Performing Department
Institute of Agriculture, Natural Resources and Extension
Non Technical Summary
Alaska's 126 million acres of forested land provides ecosystem services that benefit society at levels ranging from local to global including climate regulations. Nonetheless, accelerated climate change poses a serious threat to its health and productivity. Alaskan forest is transforming in response to persistent warming and inconsistent precipitation. Insect pests are exploding across the state forest and wildfires are consuming more forest acres than ever before. Forests in some parts of Kenai Peninsula are being permanently replaced by shrubland and grassland. A consistent statewide assessment of forest health and disturbance is lacking at present. Given consensus on the value of forest to the global climate system, Alaskan biodiversity, and local communities and economy, up to date data driven policy formulation is needed to sustain forest health and productivity, reduce insect risk, and enhance community resilience to wildfires.The overarching goal of this project is annual assessment of statewide forest health, forest damage, and forest cover change (based on satellite observations) to support forest health protection and forest management.The project will generate various map products, forest statistics and assessment for improved understanding of statewide forest health, forest damage, and forest change based on repeated satellite observations. It will provide a regionally consistent and locally relevant record of forest health, disturbance, and change at annual time scale. Also, the project will answer key questions to aid data and policy-driven improvement in forest health protection and forest management. The project findings and new knowledge will empower the U.S. Forest Service, Alaska Department of Natural Resources, and public and private land owners to implement policies and take actions for adaptive management of forest land and resources.We will create an interactive webpage for sharing project updates and annual map products with stakeholders (landowners, land managers, fire managers, timber industry, project partners) and the public. The webpage will have features to get feedback from the user that will be used to further validate/improve the product accuracy.For public outreach, we will develop and offer a workshop on Remote Sensing of Forest Health for K-12 educators, create a few short communication videos for sharing project highlights with stakeholders and the general public, participate in select public events, and contribute to a collaborative Science-Art exhibition in the final year of the project.
Animal Health Component
0%
Research Effort Categories
Basic
80%
Applied
(N/A)
Developmental
20%
Goals / Objectives
Goal:To conduct an annual assessment of statewide forest health, forest damage, and forest cover change (based on satellite observation) to support forest health protection and forest management.Research objectives: Forest health and productivity: Map forest cover and assess forest productivityusing satellite image spectral indicessince 2000Forest disturbance: Map ongoing forest damage and severity caused by insects, diseases, and wildfires using satellite image spectral indices and image classificationForest change: Map forest gain and loss including forest demography (conifer vs deciduous) using satellite image spectral indices and image classification since 2000
Project Methods
4.A. Forest health and productivity:Objective: An annual assessment of forest cover and productivity using satellite image spectral indices.Where has forest productivity (photosynthesis) been declining, stable, and increasing since 2000?What percentage of Alaska's forests are browning (languishing) vs greening (flourishing)?We can answer the above questions by studying forest productivity at Alaska scale using historic (since 2000) and current satellite image data. We will use the Google Earth Engine (GEE) cloud computing platform and its image archive (Landsat: since 1999; Sentinel 2: since 2015). GEE is an ideal platform for processing and analysis of large numbers of satellite data at regional scale, and for sharing algorithms and user-friendly applications (Gorelick et al., 2017). The Landsat pixel size of 30 m is detailed enough to study forest productivity at state scale (Verbyla, 2015; FS-R10-FHP, 2019). We will use a time series of Landsat surface reflectance data since 2000 to compute the NDVI and leaf-area index (LAI) at 30 m pixel for all forested land of Alaska. Using regression analysis, we will assess the trends in mean and peak summer NDVI values. We will identify hot spots of positive (greening) and negative (browning) NDVI trends to analyze the spatial and temporal pattern in forest productivity. We will carry out field validation for any hot spot site in the road network. We will leverage the U.S. Forest Service Aerial Detection Survey for field validation of hot spots in remote areas. We will carry out correlation study with climate variables (temperature, precipitation, and VPD), topography and geomorphology to understand and ascertain the factors or proximal causes of forest greening and browning.4.B. Forest disturbance (from insects, diseases, and fire):Objective: An annual assessment of the extent and severity of forest damage caused by insects, disease, and wildfires using satellite remote sensing.What percentage of Alaska's forest is impacted by insects, disease, and fire in a given year?With 126 million acres of forested land and few roads, monitoring forest disturbance in Alaska is challenging. Satellite remote sensing has the potential to detect forest damage and map its extent and severity statewide. One major advantage of satellite imagery is that it provides a consistent, enduring record of the landscape. Change-detection methods that use satellite imagery inputs can be documented, repeated, revisited, and adapted with new information as needed at any time.Some of the Alaska specific challenges in employing satellite remote sensing on forest disturbance are clouds, wildfire smoke, and wet puddles that are constantly changing. These three events are very common during the short growing season in Alaska. The existing tools and algorithms need improvement for better detection of clouds, smoke, and surface water so that the forest change analysis (or damage assessment caused by pests and disease) can be more accurate and reliable.Both Landsat and Sentinel satellite sensors have bands specific for aerosol (band 1), moisture (band 2), and cloud (band 9) detection. We will use these bands to improve the existing algorithms for cloud detection in the Alaskan context. Image timing is another important aspect of detecting forest damage caused by insects. For example, mid-July is the best time to see the impact of spruce bark beetles, by August leafminers are dominant. So employing both Landsat and Sentinel data will improve the temporal image coverage and enable multiple assessment within one growing season.We will use the GEE's cloud computing platform and its image archive to analyze relative change in pixel reflectance for all forested land in Alaska. Change in reflectance values can range from low-magnitude (noise and subtle defoliation) to mid-magnitude (defoliation and tree mortality) to high-magnitude (fire, landslides, and harvest). We will relate the magnitude of change to ascertain proximal cause and to assess the severity of the disturbance. The satellite detected change will be validated by aerial and ground survey information. Along with the survey data we will employ any existing field data for accuracy assessment. With enough supplemental information about disturbance timing and forest attributes, we hope to accurately identify the cause of damage. We will carry out a thorough analysis of algorithm limitations and error for reliable and responsible use of our products. We will also explore how multivariate statistical methods can relate satellite-detected change to disturbances documented by aerial surveys so that we can estimate the extent and severity of known outbreaks.4.C. Forest cover change:Objective: An annual assessment of forest gain and loss including forest cover demography using satellite remote sensing.What percentage of forest cover is conifer vs deciduous and how have their distributions changed since 2000?What is the cumulative impact of climate change, insect outbreaks, and wildfires on forest cover and demography?Any forest change (gain or loss; change from one cover type to another) due to fire, insect outbreak, and disease need to be mapped, checked for accuracy, and documented at the end of each summer to support effective pest and forest management efforts.We will use the GEE's cloud computing platform that allows ready access to Landsat and Sentinel image data along with processing algorithms and computing resources for mapping 1) forest cover, forest gain and loss, and 2) conifer and deciduous cover distribution. We will map forest cover for each year since 2000 and estimate forest cover statistics (gain and loss). We will use peak growing season NDVI, NBR, and/or similar spectral indices for mapping forest cover. We will identify hot spots of forest cover change and use existing aerial survey or field data to diagnose the proximal cause of change. For any hot spot accessible by road we will field validate the remote sensing results as well as the cause of change. Our project partners, Alaska FHP and Alaska State Division Forestry conduct extensive aerial and field surveys each summer to assess forest health and damage. Both partners have agreed to support field validation of our remote sensing results.We will map conifer and deciduous forest cover for each year since 2000 using Landsat (2000-present) and Sentinel (2015-present) image data. Conifer and deciduous vegetation have distinct spectral characteristics i.e. conifer vegetation usually have lower reflectance profile compared to deciduous vegetation. Because of their unique spectral characteristics, conifer and deciduous vegetation can be mapped reliably from Landsat/Sentinel images. We will use Random Forest machine learning image classifier to map the two vegetation types. Previous studies have demonstrated the success and effectiveness of Random Forest classifier in vegetation mapping (Chapman et al., 2010; Rodriguez-Galiano et al., 2012; Smith et al., 2021). Random Forest algorithm performs better when a large amount of training data is available, less prone to overfitting, ideal for processing large volume of data, and results in higher accuracy (Cutler et al., 2007; Chapman et al., 2010; Rodriguez-Galiano et al., 2012; Belgiu et al., 2016). We will use aerial and field survey data along with any existing ground data for Quality Assurance/Quality Control (QA/QC) of our results and map products. Also, we hope to compare our map products with global products like Global Forest Watch Open Data.The processing workflow for the three research focus areas will be tested at select study sites (e.g. Bonanza Creek Experimental Forest, Caribou-Poker Creeks Research Watershed, Kenai National Wildlife Refuge) before implementing the workflow on state-wide imagery.