Wildlife

Mexican Spotted Owl Habitat Map

The U.S. Forest Service and the U.S. Fish & Wildlife Service are jointly committed to ensuring the conservation and recovery of the Mexican spotted owl (MSO) and have begun taking several actions to meet that goal. Scientists from Rocky Mountain Research Station (USFS) along with experts from both agencies convened to produce an estimation of habitat trend spanning the last ~35 years across all Forests within Southwestern Region (Arizona and New Mexico).

From this effort, annual maps of MSO forest habitat across the Southwestern Region from 1986-2020 have been developed using Landsat imagery, climate data, topography, and a large number of known MSO nest/roost locations. The primary purpose of these maps is to identify broad-scale and long-term trends in potential MSO forest habitat. This “living map” product will continually be refined with new data and improved over time.

The living map uses actual locations of nesting and roosting owls to identify MSO forest habitat, which shows areas where owls are most likely to establish nesting territories and exhibit roosting behaviors as a function of forest vegetation, topography, climate, and their complex interactions. The living map also identifies MSO cover type, which shows forest vegetation that is similar to the type owls are known to nest and roost in (or not), but does not account for how topography and climate may constrain nesting or roosting activities. Forest vegetation is an important component of habitat that is affected by land management. The living map product is available in a Google Earth Engine web app.

While the impetus for this map product was to evaluate long-term habitat trends through time, it may also be utilized in practical applications of land use planning. However, the map is not intended to be used as a stand-alone tool for prescriptive planning but may be used as a starting point to guide project level habitat evaluation. It is recommended to be used in combination with other site-specific data and local knowledge to identify “Recovery Nest/Roost Habitat” on the landscape. For example, there may be areas within a forest or district where the map over- or under-predicts habitat; we encourage users to view the current map products as a starting point and incorporate other information when possible to meet their needs. In addition, the map may include areas that are unlikely to actually provide ‘Recovery Nest/Roost Habitat’ as defined in the 2012 Recovery Plan; the map shows areas that may be more expansive than actual on-the-ground nest/roost habitat because the map displays probable areas where this habitat could occur.

Additional detailed description of the product, guidance, and limitations are included in the following pages.

Detailed Description

Three factors of Mexican Spotted Owl habitats: topography, vegetation, and climate.
Figure 1. Topography, vegetation, and climate are the three critical factors of Mexican Spotted Owl habitats.

Overview: In April 2020, USDA Forest Service Region 3 staff convened a group of scientists and managers to form a Mexican Spotted Owl (MSO) habitat trend monitoring working group. The purpose of this working group was to use the latest tools, technology, and data to develop reliable, fine-scale annual maps of MSO nesting/roosting habitat across Region 3 (Arizona and New Mexico) from 1986-present, and use these maps to assess trends in habitat through time.

Data: We used over 2,900 MSO nest/roost locations distributed across Region 3 to create the maps. These data spanned the years 1989-2020 and were obtained from a variety of sources including past demographic studies, species recovery planning efforts, USDA Forest Service project-level surveys and protected activity center (PAC) monitoring, and several observations from non-FS ownerships (Los Alamos National Labs, Bandelier National Monument, and The Nature Conservancy lands). These data were combined with remotely sensed data sources (vegetation, topography, and climate) using machine-learning algorithms to predict MSO habitat across the region.

Products: The working group has produced two parallel, complementary products: the MSO forest habitat product and the MSO cover type product. The forest habitat product was developed using Landsat multi-spectral (i.e., vegetation) data, topography, and climate, and accounted for variation in these factors across different MSO Ecological Management Units (EMUs). The habitat product therefore displays “realized” MSO habitat, accounting for the ways habitat is constrained or amplified by topographical features and climate in the arid Southwest. The forest cover type product was developed using Landsat multi-spectral (i.e., vegetation) data as remotely sensed input variables, and displays forest vegetation along a gradient of “spectral similarity” (from not similar to very similar) to forested areas known to be used by MSO for nesting and roosting. Forest cover (i.e. vegetation) is only one part of what constitutes “habitat” (see Figure 1); but it is the part over which forest managers have the most control.

MSO Forest Habitat Product

The ‘forest habitat’ product used the Random Forests classifier to develop predictive relationships between n = 2913 Mexican spotted owl nest/roost locations and Landsat multi-spectral imagery, topographical (5), and climate (9) variables. Imagery was pre-processed through a continuous change detection and classification (CCDC) algorithm to temporally stabilize it from 1985 to current. This helped to ensure vegetation changes were due to actual forest disturbance or growth processes and not due to varying image quality from year to year. Different predictive models were developed for each of the five Ecological Management Units (EMUs) in Region 3. ‘Available’ locations for pseudo-absences were generated within a forest cover mask. The model performed well as measured by the mean area under the receiver operating curve test statistic (AUC = 0.97-0.99 across EMUs) and out-of-bag (OOB) error was minimal (3-6% across EMUs).

MSO Cover Type Product

The ‘cover type’ product used the Maxent classifier to develop predictive relationships between a subset of MSO nest/roost locations (n = 2,233) and Landsat multi-spectral imagery. Imagery was pre-processed through a continuous change detection and classification (CCDC) algorithm to temporally stabilize it from 1985 to current. This helped to ensure cover type changes were due to actual forest disturbance or growth processes and not due to varying image quality from year to year. Stationarity was assumed across the modeling region; a single global model was fitted to nest/roost locations using a random bootstrapped sample of 50% to train the model and 50% to test the prediction of each model replicate (10 replicates) compared against a random sampling of available forest pixels (n = 10,000). The model performed well as measured by the mean area under the receiver operating curve test statistic (AUC = 0.939 ± 0.001) and the mean Spearman rank of the predicted vs. expected (P/E) ratio curve (Rs = 0.996 ± 0.002).

The ‘living map’ concept: These above products can be easily updated on an annual basis into the future to ensure managers and biologists are working from the most up-to-date habitat information for project planning and other purposes. Annual updates to map products will include to (1) changes to the forest landscape (e.g., fires, insects, or other drivers of habitat change), (2) refinements and improvements to the analytical approaches used, and (3) the incorporation of new MSO nest/roost data in regions where data are currently limited (as those data become available). We anticipate that the forest cover type and habitat maps will change and improve with time, particularly with updates related to (2) and (3) above.

2 by 2 diagram of Mexican Spotted Owl nesting habitat compared to vegetation coverage.
Figure 2. Forest coverage and Mexican Spotted Owl habitat maps do not always agree. There are four possible categories resulting from comparing these maps: (1) forest cover that is not owl habitat; (2) forested owl habitat; (3) territory that is not suitable habitat; and (4) owl habitat poorly predicted by vegetation.

Intended uses, limitations, and caveats: The primary purpose of these maps is to identify broad-scale and long-term trends in potential MSO forest habitat. The maps are not intended to be used as a stand-alone tool for prescriptive planning, but may be used as a starting point to guide ground-based validation and could be beneficially used in combination with other products and local knowledge to identify recovery nest/roost habitat on the landscape. Maps are fine-resolution (30-m) and have relatively high accuracy, but significant map prediction errors can still occur at fine scales. Users must take note of the points outlined below regarding map interpretation that should help guide their appropriate application, and limit erroneous application:

  1. Forest cover type and habitat maps will not always match up with MSO Predicted Forest Recovery Habitat maps. The forest habitat and forest cover type maps described in this briefing paper were developed using different methods, different data, and at a different scale than the MSO Predicted Forest Recovery Habitat maps. It is therefore an expectation that these maps may not align with one another because they are representing different things.
  2. Forest cover type and habitat maps will not always match one another (see Figure 2). For example, in some areas, the forest cover type product may show more extensive forests that are ‘spectrally similar’ to nesting/roosting areas than does the habitat product (upper left corner of Figure 2). However, because of their topographic position or climatic profile, these areas may not be particularly suitable for MSO nesting or roosting activities, at least under current conditions. Developing and comparing the two products highlights the relative role of forest vegetation, topography, and climate in shaping MSO nesting and roosting habitat. Vegetation explains a good amount of variation in the habitat models, but so do topography and climate. We encourage users to explore both products in combination with site-specific data.
  3. Forest cover type and habitat products will not always align with local knowledge and data. For example, there may be areas within a forest or district where habitat is clearly over- or under-predicted. In these cases, we encourage you to contact the persons listed above for feedback and to provide new MSO nest/roost location data, so that the mapping team can incorporate this information into the next version update. For this reason, the forest cover type and habitat products can and should be used in combination with other spatial products and local information in any type of planning activity. As stated above, maps are not intended to be used as a stand-alone tool for prescriptive planning, but may be used as a starting point to guide ground-based validation and could be beneficially used in combination with other products and local knowledge to identify recovery nest/roost habitat on the landscape.
  4. These maps (and associated trends) of forests cover type and habitat are subject to change. As Southwestern landscapes continue to change for a variety of reasons (e.g., fire, drought, land use), as analytical methods are refined, and as new MSO nest/roost data are included, these maps may change retroactively. Updated products will be assigned a new version number. And as the version number changes, the annual maps of MSO habitat distribution will change, and so will estimated habitat trends. We do not anticipate major changes to the map products; however, continual learning about MSO habitat is anticipated, which means these products will be naturally dynamic.
  5. “Drivers of change” have not yet been ascertained. The scope of the current initial effort (April-December 2020) was to determine (1) the spatial distribution of MSO forest cover type and habitat and (2) trends in forest cover type and habitat through time. Efforts are now beginning to attribute changes in forest cover type and habitat to natural and anthropogenic disturbances (e.g., wildfire, drought mortality, disease, logging, etc.).

Next steps: There are at least four major next steps to be taken. First, RMRS scientists, regional analysts, and other partners intend to undertake a “drivers of change” analysis to understand what factors are responsible for changes to MSO cover type and habitat over the period 1986-2020. Second, RMRS scientists, regional analysts, and other partners intend to complete analyses linking MSO forest cover type and habitat to on-the-ground forest structure conditions from FIA plot data. This will help translate the remotely sensed products into information that can be used by forest managers. Third, RMRS scientists, regional analysts, and other partners intend to automate some of the analytical steps we conducted ‘manually’ in this first iteration (e.g., a pipeline to integrate new MSO nest/roost data contributed by biologists). This automation will help fulfill the vision of the ‘living map’. Fourth, RMRS scientists, regional analysts, and other partners are working on producing several peer-reviewed publications based on findings from this research-management partnership





https://www.fs.usda.gov/detailfull/r3/plants-animals/wildlife?cid=FSEPRD890979&width=full