R/EH-19

Impact/Outcome:

Hypoxia Network (HyNet) modeling

A model of hypoxia formation, propagation and structure was developed for Chesapeake Bay using network modeling approaches. We used data from a validated model of Chesapeake Bay water quality that includes daily average concentrations of dissolved oxygen from the years 1986–2015. Connection points or "nodes" of the hypoxia network (HyNet) model represent spatial cells of the Chesapeake Bay water quality model, while the lines connecting the nodes (a.k.a. "links") were defined using a measure of causality (Granger causality). Because the model was refitted each year, it allowed us to study the dynamics of the network connections and associations with other environmental conditions, 2) accounted for the annual reset of oxygen concentrations during the winter mixing in the bay, and 3) structured the analysis in blocks of years so that the network fitting approach did not depend on having a particularly long (30 years) dataset. The characteristics of the HyNet model were characterized by network statistics for each year, such as the network size (number of links) and number of isolated nodes (nodes that weren't connected to other nodes). We correlated these statistics with hypoxic areas and volumes corresponding to the given year or summer period (May–August). The strongest correlations (about -0.38) were observed between the average annual hypoxic volume and the network size, and density. This suggests that hypoxic conditions lead to fragmentation of Bay waters.

Hypoxia simulations

To examine the sensitivity of benthic invertebrate habitat to external forcing, we conducted several nutrient loading scenarios (+-50%,+-10%. +25%, -75%). These scenarios included altering nitrate loading only, as nitrate is the dominant form of nitrogen in Susquehanna River discharge and nitrogen is the primary nutrient limiting hypoxia in Chesapeake Bay. For each scenario we computed Bay-wide measures of hypoxic area and volume, and examined differences in the distributions of oxygen, chlorophyll and particulate organic carbon across the entire bottom area of the mainstem Chesapeake Bay and its tributaries. Different thresholds of low dissolved oxygen concentrations were compared across scenarios, including, less than 5 mg/L, less than 3 mg/L, and less than 2 mg/l. The different thresholds are associated with tolerances of different organisms. In addition to estimating hypoxic areas at each threshold value, we also computed hypoxic area days, which is the annual sum of all hypoxic areas, allowing for the duration of hypoxia to be included in bay-wide metrics of oxygen depletion. There was a strong, positive correlation between hypoxic area days and hypoxic volume days for all three thresholds. This suggests the variability in hypoxic volume can be accounted for by changes in hypoxic area. Reductions in nutrient loads should reduce hypoxic area and hypoxic volume at similar intervals. Therefore, hypoxic areas can be used as a key metric when analyzing benthic forage and biomass, since most are bottom dwelling organisms.

Water quality-Benthos modeling

This component of the project has focused on combining HyNet information on spatial processing of hypoxia with physical water variables from the Chesapeake Bay water quality model to predict benthic forage biomass. We used Random Forest models, which are a common and simpler kind of machine learning algorithm compared to neural networks, to determine which variables are most important to predicting benthic forage biomass, and in particular, how HyNet variables compare to traditional water quality variables. We then used those models to assess the effects of potential future nutrient loading and climate change scenarios on benthic invertebrate biomass.

We combined output from the Chesapeake Bay water quality model under various idealized future change scenarios (warming, nutrient load reductions) to threshold stress calculations and to statistical models that relate environmental conditions to benthic invertebrate biomass and distribution in the Chesapeake Bay. For each scenario, differences and distributions of oxygen thresholds for lethal and sublethal responses to hypoxia of benthic organisms were used to determine the areas of survivability for different benthic groups in the Chesapeake Bay. To assess mortality/stress thresholds, we used model simulations to create maps of these regions within Bay grid cells, indicating a threshold, below which would prevent taxonomic persistence for each organism given sufficient exposure time (TSub). Under the +2? -75% N scenario, mollusks were primarily found in inhabitable regions in the upper parts of the Bay, concentrated towards the eastern shore. This area was notably smaller compared to other warming scenarios, underscoring the role of nutrient reductions in preserving habitat.

The analysis revealed that as both temperature and nutrient levels rise, the area above TSub for mollusks diminished significantly. For instance, under the +4? +50% N scenario, the inhabitable region extended further into the Potomac River and lower Bay, while zones of unsuitability were largely situated in the upper and middle Bay. These results align with the greatest oxygen declines observed due to increased warming and nutrient input. For crustaceans, under the +2? -75% N scenario, most of the habitable regions were found in the upper and middle Bay. However, the available area under this scenario was still smaller than what was observed with just a 1? increase, highlighting the impact of warming. In the +4? +50% N scenario, the survivable area expanded within the upper Bay and stretched down into the lower Bay, with larger sections of the Potomac River becoming more favorable for crustacean survival (Figure 8d). Unsuitable regions for crustaceans were predominantly located in the middle and lower Bay, corresponding to the most severe oxygen depletion linked to warming and nutrient increases.

We then used the Random Forest models, each containing 1,000 regression trees, to forecast benthic biomass for ten taxonomic groups under 15 different nutrient/temperature scenarios. These included seven specific classes: Amphipoda, Gastropoda, Isopoda, Bivalvia (excluding Macoma spp.), Nemertea, Oligochaeta, and Polychaeta. Additionally, we aggregated worm-like invertebrates (excluding nemerteans, oligochaetes, and polychaetes) and small crustaceans (excluding amphipods and isopods) into two groups, alongside Macoma spp. clams at the genus level. The Random Forest benthic biomass models revealed complex responses of different benthic taxa to the combined effects of warming and nutrient loading. Amphipods, for example, showed slight increases in biomass in the upper Bay and tributaries under a +2°C and 75% nutrient reduction scenario, while bivalves (excluding Macoma) experienced widespread declines in the same areas. Polychaetes demonstrated mixed responses, with declines in the lower Bay and increases in the upper Bay, while small crustaceans remained relatively stable in the lower Bay and Virginia tributaries. Isopods followed trends similar to amphipods, with slight biomass increases in the upper Bay.

The combined effects of nutrient loading and warming in Chesapeake Bay had significant implications for both hypoxic conditions and benthic habitat suitability. Nutrient reduction emerged as a key factor in mitigating hypoxia and supporting the persistence of benthic organisms, particularly in the face of warming. However, temperature increases complicated these dynamics, with some taxa benefiting from moderate warming while others experience substantial declines. The results emphasize the need for integrated management approaches that address both nutrient loading and climate change to preserve benthic biodiversity and ecosystem function in the Bay.