RIVER RESEARCH AND APPLICATIONS

River Res. Applic. (2015) Published online in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/rra.2903

PREDICTING FLOODPLAIN HYPOXIA IN THE ATCHAFALAYA RIVER, LOUISIANA, USA, A LARGE, REGULATED SOUTHERN FLOODPLAIN RIVER SYSTEM T. E. PASCOa*, M. D. KALLERa, R. HARLANa, W. E. KELSOa, D. A. RUTHERFORDa AND S. ROBERTSb a

School of Renewable Natural Resources, Louisiana State University Agricultural Center, Baton Rouge, LA, USA b New Orleans District, US Army Corps of Engineers, New Orleans, LA, USA

ABSTRACT The Atchafalaya River Basin Floodway (ARBF), a regulated river/floodplain distributary of the Mississippi River, experiences an annual flood pulse that strongly influences floodplain physicochemistry. We developed several metrics to investigate the relationship between the timing and magnitude of the flood pulse and floodplain hypoxia, which in most years is a spatially extensive and temporally prolonged problem in the lower ARBF. Principal components analysis of flood metrics from 2001 to 2009 revealed contrasting flood types (early cool and late warm), but component-based general linear models were unable to predict the magnitude of hypoxia in ARBF water management areas (WMAs). Further analyses based on temperature and geographic information system-determined WMA inundation with generalized additive models (GAMs) revealed WMA-specific patterns of hypoxia, but the likelihood of hypoxia consistently increased when temperatures approached 20°C and inundation rose above 20–30%. Validation with held-out data based on logistic regression indicated that the models constructed with the 2001–2009 temperature and inundation data were able to accurately predict the probabilities of hypoxia in two WMAs based on data collected from 2010 to 2013. The GAMs were an effective tool for visualizing and predicting the probability of hypoxia based on two easily generated parameters. Our analyses indicate that modification of the Atchafalaya River flood pulse could reduce the magnitude of hypoxia within the lower ARBF, subject to engineering (control structure operation) and economic (commercial fisheries production) constraints, by minimizing floodplain inundation after water temperatures reach 20°C. Copyright © 2015 John Wiley & Sons, Ltd. key words: hypoxia; Atchafalaya River; flood pulse; general additive model; floodplain Received 4 December 2014; Revised 3 March 2015; Accepted 11 March 2015

INTRODUCTION Numerous studies conducted over the last 50 years attest to the tremendous influence of the annual flood pulse on the physicochemistry and biotic structure of large river systems (Bayley, 1991; Tockner et al., 1999) and the effects of variations in the timing, frequency, duration, and magnitude of the flood on riverine and floodplain processes (Poff et al., 1997). Anthropogenic alterations to the structural and functional integrity of the world’s large rivers from dams, levees, and tributary/distributary modifications have altered flood pulse characteristics, often at the expense of natural processes (Ward and Stanford, 1995; Bunn and Arthington, 2002; Hughes et al., 2005; Ziv et al., 2012). Most major rivers in the USA have been subject to substantial engineering modifications to facilitate power generation, navigation, flood control, and floodplain development (Nilsson et al., 2005). Despite pervasive modifications, many coastal rivers in the northern Gulf of Mexico, especially those supporting bottomland hardwood forests, maintain some level of

*Correspondence to: T. E. Pasco, School of Renewable Natural Resources, Louisiana State University Agricultural Center, Baton Rouge, LA, USA. E-mail: [email protected]

Copyright © 2015 John Wiley & Sons, Ltd.

river/floodplain connectivity and experience seasonally predictable hydrographs (Freeman et al., 2005; Hupp et al., 2008; Schramm et al., 2009). The river/floodplain connection created with the advance of the annual flood pulse can greatly enhance the diversity and productivity of both riverine and floodplain flora and fauna by providing nutrient import and export (Robertson et al., 1999; Amoros and Bornette, 2002; Burgess et al., 2013) and access to floodplain spawning and nursery areas (Ward and Stanford, 1995; Galat et al., 1998; Górski et al., 2011; but refer to Humphries et al., 1999). However, rising waters can also significantly reduce floodplain dissolved oxygen (DO) levels because of bacterial consumption of dissolved organic carbon from terrestrially derived organic matter in inundated riparian habitats (Baldwin, 1999; Whitworth et al., 2012). The latter phenomenon is particularly evident in the 5000 km2 Atchafalaya River Basin Floodway (ARBF; Hupp et al. 2008), the largest distributary of the Mississippi River located in south central Louisiana. The hydrology of the ARBF has been modified extensively over the last century by construction of internal and peripheral guide levees, numerous canal spoil banks, closure of historic floodplain distributaries, and deepening

T. E. PASCO ET AL.

of the Atchafalaya River main channel (Piazza et al., 2014). Although the timing, magnitude, and duration of rising and falling Atchafalaya River stages vary substantially from year to year, floodplain hypoxia remains a consistent and often pervasive feature of the annual flood pulse because of high decomposition rates and poor water circulation (Sabo et al., 1999a, 1999b; Kaller et al., 2011; Kroes and Kraemer, 2013). Because of these system-wide alterations, the Atchafalaya River exhibits a complex relationship with its floodplain, particularly regarding the effects of the annual flood pulse on DO and the subsequent responses of resident biota (e.g. Davidson et al., 1998; Fontenot et al., 2001; Rutherford et al., 2001; Alford and Walker, 2013). As floodwaters rise, decomposition of benthic and suspended organic matter depletes floodplain DO levels, resulting in extensive areas of hypoxia (DO ≤ 2 mg l 1) that can persist for extended periods of time if flooding continues as water temperatures rise above 20°C ( Sabo et al., 1999a, 1999b; Kaller et al., 2011). Rising floodwaters mixed with perennially hypoxic water from backwater swamps remain on the floodplain during elevated river stages, and then drain into bayous, lakes, and canals as the pulse recedes, with primary production eventually restoring normoxic conditions (Fontenot et al., 2001; unpublished data). The effects of altered hydrology on ARBF hypoxia are likely exacerbated by dense stands of exotic water hyacinth Eichhornia crassipes, hydrilla Hydrilla verticillata, and salvinia Salvinia spp. (Walley, 2007), all of which suffer substantial mortality during the winter and contribute to the decomposing organic load and declining DO levels as water levels and temperatures rise in the spring.

Atchafalaya River stage data recorded since the 1950s, and water quality data collected in the ARBF since the mid-1990s have revealed considerable variation in the characteristics of the annual flood pulse and the extent and duration of floodplain hypoxia (Kaller et al., 2011). In these analyses, we were particularly interested in the extent to which the characteristics of the flood pulse influenced the magnitude and duration of hypoxia experienced in the lower ARBF. Because the volume and timing of discharge in the Atchafalaya River is controlled through several structures at the head of the river near Simmesport, Louisiana, understanding flood pulse/hypoxia relationships might lead to alterations in discharge management and improvement in overall ARBF water quality. To investigate these relationships, we developed several simple metrics to describe the characteristics of the 2001–09 ARBF flood pulses (sensu Poff et al. 1997), modelled the relationship between these characteristics and the magnitude of hypoxia each year, and validated the model with flood pulse data collected from 2010 to 2013.

MATERIALS AND METHODS Description of Atchafalaya River Floods Previous studies have indicated simple metrics describing flood magnitude, timing, duration, and frequency can be used to differentiate flood years (Poff and Ward, 1989; Poff et al., 1997). We chose seven metrics to describe Atchafalaya River flood pulse characteristics: flood start date and flood end date, number of days of flooding, days of high-temperature or lowtemperature flooding, days of falling flood, and days of maximum flood. Metric data were accessed from the US Army

Figure 1. Example of a flood pulse with two major dewatering events. In this example from 2003, the entire flood pulse was considered to

extend from 1 January to 13 July (day 194). Copyright © 2015 John Wiley & Sons, Ltd.

River Res. Applic. (2015) DOI: 10.1002/rra

FLOODPLAIN HYPOXIA IN THE ATCHAFALAYA RIVER

Corps of Engineers’ historic daily stage data for the Atchafalaya River at Butte La Rose (http://rivergages.mvr.usace. army.mil/WaterControl/new/layout.cfm), which has been used by previous investigators to describe flood events in the ARBF (Sabo et al., 1999a, 1999b; Hupp, 2008, Kaller et al., 2011). Based on convention (Gordon et al., 2004) and previous ARBF studies (Allen et al., 2008), we assumed floodplain inundation (flood start) occurred when the Butte La Rose gage reached a minimum of 3 m for 6 consecutive days in fall or early winter (i.e. November 2004 is part of flood year 2005). Likewise, floodplain dewatering (flood end) was assumed when Butte La Rose gage fell below 3 m for 6–10 consecutive days, usually during late spring to late summer (Figure 1). These events were assigned a Julian date for each flood year, which was negative if the date was from the previous year (e.g. counting from 1 January to 13 December is 18). Total number of flood days equalled the sum of

all days from the beginning to end of the flood minus days when river stage at Butte La Rose fell below 3 m (later in the flood pulse there were periods when river stage would fall below 3 m and then rise again). Daily water temperature was determined from the US Geological Survey’s historic water quality database recorded at Melville, LA (http://nwis. waterdata.usgs.gov/nwis), with each flood day designated as high (≥20°C) or low (50% likelihood of hypoxia, blue predicts 0.82) AUC values for all of the WMAs based on comparisons of predicted and observed hypoxia (Table III). The relationship between inundation and the areal extent of hypoxia varied among the WMAs (Figure 4), whereas temperature yielded a relatively consistent impact on hypoxia across all units. Diagonal changes in coloration were indicative of interactions between temperature and inundation (Zuur et al., 2009), i.e. the probability of hypoxia increased and then decreased along a line vertical to 18°C in both Buffalo Cove and Murphy Lake. In these WMAs, inundation mediated or exacerbated temperature effects on hypoxia (Figures 4a and 4e). In Henderson, inundation over 45% had a very strong positive effect on the probability of hypoxia (Figure 4d), although this relationship was not evident in DOE/21-inch Canal or Bayou Postillion (Figures 4c and 4f). Logistic regressions indicated no statistically significant differences between predicted and observed hypoxic observations in the Buffalo Cove WMA (F1,150 = 1.37, p = 0.24) or Henderson WMA (F1,186 = 0.8, p = 0.37) for the 2010– 2013 flood pulses. Examination of the distribution of predicted and observed hypoxic observations indicated that erroneous GAM predictions tended to over-predict hypoxia in Buffalo Cove (Figure 5a) and Henderson (Figure 5b).

DISCUSSION Declines in floodplain DO levels during inundation (Howitt et al., 2007) have been reported in tropical, subtropical, and temperate river systems (Bechara, 1996; Fisher and Willis, 2000; Petry et al., 2003; Perna and Burrows, 2005), with

pervasive impacts on resident biota (Saint-Paul and Soares, 1987; Davidson et al., 1998; de Oliveira and Calheiros, 2000; Rutherford et al., 2001; King et al., 2012). In managed rivers like the Atchafalaya, the potential to manipulate the annual flood pulse makes understanding the mechanisms that determine the timing, areal extent, and duration of system-wide hypoxic events critical to mitigating these occurrences and their effects on biodiversity and productivity. We were able to describe the variability in ARBF flood pulses with simple metrics based on water temperature and the start, duration, and magnitude of floodplain inundation. Similar measures have been used to describe flooding relationships with hypoxia (Lewis, 2000; Townsend and Edwards, 2003) and the effects of flood dynamics (King et al., 2003; Balcombe et al., 2006; Beesley et al., 2012) and temperature (Gorski et al., 2011) on larval and juvenile fishes. Short-duration floods (2005 and 2006) were uncommon in our dataset, and we expected most floods to be characterized by increased incidences of hypoxia through time, given the extended residence of high-temperature water on the floodplain. Although this problem may have been periodically ameliorated by limitation of bacterial activity from reduced nutrient levels (e.g. Rejas et al., 2005; Farjalla, 2014) as floodwaters fell (P, mean 0.60 mg/l, range 0– 4.42 mg/l; N, mean 0.046 mg/l, range 0–0.56 mg/l; unpublished data, N = 130), we believed that the flooding metrics would be related to the level of hypoxia experienced in the lower ARBF during the study period. However, the PCA-defined flood types did not provide useful models for discerning flood/hypoxia relationships at finer spatial scales, likely because we constrained the relationship between flood measures and hypoxia into linearized scores that did not capture the complex and interactive nature of the flood measures. Importantly, the spline solution identified by the GAM approach robustly predicted specific temperature and inundation thresholds for increased probability of hypoxia in the WMAs. Moreover, GAMs identified conditions where temperature exerted the greatest influence on hypoxia, as well as conditions where inundation appeared to mitigate temperature effects and provided both predictive capabilities and measures of confidence and

Table III. Area under the curve (AUC) values from receiver operating characteristic curves that evaluate the fit of hypoxia values predicted by the generalized additive models (GAMs) with observed values in each water management unit/subunit (WMU/WMS) WMU/WMS Buffalo Cove Bayou Sorrel DOE/21 Inch Canal Henderson Lake Murphy Lake Bayou Postillion

AUC combined GAM

AUC temperature only

0.97 (0.02) 0.90 (0.02) 0.97 (0.02) 0.88 (0.03) 0.82 (0.03) 0.94 (0.02)

0.80 (0.06) 0.69 (0.04) 0.67 (0.07) 0.63 (0.04) 0.65 (0.04) 0.70 (0.04)

AUC inundation only 0.46 0.54 0.56 0.60 0.60 0.60

(0.07) (0.06) (0.10) (0.05) (0.05) (0.05)

p > χ2