Integrating Remote Sensing and Field Measurements to Identify Environmental Nonstationarity on Interior Alaska DoD Training Lands

Jennifer Watts | The Woods Hole Research Center



Over 60% of lands in Alaska (out of 1.72 M km2 ) are administered by the U.S. Government, with approximately 1,115,948 acres (4,516 km2 ) managed by the U.S. Department of Defense (DoD). The majority of these DoD lands are managed by the Army and are located near Fairbanks. These lands are remote, with many terrains inaccessible by roads. Off-road access is limited during the summer by extremely wet soils over permafrost (ground that has remained frozen for at least two years) and an abundance of surface water distributed across lakes, rivers and wetlands. The frozen winter period allows for better access to the landscape, but safety and ease of travel greatly depend on ice and snow conditions which are difficult to quantify remotely. The DoD has already invested heavily in infrastructure across Alaska, including Air Force radar and launch sites, Air Force and Army bases, and Army Arctic training ranges and test sites. Recent emphasis on future DoD development and Arctic operations in Alaska, identified in the 2016 DoD Report to Congress on National Security Interests in the Arctic and echoed in the FY2019 DoD Budget Request, includes new investments in Alaska’s missile defense systems and on-the-ground training exercises in addition to long-term plans for road building to access remote training areas (e.g., Tanana Flats and Yukon Training Areas in Interior Alaska) and ongoing support and maintenance for existing infrastructure.

Alaska is warming at twice the rate of the global average, and climate models now predict that temperatures across the state will increase by up to 6 ºC (10.8°F) before 2099. This warming will yield potentially severe consequences for frozen ground conditions, vegetation, hydrology, habitat characteristics, transportation and infrastructure across the state. A changing climate places Alaska ecosystems at greater risk for disturbance and, as a consequence, will increasingly affect major military installations and their terrains through nonstationarity in environmental conditions.

This one-year project was conducted to characterize contemporary heterogeneity in landscape change occurring across Alaska with a focus on DoD training lands in Interior Alaska. This study was designed to integrate observations from visible, near-infrared, thermal (VIS-NIRTIR) and microwave satellite remote sensing to detect shifts in three Earth System Indicators: 1) Terrain Thermal State; 2) Ecosystem Water Stress; 3) Vegetation State. Through statistical trend analysis, Alaska regions experiencing nonstationarity in these focus areas from < 2002 to 2017 were identified and mapped to provide the DoD with geospatial information indicating the locations of terrains experiencing longer-term ecosystem shifts. This study also explored and tested new methods for satellite data fusion (downscaling), using machine learning, to produce new Earth System Indicator records having greatly improved (30 m; 500 m) spatial resolutions to better support localized change detection and decision-making for DoD land managers. The ultimate goal of this work was to identify, test, and validate potential remotely sensed observational tools that can be used to identify areas where landscape change is occurring or may occur in the future. Due to the remote nature of many of the training ranges, remotely sensed tools and applications have the potential to save time and money and reduce environmental and safety risks for land management and training activities.

Back to Top

Technical Approach

This project included multiple remote sensing measurements obtained from a suite of U.S. supported satellite platforms available for both regional (e.g., Alaska) and global analysis. Specifically, this study used multi-spectral observations from Landsat, MODIS, SSM/I, SMMR, SMMIS, AMSR-E and AMSR2 to investigate nonstationarity in landscape frozen state, surface hydrology and vegetation over a period spanning 1979 (for Freeze/Thaw state; > 2002 for other indicators) to 2017. Image collection was daily for all sensors, except Landsat which had a 16 day repeat interval. When possible, the satellite observations were supported by in situ data provided through existing field monitoring efforts including work funded by NASA, the U.S. Army Basic Research Program, U.S. Army Alaska, and SERDP. In addition, the project team used observations from NASA’s MERRA2 reanalysis data assimilation system, from 1980 to 2017, to detect changes in air temperature and precipitation over Alaska.

A Mann-Kendall statistical time series analysis was applied to each long-term satellite and reanalysis data record on a per grid cell basis. The time series change detection analysis provided information on longer-term monotonic trends and rates of change across Alaska. In addition, the project team evaluated the use of Breaks for Additive Season and Trend (BFAST) detection in Interior Alaska to identify the occurrence of abrupt landscape change.

The project employed machine learning to integrate high temporal frequency, yet coarser spatial resolution, passive microwave remote sensing retrievals with higher spatial resolution VIS-NIR-TIR observations to examine the feasibility of providing new, spatially-improved Earth System Indicator time series records for DoD terrains in Interior Alaska. Although the change detection and data fusion methodology developed for this project were demonstrated for Alaska, they were designed to be globally applicable and readily transferable to other regions.

Back to Top


Terrain Thermal State

The change detection analysis for Alaska, using a multi-sensor suite of data from VIS-NIRTR and microwave satellite remote sensing, showed that the length of the spring (April 15th to May 31st) transitional period decreased at an average rate of six days per decade over the 16 year (2002 to 2017) observation period. For this project's purposes, the transitional state is defined as where a terrain surface had changed from thawed to frozen state, or vice versa, within a 24-hour period. In addition, the project team found that the spring thaw period had increased at an average rate of eight days per decade. These detected rates of change are higher than the ~ 3.5 days per decade increase in annual non-frozen period reported in past studies for years prior to 2006 (Smith et al. 2004; Zhang et al. 2011).

This study also observed a significant increase in the length of the autumn (September 1 to October 15) transitional period, at an average rate of two days per decade. In addition, the satellite records showed a significant increase in transitional state for Alaska during the winter period (October 16 to April 14) at a rate of 1.5 days per decade. For Interior Alaska, the project team found that the number of thawed days in autumn and early winter increased over the major DoD terrains of Donnelly Training Area and the Tanana Flats Training Area at a rate of seven days per decade, based on the AMSR-E/AMSR2 satellite record.

Ecosystem Water Stress

This study detected a significant increase in summer precipitation over 11% of Alaska’s terrains (at a rate of up to three mm per year) from 1980 to 2017 within the MERRA2 reanalysis record. This increase occurred along the Alaska Range, the southern coastal region, and YukonKuskokwim Delta. This finding is in agreement with estimates from Atmosphere-Ocean General Circulation Models which indicate that precipitation amounts are increasing within Alaska (Larsen et al. 2008). The analysis also detected a significant and multi-year decrease in summer precipitation over Interior Alaska, including on DoD terrains (at a rate of 1.5 mm per year).

The trend analysis using the 16 year (2002 to 2017) AMSR-E/AMSR2 satellite record revealed a significant increase in summer surface soil moisture over much of Alaska, including the Interior. The DoD Yukon Training Area and the Tanana Flats Training Area were observed to have significant positive trends in summer and autumn soil moisture, at a rate of 0.2 cm3 /cm3 per decade. However, the project team also detected a significant decrease in summer surface soil moisture (0.2 cm3 /cm3 per decade) occurring over the Donnelly Training Area.

The project team observed a significant increase in summer surface water inundation across coastal zones and within localized (lowland) terrains in Interior Alaska, as detected in the AMSR-E/AMSR2 25 km (2002 to 2017) and 5 km (2002 to 2015) records. This increase in standing water occurred over 21% of the state (368,000 km2 ). Within the Interior Alaska Area of Interest (AOI; centered on Fairbanks and encompassing nearby DoD lands), the project team found that 9.7% of terrains (out of 453,248 km2 ) had experienced a significant increase in surface inundation during this period. Conversely, only 5.8% of Interior AOI terrains showed a significant decrease in summer inundation, including in the western portion of Fort Wainwright, over Fort Greely and Donnelly Training Area. In autumn, substantial increases in inundation were observed over ~ 18% of the Interior AOI, primarily in lowland regions including Fort Greely.

Vegetation State

The study showed multi-year increases in biomass water content (indicated by AMSR vegetation optical depth; VOD) and vegetation greenness (indicated by MODIS vegetation indices) over Alaska’s coastal regions, occurring within the 16 year (2002 to 2017) satellite observation period. This coastal greening and increase in VOD coincide with regional increases in precipitation and summer soil moisture, in addition to a lengthening of the annual non-frozen period, which are also reported in this study.

In the Alaska Interior AOI the project team instead observed a significant decrease in biomass water content (i.e., VOD), mostly within boreal forests. These trends in VOD primarily reflect a decrease in standing live forest biomass following major fire events that have occurred over 7.6 million acres (31,000 km2 ) in this region since 2002. The project team also observed a significant decrease in VOD in spring and summer for wetlands and grass/shrub habitat within this region, coinciding with areas of terrain wetting (which could indicate the encroachment of vegetation by open water), but the terrain area affected was minimal compared to the forested terrains.

Although a significant decline in VOD was observed across much of Interior Alaska when considering the 2003 to 2017 AMSR remote sensing record, the project team also observed a strong increase in VOD from 2014 through 2017 on DoD and surrounding terrains that coincided with biomass recovery. In contrast to VOD, the MODIS vegetation indices showed significant greening occurring within the Interior region. The most substantial greening was observed within fire scars, indicating the recovery of leaf canopy and understory foliage.

The results from the case study using BFAST time series decomposition to detect severe vegetation disturbance events (e.g., fire) confirms that this approach can detect breaks in remote sensing records, but also shows that the detected break point is highly dependent on what occurred in the landscape before and after a disturbance event. This portion of the investigation also highlighted the need for an on-the-ground terrain monitoring system on Interior Alaska DoD lands, coinciding with an ongoing remote sensing based terrain disturbance detection, to provide in situ observations needed to better understand what is driving the responses observed in satellite remote sensing records.

Downscaling (Data Fusion) of Remote Sensing Records

The data fusion component in this study focused on developing and testing methodologies for downscaling Earth System Indicator metrics based on spatially coarse 25 km passive microwave retrievals (e.g. AMSR; SSMI). The methods developed here can be readily applied in Google Earth Engine over the microwave data record (< 2002 to 2017) to produce daily < 500 m estimators of terrain state over Alaska and elsewhere on Earth. These long-term, downscaled records can be used to assist in identifying more localized and spatially precise patterns of nonstationarity through the application of Mann-Kendall and BFAST change detection to detect respective monotonic trends and abrupt disturbances in the time series.

For this study, the project team developed and evaluated machine learning algorithms for the downscaling of Freeze/Thaw, surface soil moisture, open water and vegetation. They also developed and improved machine learning methods to classify open water time series within Sentinel-1 Synthetic Aperture Radar (SAR) retrievals. The resulting downscaled products, provided for Interior Alaska in this one year SEED study, have greatly improved spatial resolutions of 500 m (Freeze/Thaw, soil moisture and vegetation) and 30 m (open water). Each downscaled product produced and evaluated in this study was able to capture spatial and temporal patterns of terrain change with a documented improvement over the original, coarse resolution microwave retrievals.

The relatively high accuracies (R2 > 0.80, compared to the original microwave data) achieved in Freeze/Thaw and vegetation products indicates that this project component is ready for the next phase, which is applying the developed downscaling algorithms to the full satellite time series (2002 to present) for the Interior Alaska domain to support a finer scale change detection analysis. The SAR results also showed favorable spatial accuracy for delineating water/land patterns at 30 m resolution and the potential of near-real time flood monitoring and risk mitigation.

The slightly reduced accuracy (R2 = 0.49) for surface soil moisture is still promising, especially given the ability of the product to represent finer spatial variability in surface wetness, but indicates a need for further algorithm refinement and reference with soil moisture data from multiple in situ monitoring stations before moving forward to regional applications. Next steps here should include additional efforts to identify which terrain characteristics correspond with better or worse performance in the downscaling approach, to allow for model improvements.

Back to Top


Identifying contemporary environmental change is a critical first step in mitigating terrain risk under a warming climate. Knowledge of the location and characteristics of change occurring in Earth System Indicators of Terrain Thermal State, Ecosystem Water Stress, and Vegetation State on and surrounding DoD lands is fundamental to inform the extent and magnitude of nonstationarity that is affecting U.S. ecosystems and to provide DoD managers with the information necessary to identify and mitigate for risk to existing and future infrastructure. The remote sensing-based geospatial change detection framework developed through this SEED project provides new tools and methodologies to better support decision making for conservation and training lands management, infrastructure development, and risk assessment across the Alaska DoD domain.

This study benefits the DoD through the design and testing of a remote sensing informed geospatial change analysis platform to identify the location, trajectory, and characteristics (e.g., change distribution shape) of ecosystem shifts occurring on DoD lands. The resulting Earth System Indicator change maps and terrain delineations can be readily incorporated into DoD supported geographic systems, including the Geographic Information Supporting Military Operations (GISMO) platform and, more broadly, to the Army Geospatial Center’s various platforms, the Army Installation Management Command’s cloud based tools, and the Army Installation Atlas. Though this study focuses on Alaska, this detection framework can be readily applied to other geographic locations including Russia, Canada and the conterminous United States (with appropriate modifications).

This study also identifies the need for a coordinated network of on-the-ground observations for soil temperature, soil moisture and above-ground vegetation water content to strategically verify patterns observed in the remote sensing retrievals and as a reference for new products being produced through VIS-NIR-TIR and microwave data fusion. Though this study provides proof-of-concept for the downscaling of microwave remote sensing data, and the detection of surface water change using Sentinel-1 radar, these methods can be strengthened for Interior Alaska by linking satellite data with on-the-ground references of terrain condition (e.g. time lapse camera retrievals, thermal and moisture sensor data, and microwave derived vegetation water content) strategically positioned on DoD and neighboring terrains.

Back to Top

Points of Contact

Principal Investigator

Jennifer Watts

The Woods Hole Research Center

Phone: 508-444-1526

Program Manager

Resource Conservation and Resiliency