Andrew W. Howell and Dr. Robert J. Richardson
North Carolina State University; Dept. Crop and Soil Sciences
Why do we want to sample submersed vegetation biomass using sonar?
Invasive aquatic plants, such as non-native hydrilla (Hydrilla verticillata), negatively impact waterway systems in the southeastern United States and on a global scale. Often, these aquatic weed species impede recreational activities, power generation, and disrupt native ecological systems. Costs associated with aquatic weed management include expenses accompanied with monitoring, mapping, and implementing control measures. Prompt detection and accurate mapping of submersed aquatic vegetation (SAV) are critical components when formulating management decisions and practices. Therefore, SAV management protocols are often reliant upon the perceived extent of invasion. Traditional biomass sampling techniques have been widely utilized, but often require significant labor inputs, which limits repeatability, the scale of sampling, and the rapidness of processing. Advances in consumer available hydroacoustic technology (sonar) and data post-processing offer the opportunity to estimate SAV biomass at scale with reduced labor and economic requirements.
The objectives of this research were to document the use of an off-the-shelf consumer sonar/gps chartplotter to: 1) describe and characterize a relationship between hydroacoustic biovolume signature to measured hydrilla biomass; 2) develop algorithm for on-the-fly assessment of hydrilla biomass from interpolated biovolume records; 3) define seasonal hydrilla growth patterns at two NC piedmont reservoirs; and 4) create a visual representation of SAV development over time. From these objectives, the expected outcome was to describe a protocol for passive data collection while reducing the economic inputs associated with labor efforts involved in biomass sampling and post-processing evaluations. In our research, a Lowrance HDS-7 Gen2 was utilized to correlate biomass from monospecific stands of hydrilla within two different North Carolina piedmont reservoirs using BioBase 5.2 (now marketed as EcoSound – http://www.biobasemaps.com), cloud-based algorithm to aid in post-processing.
Methods for collecting biovolume and hydrilla biomass.
To determine the trend between biovolume and biomass, we sampled two NC piedmont reservoirs every two weeks for hydrilla biomass and acoustically-derived biovolume between mid-June and late-October 2015. The two NC piedmont waterbodies chosen for fixed sampling sites included Shearon Harris Reservoir (SH; Wake Co.; 35°38′0″N, 78°57′18″W) and Roanoke Rapids Lake (RR; Halifax Co.; 36°28’58.3″N, 77°43’38.7″W) (Figure 1). These locations were selected to simulate a range of ecological factors of bathymetric profile, littoral slope, water exchange frequency, and seasonal SAV growth rates. Based on previously conducted surveys by North Carolina State University, hydrilla has remained the dominant SAV species in both reservoirs for over a decade thus, providing an excellent scenario for testing biovolume and biomass correlations.
 |
| Figure 1. Locations of two NC piedmont reservoirs sampled for hydrilla biomass and hydroacoustic scans; Shearon Harris (bottom center) and Roanoke Rapids (top right |
Hydroacoustic sampling and processing…
…took place using a HST-WSBL 200 khz transducer. Prior to each biomass sampling, a serpentine transect with 7.5 m spacing occurred throughout the plots to determine bottom area interpolation of SAV abundance at each labeled biomass sampling point. The *.sl2 sonar log representing each location was given a unique identifier so future correlation could be prepared. All logged data were saved by the chartplotter to an internal SD memory card for further analysis and upload to BioBase cloud-based software.
Biomass plots and hydrilla collection…
…was conducted from predetermined hydrilla plots (surface area of 0.53 ha-2 each) which were georeferenced at both study locations containing n=60 evenly spaced points per plot. For each of the sampling periods, four randomly selected points were selected per plot using a random number generator without replacement. These sample points were then used for all plots at both lake locations for that sampling period (ie. SH = 8 points biweekly; RR = 12 points biweekly).
To sample biomass at each of the study sites, we used a modified version of the boat-based vertical rake method, proposed by Johnson and Newman (2011) for SAV biomass collection. We use the onboard global positioning system (GPS) from the echosounding unit to “hover” the boat over each designated random biomass point. The rake was then lowered near the boat-based transducer, spun twice, then returned back to the boat. Estimates of rake coverage and average stem length were recorded. If any non-hydrilla species were detected, those plants were separately bagged and analyzed. All hydrilla biomass collected were field washed, individually bagged for dry weight analysis, and placed in a cooler until reaching the lab where samples could then be oven dried and weighed.
Processing and data analysis…
…of all raw hydroacoustic data took place using the EcoSat algorithm to obtain biovolume estimates and depth of each sampling site. Using similar methods to Valley et al. (2015), representative plots at all sites were imported into ESRI’s ArcGIS 10.2.2 software for further post-processing and analysis. Using RStudio, the hydroacoustic and biomass dataset attributes were run through a correlation analysis and mapped to provide visual representation of temporal plant canopy development. Regression analysis and biomass prediction algorithms were also developed. Other factors such as percent area covered (PAC), depth, and seasonal biovolume development among plots at both SH and RR were also analyzed. Among all data, a false-positive limiting depth of 0.76 m-1 was assigned as the minimum depth used for correlation analysis. Removing data points from the shallow regions in this study did not impede overall analysis as plot layouts for biomass harvesting were designed to exceed 0.76 m-1 depths.
So, what are the main findings from our study?
Throughout our experimental period, biomass was collected at 84 sampled points at SH and 71 sampled points at RR to represent a wide range of littoral features and seasonal variability among each test plot. The mean seasonal biovolume percentages and biomass accumulation varied less at RR than at the SH test sites. However, both locations had similar depth variability between individual test plots.
Correlations between Biomass and Percent Biovolume
Biovolume percentages and field observed hydrilla biomass is shown in figures 5 and 6. At SH, we found that there was a strong positive trend which indicated biomass would increase with biovolume (Figure 2). Hydrilla growth at both test plots at SH contained low biovolume to biomass ratios, which may explain why there was a low association when biovolume reached ≥ 25% (Figure 2). Correlations from RR produced very strong agreement and represented high biovolume to biomass ratios well (Figure 3). Conversely, low biovolume estimates (ie. ≤ 25%) were not well represented (Figure 3). Independent correlations from both SH and RR study areas indicated strong, positive correlations. However, the pooled data provided the greatest explanation of association, as the combination of both lake data sets characterized a wide range of biovolume (0 to 100%), depth (0 to 4.74 m), and biomass ranges (0 – 446.1 g dry wt), that helped provide well spread data and the most accurate representation of seasonal hydrilla growth. The entire range of biovolume estimates were represented at both lake locations (Figure 4). Among every correlation, hydrilla biomass was often found at maximum in shallow depth locations. Overall, each biovolume to biomass relationship supports the initial hypothesis that as biovolume increases, SAV biomass should increase in a positive trend, although alterable, since bathymetry aids in defining biovolume estimations.
 |
| Figure 2. Shearon Harris correlations among estimated biovolume percentages from BioBase 5.2 and observed hydrilla biomass (Spearman rank ρ: 0.51, P < 0.001 |
 |
| Figure 3. Roanoke Rapids correlations among estimated biovolume percentages from BioBase 5.2 and observed hydrilla biomass (Spearman rank ρ: 0.73, P < 0.001). |
 |
| Figure 4. Pooled dataset correlations among estimated biovolume percentages from BioBase 5.2 and observed SAV biomass by sampling locations (Spearman rank ρ: 0.79, P < 0.001). From these observations, Shearon Harris best represents data points of low biovolume and low biomass where Roanoke Rapids more clearly depicts samples containing high biovolume and high biomass. |
Hydroacoustic biomass prediction algorithms
Study sites received predictive model parameters described by the ecological growth trends at all sampling locations since depth has been shown to limit the extent of vertical hydrilla growth in this study. Biomass prediction was strongest at SH than compared to either RR or the pooled dataset when cross-validated. Therefore, the most robust model for this study was determined using generalized additive models (GAMs) from the SH dataset, which also provided the highest range of prediction values of all models tested (Figure 5).
 |
|
Figure 5. Example of a three-dimensional, perspective plot used to represent site specific biomass (Z-axis) prediction values at Shearon Harris from generalized additive models (GAMs).
|
Seasonal hydrilla growth and temporal development…
…is a vital component of timely management applications and for studying aquatic weed composition levels within a water column (ie. biovolume). To illustrate seasonal plot biovolume accumulation, Figure 6 portrays an example of the ecological development pattern differences in biovolume over time at SH. We notice that hydrilla growth had the highest biovolume from early-September to mid-October 2015. While our growth pattern is not applicable to all growth patterns of SAV or waterbodies, lake and aquatic weed managers may find this visual appraisal extremely useful for describing seasonal growth or treatment effect patterns.
 |
|
Figure 6. Example of the temporal hydrilla growth pattern for one plot at Shearon Harris.
|
What management implications may be gleaned from this study?
Our findings are consistent with those of Stent and Hanley (1985), Duarte and Kalff (1986), and Duarte and Kalff (1990) that biomass regression analysis is a site-specific procedure due to littoral slope, turbidity, water quality, and the presence of other aquatic plants. We have also shown, that even when using monospecific stands of hydrilla, there is variation of SAV biomass among discrete waterbodies. However, using GAMs to engage vigorous statistical procedures, the power of obtaining a more precise prediction model has potential for explaining environmental factors causing deviation.
A few minor limitations involving the prediction of future hydrilla biomass were apparent in this study. These disadvantages were: 1) hydrilla biomass was highly variable as biovolume reached 100% water column occupancy; 2) once biovolume reached 100%, we were unable to predict future responses in our algorithms; 3) in areas where SAV height was at water surface, our boat was incapable of mapping those areas reliably without obstructing boat transects; and 4) we were not able to obtain biovolume estimates below 0.76 m-1 due to transducer noise. On a cautionary note, all biomass estimations occurring when biovolume is at 100% should be double checked with depth parameters to ensure model elements are not extrapolated beyond the extent of the dataset. Also, to overcome unrepresented areas containing SAV at either depths < 0.76 m-1, or areas containing 100% biovolume, spatial interpolation techniques such as kriging, IDW, bilinear interpolation, or nearest neighbor may be utilized to define those regions (Valley et al. 2005).
Although some drawbacks were present with this research, the advantages of utilizing a consumer available echosounding unit for SAV biomass assessment far out compensated the obstacles formerly described. By exercising a third-party vendor (BioBase) for managing all data recorded by the echosounder, a major reduction in post-processing time was achieved. Furthermore, our methodology proved useful in both tracking and mapping temporal changes in biovolume and biomass accumulation over time. This not only offered a repeatable, non-destructive monitoring opportunity for ecological growth patterns, but also provided visual evidence for aquatic weed management applications. Aquatic plant managers may additionally want to employ the use of the algorithms in formulating proposals for herbicide treatments, grass carp stockings, or stakeholder reports. Since this study focused solely on hydrilla, future studies may want to implicate validation of our models in similarly structured macrophytes for biomass estimation (eg. submersed Myriophyllum spp.).
In conclusion, our study defines the parallel between biovolume and hydrilla biomass– showing how technological advances used by aquatic ecologists conducting fixed point-intercept sampling protocols, may also passively record hydrilla biomass estimation using an over-the-counter echosounder. One key advantage of this echosounding unit was the ease of operation, availability, cost, and minimal training requirements. Likewise, managers could easily train employees to use this system in only a few minutes. Our findings should offer noteworthy economic efficiency with reduction to the degree of labor and time spent presenting spatiotemporal littoral zone dynamics as documented by Maceina et al. (1984). Following the results of this report, especially in lakewide survey scenarios, should offer improvements to traditional biomass sampling.
References
Duarte, C.M. 1987. Use of echosounder tracings to estimate the aboveground biomass of submerged plants in lakes. Can. J. Fish. Aquat. Sci. 44:732-735.
Duarte, C.M. and Kalff J. 1986. Littoral slope as a predictor of the maximum biomass of submerged macrophyte communities. Limnol. Oceanogr. 31:1072-1080.
Duarte, C.M. and Kalff J. 1990. Patterns in the Submerged Macrophyte Biomass of Lakes and the Importance of the Scale of Analysis in the interpretation. Can J Fish Aquat Sci. 47:357–363.
Hijmans, R.J. 2015. raster: Geographic Data Analysis and Modeling. R package version 2.5-2. http://cran.r-project.org/package=raster.
Johnson, J.A. and Newman, R.M. 2011. A comparison of two methods for sampling biomass of aquatic plants. J. Aquat. Plant Manage. 49:1-8.
Maceina, M.J. and Shireman, J.V. 1980. The use of a recording fathometer for determination of distribution and biomass of hydrilla. J. Aquat. Plant Manage. 18:34-39.
Maceina, M.J. and Shireman, J.V., Langeland, K.A. and Canfield J.R., D.E. 1984. Prediction of Submersed Plant Biomass by use of a Recording Fathometer. J. Aquat. Plant Manage. 22:35-38.
Madsen, J.D. 1993. Biomass Techniques for Monitoring and Assessing Control of Aquatic Vegetation. Lake and Reserv Manage. 7(2):141-154.
R Core Team. 2015. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. http://www.R-project.org/.
Stent, C.J. and Hanley, S. 1985. A recording echo sounder for assessing submerged aquatic plant populations in shallow lakes. Aquat Bot 21:377–394.
Thomas, G. L., Thiesfeld, S.L., Bonar, S.A. et al. 1990. Estimation of submergent plant bed biovolume using acoustic range information. Can. J. Fish. Aquat. Sci. 47: 805-812.
Valley, R.D., Drake, M.T., and Anderson, C.S. 2005. Evaluation of alternative interpolation techniques for the mapping of remotely-sensed submersed vegetation abundance. Aquatic Botany. 81:13–25.
Valley, R.D., Johnson, M.B., Dustin, D.L. et al. 2015. Combining hydroacoustic and point-intercept survey methods to assess aquatic plant species abundance patterns and community dominance. J Aquat Plant Manage. 53:121–129.
Blog represents a summary of Master’s thesis research by Andrew Howell; specifically Chapter 2: Correlation of Hydroacoustic Signature to Submersed Plant Biomass. Mr. Howell’s thesis can be downloaded at https://repository.lib.ncsu.edu/handle/1840.20/34479.
