Estimating spawning Green Sturgeon (Acipenser medirostris Ayres, 1854) abundance using side scan sonar and N-mixture models
Data files
Apr 06, 2023 version files 11.24 KB
-
Count2020.csv
2.14 KB
-
Count2021.csv
2.14 KB
-
Count2022.csv
2 KB
-
Count2022S.csv
1.99 KB
-
README.md
1.49 KB
-
UpDownStream.txt
1.48 KB
Abstract
We investigated the use of side scan sonar and N-mixture models to improve the population estimate of the threatened southern distinct population segment of the North American Green Sturgeon (Acipenser medirostris). For three years from 2020 to 2022, we collected images of all known Green Sturgeon aggregations using a Humminbird model Helix 10 with a transducer transmitting at 1.2 megahertz, while simultaneously collecting data with a DIDSON or ARIS sonar. We compared a traditional plot sampling density estimator using the DIDSON and ARIS sonar data to an N-mixture model using side scan sonar images. We investigated the use of a number of different N-mixture model distributions including Binomial and Beta-Binomial for the observation process and Poisson, Negative Binomial and their zero inflated variants for the ecological abundance process as well as the use of covariates to explain the detection process. None of these standard models were either suitable for stable enough in the parameter estimates for our overdispersed data. However, an overdispersed Poisson model with normally distributed errors in both the abundance and observation processes fit the data. Our estimated abundances with this model are 742, 1286 and 1208 for 2020–2022 respectively. Compared to the traditional methods that gave estimates of 373 and 547 for 2020 and 2021, the population estimates increased 2–3 times, mostly from improved field methods. Because of the instability of the N-mixture models in terms of both parameter estimates and differences between analysis platforms, we ultimately would recommend much simpler population estimates that provide similar central tendencies (though with the drawback of no associated measures of variability about the estimates) so as to focus future energy on reducing variability in the data collection process.
Methods
We surveyed potential adult Green Sturgeon aggregations (pools greater than 5m depth) in the upper reaches of the Sacramento River (Figure 1) from June 01-05 in 2020, June 07-18 in 2021 and July 18-21 in 2022. In 2020–2021, we surveyed with both a DIDSON or ARIS and SSS but only used the SSS in 2022. For each year we started at ~RMM 200 (322 km) and surveyed up to RMM 280 (450.5 km), 270 (434.5 km), and 296 (476.5 km) for a total of 32, 38, and 69 pools for the three years, 2020-22 respectively. In 2020-2021 we performed 7 passes with the DIDSON and then 3 passes with the SSS, because of COVID restrictions, only sites that had detected sturgeon at least once in the last 5 years were surveyed. In 2022, we performed 3 passes with the SSS and then up to 4 more if any fish were seen during the first 3 passes, as more passes may provide for better statistical results in sparse data sets (Hostetter et al., 2019). With the rare exception of two especially turbulent sections (one in which we recorded only downstream and the other only upstream), we typically recorded images in both the upstream and downstream direction at a speed between 4 and 8 knots.
We used an ~6m aluminum open hull boat with an outboard jet engine suitable for shallow rivers as a survey platform. Our side scan sonar is a Humminbird model Helix 10 with a transducer mounted on the transom transmitting at 1.2 megahertz with up to 400ft range on each side but with the 8 degrees directly below the boat, called the nadir zone, as a blind spot. We had a dual-screen setup both networked to the same transducer, one for the boat pilot for navigation and a second for recording images.
Raw sonar files were rendered into images using the SonarTRX program, controlled by the AutoIT program to automate the process. There are a number of options in the SonarTRX program to optimize or customize the final image output, we chose to keep the water column and to glue all the separate images for an individual pass at a pool into a single master image. Code for the controlling AutoIT program can be found in the online supplementary data. The master image was subsequently imported into QGIS and each pass was counted by placing a dot next to each ensonified fish image. Only fish that appeared to exhibit standard Green Sturgeon behavior were counted. For example, Green Sturgeon show a typical spacing that is dependent upon the number of animals in their vicinity, they tend to be close to the bottom, orient upstream and are greater than 1.5 meters in length. Regardless, determining what image is a fish, and what is not can be a judgment call. The dots from each pass were then exported as separate ESRI shape files (with accompanying files including a database file) for import into Microsoft Excel and R for further analysis. For 2022 we created a second more selective data set (2022S) by measuring the length of each fish in QGIS and rejected any that were less than 1.5m, as that is an approximate minimum length of adult spawners that have been observed in other systems (Beamesderfer, Simpson and Kopp, 2007).
Usage notes
Text files, any standard text editor should be able to open them.