Spatial and temporal trends of drought effects in a heterogeneous semi-arid forest ecosystem

Published Date
Forest Ecology and Management
1 April 2016, Vol.365:137–151, doi:10.1016/j.foreco.2016.01.017
Open Access, Creative Commons license

• Author 

• Timothy J. Assal ^a,b,,
• Patrick J. Anderson ^a
• Jason Sibold ^c

• aU.S. Geological Survey (USGS), Fort Collins Science Center, 2150 Centre Avenue, Fort Collins, CO 80526, USA
• bGraduate Degree Program in Ecology, Colorado State University, 1401 Campus Delivery, Fort Collins, CO 80523, USA
• cDepartment of Anthropology, Colorado State University, 1787 Campus Delivery, Fort Collins, CO 80523, USA

Received 24 September 2015. Revised 16 January 2016. Accepted 17 January 2016. Available online 1 February 2016. 

Highlights

• •
  Forest dynamics between 1985 and 2012 were modeled using vegetation indices.
• •
  Normalized Difference Moisture Index was the best indicator of forest conditions.
• •
  Linear trend analysis used to identify areas of forest vulnerable to drought.
• •
  25% of area had significant negative trend, compared to only 10% with positive trend.
• •
  Plots with negative trends had lower live tree density and more standing dead trees.

Abstract

Drought has long been recognized as a driving mechanism in the forests of western North America and drought-induced mortality has been documented across genera in recent years. Given the frequency of these events are expected to increase in the future, understanding patterns of mortality and plant response to severe drought is important to resource managers. Drought can affect the functional, physiological, structural, and demographic properties of forest ecosystems. Remote sensing studies have documented changes in forest properties due to direct and indirect effects of drought; however, few studies have addressed this at local scales needed to characterize highly heterogeneous ecosystems in the forest-shrubland ecotone. We analyzed a 22-year Landsat time series (1985–2012) to determine changes in forest in an area that experienced a relatively dry decade punctuated by two years of extreme drought. We assessed the relationship between several vegetation indices and field measured characteristics (e.g. plant area index and canopy gap fraction) and applied these indices to trend analysis to uncover the location, direction and timing of change. Finally, we assessed the interaction of climate and topography by forest functional type. The Normalized Difference Moisture Index (NDMI), a measure of canopy water content, had the strongest correlation with short-term field measures of plant area index (R² = 0.64) and canopy gap fraction (R² = 0.65). Over the entire time period, 25% of the forested area experienced a significant (p-value < 0.05) negative trend in NDMI, compared to less than 10% in a positive trend. Coniferous forests were more likely to be associated with a negative NDMI trend than deciduous forest. Forests on southern aspects were least likely to exhibit a negative trend while north aspects were most prevalent. Field plots with a negative trend had a lower live density, and higher amounts of standing dead and down trees compared to plots with no trend. Our analysis identifies spatially explicit patterns of long-term trends anchored with ground based evidence to highlight areas of forest that are resistant, persistent or vulnerable to severe drought. The results provide a long-term perspective for the resource management of this area and can be applied to similar ecosystems throughout western North America.

Keywords

Landsat time-series

Temporal trend analysis

Drought effects

Forest-shrubland ecotone

Rocky Mountain forests

---

1 Introduction

Climate shapes vegetation patterns through the balance between energy supply, moisture and the seasonal timing of the two (Stephenson, 1990). In this way, the climate of a region exerts top-down control on ecosystem pattern and process. Ecosystem disturbance, in particular large, infrequent disturbances (Turner and Dale, 1998), are also recognized as a key mechanism of landscape pattern in forests due to the enduring legacies of physical and biological structure that result from these events (Foster et al., 1998). However, disturbance also operates at less conspicuous scales and the range of disturbance impacts are best thought of along a continuum (Sousa, 1984), as legacies can persist at some level regardless of the size or frequency of the disturbance (Turner et al., 1998). Drought and desiccation stress are forms of ecosystem disturbance (Sousa, 1984), yet the spatial and temporal complexity of drought renders identification and quantification very difficult (Vicente-Serrano, 2007).

In the early 2000s, over half of the coterminous United States experienced moderate to severe drought conditions and record breaking precipitation deficits throughout the western part of the country (Cook et al., 2004). These events brought attention to drought vulnerability in semi-arid forests of western North America. Portions of the intermountain west also experienced severe to extreme drought in 2012 (NOAA, 2012). Severe drought in the early part of the last decade has been identified as the driver of tree stress, dieback and mortality across diverse forest types (Allen et al., 2010, Breshears et al., 2005, Gitlin et al., 2006 and Michaelian et al., 2011). These events also contribute to flammability of fuels and decreased snowpack, resulting in longer fire seasons (Littell et al., 2009 and Westerling et al., 2006).

The physiological drivers of tree mortality are complex (McDowell et al., 2008) and drought produces a gradient of effects on coniferous and deciduous forests in western North America. Drought can induce direct or indirect tree mortality, however, less conspicuous effects such as loss of productivity can accompany drought as well (Hogg et al., 2008). Forest response to drought is likely dependent on the spatial pattern of forest structure and function (Baguskas et al., 2014 and Hope et al., 2014), and the duration of the drought is a key element in plant response (Dorman et al., 2013). Water stress can lead to an increase in plant respiration (Jones and Vaughan, 2010), and plants cope with drought via stomatal closure and reduced leaf area index (LAI) (Hope et al., 2014). A reduction in leaf area leads to a lower photosynthetic capacity and a change in canopy structure. Collectively, these responses result in a decrease in chlorophyll and water content of plant leaves (Jones and Vaughan, 2010). Drought stress is often coupled with multiple, interacting factors (Allen et al., 2010), and lag effects of drought may lead to tree dieback and mortality several years after the drought event (Bigler et al., 2007).

Multiple studies have documented an increase in mortality rates of coniferous species throughout the western United States over the later part of the 20th century (Allen and Breshears, 1998, Breshears et al., 2005 and Van Mantgem et al., 2009). Increases in mortality rates have been reported across ecosystem type and elevation, among dominant genera and tree size, and at sites with diverse fire histories (Gitlin et al., 2006 and Van Mantgem et al., 2009). All of these mortality events were driven by increased water deficit associated with drought, but secondary agents such as bark beetle outbreaks have also contributed to conifer mortality in some areas (Bentz et al., 2010 and Van Mantgem et al., 2009). The dominant deciduous tree in western North American, quaking aspen (Populus tremuloides), may have an advantage over coniferous trees during periods of lower moisture due to its clonal root system. However, droughts of long durations are likely to affect the growth of both suckers and mature trees alike (Hessl and Graumlich, 2002). Severe drought in the boreal forest and parkland of western Canada resulted in a two-fold increase in stem mortality and a 30% decrease in regional stem growth in persistent trees (Hogg et al., 2008). Decrease in growth is the result of high levels of twig and branch dieback in the crowns of living trees and productivity is limited by carbon dioxide fixation imposed by leaf stomatal resistance during soil or atmospheric water deficits (Hogg et al., 2000). A phenomenon known as sudden aspen decline (SAD) has been documented in regional aspen forests (Worrall et al., 2008). Rapid and sudden onset of mortality is primarily caused by high temperatures, acute drought and secondary biotic agents (Worrall et al., 2008).

Disturbance alters ecosystem structure by both abrupt, obvious change and through gradual, slow change over some period of time (Assal et al., 2014). Remote sensing offers a powerful medium to capture the pre and post disturbance landscape and detect changes that might not be readily observed, such as drought stress. Spatial, temporal and spectral scales are an important consideration when using remote sensing in ecosystem disturbance studies. Two common multispectral remote sensing platforms used in drought studies are the Moderate Resolution Imaging Spectroradiometer (MODIS) (Abbas et al., 2014, Bastos et al., 2014 and Hope et al., 2014) and the Landsat satellites (Huang and Anderegg, 2012, Maselli, 2004, Vogelmann et al., 2009 and Volcani et al., 2005). Both platforms are well suited to study ecosystem dynamics at regional scales given the large coverage area per scene. However, subtle changes in forest structure and productivity are difficult to detect with satellite derived observations (Deshayes et al., 2006). Therefore, drought studies require a long-term series of observations, which makes the high temporal resolution of these satellites well suited for this application. Although MODIS has a high-temporal resolution (16-day composite product compared to 16-day revisit time for Landsat), the lower spatial resolution (250–500 m compared to 30 m) precludes its use in highly heterogeneous forest-shrubland ecotones. Trend analysis utilizing time-series of Landsat data is useful to identify, monitor, and assess both abrupt and subtle forest change (Czerwinski et al., 2014, Dorman et al., 2013, Kennedy et al., 2010 and Vogelmann et al., 2009).

Forest canopy reflectance is influenced by several biophysical parameters including crown closure, canopy and branch architecture, LAI, the chlorophyll and water content of leaves as well as the understory and exposed soil properties of the stand (Deshayes et al., 2006). Multispectral satellites have spectral bands spanning the visible and infrared wavelengths that can be combined into vegetation indices that are sensitive to differences in these biophysical parameters (Jones and Vaughan, 2010). Living vegetation absorbs radiation in portions of the visible wavelengths and reflects in the near-infrared (NIR); whereas radiation in the shortwave-infrared (SWIR) is absorbed by water content of leaves (Jones and Vaughan, 2010). The NIR and SWIR are sensitive to variations in LAI and the SWIR band is sensitive to water stress during periods of drought (Deshayes et al., 2006). Numerous spectral vegetation indices have been used in disturbance and drought studies, many of which utilize the NIR and/or the SWIR bands. The Normalized Difference Vegetation Index (NDVI) is the most widely used vegetation index to document and monitor drought and related impacts in forests (Breshears et al., 2005, Carreiras et al., 2006, DeRose et al., 2011, Lloret et al., 2007, Maselli, 2004, Volcani et al., 2005 and Weiss et al., 2004). However, other vegetation indices have utility in disturbance related vegetation dynamics including the Enhanced Vegetation Index (EVI) (Hope et al., 2014 and Tüshaus et al., 2014), the Normalized Difference Moisture Index (NDMI) (Goodwin et al., 2008 and Meddens et al., 2013), the soil adjusted vegetation index (SAVI) (Tüshaus et al., 2014), and the Tasseled Cap (Czerwinski et al., 2014).

We sought to quantify the spatial and temporal effects of drought in a semi-arid mixed forest ecosystem that is expected to be vulnerable to drought stress and climate change. The effects of climate change and variability are expected to be most rapid and extreme at ecotones, especially in semi-arid forests (Allen and Breshears, 1998and Gosz, 1992). An understanding of the link between climate variability and tree mortality for species near ecotones is an important focus of current research (Kulakowski et al., 2013). Ecotones are important barometers of climate change (NEON, 2000) and stress, dieback and mortality are expected to accompany severe drought in this arid landscape. Recent studies (Crookston et al., 2010 and Rehfeldt et al., 2009) predict the current climate profile for several prominent tree species (e.g. aspen, subalpine fir and lodgepole pine) will be greatly limited or no longer present in isolated forests of the Rocky Mountains over the course of the next century. However, regional climate can be influenced by local topography, a concept known as topoclimate (Thornthwaite, 1953). Slope, aspect and other topographic features influence air temperature, water balance, radiation, snowmelt patterns and wind exposure (Dobrowski, 2011) which amplify the effects of drought, particularly in arid landscapes.

The use of temporal remotely sensed data has been effective in monitoring drought induced changes in forests and woodlands (Maselli, 2004; Vogelmann et al., 2012). A primary challenge in spectral change analysis is to segregate long-term vegetation change from interannual phenology differences in response to climate variability. We hypothesize that the ecological consequences of drought create a landscape mosaic of drought effects and that trend analysis of vegetation indices can be used to document the distribution and magnitude of a gradient of effects (Lloret et al., 2007). The gradient of drought effects include demographic (i.e. tree mortality), structural (i.e. crown partial dieback), functional (i.e. reduction in leaf area), and physiological (i.e. temporary/permanent reduction in photosynthetic activity) properties of the forest ecosystem that result in different spectral trajectories. This study was undertaken because little is known about baseline condition in the forest, and how climate, in particular drought, affects this topographically complex ecosystem. Finally, we are interested in providing resource managers with a long-term perspective of the forest dynamics of this ecosystem with respect to variability in precipitation patterns.

Our research objectives were to:

• (1)
  Identify an appropriate vegetation index for use in temporal trend analysis based on the relationship with field measured estimates of vegetation traits,
• (2)
  Analyze the spatial location, trend direction, and timing of spectral change by forest type, and
• (3)
  Determine the correlative relationship between spectral change and tree mortality, and assess this relationship with topographic features.

2 Methods

2.1 Study area

Our study area is located in the southern part of the Wyoming Basin ecoregion, spanning parts of southwestern Wyoming, northwestern Colorado and northeastern Utah (Fig. 1). Several prominent ridges form a transition zone between basins and mountainous areas (Knight, 1994), where several species of trees are found at the xeric fringes of their respective ranges. Forests dominated by aspen (Populus tremuloides) and several coniferous species (subalpine fir (Abies lasiocarpa), lodgepole pine (Pinus contorta), and Douglas-fir (Pseudotsuga menziesii)) occur as relatively small patches on moist sites in a matrix of sagebrush steppe (Artemisia tridentata spp. vaseyana) or mixed-species shrublands. Scattered juniper (Juniperus communis var. depressa) and limber pine (Pinus flexilis) woodlands, distinct from the montane conifer forest, are found on rocky slopes at lower elevations and small patches of manzanita (Arctostaphylos patula) are found in the southern part of the study area. Most of the area is under the jurisdiction of the U.S. Bureau of Land Management, interspersed with small parcels of State and private land. The area has a midlatitude steppe climate with a substantial portion of the annual precipitation occurring as snow. Dominant land uses include livestock grazing, energy extraction and recreation. Multiple State and Federal agencies, along with the Wyoming Landscape Conservation Initiative (wlci.gov), have identified this region as a priority area for conservation. The area provides important habitat for many wildlife species including big game, migratory and resident birds, as well as domestic livestock. Management has sought to rejuvenate decadent aspen stands and reduce conifer expansion in successional aspen stands through prescribed fire and mechanical thinning. Drought related mortality of aspen is a concern in western North America (Worrall et al., 2008) and lack of aspen regeneration due to high rates of herbivory is a concern in the study area.

Fig. 1. Location and extent of forest and field plots in the study area.

2.2 Drought index calculation

To understand the effects of drought at a local level, we needed to develop a localized index of drought severity. The most widely used drought index, the standardized precipitation index (SPI) (Vicente-Serrano, 2007), was used to quantify surface water deficit and surplus. We used the SPI instead of the regional Palmer Drought Severity Index to capture local differences in annual precipitation of the relatively small and isolated study area. The SPI indicates the number of standard deviations that precipitation deviates from the long-term mean during the measured period (Vicente-Serrano, 2007). The majority of the precipitation in the study area is received in the form of snow. The flexibility of the SPI enabled us to calculate a 12-month duration equivalent to the water year (Oct–Sept) between 1975 and 2014. We obtained 4 km² SPI data from the Western Regional Climate Center (http://www.wrcc.dri.edu; accessed 15 February 2015) and calculated the water year mean for a 50 km² area that encompasses the forested ridges of the study area.

2.3 Satellite data

The growing season in the study area is short, with snow possible in mid-June and leaf senescence by late September. To select a consistent window of peak greenness, we calculated the average growing season phenology (2000–2012) of the forested portion of the study area using MODIS NDVI 16-day composite data (MOD13Q1). Predominantly cloud-free Landsat images were selected between July 01 and September 06 (day of year 182–249 in non-leap years). A total of 24 Landsat Thematic Mapper (TM), Enhanced Thematic Mapper Plus (ETM+), and Operational Land Imager (OLI) images (path 36, row 32) were acquired for analysis from the USGS EarthExplorer Archive (USGS, 2014) between 1985 and 2014 (Table 1). The imagery was processed to surface reflectance using the Landsat Ecosystem Disturbance Adaptive Processing System (LEDAPS) (Masek et al., 2006) which has been successfully used in other ecosystem change studies (McManus et al., 2012). We used the LEDAPS quality mask layer to identify pixels with clouds, cloud shadows and other unacceptable pixels that were removed from the analysis. We calculated several vegetation indices (Table 2) using the Landsat 8 OLI imagery to explore relationships between each vegetation index and the field data. We then applied the index that best explained variation in the short-term field data (see Section 3.3) to the Landsat time series.

Table 1. Acquisition dates of Landsat scenes used in the analysis (path 36, row 32). Note, the OLI image was used to establish the relationship between field measured data and vegetation indices and was not used in the trend analysis. TM = Thematic Mapper, ETM+ = Enhanced Thematic Mapper Plus, and OLI = Operational Land Imager.

For further details log on website :
http://www.sciencedirect.com/science/article/pii/S0378112716000189