Historical Loss of Native Old-Growth Grasslands on the San Juan Islands, Washington

Abstract

The San Juan Islands are one of the few places where native temperate grasslands are found in western Washington State. These ecosystems are important reservoirs of biodiversity and sources of ecosystem services, support many rare and endemic species, and have profound cultural significance to the Coast Salish peoples. These ecologically and culturally valuable ecosystems have become scarce due to the combined pressures of changes in land use, the introduction of non-native invasive species, and the exclusion of fire from the landscape. A lack of historical context and ecological baseline knowledge has made it impossible to fully understand the long-term trends in the extent and distribution of this ecosystem. To address this knowledge gap, we used historical land cover data and multispectral imagery to create a high-resolution, spatially explicit database of grassland extent on the San Juan Islands at multiple time periods since the early years of Euro-American colonization. Our spatial analysis of these data revealed significant decreases in grassland extent between time periods, with an overall 77% net decrease in the extent of non-agricultural grasslands and a loss of 93% of the area of persistent, old-growth grasslands since 1890 across the region. These changes are primarily a result of conversion to agriculture and conifer encroachment or succession to forest. The spatial data and analyses created in this study help to develop the historical baseline of native temperate grasslands on the San Juan Islands, adding to our understanding of the lingering legacy that changes in land use have had on this ecosystem, with the potential to aid in the development of effective conservation and restoration practices.

1. Introduction

Native temperate grasslands are among the most imperiled ecosystems across the globe. Grasslands provide a wealth of ecosystem services, from carbon sequestration and soil stability to flood mitigation and critical habitat, but only 4.6% of the planet’s temperate grasslands are within established protected areas [1,2,3]. Faced with the combined pressures of conversion to agriculture, urbanization, exclusion of fire from the landscape, over-grazing, increases in the presence of invasive species, encroachment by woody vegetation, and emerging threats posed by climate change, we are losing these ecosystems at a rate of eight times greater than the rate at which we are protecting them [1,2]. Evidence of this ecosystem crisis is clear in the Pacific Northwest, where studies show that native temperate grasslands throughout the region have been greatly diminished—in some cases, such as in the southern Puget Sound, to a mere 10% of their historical extent [3,4].

One area of western Washington State where fragmented remnants of old-growth grasslands can still be found is the San Juan Islands in the central Salish Sea (Figure 1). These islands experience a strong rain shadow effect from mountains on the Olympic Peninsula and Vancouver Island, which creates the relatively arid mesoclimate conditions that support grassland ecosystems. Grassland ecosystems have been a key aspect of the San Juan Islands’ ecological identity for centuries [5]. The oral traditions of Coast Salish peoples, in conjunction with studies of the fire history of the San Juans, indicate that many of the grassland ecosystems of the San Juan Islands bear the marks of thousands of years of management by Coast Salish peoples, including the Lummi, Samish, Swinomish, Saanich, and Lekwungen [6,7]. Many grasslands were traditionally carefully managed and maintained by the Coast Salish peoples using annual, low-intensity burns to support the sustainable production of bulb and root crops, such as common camas (Camassia quamash) [7,8,9]. These land management practices played a significant role in the development and maintenance of grasslands on the San Juan Islands. Grasslands in this region can thus be considered part of the cultural, as well as ecological, landscape and represent a socio-ecological system, rather than a purely non-anthropogenic “natural” ecosystem.

Native grasslands provide a wide range of ecosystem services from flood mitigation and erosion control to providing pollinator habitat and climate stabilization [10]. Furthermore, these ecosystems boast high levels of biodiversity, supporting many species that are not found in other ecosystems [11]. Grasslands cover only a small portion of the San Juan Islands, yet they are a key component of the overall biodiversity and ecological diversity of this region. Given this, as well as their cultural significance to Coast Salish peoples, it is not surprising that there has been a rise in native grassland restoration and preservation projects including targeted species reintroduction efforts, invasive species removal, and the reintroduction of fire to the landscape. However, despite the surge in initiatives to protect native grasslands, management efforts are hindered by a lack of knowledge regarding historical ecological baseline conditions and the previous extent of grasslands on the islands [12,13].

We do not have a clear picture of precisely how much native temperate grassland area has been lost on the San Juan Islands, when that loss occurred, and what the leading causes of grassland land cover loss are. Specifically, there is a distinct lack of accessible spatial data that shows how these ecosystems have changed since the arrival of Euro-American settlers in the mid-1800s and the concurrent disruption of the Coast Salish systems of land management that had maintained temperate grasslands for thousands of years [6,14]. This lack of historical reference data limits our ability to understand the lingering legacies of historical land use changes across the islands, as well as the leading causes of temperate grassland ecosystem loss [14,15]. In turn, this knowledge gap poses a challenge to conservation and restoration initiatives targeting grasslands in the San Juan Islands. Defining a historical baseline of the extent and distribution of native grasslands in the San Juan Islands is critical to fully understanding the scope of land cover change in the Pacific Northwest.

We aimed to address this knowledge gap by using Geographic Information Science (GIS) techniques, historical maps, and aerial imagery to create spatial datasets describing the locations of grassland ecosystems on the San Juan Islands over the past 133 years. We then used these datasets to determine the historical extent and distribution of grassland ecosystems at the start of widespread Euro-American colonization (approximately 1890), and how these grasslands changed in terms of patch size, total area, and level of fragmentation. We also examined which land cover types replaced grassland ecosystems during the study period to better understand land cover change trajectories.

2. Materials and Methods

We developed spatial datasets delineating the locations and extent of grassland ecosystems on the San Juan Islands using historical maps for the late 1800s, historical black and white aerial photographs for the early 1900s, and multiband aerial imagery for the early 2000s. We used ESRI’s ArcGIS Pro 2.9 for all analysis.

2.1. Creating Historical Land Cover Datasets: 1890

To delineate 1890s grassland extent, we digitized historical hand-drawn maps known as “T-sheets”, which were created by the United States Coastal Survey (later renamed to the United States Coast and Geodetic Survey) in the late 1800s (1852–1926). The maps that cover the extent of the San Juan Islands were completed during 1888–1897. Each island in the study area was mapped at a scale of 1:10,000. The maps include 20 ft contour lines, the total length of shoreline and roads for each island, the area of the island in square miles, as well as buildings, fence lines, and at least eight different categories of land cover represented by unique patterns: forest, sparse or slashed forest, grassland, cultivated field, orchard, marsh, wooded marsh, and bluff/beach (

Table 1. Land cover symbology and descriptions from the “T-Sheet” historical maps used in our analysis.

Land Cover—1890s
Forest	Sparse Forest	Grassy Forest	Grassland	Cultivated Field
Forested land cover	Landscapes with sparse tree/canopy cover, or recently cleared/logged areas	Landscapes with sparse tree cover and grassland intermixed. Includes Garry-Oak savannahs and woodlands	Includes prairies, meadows, and fenced pastures	Farmland—includes cropland and homesteads
Orchard	Marsh	Wooded Marsh	Beach/Bluff
Farmland in production of tree fruits	Includes marshes and other wetlands	Marshes and wetlands with some tree cover	Beaches, rocky bluffs, prominent sandbars, rocky protrusions

). In some areas, multiple land cover types appear to be intermixed, likely representing ecotones or more nuanced ecosystems such as oak-savannah grasslands. Although the original T-Sheets did not include a legend, we relied on the most commonly used and widely accepted legend for these documents, which was developed in 2000 by Tom Schroeder based on a US Coast and Geodetic Survey Report that includes an appendix on topographic symbols used by the organization at that time [16].

We downloaded scanned versions of the T-Sheets from the Puget Sound River History Project’s (PSRHP) data portal [17]. These image layers were originally georeferenced by PSRHP researchers using the NAD 27 UTM Zone 10 North Projected Coordinate System (PCS). We re-projected the T-Sheets into the NAD 83 Washington State Plane North (US Feet) PCS for consistency with our other primary and ancillary spatial datasets.

We used manual digitization at a scale of 1:1500 to create geospatial datasets consisting of polygons, lines, and points representing different land cover categories (Table 1) based on visual assessment of the land cover patterns in the historical maps. To improve the efficiency of the digitization process, we used the Washington State Department of Natural Resources ShoreZone Inventory Program’s coastline dataset as a guide for the location of each island’s coastline and digitized land cover types within the outline of each island derived from the ShoreZone data [18]. While the coastline has changed in some places due to erosion or deposition, such changes would have occurred primarily in the “marsh” and “beach/bluff” land cover classes and therefore the grassland areas that are our focus were far enough inland that these processes are unlikely to have led to any significant change grassland area. We applied map topology rules to the digitization process, ensuring that newly created features would be aligned without gaps or overlaps between digitized polygons. For subsequent analyses, we used the Grassland and Grassy Forest land cover categories to represent grassland extent.

2.2. Creating Historical Land Cover Datasets-1932

In 1932, the Canadian military conducted an aerial survey of the San Juan Islands, capturing a set of approximately 500 black and white photographs of these islands. These photographs provide a detailed source of information about land cover, land use, and development on the islands at the start of the 1930s. Through a partnership between San Juan County’s Public Works department and the Washington Department of Natural Resources, the photos have been scanned, georeferenced and mosaiced into a single imagery layer that is hosted as a basemap on the county’s online GIS interface, Polaris, as well as on ArcGIS Online.

We used the georeferenced and mosaiced aerial imagery basemap hosted by San Juan County to carefully interpret and manually digitize land cover classes at a scale of 1:1500. We digitized ten different categories (

Table 2. Land cover categories derived from 1932 aerial imagery for our analysis, including descriptions and example images.

Land Cover—1932
Forest	Sparse Forest	Grassy Forest	Grassland	Cultivated Field
Forested landscapes	Landscapes with sparse tree/canopy cover	Landscapes with intermixed grassland and sparse tree cover. Includes Garry-Oak savannahs/ woodlands	Includes prairies, meadows, balds, and fenced pastures	Farmland—includes cropland, mowed/sown pastures, and homesteads
Orchard	Marsh/Wooded Marsh	Beach/Bluff	Cleared Land	Developed
Orchards or berry fields	Marshes, wooded marshes, and other wetlands	Beaches, rocky bluffs, prominent sandbars, rocky protrusions	Areas that were forested in 1890, but are cleared of trees in 1932	Areas with a high concentration of buildings and other infrastructure

) based on extensive visual assessment of the aerial imagery and consideration of seven visual cues: shape, size, pattern, tone, texture, shadow, and context. We conducted our digitization in the NAD 1983 StatePlane Washington North FIPS 4601 (US Feet) because that is the PCS the 1932 aerial imagery basemap was already available in.

There were some instances in which distinguishing between grassland and farmland was exceptionally challenging. While comparing the land cover conditions in 1932 to those in 1890 would provide clarification in instances in which previously forested areas were cleared, there were many cases in which former grasslands were converted to agriculture or pasture. In these situations, it was necessary to thoroughly consider the characteristics of the patch, as well as the surrounding region (Figure 2,

Table 3. Visual cues used to assist in distinguishing grassland from farmland in the 1932 aerial imagery used in our analysis.

Visual Cue	Indicator of Grassland	Indicator of Farmland/Agriculture
Shape	Area boundary is irregular, with a gradual transition to other land cover types	Area has sharp, clear borders
Context	Area is unfenced	Area is fenced
Context	No nearby structures	Nearby structures such as dwellings or barns
Pattern	No regular patterns	Repeating patterns of ripples or grooves (such as furrows from a tractor, plow, or from harvesting)
Texture/Tone	Texture/tone is irregular, suggesting presence of several plant species growing at different paces	Texture/tone of area is fairly homogeneous (suggesting dominance of a singular plant species)

).

2.3. Creating a Contemporary Land Cover Dataset Using Multi-Spectral Imagery

We acquired aerial imagery with a spatial resolution of 60 × 60 cm through the United States Department of Agriculture (USDA) National Aerial Imagery Program (NAIP) [19]. This imagery includes four spectral bands: red, blue, green, and near-infrared (NIR). We imported the multispectral imagery into ArcGIS Pro 2.9, clipped it to the extent of the study area, and re-projected it into the NAD 1983 State Plane Washington North FIPS 4601 (US Feet) PCS. We created a classification schema that closely matched the land cover classification categories used for the 1932 aerial imagery, with the addition of a “Deciduous Forest” category (this was added after an initial classification attempt failed to accurately distinguish between deciduous forests and wetlands, indicating a need for more targeted training samples).

To minimize bias in sample size during the image classification process, we created an equal number of similarly sized training samples for each land cover group. Samples were evenly dispersed across each of the four islands in the study groups (with the exception of Shaw Island, which received fewer samples than San Juan, Orcas, and Lopez islands due its smaller size). Samples were placed with consideration of known land cover locations, meaning that only areas where land cover was definitively known were used as training sample sites.

Following precedent established by United States Department of Fish and Wildlife (USDFW) in the creation of a High-Resolution Land Cover Data layer using 4-band NAIP imagery, we conducted classification on a 7-band imagery stack instead of the single 4-band NAIP raster [20]. The stack consisted of the original 4-band imagery and three indices generated from the original raster: Normalized Difference Vegetation Index (NDVI), Soil-Adjusted Vegetation Index (SAVI), and a variance raster to emphasize texture, generated from the NIR band. This imagery stack was then segmented and evaluated for classification using the Support Vector Model Object-Based Image Analysis (SVM OBIA) method, to minimize noise and account for any unintended variations in training sample sizes between classes. We reviewed the resulting land cover raster and qualitatively assessed the overall accuracy by conducting a visual comparison of concurrence between our classification results and other land cover datasets, including layers representing wetlands [21], areas zoned for agricultural use in the Salish Sea bioregion [22], oak woodlands and grasslands of the Puget Trough ecoregion [23], and forested lands derived from a 2016 land cover raster [24]. Areas classified as potential grasslands were compared with multispectral NAIP aerial imagery from 2021, Google Earth imagery, and data layers representing wetlands [21], parks/protected areas and agriculture areas [25], the locations of oaks and grasslands [23], and zoning/tax parcel data within San Juan County [22] to determine if the polygon represented grassland, farmland, or lawns and other vegetation. Where these ancillary datasets indicated misclassification, we adjusted the original output from our land cover classification accordingly.

2.4. Calculation of Change in Grassland Ecosystems

To quantify changes in the area, fragmentation, and isolation of non-agricultural grassland ecosystems on the San Juan Islands since 1890, we conducted statistical analyses of our land cover datasets in ArcGIS Pro 2.9 and Excel. We used the combined area of the Grassland and Grassy Forest land cover categories in each time period to represent non-agricultural grassland extent. We assessed changes in land cover by calculating the area of grasslands that were replaced by other land cover types between time periods. We calculated changes in grassland area over time using the mean, minimum, and maximum area of grassland patches on each island. We defined fragmentation as a function of the ratio between grassland edge (patch perimeter) to core habitat (patch area). We calculated the edge:core ratio for each grassland patch during each time period and summarized those results by island. We created a distance accumulation raster to calculate Euclidean distance between grassland patches to represent changes in isolation.

3. Results

Between 1890 and 1932, 27.9 square kilometers of grassland were lost (

Table 4. Total grassland loss and gain reported as area (square kilometers) by time period.

	1890s Grassland Area	1890s to 1932		1932 to 2021		1890s to 2021
		Area Lost	Area Gained	Area Lost	Area Gained	Area Lost	Area Gained
San Juan	23.8	13.7	5.2	12.3	2.5	26.1	4.9
Orcas	9.4	7.4	4.7	5.8	1.3	13.2	7.7
Lopez	7.0	6.0	1.9	2.5	1.0	8.5	7.7
Shaw	1.1	0.8	0.4	0.6	0.2	1.5	0.7
Total	41.5	27.9	12.1	21.2	5.0	49.2	17.1

; Figure 3). Only 33% of the square kilometers of grassland patches present in the 1890s were still grassland in 1932. In terms of net changes in the total area of grassland land cover, the loss was partially offset by a gain of 12.1 square kilometers of grassland. Overall, these changes equate to approximately a net 38% decrease in the total area of grassland land cover across the San Juan Islands (

Table 5. Net grassland loss reported as area (square kilometers) and as percent of grassland existing in the previous time period.

	1890s to 1932		1932 to 2021		1890s to 2021
	Net Area Lost	Net Percent Lost	Net Area Lost	Net Percent Lost	Net Area Lost	Net Percent Lost
San Juan	8.5	36%	9.8	64%	18.4	77%
Orcas	2.7	28%	4.5	66%	7.1	75%
Lopez	4.2	59%	1.5	53%	5.7	81%
Shaw	0.4	39%	0.4	63%	0.9	77%
Total	15.8	38%	16.2	63%	32.1	77%

), with the biggest net decreases occurring on Lopez Island (59%) and Shaw Island (39%).

Between 1932 and 2021, there was a further loss of 21.2 square kilometers of grassland. Of the total area of grassland patches present in 1932, only 17% were still grassland by 2021. Over this time period, there was also a slight gain of 5.0 square kilometers, equating to a net loss of 63% of the remaining grassland land cover. The largest net losses during this period occurred on Orcas Island (66%) and San Juan Island (64%).

Overall, the San Juan Islands experienced a net loss of 77% of the area of grassland land cover between 1890 and 2021. Of the individual square kilometers of grassland present in the 1890s, only 33% remained grassland by 1932, and only 7.3% were still grassland in 2021. This represents a total loss of at least 92.7% of the area of old-growth, persistent grassland ecosystems on the San Juan Islands.

In addition to a dramatic reduction in total grassland area from the 1890s to 2021, our results also revealed an increase in the fragmentation of remaining grassland patches (

Table 6. Changes in grassland fragmentation during study period (1890–2021).

	1890s	2021
Average grassland habitat patch size (square kilometers)	0.052	0.078
Average distance between grassland habitat patches (meters)	403.976	193.825
Average ratio of perimeter to core area in grassland habitat patches	0.03	0.67

). Compared to the 1890s, by 2021 the remaining individual grassland habitat patches were, on average, only 10% as large and 49% as far apart, and their edge:core ratio was 22 times higher.

Grasslands covered 10.64% of land on the four largest islands in the San Juan Island archipelago in the 1890s, but only 2.41% by 2021 (Figure 4). From the 1890s to 1932, 47% of the grassland area initially present was replaced with farmland (including cultivated fields and orchards) and 18% was replaced by forest and sparse forest (Figure 5). Conversion to agriculture was the most prevalent reason for loss of grasslands on all islands, with the highest rate on Lopez Island, where 70% of the initial grassland area was converted to agriculture by 1932.

After 1932, conversion to forest replaced agricultural conversion as the leading cause of grassland loss. From 1932 to 2021, 55% of the grassland area present in 1932 across the San Juan Islands was converted to forest and sparse forests. Conversion to forested land cover was responsible for over half of all grassland loss on every island. Conversion to agriculture was less widespread than it had been before 1932, but it still accounted for a loss of 11% of the 1932 grassland area by 2021. Conversion of grasslands to lawns, turf, and cleared or developed land was responsible for the loss of another 13% of the 1932 grassland area.

4. Discussion

The results of our land cover analysis indicate that grassland ecosystems have changed dramatically since 1890, decreasing by a net 77% from their extent at that time and leaving only 7.3% of the area of persistent, old-growth grassland patches that were present in the 1890s. A net total of at least 32.06 square kilometers of grassland was lost during this time period. Grassland was primarily replaced by farmland and forest, with farmland being the primary cause of loss between 1890–1932, and afforestation being the biggest cause of loss from 1932–2021 (Appendix A, Figure A1, Figure A2 and Figure A3). These changing land cover trajectories reflect historical changes in the prevalence of agricultural land use in the San Juan Islands [9].

While agriculture was the most significant driver in grassland loss prior to 1932, it was replaced by afforestation as the leading cause of grassland loss between 1932–2021. This can be explained by the decline in agriculture that occurred throughout the 1900s [9], in combination with a continuation of fire exclusion policies. Reaching its peak in 1920, the number of farms and total acreage of farmland on the San Juan Islands declined steadily through the 1970s [9]. With this decrease in agricultural activity, formerly cultivated fields fell out of production. Combined with fire exclusion through both active fire suppression and banning of traditional cultural burning practices, this allowed for increased rates of encroachment from Douglas-fir (Pseudotsuga menziesii) and other coniferous trees, shrub species such as Nootka rose (Rosa nutkana) and common snowberry (Symphoricarpos albus), and non-native invasive forbs such as Scot’s broom (Cytisus scoparius) and Himalayan blackberry [11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28].

The grasslands that have remained between each of the study years show signs of increasing fragmentation, as shown by the increase in edge:core ratio. The increasing ratio indicates an increase in the amount of edge habitat per unit of area over time [26]. Another way to reveal the increasing fragmentation of grasslands is to examine changes in mean distance between grassland patches, as well as mean patch size. Although the mean distance between individual grasslands has decreased since 1890, so has total grassland area. This is due to a “splintering” effect; as grasslands are broken into increasingly smaller and less resilient patches, the distance between grasslands decreases along with the average patch size.

While there was a net loss of grassland in each time period we analyzed, there were also significant areas of grassland “gained”. 71% of the new grassland area was gained between the 1890s and 1932 (Appendix A, Figure A4 and Figure A5). These new areas of grassland gained by 1932 were primarily areas of forest clearance. Additional, generally smaller, areas of grassland were “gained” simply as artifacts of the data sources. The 1890s grasslands extent was based on relatively coarser resolution, more generalized land cover polygons derived from the hand-drawn T-Sheets and may therefore have had inaccuracies in the exact location of the edges of grassland patches. The 1932 data were derived from higher resolution aerial photographs that offered more comprehensive coverage of the islands compared to the T-sheets, which were in some instances based on surveys conducted offshore and could have resulted in an over-simplification of landcover or even misclassification in some cases [17]. Grassland area gained between 1932 and 2021 generally occurred in smaller, isolated patches compared to the gains before 1932. The largest gain of grassland after 1932 was created by the conversion of land previously used for agriculture back to grassland on the southern edge of San Juan Island as part of the San Juan Island National Historical Park’s ecosystem restoration plans.

The combined pressures of agriculture, fire exclusion, forest encroachment, and human development have severely diminished and altered the grassland ecosystems of the San Juan Islands. For example, the dry-prairie community Festuca roemeri–Camassia quamash–Cerastium arvense is so fragmented that it is considered functionally extirpated in Washington State [29]. Such reductions in viable grassland habitat disproportionately affect native species with high habitat fidelity; as a result, these species are experiencing regional declines and, in some cases, extirpation [30]. A natural resource condition assessment of San Juan National Historical Park conducted in 2020 found that, “in the absence of significant and aggressive restoration actions, the current trend in the areal extent of actual native prairie is likely to be a gradual loss” [29]. Given the ominous trajectory of loss and fragmentation, it seems clear that substantial conservation and restoration efforts must be implemented to preserve remaining grasslands and restore those that are degraded or altered.

Our findings can provide site specific insight into to how grassland ecosystems have been affected by the changes in land use that we know to have occurred since 1890. Such information can be used to understand the historical legacy of changing land use priorities and improve our understanding of the ecological traits, tendencies, or even constraints that may be present, therefore informing restoration practices. In addition to providing historical reference, these datasets can be used to enhance our understanding of the human history of the San Juan Islands and how post-settlement practices impacted both grassland ecosystems and Coast Salish traditional ways of life. As local governments, non-profits, national parks, and other land management groups begin (and hopefully continue) to recognize the grasslands of the San Juan Islands as a cultural landscape and the role that traditional land stewardship practices had in shaping these ecosystems, spatial data that highlights where these ecosystems were and when they were lost will be critical in ecosystem recovery, as well as the revival of traditional practices such as regular burning.

While grasslands occurring within National or State Park boundaries and other protected areas benefit from regular monitoring and restoration programs that include invasive species removal, mowing, and planting of native species, this only accounts for 46% of current grasslands. The remaining 54% do not receive the same degree of protection, placing them at a higher risk of further degradation and succession to forest, after which restoration becomes extremely difficult and expensive [11,31]. Active monitoring of these grassland sites is needed to assess their condition and vulnerability and is a precursor to the development of appropriate and effective conservation strategies. Collaborating with landowners to develop conservation strategies and to protect grassland remnants that occur on private property is essential to ensuring the resilience and longevity of old-growth grassland ecosystems on the San Juan Islands.

Expanding services to increase outreach and education to private landowners with grassland remnants on their property could enhance the effectiveness of grassland conservation and restoration efforts. Therefore, it is our recommendation that grassland conservation efforts on the San Juan Islands should include a thorough evaluation of high priority grassland sites that have been present since at least 1890; contain contiguous patches of grassland, as opposed to minute fragments; and primarily occur on unprotected land (Appendix A, Figure A6), followed by the implementation of a conservation or restoration plan suited to the needs of the individual site.

This study makes an important contribution to our understanding of the historical baseline of grassland ecosystems on the San Juan Islands, but further research is clearly still needed. Future research focused on refining imagery classification results, evaluating current grassland conditions, or extending the known historical baseline further back in time could improve the efficacy of restoration initiatives. Future studies could refine current land cover datasets by employing more advanced imagery analysis software and additional spectral bands to more accurately distinguish native grassland ecosystems from other non-forested ecosystems, such as farmland. A mixed methods approach that combines imagery analysis with remote sensing techniques, followed by rigorous ground truthing to determine the composition of current grasslands, evaluate the presence of invasive species, and make note of potential stressors or threats to grassland health and stability, would likely be the most effective way to create a highly accurate and comprehensive contemporary grassland land cover dataset.

While this study tracks the changes in extent and distribution that occurred prior to the height of the Euro-American settlement period, many changes had undoubtedly already been wrought before 1890. Additionally, even where grassland ecosystems have persisted, they have been altered by the introduction of non-native species and by changes to climatic conditions and disturbance regimes resulting from human activities. Therefore, the results of this study are inevitably an underestimation of how much grasslands have decreased from their historical extent on the San Juan Islands. Future research could incorporate paleoecological records derived from soil samples, sediment cores, and tree-ring records to add additional context about the historical extent of grasslands. By collaborating with the Coast Salish tribes that have lived in the San Juan Islands since time immemorial, and incorporating Traditional Ecological Knowledge, oral histories, and participatory mapping, it could be possible to reveal additional changes that had already occurred by 1890.

5. Conclusions

We used historical maps, aerial photographs, and multispectral imagery to map and quantify changes in the extent of grasslands on the San Juan Islands in Washington State. This paper contributes to the body of work on reconstructing historical socio-ecological landscapes and tracing the changing land use priorities that have led to rapid ecological change since Euro-American settlement of coastal Washington. Extensive changes in land cover and land use have occurred on the San Juan Islands since Euro-American settlement in the 19th century. These changes include a dramatic reduction in the extent and distribution of native temperate grasslands. Due primarily to changes in land use priorities and fire exclusion policies, native temperate grasslands on the San Juan Islands have experienced a net decrease to less than 23% of their historical extent in 1890. Only 7% of the original grasslands present in the 1890s remain grasslands today. This represents a loss of 93% of the area of old-growth grassland ecosystems present in the 1890s. As alarming as this statistic is, it underestimates the full extent of grassland ecosystem loss that has occurred since settlement of the San Juan Islands because some conversion of grasslands to agriculture had already occurred by 1890, and because the grasslands that do remain have been altered by the introduction of non-native species and by changes to climate conditions and disturbance regimes that have resulted from recent anthropogenic activities. Therefore, this study represents a conservative estimate of the full extent of grassland loss following Euro-American settlement on the San Juan Islands of Washington State.