Background & Summary

Even though algae are at the nexus of the next generation of agricultural innovations for food and fuel production, significant challenges associated with consistent year-round productivity demonstrations at scale remain. Perhaps one of the largest challenges is associated with the lack of a consistent framework for standardized data collection and analysis such that diverse cultivation or pond operational strategies, pond or equipment design, or other associated variables can be compared on a level playing field. Although there are a wide variety of algal cultivation data available in the scientific literature, for example, the Aquatic Species Program closeout report1 reviewed algal cultivation research sponsored by the US Department of Energy, none of the literature describes data from the same species grown during the same season at different geographic or climatic conditions or even collected data for a consistent set of metrics. Even though several reports have called for increased research and development of algae technology at scale to address the data gap between assumed and actual experimental values of outdoor pond performance and algal productivity to reduce risk and uncertainty in large-scale deployment, few coordinated projects have been funded2–5. Thus the goal of the Algae Testbed Public Private Partnership (ATP3) was to provide year-over-year, well-characterized, algal cultivation datasets to the research, development, and commercialization community to help improve the understanding of algae biomass production in terms of geographic and seasonal variation in cultivation, harvest, and productivity to allow for improved techno-economic analysis (TEA), life cycle analysis (LCA), resource assessment (RA), predictive growth modelling, and crop protection strategies for algae cultivation6. With the successful establishment of the ATP3 algae culturing testbed network, the objective of the Unified Field Studies (UFS) was to address this knowledge gap and generate a robust dataset of algae growth metrics in outdoor open ponds with a focus on the comparison of algae biomass productivity in identical ponds under different seasonal, climatic, and operational conditions at 1000 L scale by controlling the non-geographical related variables of inoculum production, biomass production systems, processes and protocols, system scale, and algae strain.

We report here the datasets from the setup and standardization of the testbed network and the subsequent harvest operations (UFS) of identical algae growth systems across 5 geographic regions over the course of 19 months. This is the first ever demonstration of a multi-site coordinated algae productivity experimental design that required tight management of operational parameters for the cultivation studies, as well as a highly flexible and dynamic environment for data collection that allowed rapid adaptation to user (algae cultivation scientists) feedback. The goal of this paper is to present the complete datasets (Table 1 (available online only)) as well as the cultivation and analytical data collection and curation methods to establish a standardized data collection and curation benchmark for future algae cultivation. By establishing this benchmark, we intend to contribute to the implementation of a set of standardized data metrics for understanding the critical aspects of algae cultivation. A common set of standardized data will facilitate further analysis, discussion, and optimization of algae cultivation as a productive and widespread agricultural crop.

Table 1 Listing of complete UFS experimental protocols and data sets.
Full size table

Methods

We recently published a detailed description of the ATP3 framework which covers the experimental and operational alignment for data collection5,6. A strong emphasis on the initial validation phase of the project was to ensure that laboratories at different sites were able to carry out the implementation of analytical procedures for data collection in a consistent manner6,7. Here, we briefly summarize the essential features of the experiments that produced the data from the UFS, including the location of the sites, the general design of the studies, the equipment used, the algae strains tested, and the typical pond operations during these studies.

Testbed sites

Five testbed sites were originally chosen for their differing climatic and geographic locations; Southwest, desert (AzCATI, Arizona State University (ASU), Mesa, AZ); Western, coastal (California Polytechnic State University (CP), San Luis Obispo, CA); Southeast, inland (Georgia Institute of Technology (GT), Atlanta, GA); Pacific, tropical (Cellana LLC (CELL), Kona, HI), and indoor greenhouse (Touchstone Research Labs, TRL); (Fig. 1a–f). Note that TRL was replaced in the ATP3 consortium by Southeast, coastal region (Florida Algae (FA), Vero Beach, FL) after the UFS-1.

Figure 1: Geographic location, site overview showing pond orientation, installed pond layout (inset diagram), and latitude and longitude of the ATP3 testbed facilities.
figure 1

(a) ASU: 33.304294, −111.673536, (b) CP: 35.254055, −120.674553, (c) TRL: 40.824708, −81.862825, (d) GT: 33.770844, −84.403457, (e) FA: 27.675673, −80.362776, and (f) CELL: 19.734646, −156.053119.

Full size image

Experimental Design

The ponds were installed and a baselining growth experiment without harvesting operations was performed during October to December 20136. With the successful build-out and initial operation of the testbed ponds, the UFS were initiated to determine algae biomass production at each of the geographically diverse sites with three different strains during all four seasons. The experimental design of the UFS was to cultivate algae across the five different geographic regions in the exact same manner. In this way, seasonal and geographic effects could be observed in algae performance data. Pond production experiments were run during each season (identified as approximately 13 week periods) with each individual cultivation experiment lasting at least 8 weeks. The UFS began in April 2014 and ran through July 2015 to conclude this phase of the ATP3 project (Table 2). For the purposes of the UFS, we did not attempt to run continuously throughout the entire year to determine the number of pond operational days that could be achieved in a one-year period as this was outside of the scope of the UFS. Rather we intended to explore the seasonal impacts on algae cultivation across the testbed network by running cultivation experiments during each season. Pond crashes caused by fungus (e.g. Chytridiomycota), bacteria, other algae (predatory, e.g. Poterioochromonas, or competitive, e.g. other eukaryotic algae), and predators (e. g. amoeba) are indicated by underlined values for each experiment (Table 2) and noted in a separate column as a crash in the summary harvest data. The data were collected continuously throughout the experimental duration, periodically checked by each of the site representatives, and ultimately subjected to rigorous quality control (as discussed below in the Data Collection, Review, and Compilation section) when the experiment was completed.

Table 2 Number of UFS pond operational days per strain, site, year, and season.
Full size table

Algae Strains

The performance of three different algae strains are included in the data reported here: Nannochloropsis oceanica KA32 (source: Cellana) grown in a saltwater medium, Chlorella vulgaris LRB-AZ-1201 (source: ASU) grown in a freshwater medium, and Desmodesmus sp. C046 (source: Cellana), grown in both saltwater and freshwater media. Inoculum for these algae strains was produced in indoor columns using appropriate respective growth media, scaled up, and transferred to outdoor ponds as previously described6.

Pond Design & Operation

At each of the testbed sites, 6 identical raceway ponds measuring 3.5 m by 1.5 m and 35 cm in depth (1025 L nominal volume at 25 cm working depth, with a surface area of 4.2 m2 including the paddlewheel area) were installed. Each UFS experiment was performed according to the specific experimental protocol which is included with the dataset as listed in Table 1 (available online only). In general, the experimental plan was designed around a semi-continuous, also known as a drain-and-fill, strategy with harvesting operations to commence once cultures reached a biomass concentration of 0.5 g ash-free dry weight (AFDW) L−1. Ponds were operated with 24-hour paddlewheel mixing at 7.2 rpms creating an average flow rate of 9.3 cm s−1 with on-demand CO2 sparging at 5 L min−1 based on pH. Although carbon uptake efficiency is an important metric in terms of algal cultivation, the UFS was setup to ensure CO2 was not a limiting reagent and was thus supplied in excess. While the flow rate of CO2 to the system was controlled through a pH setpoint, actual volume delivered was not quantified. The initial grow-out period varied but typically took approximately 2 weeks. Harvesting operations to achieve the dilution rate as specified for a given experiment typically involved removing either 25% or 50% of the pond volume three times per week, or 75% of the pond volume once per week. The ponds were harvested using a portable pump and drained to a specified depth depending on the harvesting regime detailed in the protocol for each experiment. Depending on the amount harvested, the time to reset a triplicate set of ponds was typically 45 min including sampling. Since the algal biomass was typically disposed of and not used for downstream processing, harvesting efficiency and percent solids harvested were not calculated and was outside of the scope of the UFS. After routine harvesting, the ponds were re-filled with fresh media to continue the cultivation trial. Ponds were also harvested in response to contamination events and occasionally pre-emptively due to large amounts of rain. Pond samples were taken daily and analysed for water quality and biochemical composition in triplicate according to Table 3.

Table 3 Sample data, units, schedule, and method for data collection.
Full size table

Compositional analysis methods for total protein, total carbohydrate, and total fatty acid methyl esters (FAMEs) followed consensus methods that were validated and implemented across the ATP3 testbed network. These analytical procedures are complex and built on years of experimental analytical biochemistry research and were recently described in detail elsewhere7,8. Detailed analysis protocols are available at https://www.nrel.gov/bioenergy/microalgae-analysis.html.

Data Collection, Review, and Compilation

Researchers at each site were responsible for recording primary experimental data in pre-formatted spreadsheets (available upon request). For any individual experiment (e.g., UFS-1) all spreadsheet formats were the same, but the standard spreadsheet format was modified over time to improve fidelity in recording data. For example, the spreadsheets were set up to automatically highlight outliers (e. g. samples having a relative standard deviation greater than 10%) or identify user input errors (e. g. negative values for pH measurements). The spreadsheets were also occasionally modified to facilitate data entry due to changes in experimental design. For example, the original version of the spreadsheet tracked only the nitrate concentration in ponds while later versions tracked both ammonia and nitrate concentrations (variation in nitrogen source was an integral part of a later experimental design), and the original version of the spreadsheet tracked the optical density of the algae ponds at 750 nm (OD750) while later versions tracked both OD750 and OD680. The final datasets provide columns for nitrate and ammonia and both wavelengths, with “NA” inserted for experiments where these measurements were not taken.

The spreadsheets were shared among ATP3 researchers, and resulting datasets underwent multiple rounds of review and correction over the course of the experiment. These reviews identified missing and incorrectly formatted data, and data entry errors. At the conclusion of each experiment, the spreadsheets were transferred to a centralized data repository that was accessible to the core group of data analysts at ATP3.

We used the statistical programming language R9 (www.R-project.org) to process the data in the spreadsheets into text files. Primary “builder” scripts (one for each data sheet) combined data from the different spreadsheets. Secondary scripts plotted these data to simplify data review and correction. A final script parsed the combined data into separate files for each experiment.

Code availability

Custom code was written in R to both perform quality control and collate all the data from different spreadsheets. The scripts used to combine, review, and parse the primary spreadsheet data, along with individual primary spreadsheets, are available upon request.

Data Records

The data from the ATP3 UFS are organized into five different types as shown in Table 1 and described in detail below. These data include measurements from routine pond operation, measurements from automated instrumentation, weather data, harvest data, and algae compositional analysis data (Data Citation 1). The instrumentation and weather data are available either as complete data sets or as daily-averaged data.

Datatype 1: Pond Operational Data

Pond operational data include the measured pond depth (cm), pH, salinity (g L−1), pond water temperature (oC), nitrogen concentration (mg L−1 N), phosphorus concentration (mg L−1 P), N:P ratio (molar), the sample ID and tracking ID of a physical sample (if taken), the algae concentration (g L−1) as dry weight and ash-free dry weight (AFDW), the ash content of the algal biomass (%), and the OD750. The OD680 is also available only for UFS-7.

Datatype 2: Summary Harvest Data

Complete harvest data are available for each experiment at each site. These data include date, strain, batch and source ID, pond treatment, and depth (cm). Harvest data include harvest number, time between harvests, harvest volume (L), amount harvested (g), ash-free dry weight (AFDW; g L−1) at the time of harvest, and an indication of a pond crash. The UFS-1 Baseline Experiment was an initial start-up, equipment, and operator trial run thus no harvest data were collected.

Datatype 3: Compositional Analysis Data

Compositional analysis data were generated on samples taken at regular intervals during each experiment, including all harvest operations. These data include total protein, total carbohydrate, and FAMEs. Individual samples that did not meet the QC requirements or that experienced experimental errors during analysis are not included7.

Datatype 4: Instrumentation Data

The complete and daily-averaged YSI-5200 instrumentation data are available for each experiment at each site. These data include continuous measurement of pH, pond water temperature (oC), pond conductivity (ms cm−1), dissolved oxygen concentration (DO, mg L−1), dissolved oxygen saturation (%), salinity (g L−1), and photosynthetically active radiation (PAR, μmol m−2 s−1). The PAR sensor was connected to only one of the ponds, thus only one pond collected light intensity data, and these data are used to characterize the entire site.

Datatype 5: Weather Data

The complete and daily-averaged weather data are available for each experiment at each site. These data were collected either on site (ASU and CELL) or using publicly available data in close proximity (within 6 miles) to the site. For the CP site, data from the California Irrigation Management Information System, San Luis Obispo station #52 were used. For the GT site, data from the Clark Atlanta University weather station (http://weather.uga.edu) were used. For the FA site, data from Weather Underground, station: KFLVEROB15 were used. For the TRL site, data from the National Solar Radiation Database delineated by the latitude and longitude of the TRL site were used10. These data include air temperature (oC), relative humidity (%RH), Global Light Intensity (W m−2 s−1), daily precipitation (cm), wind speed (km hr-1), and wind direction (degrees).

Datatype 6: Summary Combined Data

The primary Pond Operational Data (Datatype 1) and Summary Harvest Data (Datatype 2) were combined into one spreadsheet for each experiment to aid in the further analysis of these datasets. Additional metrics beyond primary data collection were also calculated to provide greater utility to the dataset. These calculated metrics include nutrients supplied (g) (nitrate (NO3), ammonium (NH4), and phosphorous (P)), nutrients utilized (g), nutrient utilization efficiency (%), and nutrient demand (g N or g P g AFDW−1). Daily evaporation rate (cm day−1) and biomass per unit energy (mg AFDW mol photons-1 and mg AFDW kW−1) are also included. In addition, pond operator comments are included in these spreadsheets to provide information about what was being observed in the ponds.

Technical Validation

The experimental protocol document for each UFS provides details regarding procedures on data collection. Routine pond operational measurements (e. g. pH, temperature, optical density) were collected using standard laboratory instruments subjected to regular calibration. Physical samples for AFDW and ash content were collected and analysed in triplicate. The probes used to collect the instrumentation data were calibrated regularly or if a probe was determined to be giving erroneous values. The raw data were plotted using R scripts as noted above as part of the technical validation process. These plots were then manually examined for obvious outliers, for example, a series of pH values rapidly and temporarily changing to values of approximately 4 or 10 indicate pH probe calibration data that were accidentally included in the automated data collection run, or a sudden deviation in temperature when the pond is stable at 30 oC indicating the temperature probe was briefly removed from the pond for cleaning without interrupting data collection. These values were typically flagged with a comment in the primary datasheet. The weather data collected from independently-operated weather stations was not curated by ATP3 researchers. Further details of the initial harmonization across the sites and of the technical validation associated with the algae compositional analysis data are presented elsewhere6,7.

We have deliberately not removed large amounts of data, since it is quite difficult to determine with high confidence which data are “good” and which are “bad”. Some of the challenges we experienced in curating these data are illustrated in Fig. 2. The figure shows the pH data collected for three ponds at ASU during the UFS-1 experiment. The black points show the pH measured at 15-minute increments using a pH probe connected to an automated data logger. The magenta points show the daily average value of these pH measurements. The blue points show the manually-recorded pH measurements, collected twice per day (just after sunrise and just before sunset) Monday through Friday along with other pond operational data.

Figure 2
figure 2

Auto-logged (black), manually-logged (blue), and daily average (magenta) pH in 3 ponds during the UFS-1 experiment at ASU.

Full size image

The pH data from the auto-logger (black points) show a small diurnal oscillation due to the change in algae metabolism under illumination and darkness. The daily-averaged data (magenta points) reduce this oscillation. The manually-recorded pH data (blue points) show slightly more variability than the daily-averaged auto-logger data. In the top figure, the auto-logger data show a number of excursions from the apparent “true value” of approximately 8.5. These excursions are highlighted with red boxes in the figure. These excursions in measured pH are due to the inadvertent logging of the pH probe recalibration process, where the probe is removed from the pond and immersed in a series of pH standards. The experimental protocol called for the auto-logger to be paused for the recalibration process. These inadvertent measurements result in slightly lower daily-averaged values as well. The manually-recorded pH data do not show this effect. There appears to be a slight positive bias in the manually-recorded pH data compared to the auto-logger pH data as well, despite being calibrated regularly.

In Fig. 3, we show the corresponding pond temperature data, to the pH data in Fig. 2. The black points in Fig. 3 show the temperature measured at 15-minute increments using a temperature probe connected to an automated data-logger. The magenta points show the daily average value of these temperature measurements. The blue points show the manually-recorded temperature measurements, collected twice per day Monday through Friday (morning and evening) along with other pond operational data. The diurnal variation in pond temperature is evident, and the daily-averaged temperature value reduces this variation. The manual temperature measurements are very close to the extreme daily values, since these were made shortly after sunrise and shortly before sunset. Due to a malfunction in the pond P2 auto-logger there are no auto-logger data for pond P2 for this experiment. There is overall a good agreement between the temperature data collected for each of the three ponds, with minimal pond-to-pond variability. There is much better agreement in temperature values as opposed to pH between the manually-recorded data and the auto-logger data.

Figure 3
figure 3

Auto-logged (black), manually-logged (blue), and daily average (magenta) temperature in 3 ponds during the UFS-1 experiment ASU.

Full size image

Usage Notes

The complete description of a specific UFS experiment requires all five data types as well as the operational protocols (as discussed in the Data Records section). To effectively use these data, it is necessary to link the data from each file. Pond operational data, instrumentation data, and summary harvest data are linked through the site location, pond number, and date and time. Because the weather data and the PAR data characterize the entire site, they can be linked to the pond operational and harvest data through the site location and date and time. Pond operational data and harvest data are linked to the algae compositional data by the use of a tracking ID recorded on the pond operations spreadsheet at the time of harvest, since each biomass sample with compositional analysis data has a unique tracking ID.

Though the datasets were curated for outliers, typographic errors upon data entry, and instrumentation sensor data that was clearly due to a bad sensor, we felt that too much curation could jeopardize the notion that these datasets are purely objective (not selective) datasets. Thus, for example, present in the YSI pond temperature data there are occasional rapid spikes to 40 oC. This represents times when during harvest or other pond operations, the temperature probe was removed and set in a cup of water by the pond. Being in such a small volume of water, the temperature invariably spiked upwards on sunny days. Similarly, as described above, there are multiple instances of the pH measurement for a pond spiking to pH10 or pH4 for several minutes. This was caused by the periodic recalibration of the pH probes while the instrumentation was collecting data. While the protocol called for the instrument to be placed in “standby mode” during the recalibration, this did not always occur.

It is anticipated that many users of these data will be interested in algae productivity; the rate at which algae biomass is either produced or harvested from outdoor ponds11. Algae productivity is generally calculated either as the increase in biomass or the summation of biomass harvested over time. Although a reliable representative measure of growth, biomass productivity calculated as an increase in biomass concentration over time is mainly of academic interest in the context of a biomass production scheme for conversion to biofuels, since it is calculated in such a way that does not account for how much biomass can be or was removed from the pond for conversion to biofuels and may not take into account the subsequent effects of biomass removal. Thus the algae productivity calculation that is more relevant is the calculation based on the actual amount of biomass harvested from the pond and available for conversion to biofuels during the course of a given production run. Usage of the datasets requires careful selection of the appropriate data and additional curation by researchers seeking to use average values for predictive modelling or geographical projections. We have previously presented guidelines and examples on the proper use of the ATP3 data to calculate the relevant harvest yield productivity from these datasets11 and the harvest yield productivity calculated from these datasets for the algal strain Nannochloropsis KA32 is presented (Fig. 4). Clearly the biggest driver of areal harvest yield productivity in Nannochloropsis is seasonal with the decrease in temperature and available sunlight with the change from summer into winter. This effect is tempered at the coastal sites of Cellana and Florida Algae. In addition to areal harvest yield productivity, the composition of the algae is a critical parameter in modelling the value and potential of algae cultivation. The algal biomass composition made up of protein, carbohydrate, and lipids as fatty acid methyl esters (FAME) as % AFDW, over the course of the UFS shows how the various components changed through the course of an experiment, across the seasons, and at the various sites (Fig. 5). In general, the UFS was designed to provide baseline growth metrics rather than nutrient deprivation studies. This is evident in the composition as protein is usually high while FAME and carbohydrates are lower. There are examples where nitrogen became depleted with the resulting decrease in protein and increase in FAME and carbohydrates.

Figure 4: Areal harvest yield productivity for N. oceanica KA32 during the entirety of the UFS.
figure 4

Productivity is shown as calculated across the entire experimental duration (blue) and for just the continuous portion of the experiment (orange).

Full size image
Figure 5: Biomass composition, protein (black), carbohydrate (green), and FAME lipids (red) as % AFDW.
figure 5

The strain of algae is denoted by colored blocks on the x-axis.

Full size image

Finally, the UFS encompasses typical algal growth at 5 different sites across the United States in regions identified as amenable to outdoor algal cultivation. However, as stated in the introduction, the goal of the Algae Testbed Public Private Partnership (ATP3) was to provide year-over-year, well-characterized, algal cultivation datasets. The comprehensive operational, harvest, and composition data, will be useful to the algae research community even though we did not optimize our experiments for either biomass accumulation nor algal compositional shifts, e. g. nitrogen stress to induce the accumulation of lipids. These datasets represent a conservative, non-optimized, estimation of typical algal areal harvest yield productivities and compositions that can be achieved with these strains at the scale, locations, and operational parameters chosen.

Additional information

How to cite this article: Knoshaug, E. P. et al. Unified field studies of the algae testbed public-private partnership as the benchmark for algae agronomics Data. 5:180267 doi: 10.1038/sdata.2018.267 (2018).

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.