Michael A. Dallosch

Headwater non-floodplain wetlands are often overlooked in nutrient-management strategies despite their potential to regulate downstream water quality. Here we test whether headwater non-floodplain wetlands act as nutrient gatekeepers in the Lake Winnipeg Watershed. We define wetlands as seasonally-to-persistently inundated surface-water areas mapped from satellite imagery, and combine multi-decadal observations of wetland inundation and wetland–stream connectivity (1984–2020) with observed annual total nitrogen and total phosphorus loads (1994–2020) and random forest models. Nutrient export is lowest when headwater non-floodplain wetlands remain hydrologically disconnected and increases during episodic reconnection, indicating a threshold-like shift from nutrient storage to export. The ratio of disconnected non-floodplain wetland area is the strongest negative predictor of nutrient delivery, whereas mid/high-order reaches are weaker and less consistent. These results identify intermittently connected headwater non-floodplain wetlands as basin-scale regulators of nutrient pathways and highlight protecting them as a scalable strategy to reduce eutrophication risk in climate-sensitive watersheds.

The application of remote sensing data to empirical models of inland surface water chlorophyll-a concentrations (chl-a) has been in development since the launch of the Landsat 4 satellite series in 1982. However, establishing an empirical model using a chl-a retrieval algorithm is difficult due to the spatial heterogeneity of inland lake water properties. Classification of optical water types (OWTs; i.e., differentially observed water spectra due to differences in water properties) has grown in favour in recent years over traditional non-turbid vs. turbid classifications. This study examined whether top-of-atmosphere reflectance observations in visible to near-infrared bands from Landsat 4, 5, 7, and 8 sensors can be used to identify unique OWTs using a guided unsupervised classification approach in which OWTs are defined through both remotely sensed reflectance and surface water chemistry data taken from samples in North American and Swedish lakes. Linear regressions of algorithms (Landsat reflectance bands, band ratios, products, or combinations) to lake surface water chl-a were built for each OWT. The performances of chl-a retrieval algorithms within each OWT were compared to those of global chl-a algorithms to test the effectiveness of OWT classification. Seven unique OWTs were identified and then fit into four categories with varying degrees of brightness as follows: turbid lakes with a low chl-a:turbidity ratio; turbid lakes with a mixture of high chl-a and turbidity measurements; oligotrophic or mesotrophic lakes with a mixture of low chl-a and turbidity measurements; and eutrophic lakes with a high chl-a:turbidity ratio. With one exception (r² = 0.26, p = 0.08), the best performing algorithm in each OWT showed improvement (r² = 0.69–0.91, p < 0.05), compared with the best performing algorithm for all lakes combined (r² = 0.52, p < 0.05). Landsat reflectance can be used to extract OWTs in inland lakes to provide improved prediction of chl-a over large extents and long time series, giving researchers an opportunity to study the trophic states of unmonitored lakes.