Catching up with 'Observe'

Every 3 - 4 months, Silva21 gathers to hear project updates from our highly qualified personnel (HQP). Last year, we organized these meetings by hub to highlight research taking place in close geographical proximity. This time around, we organized these meetings by Silva21's three research themes: Observe, Anticipate, Adapt.

---

In OBSERVE, we collect data using innovative tools to assess the growth and vigor of trees and thus allow more flexible and adaptive management strategies in the face of climatic stresses and disturbances

---

On Tuesday, November 15th, we heard from our following HQPs with projects under the Observe theme:

Click on a name to read their summary:

Liam Irwin: Project OB2: Advanced remote sensing

My project focuses on developing a performance monitoring framework for mid-rotation stands, attempting to answer the need for better silvicultural information with the increasing availability of fine-scale remotely sensed data from drone-based sensors. Over the summer I conducted extensive field sampling in British Columbia, Ontario, and Quebec to investigate topic.

In Quesnel we collected around 250 hectares of high density lidar, and 1cm RGB imagery before thinning. In addition to remote sensing data, we conducted stem mapping of around 1000 trees in plots distributed across the five commercial thinning blocks sampled. This fall, I returned to several of the now thinned blocks and collected a sample of 70 tree cores across my stem mapped plots, as well as 1cm RGB imagery. With these cores I aim to correlate observed growth in diameter with metrics computed from the lidar and RGB tree approximations. These metrics will be applied to explain diameter growth and quantify the degree of competition experienced by each approximated tree in the stand. Doing so will allow us to create a species specific estimated incremental growth map which could be used to prioritize and evaluate the thinning treatment.

Out east we visited Timmins, Ontario and worked in the Romeo-Malette Forest. In the forest we mainly sampled block 18; a large enhanced forest productivity demonstration site. The site was planted with varying species mixes of six local conifers back in 2006 and serves as a diverse area to test tree-level approximation from remote sensing. An additional draw to this site is the existence of high density 2018 and 2020 ALS datasets. By using a combination of tree approximation and species classification with the 2022 drone datasets I hope to backcast these results together wtith previous ALS acquisitions to observe incremental height growth during these periods. I collected around 300 hectares of drone lidar and 1cm RGB imagery to attempt this; as well as measuring around 750 trees in existing permanent growth plots distributed across the area. These data will help answer questions around evaluating species level height growth performance with a drone-based reference, across a diverse boreal site.

During the trip we had the chance to meet with partners at the Green First Forest Products office in Timmins for an evening meal and presentations, this was a very useful exchange of local knowledge and the latest science, thanks for hosting us! In Quebec we also visited the Montmorency Research Forest where we held presentations and had a drone demonstration including the acquisition of drone lidar across a large balsam fir study site.

Liam Irwin (UBC)

PhD student

liamkirwin@gmail.com

Sarah Smith-Tripp: OB1a: Regeneration after catastrophic disturbance

After collecting over a thousand hectares of RPA lidar in Quesnel this summer Sarah had developed linear relationships between the field measurements and the models of interest. She used wall-to-wall estimates of field measurements to identify unique forest structural types on the landscape after severe disturbances using the lidar based estimates. The lidar estimates are then used as goals to achieve using satellite data. Next steps are to assess the patterns as they are modified by time since the disturbance

Sarah Smith-Tripp

PhD Student (UBC)

ssmithtr@student.ubc.ca

Gabrielle Thibeault: OB1b - Optimization of the characterization of burning patterns

I am doing a master’s degree in Forest Science at Laval University under the supervision of Alexis Achim. For my project I’m trying to optimise the characterization process of burn pattern in the boreal forest. The burn pattern is the impact of the fire on the landscape, and the characterization of that pattern is used for the salvage operation after fire.

I am presently trying to classify each of the pixels of some high-resolution satellite images by using classification models. The images I am using are really diversified since they are from three different satellite sensors (Skysat, Geoeye, Pleiades), with resolution between 0,5m and 1,71m and they don’t have all the same bands.

I am trying to identify the complications that this heterogeneity could bring to the classification process. Three classification models (featureless, knn and randomforest) were tested to determine which one would be the most accurate. Those models were train with three different training data set to evaluate if they would be good for generalisation. That test show that the random forest model is the most accurate. The first training dataset (called « other ») train the model on every image except the one I want to classify. The second dataset (called « same ») train the model with data from the image I want to classify. The last dataset (called « all ») trains the model with data from all the images including the one I want to classify. The results show that it’s harder to generalise the classification to a new image that the model has never seen. To increase the accuracy of the classification, some training data must be taken from the image we want to classify.

The next question would be: How much training data do I need to collect from the image I want to classify

Gabrielle Thibeault

M.Sc. student (ULaval)

gabrielle.thibault.4@ulaval.ca

Alexandre Morin-Bernard: OB 3b - Early alert systems for forest management

The response of boreal forests to climate change is highly variable depending on species composition, but also on the region and site characteristics. There is accumulating evidence that warmer temperatures are associated with a decrease in the growth rate of black spruce forests where water is limited. However, the magnitude and location of such gradual changes in forest productivity remains unknown. Remote sensing from satellite imagery is now widely used to study and monitor forest ecosystems, but has yet mainly focused on detecting abrupt changes, while gradual change had received considerably less attention. The objective of this part of the project was to propose a modelling approach to predict the net growth in boreal forests using Landsat-derived vegetation indices and to apply the approach to characterize trends in growth over the last decades over a forest management unit in Canada.

We used data from 162 permanent sample plots located within and around the Romeo Malette Forest to derive the average net basal area growth rate for the interval between two consecutive field measurements. For each plot and associated interval between field measurements, we created time series of Landsat-derived vegetation indices for a 3 x 3 pixel window enclosing the plot. We used ordinary least square regression to model the net growth from Landsat-derived predictors using two thirds of the entire dataset, and relied on a model selection procedure to identify the best model and associated predictors. The predictors retained in the best-ranked model comprised the Theil-Sen slope of the Tasseled Cap Wetness as well as the values of the TCW and NBR index at the first field measurement. To investigate long-term trends over the study site, we first mapped the predicted the net growth over all coniferous stands in the study site that have not been subjects to a stand-replacing disturbance between 1984 and 2021. To investigate how the growth trend changed over time, we applied the model to predict the net growth over a five-year moving window, resulting in 34 windows. Decomposing the time series shorter intervals allows seeing the evolution of the predicted growth rate through time. To get a better understanding of the growth dynamics in the study site, we performed a cluster analysis of the Tasseled Cap Wetness time series to identify exemplars of growth trajectories during the investigated period. The next step of the project will be dedicated to linking fine scale growth and biophysical data such as growth from increment cores and LiDAR data to predictions of the model with the aim of identifying characteristics shared by stands showing a similar growth trend.

Alexandre Morin-Bernard

PhD student (ULaval)

alexandre.morin-bernard1@ulaval.ca

Chris Mulverhill: OB 5a: continuous forest inventory framework

Enhanced forest inventories (EFIs) based on airborne lidar data form a fundamental component forest management in the 21st century. However, due to high cost of data acquisition, EFIs are only updated on 5- or 10-year repeat intervals. A continuous forest inventory framework has been proposed in order to allow EFIs to be updated on a shorter repeat interval. The components of this framework include establishing the initial EFI, continuously monitoring for change using optical satellite data, updating changed cells, and forecasting growth in unchanged cells. Two manuscripts about using satellite data for change detection have been submitted and are in review.

Current work on this project is on updating EFI attributes in changed pixels. To do this, pixels are first grouped into strata of similar dominant species and site indices (taken from inventory polygons). Next, models are developed to estimate the difference in forest attributes (such as basal area or canopy cover) as a function of difference in spectral index values. Initial results suggest that attributes such as canopy cover have the highest estimation accuracy (r2 ≈ 0.90, 10.4 relative root mean square error; RMSE%), while attributes such as basal area are less accurately predicted using optical data (r2 ≈ 0.57, 14.2 RMSE%). Future work will aim to continue developing these models and test their accuracies in the context of a continuous forest inventory.

Chris Mulverhill

Postdoctoral Fellow (UBC)

chrismulverhill@gmail.com

José Riofrio: OB 7: Climate-sensitive growth modeling in Ontario

Tree mortality is a complex multifactorial process with short- and long-term impacts on different forest attributes. Empirical mortality models make it possible to predict the probability of mortality of individual trees in function of different tree and stand or site characteristics. Mortality prediction is an essential component of individualbased forest growth simulators. However, in most empirical simulators, mortality is predicted without consideration for climate variables. Given the ongoing climate change, having climate-sensitive mortality models would help understand and predict the functioning and dynamics of forest ecosystems under changing environmental conditions.

This project aims to develop species-specific climate-sensitive tree mortality models for 30 tree species in Ontario. To do this, we first homogenize the data from the Ontario Forest Growth and Yield Program in a database structured to track the individual tree status (alive or dead) along successive measurements of each permanent plot in nonoverlapping intervals. Then, we will fit the species-specific mortality models using logistic regression. In a first step, we will fit a basic version of our mortality models by considering only the tree- and stand-level variables. In a second step, we will add climate variables in these basic models and measured the gain of precision.

José Riofrio

Postdoctoral Fellow (UBC)

jose.g.riofrio@gmail.cm

---

If you are a member of the Silva21 and would like to receive a copy of all sides, please email our scientific coordinator (amy.wotherspoon@ubc.ca)

---

Our ADAPT and ANTICIPATE meetings are taking place later this November, so we'll be sure to post summary reports of those as well.