Using a collaborative data collection method to update life-history values for snapper and grouper in Indonesia’s deep-slope demersal fishery

The deep-slope demersal fishery that targets snapper and grouper species is an important fishery in Indonesia. Boats operate at depths between 50-500 m using drop lines and bottom long lines. There are few data, however, on the basic characteristics of the fishery which impedes accurate stock assessments and the establishment of harvest control rules. To address this gap, we developed a collaborative data collection and recording system for species and length composition of commercial catches. The Crew-Operated Data Recording System (CODRS) involves fishers who take photos of each individual fish in the catch along with a low-cost vessel tracking system. As it relies on fisher’s collaboration and willingness to share data, CODRS is comparable with a logbook system but enables verification of species identification with greater spatial resolution. We implemented this system from 2015 to 2018 and gathered data from 251 captains and 2,707 fishing trips, which yielded more than one million individual fish, or 2,680 tons. While there were over 100 species in the fishery, we found that the top five species accounted for approximately half of the total catch. We also unveiled fifteen species previously not associated with the fishery due to the fish being eaten on-board, used as bait, or sold prior to being recorded by traders. Using these data, we updated life-history parameters (length at maturity, optimum fishing length, asymptotic length, and maximum length) of the top 50 species in the fishery based on the maximum observed length; this study resulted in higher estimates for maximum length, most likely due to the high sampling size. For some species, the discrepancies between different sources were large, whereas others were not. This collaborative data collection method and findings are useful for scientists and managers interested in conducting length-based stock assessments to establish harvest control rules for data-poor fisheries.


Introduction 40
In multi-species fisheries, conventional fishery-dependent data collection methods (port 41 sampling, logbooks, and observers) are often viewed as the best way to understand the fishery.  Data collection for each trip began when the boat left port with the GPS automatically 149 recording vessel tracks (Fig. 2). After reaching the fishing grounds, crew would usually fish for a 150 couple of hours, temporarily storing fish on the deck or in chillers. Crew would then take 151 pictures of each fish during the packing process of putting the fish in the hold: one crew member 152 collected fish from the deck and put it on the measuring board, where another crew member took 153 the picture. Thereafter, the fish were stored in the hold. For very small fishing vessels (<5 GT), 154 the process was slightly different: they took pictures upon reaching land instead of at sea. 155 Combined with the location GPS data, the timestamps of the photographs were recorded and 156 used to match each picture with an approximate position.
At the end of each fishing trip, which varied between two days and two months 158 depending on vessel size, captains transferred the memory card containing the photographs of 159 their catch to the research technicians at port. One research technician then identified fish species 160 and another one determined the total length (TL; cm) from the pictures. An experienced third 161 research technician examined the species identification and TL results for accuracy. A senior 162 fisheries scientist verified the pictures of any specimens that exceed the largest fish in our 163 database. To determine weight (kg), allometric length-weight relationships were obtained from 164 the literature (S1 Table). When no values were found for a species, we used morphologically 165 similar species to obtain the length-weight coefficients. 166 Catches that were abnormally low, had low quality photographs and/or only represented 167 the first day of fishing from a multi-day fishing trip were flagged as incomplete and removed 168 from the dataset. Catch and location data were then uploaded to a database (online) where vessel 169 owners, captains, and researchers had access to the contents, each with different viewing 170 privileges. For instance, captains were not able to see the fishing grounds and corresponding 171 catches of other captains, but researchers were. Based on the quality of the photographs, research 172 technicians provided feedback to the captains and/or crew to improve data quality on subsequent 173 trips (Fig. 2). 174 175 Data Accuracy and Catch Composition 176 Receipts or ledgers represented an estimate of total catch weight that was independent 177 from CODRS. Other studies [e.g., 23] have found that sales records represent a reliable estimate 178 of the total catch weight. To test this hypothesis, we collected receipts from fish traders that 179 purchased fish from our partner vessels from August to November 2017. We compared these data to catch estimates from the CODRS system using paired t-tests and linear regression. Data 181 were inspected for normality and homogeneity of variance using a Shapiro-Wilks test. We used 182 descriptive comparisons to determine the most frequently caught species in this fishery by 183 frequency and biomass. represented), (iii) was conducted at a comparable latitude to Indonesia, and (iv) had verifiable 192 species identification (i.e., photograph available, species is distinct and less likely to be 193 misidentified, species exists in the area) due to the high probability of misidentification. For 194 studies that only estimated L inf and not L max , we converted L inf into L max using the following 195 conversion: L max = L inf * 1.1 [24]. Also, if fish length from literature was recorded as fork length 196 or standard length, we converted it into total length using published conversion ratios. If L x-CODRS 197 was chosen as the new L max for a species, then the photograph was reviewed by two or more 198 research technicians and a senior fishery scientist to ensure correct species identification.

199
To further verify our updated L max values, we searched the Internet for angling 200 photographs for each species from comparable latitudes using key words that contained: (i) 201 scientific name of the species of interest, (ii) scientific name of similar species, or (iii) common 202 names from different regions. We then identified the catch species and searched for accompanying descriptive text to determine the catch area. To determine the estimated length of 204 the fish, we used reference objects in the photograph (usually the angler's hands) and measured 205 the TL of the fish. Even though this approach may be less accurate, the photographs gave us a 206 representation of the possible upper ranges of fish sizes that can help assess the plausibility of 207 published L inf or L max values and the values from our CODRS database. We also compared L mat 208 values from our calculation with maturity studies that determined the length at which 50% of the 209 population matures (of the top 15 species in the catch). We excluded studies that published 210 values for length at first maturity. We compared L mat values from areas with similar latitudes 211 (15 o S -15 o N); when not available, we included studies from other latitudes. 212 We calculated L inf , L mat , and L opt using known relationships between the parameters and 213 the accepted L max value as described above. For all families we used L inf z = 0.9 * L max [22]. L mat 214 calculations differed based on the family -for Lutjanidae, L mat = 0.59 * L inf ; for Epinephelidae,  angling records for each of the 25 most common species. This is a result of the efficiency of a 233 collaborative data collection system that involves hundreds of fishers who were able to capture 234 verifiable data. 235 We used total weights from catch receipts as our control dataset to compare with 236 CODRS. We obtained receipts from 41 captains with boats <30 GT, and from 3 captains with 237 boats >30 GT. Because of the small sample size for large boats >30 GT, we did not use the data 238 in our analysis. We found a statistically significant difference for the total catch weight per trip 239 between data collected from receipts and CODRS (p < 0.001, t = 5.5243). Our CODRS dataset 240 also recorded more fish per catch than the receipts and this became more pronounced as the catch 241 got larger (Fig. 3). The estimates of total catch by CODRS appeared higher than estimates of 242 total catch from the receipts and the variation was substantial. Receipts that indicated a total 243 catch in the 10-500 kg range were associated with CODRS data indicating a catch of up to 1.5 244 metric tons. In the 500 kg -2,500 kg per trip category, CODRS appeared to indicate a total catch 245 that was around 50% lower than the figures indicated on the receipts. This is in contrast to the 246 largest catches (> 2,500 kg) where there was a high correlation between CODRS and the 247 receipts. This discrepancy was due to some fish being used as bait, eaten on-board, sold directly to individual buyers (without any receipts), or even "cheating" (rigging weighing scales to record 249 lower weights). 250 It remains speculative which method provided the most accurate data for each landing, 251 but it is remarkable that even a relatively simple observation such as total catch may easily be 252 20-50% higher or lower depending on the method used (ledgers versus CODRS). The problem is 253 not with the estimation of the amount of fish in the hold at any one time. Rather, the problem is 254 with the operational practices that affect the amount of fish in the hold as compared to the 255 amount of fish that was actually caught. The implication is that sources of variation such as 256 (unobserved) offloading at sea, reporting by fishers of "commercial" catch vs. catch for the local 257 market, consumption by crew, etc., may be orders of magnitude higher than measurement errors 258 in total catch weight at the moment that the boat is landing. These observations serve as further 259 evidence of the importance of an on-board data collection system for this fishery as opposed to 260 post-landing data collection methods. Recording System). Black line denotes 1:1 ratio between receipts and CODRS total weight; 264 blue line denotes fitted linear regression with 95% confidence interval in grey.

266
The cost to implement CODRS per year was approximately $3,600-$6,300 per vessel 267 (depending on vessel size). This is substantially more expensive than that of logbooks ($42) but 268 not observers ($2,700 per observer trip). However, given the amount of data obtained from 269 CODRS and its accuracy, the value of this method far exceeds that of other methods. Logbooks,  We expect that technological improvements will enhance scaleability and applicability of 305 CODRS to other fisheries. This may include things such access to cheaper high-quality cameras 306 and an automated fish identification system [2]. Currently, photographs can be blurry especially 307 if the photograph were taken in rough seas and/or during the nighttime. We expect that 308 automation of image analysis through artificial intelligence will expedite the species 309 identification process and remove may of the technical barriers to data analysis [38]. Although  Our findings show that the deep-slope demersal fishery exploited more than 100 species 315 of fish (S2 Table). Half of the total catch, however, belonged to only five species ( using L x-CODRS to determine L max was not an 'anomalous' fish; as illustrated through the length-frequency distributions of the top four species, large sizes were less prevalent, but not anomalous 372 (Fig 4). Photographs of L x-CODRS act as a verifiable evidence of the lengths that these species can 373 attain. In addition, large size ranges in the database also ensured that the data collection had 374 broad selectivity from multiple gear types and multiple vessel sizes.    species. However, the L mat between the largest and smallest species differs by up to 12 cm.

527
Managing these species as one group would lead to overfishing of the largest growing species (P. In Indonesia, a multi-species data collection program of this scale has never been 535 documented before. Our crew operated data recording system (CODRS) as a method proved to 536 be an accurate and effective system to gather catch and effort data for the deep-slope demersal 537 fishery in Indonesia. In addition to collecting high-volume data, CODRS may also act as a first 538 step to collaborative fishery management by engaging fishers in data collection and providing 539 constant feedback between researcher and fisher. The quantity of verifiable length measurements enabled us to compare catch composition between gear types and update important life-history 541 parameters such as maximum length (L max ) and others which will be important for length-based 542 stock assessments. We hope that the ability of CODRS to gather the high amount of species-543 specific catch and effort data in this pilot study can empower other fishery scientists and 544 managers to replicate and improve this system in other data-poor multi-species fisheries.