The authors have declared that no competing interests exist.
This study investigated the contribution of shrimp stocks in supporting the production of valuable predator species. Fishery-independent data on white shrimp, brown shrimp, and selected fish species (spotted seatrout, red drum, and southern flounder) were collected from 1986 to 2014 by the Texas Parks and Wildlife Department, and converted to catch-per-unit effort (CPUE). Here, the associations between the CPUEs of fish species as predators and those of shrimp species as prey in each sampled bay and sampling season were analyzed using co-integration analysis and Partial Least Squares Regression (PLSR). Co-integration analysis revealed significant associations between 31 of 70 possible fish/shrimp pairings. The analysis also revealed discernible seasonal and spatial patterns. White shrimp in August and brown shrimp in May were associated with fish CPUEs in bays located along the lower coast of Texas, whereas white shrimp in November was more strongly associated with fish CPUEs in bays located on the upper coast. This suggests the possible influence of latitudinal environmental gradient. The results of the PLSR, on the other hand, were not conclusive. This may reflect the high statistical error rates inherent to the analysis of short non-stationary time series. Co-integration is a robust method when analyzing non-stationary time series, and a majority of time series in this study was non-stationary. Based on our co-integration results, we conclude that the CPUE data show significant associations between shrimp abundance and the three predator fish species in the tested regions.
The shrimp industry is the most valuable fishery in the Gulf of Mexico; it is worth $588 million USD and accounts for 65% (by weight) of the total US shrimp landings [
In fisheries, species are considered forage species when they play a role as necessary prey for larger predators, such as larger fish, marine mammals, and seabirds [
Major predators of juvenile penaeid shrimp in Galveston Bay, TX, include southern flounder (
As a part of a project aimed at evaluating the importance of penaeid shrimp as a forage species, we investigated the statistical association between time series (1986–2014) data on the catch-per-unit effort (CPUE) of shrimp (white shrimp and brown shrimp) and fish (spotted seatrout, red drum, and southern flounder). Both white shrimp and brown shrimp exhibit an annual life history and utilize coastal marshes and estuaries during their juvenile stages, which are preyed upon by the juvenile and young adult stages of the studied fishes. The data used in this study targeted these stages of shrimp and fishes, and samples were collected in the major bays along the Texas coast as a part of a long-term monitoring program that was independent of any fishery. As co-linearity and non-stationarity are frequent problems in statistical associations of time series data, we used partial least squares regression (PLSR) [
Fishery-independent data were collected by the Coastal Fisheries Division of the Texas Parks and Wildlife Department as part of the Marine Resource Monitoring Program conducted in major coastal bays of Texas. In the present study, we analyzed data obtained from Galveston Bay, Matagorda Bay, San Antonio Bay, Aransas Bay, Corpus Christi Bay, and the upper Laguna Madre (
For fish, gill net sampling was conducted twice a year, for the 10 weeks following the first full week of April (spring sampling) and the 10 weeks following the first full week of September (fall sampling). A total of 45 gill nets were set in each bay system per season, with no less than three gill nets set each week. The gill nets covered the water column from the seafloor to 1.2 meters above the bottom, had a total length of 182.9 m, and were constructed of four continuous 45.7 m-long panels of stretched mesh monofilament webbing of 152 mm, 127 mm, 102 mm, and 76 mm in size. Nets were set perpendicular to the shore, with the smallest mesh (76 mm) nearest the shore; they were deployed around sunset and retrieved around sunrise each day. Organisms greater than 5 mm in total length were identified to the lowest taxonomic level [
For shrimp, bay trawl surveys were conducted monthly. To ensure proper spatial representation, larger bays (Galveston, West Matagorda, San Antonio, Aransas, and Corpus Christi) were stratified into approximately equal-sized upper and lower areas. A total of 20 trawls were conducted in each bay system per month, and scheduled to ensure the temporal representation of samples within a month. The first set of samples was collected during the first half of the month, and the remaining samples were collected during the second half of the month. Sampling was conducted during daylight hours from 30 minutes before sunset to 30 minutes after sunset. The utilized otter trawls had 6.1-m openings and were made of 38-mm stretched nylon multifilament mesh. They were towed for 10 minutes at 3 mph in a semi-circular manner. As with the gill net sampling, captured organisms greater than 5 mm in total length were identified to the lowest taxonomic level [
For most locations and years, the seasonal peak of brown shrimp was May and those for white shrimp were August and November. Therefore, the CPUEs of shrimp species from those months were used for the analysis (
CPUEs of brown shrimp in May (panels a, c, e, g, i, and k) and white shrimp (panels b, d, f, h, j, and l) in August (circle) and November (triangle). Locations: (a-b) Galveston Bay, (c-d) Matagorda Bay, (e-f) San Antonio Bay, (g-h) Aransas Bay, (i-j) Corpus Christi Bay, and (k-l) Upper Laguna Madre. The original CPUEs were transformed by taking the square root and then standardized by taking the Z-score.
Left column: spotted seatrout; middle column: red drum; right column: southern flounder. Circles indicate spring, and x indicate fall. Locations: (a-c) Galveston Bay, (d-f) Matagorda Bay, (f-i) San Antonio Bay, (j-l) Aransas Bay, (m-o) Corpus Christi Bay, and (p-r) Upper Laguna Madre. The original CPUEs were transformed by taking the square root and then standardized by taking the Z-score.
Associations between the CPUEs of each fish species and those of each shrimp species for each bay in each fish-sampling season were analyzed using PLSR. This analysis is similar to canonical correlation analysis except that the error term in the PLSR analysis is a univariate, so the analysis assumes the direction of dependency for the variables. As this study aimed to investigate the importance of shrimp as forage species for larger fish species, the fish time series were treated as dependent variables while the shrimp time series were treated as independent variables (
A ratio > 1 indicates a significant association (emphasized with bold and asterisk *) between the fish and shrimp time series (brown and white shrimp combined).
Location | Spring Fish Data | Fall Fish Data | ||||
---|---|---|---|---|---|---|
Spotted Seatrout | Red Drum | Southern Flounder | Spotted Seatrout | Red Drum | Southern Flounder | |
Galveston | 0.93 | 0.84 | 0.79 | 0.73 | 0.85 | 0.91 |
Matagorda | 0.89 | 0.87 | 0.86 | 0.95 | 0.74 | |
San Antonio | 0.96 | 0.85 | 1.02 | 0.92 | 0.89 | 0.73 |
Aransas | 0.91 | 0.80 | 0.93 | 0.88 | 0.95 | |
Corpus Christi | 0.77 | 0.89 | 0.97 | |||
Upper Laguna Madre | 0.94 | 0.74 | 0.80 | 0.74 |
Co-integration analysis [
Biologically, co-integrated population time series can result when the populations are regulated together but are not necessarily at an equilibrium point (e.g., they could be gradually recovering from past reduced abundance or declining due to over-exploitation). Co-integration can find associations among time series without identifying the source of a non-stationary pattern; this is advantageous in population time series analysis because such a pattern could be produced by many potential processes. Determining the source of non-stationary patterns using short time series is often very difficult. The co-integration method circumvents this problem. A more detailed description of the application of this method in population ecology is described in [
Because the power of co-integration analysis is weak [
Although co-integration analysis is a relatively new statistical method in population time series analysis, the method is well established in econometrics [
A total of 54 time series were analyzed (Figs
An Augmented Dickey-Fuller test suggested that a large number of the time series were non-stationary, but the lag-one differences of the time series were stationary. Because we were interested in finding non-stationary times series that become stationary through their linear combination (i.e., co-integrated time series), we eliminated the stationary time series from further analysis. In total, we compared 70 pairs (out of 108 possible pairs) of shrimp and fish time series using the Engle-Granger co-integration test. Under all of the comparisons, AICC suggested the time lag of two for residual regression. Of them, 31 were significant (
Panels show: (a) co-integrated time series from Upper Laguna Madre; (b) white shrimp in August (circle) and spotted seatrout in spring (X) CPUEs in Upper Laguna Madre; (c) co-integrated time series from Galveston Bay; (d) white shrimp in November (circle) and spotted seatrout in spring (X) CPUEs in Galveston Bay; (e) co-integrated time series from Upper Laguna Madre; and (f) brown shrimp in May (circle) and spotted seatrout in spring (X) CPUEs. CPUEs were transformed by taking the square root and then standardized by taking the Z-score. The co-integrated time series shown in the left panels are the weighted linear combination of the time series shown in the right panels.
The panels show: (a) co-integrated time series from Upper Laguna Madre; (b) white shrimp in August (circle) and red drum in fall (X) CPUEs in Upper Laguna Madre; (c) co-integrated time series from San Antonio Bay; (d) white shrimp in November (circle) and red drum in fall (X) CPUEs in San Antonio Bay; (e) co-integrated time series from Upper Laguna Madre; and (f) brown shrimp in May (circle) and red drum CPUEs in fall (X). CPUEs were transformed by taking the square root and then standardized by taking the Z-score. The co-integrated time series shown in the left panels are the weighted linear combinations of the time series shown in the right panels.
The
Location | Spring Fish Data | Fall Fish Data | ||||
---|---|---|---|---|---|---|
Spotted Seatrout | Red Drum | Southern Flounder | Spotted Seatrout | Red Drum | Southern Flounder | |
Comparison with White Shrimp in August | ||||||
Galveston | 0.23 | 0.15 | 0.22 | 0.17 | 0.20 | 0.15 |
Matagorda | -- | -- | -- | -- | -- | -- |
San Antonio | 0.36 | 0.42 | 0.43 | 0.41 | 0.41 | 0.43 |
Aransas | -- | -- | -- | -- | -- | -- |
Corpus Christi | -- | -- | -- | -- | -- | -- |
Upper Laguna Madre | ||||||
Comparison with White Shrimp in November | ||||||
Galveston | ||||||
Matagorda | 0.06 | |||||
San Antonio | ||||||
Aransas | -- | -- | -- | -- | -- | -- |
Corpus Christi | -- | -- | -- | -- | -- | -- |
Upper Laguna Madre | 0.09 | 0.15 | 0.22 | 0.15 | 0.16 | -- |
Comparison with Brown Shrimp in May | ||||||
Galveston | 0.23 | 0.15 | 0.22 | 0.17 | 0.21 | 0.15 |
Matagorda | 0.14 | 0.18 | 0.10 | 0.14 | 0.17 | 0.07 |
San Antonio | 0.36 | 0.42 | 0.43 | 0.41 | 0.41 | 0.43 |
Aransas | 0.08 | 0.07 | 0.11 | |||
Corpus Christi | -- | -- | -- | -- | -- | -- |
Upper Laguna Madre | -- |
The objective of this study was to seek evidence supporting the importance of penaeid shrimp as a forage species in the Gulf of Mexico. We hypothesized that if the availability of two of the most abundant shrimp species in the region was vital for the sustainability of fish species, the fluctuations in prey and predator species abundance should have a statistical association. Although one should not conclude a trophic interaction based solely on a statistical association, we propose that such an association would support the existence of significant trophic interactions when combined with the knowledge that shrimp are commonly found in the stomachs of these fish species [
The results from our PLSR analysis suggested that there were some statistical associations between the shrimp and fish CPUEs, but we could find no discernible pattern with respect to space or species. Similar to linear regressions and associated analyses, the PLSR analysis is known to exhibit an inflated Type 1 error (false positive) when applied to non-stationary time series. As a large number of the time series in this study was non-stationary, the associations of 22% of the tested pairs (8 of 36 comparisons) could reflect such error. Conversely, time series data often suffer from large sampling errors, which can obscure the associations and increase Type 2 error (false negative). Therefore, we could not draw any conclusion based solely on the results from the PLSR analysis.
Unlike PLSR analysis, co-integration analysis is robust when used on non-stationary time series. Interestingly, our co-integration revealed that 31 of 70 pairs exhibited significant associations, and we could identify a discernible spatial pattern. Both white shrimp in August and brown shrimp in May were important in the bays of the south, whereas white shrimp in November was important in the bays of the north. This suggests that there may be a latitudinal influence. For example, temperature is known to influence the survival and growth of shrimp in the Gulf of Mexico [
As co-integration identifies associations by finding a linear combination of data that produces a stationary time series, an identified association will not come from a year-to-year (i.e., high frequency) fluctuation. Instead, the association comes from a slowly changing (i.e. low frequency) pattern. Many potential causes of the low frequency pattern exist. For example, change in shrimping effort [
On the other hand, high frequency fluctuations in shrimp CPUEs are likely from environmental fluctuations because both brown shrimp and white shrimp in the Gulf of Mexico exhibit weak stock-recruitment relationships [
As a part of our co-integration analysis, we also fitted a vector autoregressive (VAR) model with co-integrated variables to the data to generate a vector-error corrected (VEC) model [
Behavioral differences between brown and white shrimp have been suggested to create differences in their vulnerability to potential predators. For example, brown shrimp were found to be more abundant in vegetated than non-vegetated areas, whereas white shrimp were less selective among habitat types [
Commercial and/or recreational fishing for these species has a measurable impact on their populations and corresponding roles as prey or predators. Management actions have been required in response to shifts in population trends. By the mid 90’s the commercial shrimp fishery in Texas was overcapitalized and experiencing an excessive fishing effort, in order to address this situation, the Texas Parks and Wildlife Department introduced a limited entry and buyback program, which ended the sale of new commercial shrimp fishing licenses and provided funds for the purchase of existing licenses [
Management actions have also affected populations of red drum, spotted seatrout, and southern flounder. Some of these actions include gear restrictions, such as a gill net ban in place since 1980 that benefited most fish species, or a more recent gigging ban in November, in response to a declining trend in southern flounder catch rates and designed to reduce fishing effort during their annual migration to the Gulf. Other actions include special designations, such as gamefish for red drum and spotted seatrout, which effectively ended the commercial fishery for these two species. Stock enhancement is another management measure that benefited red drum and spotted seatrout with about 25 million red drum and spotted seatrout juveniles produced and released every year into bay systems along the Texas’ coast in order to supplement their natural populations [
When the importance of a fishery stock is evaluated, the analysis often considers only the market value of the landed mass. However, our analysis supports the idea that penaeid shrimp in the Gulf of Mexico are also important as forage species for the three fish species that constitute an important resource for recreational fisheries in the region. These are the most sought after fish species in Texas, with spotted seatrout and red drum standing out as the two most frequently landed species in the coastal waters of Texas [
Any statistical analyses have associated potential errors, and co-integration analysis is not an exception. For example, an Augmented Dickey-Fuller test may spuriously categorize stationary time series as non-stationary. Then, a subsequent Engle-Granger co-integration test will conclude the existence of co-integration spuriously. However, in our analysis, the visual inspection of the original time series used in an Engle-Granger co-integration test suggests that many of them appear non-stationary. For example, the right panels of Figs
Here, by combining the results from PLSR and co-integration analysis, we conclude that the CPUE data suggest a significant association between shrimp and fish in the Gulf of Mexico. The abundances of shrimp and fish both exhibit low frequency patterns (i.e., non-stationary), and the association is found at this frequency. In contrast, year-to-year fluctuations do not necessarily show any significant association, likely because the responses of fish populations are not immediate (i.e., low inertia; [
Catch per unit effort (CPUE) of fish data collected in fall, fish data collected in spring, white shrimp data collected in August, white shrimp data collected in November, and brown shrimp data collected in May.
(XLSX)
We thank Amanda Stoner for preparing