; in %); (2) elevation above sea level (Elevation; in m); (3) the component of the horizontal wind towards east, i.e. zonal velocity (u-speed ; in m s -1); (4) the component of the horizontal wind towards north, i.e. meridional velocity (v-speed ; in m s -1); (5) the number of days not affected by cloud cover (clear days; in days yr -1); (6) the number of days with precipitation above 20 mm (days > 20mm; in days yr -1 ); (7) the number of months with precipitation below 100 mm (months < 100mm; in months yr -1 ) ; (8) lightning frequency (flashes rate); (9) annual precipitation (in mm); (10) potential evapotranspiration (in mm); (11) coefficient of variation of precipitation (precipitation seasonality; in %); (12) amount of precipitation on the wettest month (precip. wettest; in mm); (13) amount of precipitation on the driest month (precip. driest; in mm); (14) mean annual temperature (in °C); (15) standard deviation of temperature (temp. seasonality; in °C); (16) annual maximum temperature (in °C); (17) soil clay content (in %); and (18) soil water content (in %). 
Type Of Material Database/Collection of data 
Year Produced 2020 
Provided To Others? Yes  
URL https://zenodo.org/record/4037100
 
Title Maximum tree height map for the Brazilian Amazon 
Description Maximum tree height distribution estimated by the Random Forest model based on the environmental variables. To explore the influence and importance of the environmental variables for development in tree height, we employed Random Forest modeling, which consists of generating a large number of regression trees, each constructed considering a random data subset. The regression trees are used to identify the best sequence to split the solution space to estimate the output. Were considered 18 environmental variables: (1) fraction of absorbed photosynthetically active radiation (FAPAR; in %); (2) elevation above sea level (Elevation; in m); (3) the component of the horizontal wind towards east, i.e. zonal velocity (u-speed ; in m s -1); (4) the component of the horizontal wind towards north, i.e. meridional velocity (v-speed ; in m s -1); (5) the number of days not affected by cloud cover (clear days; in days yr -1); (6) the number of days with precipitation above 20 mm (days > 20mm; in days yr -1 ); (7) the number of months with precipitation below 100 mm (months < 100mm; in months yr -1 ) ; (8) lightning frequency (flashes rate); (9) annual precipitation (in mm); (10) potential evapotranspiration (in mm); (11) coefficient of variation of precipitation (precipitation seasonality; in %); (12) amount of precipitation on the wettest month (precip. wettest; in mm); (13) amount of precipitation on the driest month (precip. driest; in mm); (14) mean annual temperature (in °C); (15) standard deviation of temperature (temp. seasonality; in °C); (16) annual maximum temperature (in °C); (17) soil clay content (in %); and (18) soil water content (in %). 
Type Of Material Database/Collection of data 
Year Produced 2020 
Provided To Others? Yes  
URL https://zenodo.org/record/4036988
 
Title Maximum tree height map for the Brazilian Amazon 
Description Maximum tree height distribution estimated by the Random Forest model based on the environmental variables. To explore the influence and importance of the environmental variables for development in tree height, we employed Random Forest modeling, which consists of generating a large number of regression trees, each constructed considering a random data subset. The regression trees are used to identify the best sequence to split the solution space to estimate the output. Were considered 18 environmental variables: (1) fraction of absorbed photosynthetically active radiation (FAPAR; in %); (2) elevation above sea level (Elevation; in m); (3) the component of the horizontal wind towards east, i.e. zonal velocity (u-speed ; in m s -1); (4) the component of the horizontal wind towards north, i.e. meridional velocity (v-speed ; in m s -1); (5) the number of days not affected by cloud cover (clear days; in days yr -1); (6) the number of days with precipitation above 20 mm (days > 20mm; in days yr -1 ); (7) the number of months with precipitation below 100 mm (months < 100mm; in months yr -1 ) ; (8) lightning frequency (flashes rate); (9) annual precipitation (in mm); (10) potential evapotranspiration (in mm); (11) coefficient of variation of precipitation (precipitation seasonality; in %); (12) amount of precipitation on the wettest month (precip. wettest; in mm); (13) amount of precipitation on the driest month (precip. driest; in mm); (14) mean annual temperature (in °C); (15) standard deviation of temperature (temp. seasonality; in °C); (16) annual maximum temperature (in °C); (17) soil clay content (in %); and (18) soil water content (in %). 
Type Of Material Database/Collection of data 
Year Produced 2020 
Provided To Others? Yes  
URL https://zenodo.org/record/4036987
 
Description Eric Gorgen 
Organisation Federal University of Jequitinhonha and Mucuri Valleys
Country Brazil 
Sector Academic/University 
PI Contribution We provided expertise in lidar process and tropical ecology
Collaborator Contribution The provided access to an airborne lidar survey of brazil which cost over US$ 10,000,000 to collect and hasn't been that extensively utilised since it was collected
Impact My postdoc on the project - Toby Jackson - travelled to Brazil to film the largest tree ever discovered in South America and record aspects of its ecology. We have two papers published and another couple almost ready from this work. It has helped form a trusting relationship with INPE and Brazilian scientists which has made access to further datasets more possible.
Start Year 2019
 
Title Maximum entropy model trained to estimate probability to host trees taller then 70 m based on enviromental factors 
Description Focusing only on the tallest trees - those over 70 m in height - we built an environmental envelope model to assess the conditions which allow them to occur. We employed the maximum entropy approach (MaxEnt) commonly applied to modelling species geographic distributions with presence-only data to discriminate suitable versus unsuitable areas for the species. We initially considered a total of 18 environmental variables: (1) fraction of absorbed photosynthetically active radiation (FAPAR; in %); (2) elevation above sea level (Elevation; in m); (3) the component of the horizontal wind towards east, i.e. zonal velocity (u-speed ; in m s -1); (4) the component of the horizontal wind towards north, i.e. meridional velocity (v-speed ; in m s -1); (5) the number of days not affected by cloud cover (clear days; in days yr -1); (6) the number of days with precipitation above 20 mm (days > 20mm; in days yr -1 ); (7) lightning frequency (flashes rate); (8) annual precipitation (in mm); (9) potential evapotranspiration (in mm); (10) coefficient of variation of precipitation (precipitation seasonality; in %); (11) amount of precipitation on the wettest month (precip. wettest; in mm); (12) mean annual temperature (in °C); (13) standard deviation of temperature (temp. seasonality; in °C); (14) annual maximum temperature (in °C); (15) soil clay content (in %); and (16) soil water content (in %). 
Type Of Technology Software 
Year Produced 2020 
Open Source License? Yes  
URL https://zenodo.org/record/4066652
 
Title Maximum entropy model trained to estimate probability to host trees taller then 70 m based on enviromental factors 
Description Focusing only on the tallest trees - those over 70 m in height - we built an environmental envelope model to assess the conditions which allow them to occur. We employed the maximum entropy approach (MaxEnt) commonly applied to modelling species geographic distributions with presence-only data to discriminate suitable versus unsuitable areas for the species. We initially considered a total of 18 environmental variables: (1) fraction of absorbed photosynthetically active radiation (FAPAR; in %); (2) elevation above sea level (Elevation; in m); (3) the component of the horizontal wind towards east, i.e. zonal velocity (u-speed ; in m s -1); (4) the component of the horizontal wind towards north, i.e. meridional velocity (v-speed ; in m s -1); (5) the number of days not affected by cloud cover (clear days; in days yr -1); (6) the number of days with precipitation above 20 mm (days > 20mm; in days yr -1 ); (7) lightning frequency (flashes rate); (8) annual precipitation (in mm); (9) potential evapotranspiration (in mm); (10) coefficient of variation of precipitation (precipitation seasonality; in %); (11) amount of precipitation on the wettest month (precip. wettest; in mm); (12) mean annual temperature (in °C); (13) standard deviation of temperature (temp. seasonality; in °C); (14) annual maximum temperature (in °C); (15) soil clay content (in %); and (16) soil water content (in %). 
Type Of Technology Software 
Year Produced 2020 
Open Source License? Yes  
URL https://zenodo.org/record/4066653
 
Title Random Forest trained to estimate Amazon maximum height based on enviromental factors 
Description The Random Forest model obtained MAE = 3.62 m, RMSE = 4.92 m, and R² = 0.735. we initially considered a total of 18 environmental variables: (1) fraction of absorbed photosynthetically active radiation (FAPAR; in %); (2) elevation above sea level (Elevation; in m); (3) the component of the horizontal wind towards east, i.e. zonal velocity (u-speed ; in m s -1); (4) the component of the horizontal wind towards north, i.e. meridional velocity (v-speed ; in m s -1); (5) the number of days not affected by cloud cover (clear days; in days yr -1); (6) the number of days with precipitation above 20 mm (days > 20mm; in days yr -1 ); (7) the number of months with precipitation below 100 mm (months < 100mm; in months yr -1 ) ; (8) lightning frequency (flashes rate); (9) annual precipitation (in mm); (10) potential evapotranspiration (in mm); (11) coefficient of variation of precipitation (precipitation seasonality; in %); (12) amount of precipitation on the wettest month (precip. wettest; in mm); (13) amount of precipitation on the driest month (precip. driest; in mm); (14) mean annual temperature (in °C); (15) standard deviation of temperature (temp. seasonality; in °C); (16) annual maximum temperature (in °C); (17) soil clay content (in %); and (18) soil water content (in %). Among the initial 18 environmental variables, two of them (precipitation on driest month and months < 100mm) were excluded due to high correlation (> 0.80) to other independent variables. 
Type Of Technology Software 
Year Produced 2020 
Open Source License? Yes  
URL https://zenodo.org/record/4061838
 
Title Random Forest trained to estimate Amazon maximum height based on enviromental factors 
Description The Random Forest model obtained MAE = 3.62 m, RMSE = 4.92 m, and R² = 0.735. we initially considered a total of 18 environmental variables: (1) fraction of absorbed photosynthetically active radiation (FAPAR; in %); (2) elevation above sea level (Elevation; in m); (3) the component of the horizontal wind towards east, i.e. zonal velocity (u-speed ; in m s -1); (4) the component of the horizontal wind towards north, i.e. meridional velocity (v-speed ; in m s -1); (5) the number of days not affected by cloud cover (clear days; in days yr -1); (6) the number of days with precipitation above 20 mm (days > 20mm; in days yr -1 ); (7) the number of months with precipitation below 100 mm (months < 100mm; in months yr -1 ) ; (8) lightning frequency (flashes rate); (9) annual precipitation (in mm); (10) potential evapotranspiration (in mm); (11) coefficient of variation of precipitation (precipitation seasonality; in %); (12) amount of precipitation on the wettest month (precip. wettest; in mm); (13) amount of precipitation on the driest month (precip. driest; in mm); (14) mean annual temperature (in °C); (15) standard deviation of temperature (temp. seasonality; in °C); (16) annual maximum temperature (in °C); (17) soil clay content (in %); and (18) soil water content (in %). Among the initial 18 environmental variables, two of them (precipitation on driest month and months < 100mm) were excluded due to high correlation (> 0.80) to other independent variables. 
Type Of Technology Software 
Year Produced 2020 
Open Source License? Yes  
URL https://zenodo.org/record/4061837
 
Description Video released on University Webite 
Form Of Engagement Activity Engagement focused website, blog or social media channel
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Public/other audiences
Results and Impact The University produced a video explaining how Toby Jackson had travelled to a remote region of Brazil to survey a massive tropical tree discovered by laser scanning
Year(s) Of Engagement Activity 2019
URL https://www.cam.ac.uk/research/news/expedition-finds-tallest-tree-in-the-amazon