While subsidence is widely recognized as a driver of geomorphic change in the northern Gulf of Mexico (GOM), there is considerable disagreement over the rates of subsidence and the interpreted variability in these rates, which leads to controversies over the impacts of subsidence on surface land area change. Here we present a new method to calculate subsidence rates from the tide gauge record that is based on an understanding of the meteorological drivers of inter-annual sea-level change. In Grand Isle, LA and Galveston, TX, we explicitly show that temporal patterns of subsidence are closely linked to subsurface fluid withdrawal and coastal land loss, and suggest changes in withdrawal rates can both increase and decrease rates of subsidence and wetland loss. Our results also imply that the volume of sediment needed to rebuild GOM wetlands may currently fall within the low end of some restoration scenarios.
 While it is widely recognized that subsidence can be a major driver of land loss [Day et al., 2007; Morton et al., 2002; Reed, 2002], accurate determinations of subsidence rates on sedimentary shorelines over decadal and centennial time scales have largely remained elusive. This results primarily from two factors. First, inter-annual and decadal scale variability in sea level is orders of magnitude greater than long-term trends [Emery and Aubrey, 1991; Kolker and Hameed, 2007; Kolker et al., 2009]. Second, a dearth of stable monuments along many sedimentary shorelines complicates efforts to directly measure subsidence rates, which is particularly problematic where subsidence is not primarily a result of glacial isostatic adjustments (GIA), which can be modeled [Peltier, 2004]. One area where subsidence is a societal concern and difficult to measure is the northern Gulf of Mexico (GOM). In the Mississippi River Delta (MRD), one of this system's largest features, various investigators reported subsidence rates that vary from <5 to >16 mm yr−1 [Dokka, 2006; Törnqvist et al., 2008].
 Our understanding of the patterns in spatial and temporal variability in subsidence rates in the MRD remains limited, and this variability makes it difficult to quantify the role that subsidence has played in coastal change. Thus, it is not clear how much subsidence has contributed to the loss of 4500 km2 of wetlands in the MRD during the 20th Century, or the loss of 140 km2 in Galveston Bay between 1952 and 1989- though it is suspected to be important in both locations [Day et al., 2007; Reed, 2002; White and Tremblay, 1995]. This is an important research gap as wetland loss depletes habitat for fisheries [Day et al., 2007], releases organic carbon to the coastal zone [Bianchi et al., 2011; Wilson and Allison, 2008], alters the hydrodynamics of coastal bays [Fitzgerald et al., 2004], and increase the vulnerability of coastal communities to tropical cyclones such as Hurricane Katrina – an event that cost $80–$150 billion dollars in damages [Day et al., 2007].
 Here we present a new analysis of tide gauge records from the northern Gulf of Mexico that utilizes an understanding of the dynamical drivers of sea level change [Kolker and Hameed, 2007; Kolker et al., 2009] to demonstrate that rates of subsidence in Grand Isle, LA and Galveston, TX are strongly related to patterns of subsurface fluid withdrawal and patterns of wetland loss. While this has long been suspected, we show a close sub-decadal scale coupling between the processes, indicating a strong relationship between anthropogenic changes in the subsurface and earth surface changes.
 We used data from the Permanent Service for Mean Sea Level (PSMSL) for three tide gauges from the northern GOM, Pensacola, FL; Grand Isle, LA and Galveston, TX [Woodworth and Player, 2003]. These were chosen to be representative of different depositional environments with well-studied histories of wetland loss and fluid withdrawal within a similar climatic region (Figures 1 and 2). These gauges are stationed away from the mouth of the Mississippi River, and are not likely to be influenced by changes in stage in the continent's largest river. They are also among the longest continually running tide gauges in the northern GOM for which data is available and standardized by the PSMSL, thereby facilitating inter-gauge comparisons. The Pensacola, FL tide gauge sits on a carbonate platform on the stable North American continent [Gonzalez and Tornqvist, 2006; Peltier, 2004], thereby providing a record that consists mostly of the oceanographic and atmospheric drivers of relative sea-level rise (RSLR), with minimal (<0.5 mm yr–1) contributions from GIA [Gonzalez and Tornqvist, 2006; Peltier, 2004]. The Grand Isle, LA tide gauge sits atop a barrier island at the southern edge of the Barataria Bay, one of the large interdistributary bays in the MRD. This gauge provides a record of RSLR in a sedimentary environment with extensive human impacts in a region that has experienced extensive wetland loss. The Galveston, TX Pier 21 gauge is located in Galveston Bay, TX, a back-barrier coastal lagoon, and is also a sedimentary environment that has experienced considerable human impacts and wetland loss in the past century [White and Tremblay, 1995].
3. Results and Discussion
3.1. Sea Level Variability and Trends
 The long-term linear rate of RSLR was 2.15 ± 0.15 mm yr−1 at Pensacola, 6.38 ± 0.16 at Galveston and 9.27 ± 0.34 mm yr−1 at Grand Isle (Table 1 and Figure 2a). Despite these differences in trends, the year-to-year sea level variability at these stations was similar. To demonstrate this, we removed the linear trend, which is a combination of subsidence, isostacy, eustacy and long-term dynamical shifts, and plotted the three gauges against each other (Figure 2b). A linear, least squares fit indicates that the variance in the detrended Pensacola tide gauge can explain 61% of the variance in the detrended Grand Isle and Galveston gauges, while a linear model of the variance in the Galveston tide gauge can explain 82% of the variance in the Grand Isle tide gauge. Because these gauges are highly correlated, the drivers of inter-annual variability are likely to be similar [Kolker and Hameed, 2007; Kolker et al., 2009].
Table 1. Rates of Relative Sea Level Rise and Inferred Subsidence at Grand Isle, LA and Galveston, TXa
RSL denotes rates of RSLR as determined from a linear fit of the Permanent Service for Mean Sea Level data, while all the other rates are inferred subsidence rates as determined by subtracting the local gauge from the Pensacola gauge. All rates are statistically significant at the p < 0.01 level or greater unless they are bolded. The r2 value is the amount of variance explained by a linear fit of the data, be it the relative sea-level rise or the inferred subsidence.
9.27 ± 0.34
6.24 ± 0.26
7.59 ± 0.23
4.722 ± 0.21
3.16 ± 1.00
2.55 ± 2.17
9.82 ± 0.33
6.18 ± 0.342
12.64 ± 0.69
7.14 ± 1.06
8.59 ± 0.75
7.05 ± 0.87
1.04 ± 0.97
−1.987 ± 1.41
3.2. Atmospheric Drivers of Interannual Sea Level Change
 To understand the drivers of this inter-annual variability, we examined wind velocity, atmospheric pressure at sea level, and skin (i.e., sea) surface temperature (SST) anomalies (Figure 3), during years in which sea level at the geologically stable Pensacola gauge was 1σ above the mean and 1 σ below the mean, using the NCEP/NCAR reanalysis tool [Kalnay et al., 1996]. These anomalies are relative to the period 1981–2010, a procedure that follows earlier work [Kolker and Hameed, 2007; Kolker et al., 2009; Piontkovski and Hameed, 2002]. During years in which detrended sea level anomalies in Pensacola are high, there is a large low atmospheric pressure anomaly over the western United States and wster GOM, which contributes to a wind anomaly directing water towards the northern GOM. These are also years with SST anomalies of 0.2 ∼ 0.3oC. During years with low sea levels at Pensacola the opposite pattern emerges. Sea level pressures are high over the western and northern GOM, wind anomalies are pointed away from the northern Gulf with a limited rotation, and there is a 0.0 to −0.1oC SST anomaly. Most likely, the differences between high and low periods can be explained by shifts in atmospheric pressure, which control wind fields that drive water on or offshore. In years of high pressure, this results in warm water transport from the southern GOM towards the northern GOM. In years of low sea level, northern GOM waters are blown southward, potentially leading to upwelling of colder deeper waters.
3.3. Determination of Subsidence Rates
 If we are correct that the interannual variability observed in Figure 2b results from the meteorological factors presented in Figure 3, then we can remove the correlated interannual variability and isolate subsidence rates at each gauge by subtracting the Pensacola record from Grand Isle and Galveston gauges. Since Pensacola sits on a stable carbonate platform where vertical land movements are minimal, these new records provide estimates of subsidence at Grand Isle and Galveston. The long-term rate of subsidence for Grand Isle and Galveston were determined to be 7.59 ± 0.23 mm yr−1 and 4.71 ± 0.21 mm/yr respectively. These new subsidence records can be viewed in a number of different ways. The inferred Grand Isle subsidence curve appears to experience three distinct phases. Phase 1 lasts from 1948 to 1958 and has a trend of 3.16 ± 1.0 mm yr−1. Phase 2 lasts from 1958 to 1991, and has a trend of 9.82 ± 0.33 mm yr−1, while phase 3 lasts from 1992–2006 and has a rate of 1.04 ± 0.97 mm yr−1. Galveston too can be broken into three similar sections with rates that are 2.55 mm yr−1 ± 2.15 for 1947–1958, 6.18 ± 0.34 mm yr−1 for 1959–1991 and −1.99 ± 1.41 mm yr−1 for 1992–2006.
 Alternatively one can calculate subsidence rates for successive 6-year periods. This interval is a practical consideration that is long enough to reduce biases related to the remaining inter-annual variability and short enough to allow one to see changes in subsidence rates over time (Figures 2d and 2e). Since 6 is a multiple of 18, this approach partially allows us to account for the 18.6 lunar nodal cycle [Gratiot et al., 2008]. Rates of subsidence in Grand Isle follow a quasi-parabolic pattern. They start at 3.52 ± 2.79 mm yr−1 in the 1947–1952 period, reach their maximum of 15.83 ± 3.06 mm yr−1 in the 1965–1970 period and then decline to −1.54 ± 6.20 mm yr−1 in the 2001–2006 period. In Galveston, the inferred pattern of subsidence is more erratic, particularly if pre-1947 rates are included. Inferred rates of subsidence range from +13.62 mm/yr from 1925–1930 to −12.93 ± 7.08 mm yr−1 for the period 2001–2006 (Figure 2f). Like Grand Isle, there is a steady decline in rates of subsidence after the mid-1970s. Both analyses yield negative rates of subsidence at Galveston, which probably results from differences in dynamically driven sea level changes between Galveston and Pensacola [Kolker and Hameed, 2007], resulting from the differences in atmospheric pressure, wind and SST anomalies between these sites (Figure 3).
 These rates can be compared to published rates of subsidence and relative sea level change. A study of a relict swamp in south Louisiana suggests that subsidence in the MRD is driven largely by compaction of peat layers, indicating subsidence rates that are typically <5 mm yr−1 [Törnqvist et al., 2008]. An alternative view, based on releveling surveys, suggests that rates in eastern New Orleans were as high as 16.9 mm yr−1, at least for short intervals of time [Dokka, 2006]. Our results essentially suggest that both authors have valid points – subsidence before oil and gas withdrawal accelerated were close to the Törnqvist et al.  rates while rates observed during the period of maximum fluid withdrawal were closer to Dokka's  higher rates. The patterns in subsidence we observe in Grand Isle are similar to Morton and Bernier's  tide gauge analysis, which indicated that rates of RSLR rates in the MRD varied over the past century, with the greatest rates (10.3 mm yr−1) occurring between 1965–1993, and slower rates occurring before (3.3 mm yr−1; 1947–1965) and after (4.1 mm yr−1; 1993–2006). While Morton and Bernier  were unable to account for the effects of eustacy and inter-annual variability on RSLR trends, our results corroborate their analyses, and advance on their work by removing the oceanographic and atmospheric-derived variability processes, thereby highlighting the impacts of subsidence on relative sea level trends in the northern GOM.
3.4. Drivers of Subsidence
 One likely driver of subsidence in the northern GOM is subsurface fluid withdrawal, which can drive subsidence by decreasing the pressure associated with these fluids, thereby altering grain-to-grain contacts in the sediments [Holzer and Galloway, 2005; Mallman and Zoback, 2007; Morton et al., 2002; Morton and Bernier, 2010]. Onshore oil production in south Louisiana stood at 1.14 × 108 barrels in 1945, reached a maximum of 4.37 × 108 barrels in 1968 and declined to 5.55 × 107 barrels in 2005 [Meckel, 2008]. Interestingly, this pattern in oil production shows a close correspondence to changes in subsidence rates (Figure 2e), suggesting an increase in subsidence as production increased and a decrease in subsidence as production slowed down. At Galveston, where subsidence is generally recognized to be influenced by groundwater withdrawal [White and Tremblay, 1995; White and Mortan, 1997], a similar phenomenon emerges (Figure 2f). Calculated rates of subsidence are greatest during the early-mid 20th century when rates of fluid withdrawal were greatest and decline after the mid-1970s as groundwater withdrawals decreased [Meckel, 2008]. We therefore link changes in subsidence to human activity, and suggest that changes in subsidence rates can be brought about by changes in the volume of fluid removed from reservoirs or aquifers. While others have noted this before [Morton et al., 2002], our analysis suggests a tight temporal coupling between the two, indicating that subsurface anthropogenic activities can have a rapid influence on earth surface processes.
 It is likely that these trends are not artifacts of the methods used or contaminated by broader-scale processes. If the changes observed were caused by the establishment of the tide gauge monument, then the rate of subsidence would be greatest at the beginning of the record and should decelerate thereafter, which is not observed here. The differences in subsidence rates calculated are also unlikely to be a function of the measurement length [Meckel, 2008; Sadler, 1981], as the gauges can all be broken into time slices of similar lengths. These changes are unlikely to be a function of global sea level rise- as those would be a) accounted for in the Pensacola gauge a b) should show an acceleration after ∼1993 [Merrifield et al., 2009], rather than a deceleration. GIA rates along the northern GOM range between 0.1 to 0.2 mm yr−1 [Peltier, 2004], which is one to two orders of magnitude less than the rates of subsidence calculated here. Changes in subsidence rates due to the loading of sediments along the Mississippi River valley [Blum et al., 2008], are also unlikely to be important on the temporal scale measured here.
 In the northern GOM there are numerous faults, which some authors have suggested are partially responsible for land loss [Dokka, 2006; White and Tremblay, 1995]. For example, Dokka  suggested that the Michaud fault in eastern New Orleans experienced a substantial slip of 16.9 mm yr−1 from 1969–1971, which was followed by a period of relatively rapid subsidence (7.1 mm yr−1 from 1971–1977), and then a quasi-asymptotic deceleration until the early 1990s. While there are some similarities between the rates we report and Dokka's  rates, our subsidence curve suggests relatively systematic changes for over fifteen years (1959–1974), which are unlikely to be caused by a slip event. In Galveston, faulting appeared to be most active in the 1960s and 1970s, which was a period of high but variable subsidence, suggesting that slippage may have played a more important role here.
3.5. Influence of Subsidence on Land Loss
 To understand the implications of subsidence on coastal processes, we compared subsidence rates at Grand Isle, LA to measurements of persistent land loss in Barataria Bay (Figure 2e) [Couvillion et al., 2011]. The definition of persistent land loss is derived from an algorithm that looks for land loss that is observable from maps and aerial photographs over multiple time periods, and as such, is relatively free from the influence of local water level changes [Couvillion et al., 2011]. Land loss was relatively slow during the period 1932–1956 (6.1 km2 yr−1; 146 km total), reached a maximum during the period 1973–1975 (92.4 km2 yr−1; 166 km2 total) and then generally declined, reaching a total of 9.3 km2 yr−1 from 2008–2010 [Couvillion et al., 2011]. While there are almost certainly spatial variations in subsidence rates, similar patterns emerge when one compares rates of subsidence in Grand Isle to rates of land loss across the entire MRD [Couvillion et al., 2011]. Galveston Bay is also an area that has experienced rapid wetland loss that has been linked to subsidence. Though temporal patterns in wetland loss in Galveston have not be quantified as precisely as in Louisiana, spatial patterns in subsidence across Galveston Bay tend to match the spatial patterns in wetland loss [White and Tremblay, 1995].
 Taken together, these findings point to a tight coupling between fluid withdrawal, subsidence rates, and wetland loss. Subsidence, coupled with reduced sediment loads and global sea level rise, leads to an elevation deficit, leading to marsh submergence of marshes and conversion of land into open water [Day et al., 2007; Morton et al., 2002; Reed, 2002; Roberts, 1997]. Similar processes appear to be operating in the MRD and GB, despite different types of fluid withdrawal [Meckel, 2008; Morton and Bernier, 2010; White and Tremblay, 1995; White and Mortan, 1997]. It should be noted that subsidence is not the only process driving land loss in the MRD, GB or elsewhere. Surface processes including canal construction, sediment compaction saltwater intrusion, reduced sediment loads, changing biogeochemical regimes and invasive species have all played an important role in coastal land loss [Day et al., 2007; Gagliano et al., 1981; Meckel et al., 2006; Törnqvist et al., 2008; Turner, 1997]. However, subsidence can make coastal wetlands more vulnerable to surface impacts, as it lowers the elevation of wetlands to a point that is outside of the range of tolerances for vegetation, increases the amount of material needed for marsh surfaces to maintain an elevation equilibrium with sea level, and increases the amount of open water in a bay, which increases fetch and erosive processes [Reed, 2002].
3.6. Implications for Coastal Restoration
 Our work has important implications for the restoration the MRD and other coastal systems, as the subsidence rate is a key variable entered into calculations of coastal sustainability [Reed, 2002]. We suggest that subsidence can be exacerbated or mitigated by anthropogenic activities on sub-decadal time scales. These results are limited in spatial extent, and do not cover the Bird's Foot region of the MRD where subsidence rates are greatest and linked to the thickness of Holocene sediments. These findings can also be compared to projections of land change in the MRD. One recent study modeled the survivability of the MRD, using subsidence rates that ranged from 3–8 mm yr−1 [Blum and Roberts, 2009], and concluded that large areas of the MRD would drown under high rates of subsidence and global sea-level rise. However that report also suggested that with low rates of subsidence and global sea level rise, coupled with high rates of sediment trapping, net accretion could occur [Blum and Roberts, 2009]. The subsidence rates we calculated for Grand Isle are presently lower than Blum and Roberts' low-end subsidence scenario, though they once exceeded their high-end scenario. If these findings are confirmed by future studies conducted across wider areas, future wetland losses associated with subsidence could be limited.
 We thank K. Straub, K. Williams and N. Gasparini for a productive conversation that helped kick-off this work. We thank T. Meckel for providing the estimates of oil production in coastal Louisiana and water withdrawal in Galveston Bay that were originally presented by Meckel . R. Morton helped provide background information on subsidence in Galveston Bay. T. Meckel and S. Bentley provided thoughtful reviews which improved the paper.
 The Editor thanks Sam Bentley and Timothy Meckel for their assistance in evaluating this paper.