Depth and seasonal biases in organic temperature proxies: a modelling study

<p>For more than a decade TEX<sub>86</sub> and U<sup>K’</sup><sub>37</sub>, derived from ratios of biomarker lipids have widely been used as organic paleotemperature proxies. Yet, these proxies, especially TEX<sub>86</sub>, have several uncertainties associated with factors such as depth and seasonal biases which are complicating its application as an annual mean sea-surface temperature (SST) proxy. To constrain this impact, we performed a relatively simple modelling exercise where we use instrumental temperature and nutrient data from 40 locations across the globe to predict theoretical proxy values and compare them with measured core-top proxy values.</p><p>The model first uses instrumental nutrient and temperature data, and probability density functions to predict the theoretical depth occurrence of the source organisms of the two proxies. Additionally, seasonal bias was introduced by predicting seasonal occurrences using instrumental nutrient and chlorophyll data. This was used to calculate the depth- and season weighed temperature signal annually deposited in the sediment, which in turn was converted to theoretical proxy values using culture or mesocosm calibrations. This showed, as expected, that depth and seasonal biases introduced scatter in the correlation between theoretical proxy values and annual mean SST but still highly significant for both U<sup>K’</sup><sub>37</sub> (r<sup>2</sup>= 0.96), and TEX<sub>86</sub> (r<sup>2</sup>= 0.77). We find that the theoretical proxy values are much lower than measured proxy value for TEX<sub>86</sub>, which tentatively suggests that TEX<sub>86 </sub>might in fact be coming from shallower depths or that the mesocosm calibration is incorrect. Our model for U<sup>K’</sup><sub>37</sub> results in theoretical values similar to measured values except for low temperature locations. This might suggest an influence of seasonal bias towards more warmer summer seasons which is more pronounced in high latitudes than in tropics.</p>

ML-SWAN-v1: a hybrid machine learning framework for the concentration prediction and discovery of transport pathways of surface water nutrients

Nutrient data from catchments discharging to receiving waters are monitored for catchment management. However, nutrient data are often sparse in time and space and have non-linear responses to environmental factors, making it difficult to systematically analyse long- and short-term trends and undertake nutrient budgets. To address these challenges, we developed a hybrid machine learning (ML) framework that first separated baseflow and quickflow from total flow, generated data for missing nutrient species, and then utilised the pre-generated nutrient data as additional variables in a final simulation of tributary water quality. Hybrid random forest (RF) and gradient boosting machine (GBM) models were employed and their performance compared with a linear model, a multivariate weighted regression model, and stand-alone RF and GBM models that did not pre-generate nutrient data. The six models were used to predict six different nutrients discharged from two study sites in Western Australia: Ellen Brook (small and ephemeral) and the Murray River (large and perennial). Our results showed that the hybrid RF and GBM models had significantly higher accuracy and lower prediction uncertainty for almost all nutrient species across the two sites. The pre-generated nutrient and hydrological data were highlighted as the most important components of the hybrid model. The model results also indicated different hydrological transport pathways for total nitrogen (TN) export from two tributary catchments. We demonstrated that the hybrid model provides a flexible method to combine data of varied resolution and quality and is accurate for the prediction of responses of surface water nutrient concentrations to hydrologic variability.

ML-SWAN-v1: a hybrid machine learning framework for the prediction of daily surface water nutrient concentrations

Nutrient data from catchments discharging to receiving waters are necessary to monitor and manage water quality, however, they are often sparse in time and space and have non-linear responses to environmental factors, making it difficult to systematically analyse long- and short-term trends and undertake nutrient budgets. To address these challenges, we developed a hybrid machine learning (ML) framework that first separated baseflow and quickflow from total flow, and then generated data for missing nutrient species, using relationships with hydrological data, rainfall, and temporal data. The generated nutrient data were then included as additional variables in a final simulation of tributary water quality. Hybrid random forest (RF) and gradient boosting machines (GBM) models were employed and their performance compared with a linear model, a multivariate weighted regression model and stand-alone RF and GBM models that did not pre-generate nutrient data. The six models were used to predict TN, TP, NH3, dissolved organic carbon (DOC), dissolved organic nitrogen (DON), and filterable reactive phosphorus (FRP) discharged from two study sites in Western Australia: Ellen Brook (small and ephemeral) and the Murray River (large and perennial). Our results showed that the hybrid RF and GBM models had significantly higher accuracy and lower prediction uncertainty for almost all nutrient species across the two sites. We demonstrated that the hybrid model provides a flexible method to combine data of varied resolution and quality, and is accurate for the prediction of responses of surface water nutrient concentrations to hydrologic variability.

GOANA, a Global Ocean Atlas, Neutrally Averaged

<p>Isopycnally averaged hydrographic data gives results that are significantly different to the standard method of averaging at constant depth. The act of averaging isopycnally ensures that water masses are neither created or destroyed.  We average using the weighted least squares quadratic (or LOESS) fitting method of Chelton and Schlax (1994) and Ridgway et al. (2002) along appropriately defined density surfaces.  This produces an gridded oceanographic atlas that is composed of the Fourier coefficients of the mean temporal trend, the strength of the semi-annual and seasonal cycle allowing the user to reconstruct a climatology at any temporal resolution. Initially we are producing an atlas consisting of Absolute Salinty and Conservative Temperature but in the future we aim to include nutrient data.</p>

Nutrient composition of commercially available compared with prescription only gluten-free bread and flour mixes in England

There has been a global increase in gluten-free foods market and perceived health benefits are portrayed in the media. UK legislation requires wheat flour to be fortified with calcium, this is not the case for gluten-free (GF) flours. Calcium intake has been reported to be low in those consuming a GF diet. The only treatment for people with coeliac disease is a GF diet and they have a greater risk of osteoporosis. There is variable access to GF foods on prescription for this patient population in the UK. The study aimed to compare the nutrient profile of commercially available and prescribable GF breads and flour mixes.Nutrient data were collated from the nutritional information and ingredients lists from manufacturers and online stores; 122 GF breads and 17 GF flour mixes. White GF breads (n = 52), GF brown and multi-seeded GF breads (n = 70). Data collated from 17 GF flour mixes (12 white and 5 brown/wholemeal). Nutrient data from the nutritional information and ingredients lists were collated. Data is presented as mean ± standard deviation.The energy and macronutrient composition of commercially available (CA) compared with prescribable (P) white GF breads was similar. Fifty-three percent of prescribable GF white breads were fortified with calcium (8 out of15), whereas only 16 % of CA (6 out of 37). Calcium content were similar in fortified GF white breads (CA: 178 ± 91mg/100 g and P:181 ± 142mg/100 g; NS). Only 4 of the 12 white flour mixes, for bread making, were fortified with calcium. Commercially available brown and multi-seeded breads had significantly less saturated fat and sugar compared with prescribable (saturated fat: CA: 0.6 ± 0.3g/100 g and P: 0.9 ± 0.4g/100 g, p = 0.02; sugar: CA: 2.2 ± 2.3g/100 g and P: 3.8 ± 1.8g/100 g, p = 0.01). Sixty-seven percent of prescribable brown and multi-seeded breads were fortified with calcium (10 out of15), whereas only 13% of CA (7 out of 55) and calcium values were similar in those fortified (CA: 215 ± 178mg/100 g and P: 194 ± 136mg/100 g; NS).This study highlights an inadequate proportion of commercially available GF breads fortified with calcium. There is a need for more GF breads to have their calcium content presented within the nutritional information on the food labels to enable people following a gluten free diet to make informed decisions. We recommend mandatory calcium fortification of GF flours to improve calcium intake in a population with an increased risk of fractures.

Menu Descriptions and Nutrient Content of Food Items in U.S. Restaurants (FS02-01-19)

Objectives This study concerns how the description of foods on restaurant menus relates to their nutrient content as disclosed on company websites. We aimed to test halo effects, regarding how claims about some desirable features might be associated with the presence of other attributes. Methods We used item descriptions and nutrient data for food items (n = 92,949) at the top-selling restaurant chains (n = 92) from 2012 through 2017 in the United States, compiled by the MenuStat project. We classified items into 4 types (mains, appetizers, desserts, sides) and claims into 3 groups using 29 search terms based on consumer interests in health (e.g., “nutritious”), product sourcing (e.g., “local” or “organic”), and vegetal items (vegetarian or vegan). Nutrient data focus on 4 dietary recommendations to limit sodium (mg), trans-fat (g) and saturated fats (% of energy), and to increase fiber (g). We also report calories per item (kcal) and its share from carbohydrates, protein and total fat (%). We used multiple regression to test whether nutrient content was associated with menu claims, controlling for year and restaurant brand, the item being marked as “shareable”, on a kid's menu, or regional and limited-time offerings. Methods and hypotheses were preregistered on As-Predicted.com. Results Contrary to our prediction, nutrient content was more often aligned with U.S. dietary guidelines when their description did include claims. With 3 claim types, 4 food types and 4 recommendations we test 48 possible cases. In 25 (52%) we found alignment between claims and nutrient recommendations, e.g., main dishes with health-related claims had 2% less calories from saturated fat (P < 0.01) and 142 mg less sodium (P < 0.01). In 3 of 48 cases (7%), claims were contrary to recommendations, all of which were desserts with sourcing claims which had more sodium, more trans-fat and more saturated fat than other desserts (all P < 0.01). In 20 of 48 cases (42%) there was no significant difference between items with and without claims. Conclusions Items described as vegetarian/vegan or with sourcing and health claims had nutrient contents that were more often aligned with dietary guidelines than other items. Menu labeling that communicates meal content more directly, such as nutrient fact panels, could inform choice and build trust in restaurant meals. Funding Sources None.

Developing an Ingredient Database of Commercially Packaged Retail Foods (IngIDB), Using the USDA Branded Food Products Database, Illustrated with Food Category ‘Cookies’ (P04-131-19)

Objectives To develop a database of ingredients used in high-selling commercially packaged retail foods, using the publicly available USDA Branded Food Products Database (BFPD). Methods Nutrient Data Laboratory analyzed 2016 IRI (Information Resources Inc.) market sales data, and identified 32 top food categories that contribute > = 1% based on either total volume/weight or number of products sold in the U.S. These include 5 baked product categories – cookies, crackers, fresh breads and rolls, pastries and doughnuts, and pies and cakes. Ingredient lists for the products contributing over 0.001% of the total weight sold for the category were obtained from BFPD. If information for the product was not available in BFPD, other sources such as manufacturer's website were used for top-selling products (contributing over 0.1% of the total weight sold for the category), with the goal to obtain ingredient lists for at least 70–80% of the products sold by total weight for the category. Individual ingredients listed in the ingredients lists were parsed and reviewed. Many of these ingredients were synonyms, necessitating the need for a thesaurus, to facilitate combining same ingredients described otherwise. A prototype thesaurus was developed for flour and fat ingredients in the food category ‘cookies’. Results For food category ‘cookies’, there are ∼13,500 products available, of which ingredient lists were obtained for 1718 products representing ∼84% of the total cookies sold by weight (BFPD: 1699 products (79%); other sources: 19 top-selling products (4%)), hence, meeting our goal. These 1718 cookie products use about 2500 uniquely described ingredients, as per the ingredients lists on the labels. These ingredients include about 30 different types of enriched flours described in ∼ 220 unique ways, including spelling errors etc on the ingredient lists. The thesaurus will need to be expanded to other ingredients. Conclusions The publicly available BFPD can be a useful resource for developing an Ingredient Database for commercially packaged retail foods. The database will be used to prioritize ingredients that need to be chemically analyzed for nutrient information, and provide insights on ‘What is in the foods we eat in America? ’ Funding Sources N/A.