Vernal pools are frequently considered a type of so-called “geographically isolated wetland”, temporary or semi-permanent pools typically defined as occurring in a confined basin depression without a permanently flowing outlet [1
]. These systems can become inundated pools in the fall, winter, or spring, and frequently dry completely in the summer [3
]. Vernal pools are also known as ephemeral ponded wetlands, temporal ponds, or seasonal woodland ponds and are common in the glaciated portions of North America. Normally free of reproducing fish populations, vernal pools and vernal pool complexes are important breeding habitat for numerous amphibians and reptiles adapted to reproduction in temporary and fishless habitats. However, due in part to their small size and short hydroperiods, as well as the relative lack of resilience to hydrologic alterations to the surrounding landscape, vernal pool wetlands are at risk of destruction across the northeastern United States [2
Some geopolitical entities of the United States have developed guidance and regulations for vernal pool protection, including New Jersey [4
], Maine [8
], and Massachusetts [9
], as well as smaller municipalities such as towns of Farmington, Simsbury, and Suffield in Connecticut [11
]. Massachusetts provides a useful model for vernal pool protection. The protection of vernal pools in Massachusetts involves a certification process in which biological data are collected by citizens to demonstrate that a wetland provides vernal pool functions. The Massachusetts Wetland Protection Act does not specifically protect vernal pools unless they are certified by a field verification process to ascertain that they serve as habitat for obligate or facultative vernal pool amphibian species [5
]. Once certified, regulatory restrictions are placed on the development and other activities affecting a pool [13
]. However, the protection of vernal pools and other small, geographically isolated wetlands suffers from a lack of effective methods for identifying these features in the landscape. Current methods for identifying potential vernal pools include visual interpretation and digitization on aerial photographs and subsequent field verification [4
], as well as statistical modeling approaches to map or predict vernal pool locations [5
]. Several studies have been conducted in the northeastern United States to map potential vernal pools on aerial photographs, yet their accuracy has been highly variable. The reported commission errors ranged from 3% to 90%, as the visual interpretation process was largely dependent on the skill of the photo interpreters, imagery types, imagery acquisition dates, minimum mapping unit size, extent and type of forest cover, and topography [14
Recent advances in remote sensing technology hold great potential for enhanced local, regional and national wetland mapping and inventory. High-resolution light detection and ranging (LiDAR) data can facilitate the detection of wetlands that are normally difficult to identify, such as vernal pools. A number of studies have demonstrated the promise of LiDAR for enhanced wetland mapping and inventory [20
]; however, few studies have specifically used LiDAR data for identifying small, geographically isolated wetlands, such as vernal pools. In this paper, we propose an efficient and effective method for identifying potential vernal pools with high-resolution LiDAR data and near-infrared aerial photographs. We accepted the vernal pool definition of Calhoun et al.
] and Carpenter et al.
], namely that a vernal pool is a geospatially isolated or depressional basin with no permanent inlet or outlet and is typically not connected through surface water to or immediately adjacent to other water bodies (see Section 2.5
below). Vernal pool habitats generally have a water regime of seasonally to semi-permanently inundated, meaning that they hold water for at least two continuous months (though we note that vernal pools do, on occasion, stay inundated year-round [3
]). To standardize the types of habitats that were designated as vernal pools across our study area, we defined vernal pools as confined surface depressions with no permanent surface inflow or outflow with a total surface area larger than 50 m2
(the minimum size that could be confidently identified using our datasets), which is much smaller than the reported minimum mapping unit (200–250 m2
) that has been reliably identified with leaf-off color-infrared (CIR) aerial photographs in previous studies on vernal pools [4
]. A stochastic depression analysis method [25
] using a Monte Carlo approach was developed to extract surface depressions from a 1-m resolution LiDAR digital elevation model (DEM). By applying a Monte Carlo approach, we estimated the likelihood of a topographic depression actually occurring in the landscape, given the degree of uncertainty in the LiDAR DEM. With this methodology, actual surface depressions can be distinguished from artifact depressions. These derived actual surface depressions can then be further refined using ancillary data, such as depression area, the National Hydrography Dataset (NHD) [26
], land use and land cover type, and the Normalized Difference Water Index (NDWI) [27
], to identify potential vernal pools. Successful application of this approach to other portions of the glaciated northeastern North America, using data and methods developed from the localities of Norton and Attleboro in eastern Massachusetts, would suggest that potential vernal pools can be detected efficiently and objectively with high accuracy and low commission and omission errors, thereby increasing our understanding of the extent of these important ecosystems.
Vernal pools are important landscape elements that contribute to amphibian metapopulation dynamics [49
], the maintenance of threatened and endangered plant and animal species [52
], water storage [53
], and local biogeochemical reactions [55
]. However, wetlands are also typically small and shallow systems that are easily overlooked and frequently unprotected by local, state, and federal regulations [57
]; thus, they are often disturbed and destroyed [58
]. Understanding the location of these systems is frequently the first step towards better understanding of their functions, connections to other waters and systems [60
], and potential impacts of human alterations on the maintenance of system integrity.
In the past, aerial photography was the best available data for discerning vernal pools, but with the advent of readily available LIDAR data and increasingly robust geostatistical processes [5
], as well as increasing availability of repeated satellite imagery (though not applied in this study; see [61
]), we have an improved opportunity to discern extant vernal pools of formerly glaciated terrain. Our approach was successful in identifying over 1800 additional vernal pools in the study area, which (assuming their veracity as extant systems) would almost triple the number of potential vernal pools previously identified in the study area (when coupled with the work by Burne et al.
]). As shown in Table 5
, 1832 depressions in Class 6 were identified as putative vernal pools with our method, but not related to any vernal pool in the statewide vernal pool databases. The 1832 putative vernal pools mapped by our stochastic depression analysis method were in part related to differences in minimum mapping units. Most authors concluded that only pools greater than 200–250 m2
circular area (0.020–0.025 ha) could be reliably identified with leaf-off CIR aerial photographs [4
]. As pointed out by Burne [10
] in his statewide potential vernal pool database, only pools of 125 feet (38.1 meter) in diameter and larger on the photographs could be reliably identified when photos were of fair to excellent quality, and where evergreen trees were not dominant. The New Jersey vernal pool database developed by Lathrop et al.
] used the minimum detectable pool size of 200 m2
. Only the pool centroid point location was digitized on-screen based on the CIR aerial photographs. As stated in the NHESP’s vernal pool certification guidelines [13
], the NHESP does not establish a physical, on-the-ground vernal pool boundary during the certification process. The minimum mapping unit for our depression analysis method was 50 m2
. Of the 1832 depressions detected in Class 6914 (50.0%) were within the area from 50 m2
to 250 m2
. This means that our method was able to detect small vernal pools (< 250 m2
) that could not be reliably identified in previously studies. This also indicated that photographic interpretation may considerably underestimate the number of vernal pools.
The 59 (2.5%) false positive identifications of vernal pools in Class 9 could have resulted in part from the confusion between tree shadows and water bodies, due to their similar spectral signatures derived from the CIR aerial photographs. As apparent from an example of Class 9 systems (see Figure 10
), the depression had tree shadows in it, which appeared as dark features on the CIR aerial orthoimagery. This resulted in positive NDWI values in the depression that caused it to be falsely identified as a vernal pool.
The acquisition date differences between LIDAR, CIR aerial photographs, and land use data may have also contributed to the false identification of vernal pools. The LIDAR data and CIR aerial photographs used in our study were acquired in 2010 and 2013, respectively. The land use data were derived from the four-band orthoimagery acquired in 2005. For the purposes of our study we assumed that depressional vernal pools occur infrequently on developed land [5
], and the land use data in 2005 was used to refine the depressions. In this case, land use and land cover change after 2005 would not be reflected in the 2005 land use data. Human development and land use practices, including new residential or commercial development since 2005, could have resulted in removal of some pools.
The extraction of vernal pool boundaries from remotely sensed imagery is highly dependent on the water level at the time of image acquisition. Although the CIR aerial photographs used in our study were acquired in generally leaf-off conditions during mid- to late April 2013, it should be noted that not all potential vernal pools were flooded or holding water during the imagery acquisition time. Some pools might already have dried out in the early spring or were just beginning to form in the early summer. The estimated statewide average precipitation in April 2013, was only 4.93 cm, which was 51 percent of the long-term average for the month (9.65 cm). According to the National Drought Mitigation Center’s 30 April 2013, Drought Monitor Map, 70% percent of Massachusetts was in abnormally dry conditions [28
]. As a result, many small intermittent pools that normally stay saturated during the spring seasons were dry. In other words, our method might not have captured all the vernal pools solely based on the dry conditions represented in the aerial photographs.
In contrasting our results with that of the statewide CVP/PVP databases, we determined that approximately 52% of the 913 CVPs + PVPs within the study area occurred within discrete depressions. However, 23% of the CVPs + PVPs were not associated with depressions, but were visually confirmed based on the 2013 CIR orthoimagery (see Table 4
) and were therefore considered as errors of omission in our study. By inspecting these omitted pools and their associated LIDAR DEM, NHD, and depression layers, these omissions could be subdivided into two types. The first type were depressions associated with omitted pools that were not detected with our stochastic depression analysis in the first step (see Section 3.1
). These could be considered true errors of omission and may have been due to the RMSE error for the LIDAR data (0.095 m) being greater than the average depth of the vernal pool system. We found that 152 vernal pools belonged to this case, which accounted for 71% of the undetected vernal pools. The second type are omissions which emanate from analytical processes in which these omitted pools were initially detected using the stochastic depression analysis method, then eliminated because as they did not meet our refining criteria (see sections 3.2 and 3.3). In particular, the adjacency criteria of > 10 m from NHD features seemed to account for most of these eliminations. Through our stepwise process, we identified and removed 32 vernal pools that were within 10-m buffer zones of NHD features, which could have accounted for up to 15% of our errors of omission. We note that Lang et al.
] found that the horizontal accuracy of NHD maps was > 18 m, which is substantially greater than the 10-m buffer distance of NHD flowlines and areas used by Reif et al.
] and Lane et al.
] and applied in this study. However, Lane et al.
] doubled the buffer width to 20 m in a 8600 km2
subsample of their study extent and reported an areal change in GIWs of approximately 3%, suggesting that the horizontal accuracy error in the NHD may play a small (though not insignificant) role in the outcome of the current study. Further refinement of our data required an abundance of water pixels to be identified within each depression. If this criterion was not met, the depression was rejected as a potential vernal pool.
Coupling our depression analyses with soil wetness features, along with other ancillary GIS data sets and algorithms, such as performed by Grant (2005) [5
], could result in increased precision in the number of CVPs and PVPs identified. However, the majority of the vernal pools in our study were located within depressions (or within 15 m/5 m of a depression), suggesting that focusing on depression systems can improve the precision of the studies and better discern extant vernal pools on the changing landscape.