Soil erosion of agricultural lands is a major contemporary global environmental problem. Arable soils provide important ecosystem services, with approximately 15 million km2
used for crop production [1
]. The demand for increases in farm land and crop yields is expected to rise with an estimated World population of 9.7 billion by the year 2050 [2
]. Maintaining a healthy and productive soil is a critical issue for sustaining food production and other agricultural activities when meeting society’s needs [1
In the Midwestern U.S., the 1930s are remembered as a period of extreme soil loss when drought exposed top soil, resulting in high winds carrying huge volumes eastward, and depositing it into the Atlantic Ocean [3
]. Dust storms experienced during the 1930s have been considered the worst human-driven environmental problem that the U.S. has faced [4
]. Since then, changes in agricultural practices, such as the implementation of conservation tillage and the use of shelterbelts, have reduced the potential for wind erosion [5
]. Nevertheless, recent estimates of wind erosion in U.S. cropland average 4.3 Mg/ha, while soil renewal rates are less than 1 Mg/ha. These data suggest that policymakers and land managers should extend and strengthen the mitigation efforts to reduce wind erosion and preserve soil productivity [6
Shelterbelts—sometimes-called windbreaks, hedgerows, or fencerows—have been used to mitigate wind erosion since the 1450s [7
]. In the 1930s, the U.S. President Franklin Roosevelt established the Prairie States Forestry Project, with the goal of planting shelterbelts from North Dakota to Texas [7
]. In addition to the reduced rates of soil erosion by wind, shelterbelts can maintain soil moisture which benefits crop yields, even outweighing the loss of acreages used for planting the shelterbelt [9
]. Shelterbelts also improve crop water usage during drought periods by reducing evaporation rates; they reduce wind-chill impacts on livestock during winter, and improve livestock health; and reduce stress and damages on people and properties by providing protection from high winds [10
]. Shelterbelts maintenance is important for preserving healthy trees, otherwise landowners must decide to renovate or remove tree’s during their later life-stages [11
]. In the U.S., many shelterbelts have been planted since the 1930s [8
], species composition and tree density have experienced significant changes over the years. Baltensperger [12
] measured the linear distances of shelterbelts in Iowa and Kansas and found that, in both states between the 1880s and the 1970s, shelterbelts decreased from 1600 km to 72 km in Iowa, and from 3000 km to 1100 km in Kansas. Schaefer, Dronen, and Erickson [11
] surveyed 2875 shelterbelts in South Dakota and found that only 1150 were in a healthy condition, having no need for renovation.
Grand Forks County, North Dakota, once had a very high concentration of shelterbelts, and planting tree shelterbelts was a popular practice among the local communities to protect the soil (Figure 1
]. However, this point of view may have changed, recent interviews with land managers suggest that that the rate of shelterbelt removal in this region has increased [13
]. Many shelterbelts have reached their life expectancy, and are in need of maintenance [15
]. Agricultural producers are choosing to remove the trees without replanting, to alleviate the burden of maintenance costs [13
]. This raises concerns that wind erosion could increase, especially during dry climate conditions. Soil erosion can occur at wind speeds above 6.2 m/s, which have been recorded in the county [19
]. Understanding how shelterbelts have changed, and what may have driven these changes, in this region could provide scientific evidence and determine if extra policy measures are needed to cope with such changes.
To investigate this spatial shift, we adopted remote sensing techniques for data acquisition, and used Geographic Information Systems (GIS) to analyze spatial data. Shelterbelts can be digitized using aerial imagery over multiple years, and changes in shelterbelt densities can be measured [20
]. While several studies have used either aerial or satellite imagery to classify shelterbelts, e.g., [21
], none have attempted to detect change in shelterbelt densities over time. The ability to detect changing shelterbelt densities can provide insight into changes in landowner decision-making and support future policymaking on mitigating wind erosion.
Using manual digitization and geographic object-based image analysis, we measured shelterbelt densities in GFC with multiple years of high-resolution aerial imagery. The result of density change between years was validated with an error of 4.4%. This amount of error is considered acceptable when compared with 4–8% errors produced by other studies for a single year classification [21
Our results show a doubling of shelterbelt densities from 1962 to 2014, with an increase of 6402 m2
over the 52 years (123 m2
/year; Figure 5
). This large growth in shelterbelt density was surprising since reports before 2014 found that shelterbelt removal was becoming common practice [13
]. These data suggest that shelterbelt density was much higher prior to our 2014 measurement and has started declining recently. Although the change is subjected to measurement errors, from 2014 to 2016, we detected 1,040,178 m2
(2.4%) of shelterbelt area removed from the county, creating a density loss of −157 m2
Depending on individual species and their growth, it is possible that between 2014 and 2016 some tree planting was not detectable in the NAIP imagery, since small tree sapling may not be visible from remotely sensed imagery. However, other studies have documented that the number of tree planting since 2010 has declined to less than 20% the annual amounts planted in the 1990s [20
], which has resulted in a 70% drop in tree sales in North Dakota between 2002 and 2013 [16
]. Therefore, a more reasonable explanation for the lower density is that less planting has occurred.
The use of shelterbelt areas as a measurement for examining shelterbelt change has a few challenges. In Section 2.4
we addressed one of these challenges by removing changes in shelterbelt areas caused by increases or decreases in the tree canopy. Some studies have used linear lengths of shelterbelts as a metric for measurement [11
], however these measurements were done in the field, or manually digitized with a much smaller sample. Using GEOBIA allows for the automation of shelterbelt feature extraction from imagery over large areas. The produced GEOBIA shelterbelts are polygons instead of lines and are measured in terms of area [21
]. We further conducted a statistical analysis and validated that shelterbelt area (m2
) can serve as a close proxy of shelterbelt length (m) for measuring temporal changes (Figure A1
). Our results also show that average shelterbelt length was consistent across all three years ranging from 436.2 m to 450.9 m (Table A1
An interesting discovery from our shelterbelt classification was the spatial clustering of shelterbelts within GFC. Since most of the county is agricultural land, we assumed that shelterbelts would be evenly distributed throughout the region. Using soil pH with the Bivariate Local Moran’s I (Figure 6
) we found local regions of GFC that have significant spatial autocorrelation (p
≤ 0.05) with shelterbelt density. Much of the eastern side of the county, making up between 18.7–21.1% of sections, contains regions of high soil alkalinity and low shelterbelt density. These regions are likely not well suited for tree growth and shelterbelt establishment [38
]. The majority of sections with a high shelterbelt density consisted in areas of low alkalinity, between 9.9–10.6% of sections, though regions of high density also exist in areas with alkaline soils, making up around 6.8–8.0% of sections. The results from analyzing soil pH data indicates that parts of eastern GFC would be more difficult for establishing shelterbelts or would require tolerant species.
Our analysis of surface geology (Figure 7
) in GFC proved to have a consistent relationship with shelterbelt density. The regions of sand, cross-bedded sand, clay, and till in the county occur in a band running directly parallel with the locations of majority of shelterbelts. This is not a coincidence since this north-south band of clay and sandy soils are likely highly erodible soils made up of the beach ridges formed by Glacial Lake Agassiz [29
]. Results from the ANOVA tests showed a significant relationship (p
≤ 0.05) between the highest shelterbelt density and the sand surface geology. This relationship was consistent across all three years, with cross-bedded sand also falling into the highest shelterbelt density category for 1962 only. This suggests that agricultural operators likely plant shelterbelts in sandy soil due to its vulnerability to wind erosion [43
]. The lowest shelterbelt densities across all three years of data (p
≤ 0.05) occur in the silty soils found along the eastern side of GFC. These soils are likely more resistant to wind erosion than the sandy soils [43
], and many of them are also alkaline soils which may not be suitable for shelterbelt establishment, particularly if alkalinity levels are high enough to hinder the growth of tolerant species.
Perhaps the most important factor in both understanding the placement of shelterbelts and their density dynamics is human-decision making, as Schaefer, Dronen, and Erickson [11
] found in their examination of shelterbelts in Kansas. Shelterbelts that were well maintained, especially during their first few years of growth, were more likely to remain in a healthy condition longer, and those that were not cared for, were likely to need replacement or removal in later stages of life. We examined the ownership of shelterbelts in GFC (Figure 8
), and found a heavy skew towards a few agricultural operators owning the highest densities of shelterbelts. In 2014, 55% of operators did not have a shelterbelt, while 40% of landowners did not have one. Of the remaining percentages that own a shelterbelt, the majority had a density less than 15,000 m2
, while only 1.3% of operators had densities above 100,000 m2
. This skewed distribution between both operators and landowners suggests that decision-making for the planting, maintenance, and removal of shelterbelts is likely influenced by personal preference and knowledge, as well as the physical condition of the soils owned.
In 1985 the Conservation Reserve Program (CRP) was initiated by the USDA FSA, which allowed land owners to voluntarily place land into conservation efforts over a 10–15 year period, and receive financial assistance through contract [44
]. Under CRP, shelterbelts can fall into two categories: field windbreak (CP5) and shelterbelts (CP16), both serving the purpose of wind erosion control for crops, livestock, homesteads, and snow accumulation [45
]. Data available from the USDA (Figure 9
a) shows a steady increase in the hectares of shelterbelts enrolled in CRP contracts in North Dakota from the early 2000s until ~2010–2012. After 2012, shelterbelt enrollment declined across the state. This statewide summary aligns well with our detected shelterbelt change in GFC, showing an increase in shelterbelts up 2010, and then a sudden shift to a decrease. Recent changes in agricultural and energy policies to increase the use of biofuels created a greater demand for corn and soybeans in the Midwestern region of U.S. [46
]. Following a major agricultural drought event in 2012, the prices of corn and soybeans have doubled in the region (Figure 9
b). The surge in crop prices resulted in less interests in CRP enrollments, and in North Dakota an estimated 89,000 ha of grassland was converted to corn, and 220,000 ha of grassland was converted to soybeans [46
]. The loss of CRP contracts during the high commodity prices in 2012 has been reflected by the change in the CRP shelterbelts (Figure 9
a), suggesting that trees were removed to increase the amount of available land for corn and soybeans.
Congruently with the increase in crop prices, a shift in agricultural practice can also partially explain shelterbelt removal in GFC. In North Dakota, between 1982 and 2007, the midpoint acreage of farm sizes more than doubled [47
]. In GFC, the average farm operation sizes increased from 32,784 ha in 1997 to 59,255 ha by 2012 [31
]. This increase in farm size created the need for larger, more efficient machinery [47
]. As agricultural machinery has gotten larger, it is more difficult to move in and out of fields, or to turn equipment at the edge of the fields when trees are in the way. Mature trees can get in the way of operation when branches or entire trees fall onto the field. In addition, herbicides in the U.S. were applied to 5–10% of cropland in the 1950s, and this increased to 90–99% by the 1980s [48
]. During herbicide application, particularly with aerial sprays, herbicides could land on and damage the adjacent shelterbelts. The issues related to farming scale as well as management practices likely will influence landowner’s decision-making related to shelterbelts’ planting, maintenance, and removal [13
], and ultimately has an effect on wind erosion in the future. On the other hand, outreach and information sharing with landowners and agricultural operators through extension services and government agencies could offer future opportunities in promoting shelterbelts planting as a conservation practices to mitigate soil erosion. In particular, financial support from government agencies can provide incentives to agricultural operators for shelterbelts planting, maintenance, and renovation. In 2015, the North Dakota Outdoor Heritage Fund awarded the North Dakota Forest Service $
1.8 million to renovate shelterbelts in the state [18
]. Continued monitoring of shelterbelts at a statewide level is needed to evaluate if conservation efforts like this can generate a substantial and long-term impact.
The aim of this study was to detect shelterbelt density change through time and determine if recent claims of tree removal should be of concern. We were able to use high-resolution aerial photography to both classify and detect change in shelterbelt densities within GFC. From 1962 to 2014—a total of 52 years—we found that shelterbelt densities doubled, and this provides great praise for the shelterbelt planting efforts that have taken place in the county. However, our more recent change investigation from 2014 to 2016 suggests that the rate of shelterbelt planting in GFC has slowed, and more removal is taking place. While our change detection between 2014 and 2016 could have resulted from the 3% measured error in detecting shelterbelt removal (Table 1
), the decline in CRP enrollment from 2012 to 2018 (Figure 9
a) agrees with this trend. The decline in trees sales in North Dakota [42
] also agrees with our finding that tree planting has declined. This raises concern for the potential risk of wind erosion in the future. If the increased rate of shelterbelt removal does not slow, or is not offset by an increase in planting, we could see increasing amounts of top soil loss during high wind events, particularly when the soil is exposed during crop planting and harvest.
The methods we developed for this study can be applied to measure shelterbelt densities in other parts of the Midwestern U.S. wherever NAIP imagery is available. Unfortunately, while the spectral, spatial, and relational object properties chosen for the 2014 and 2016 NAIP imagery can be used as a framework for automating the extraction of shelterbelt polygons from NAIP imagery (Figure 3
), the exact values cannot be used directly since pixel values are not standardized. It would be beneficial to improve the GEOBIA process by developing a system that automatically adjusts threshold values for various imagery. This would simplify sharing and widespread use of the developed GEOBIA classification systems.
From a policy perspective, our findings indicate knowledge gaps in understanding changing agricultural practices related to shelterbelts. Our results show that individual landowner’s and farm operator’s preferences likely have a great influence on the number of shelterbelts planted in any particular field. To identify further the driving factors of decision-making on shelterbelts planting, a survey or other method of outreach will need to be incorporated. It is important to understand why agricultural producers that had shelterbelts in the past are no longer choosing to replace them. If changes in the farming scale contribute to the recent shelterbelt removal, policymakers should provide additional guidelines and incentives to balance the tradeoffs between soil erosion and agricultural intensification. Improvements in our understanding of changing agricultural technology will also be helpful to sustain the efforts in preserving shelterbelts and mitigating wind erosion, to ultimately prevent future conditions like those seen during the 1930s.