Quantifying Drought Effects on Waterfowl Food Resources in California's Central Valley Through Time-Series Remote Sensing and Machine Learning

California's Central Valley, USA is a critical component of the Pacific Flyway despite loss of more than 90% of its wetlands since the 1850s and its variable supply of freshwater. Now managers produce moist soil seed (MSS) plants by flooding managed wetlands to provide an essential food resource for waterfowl. Managers need MSS plant area and productivity estimates to support waterfowl conservation, yet this remains unknown at the landscape scale. MSS plant production also requires a stable water supply, and the effects of recent drought on production have not been quantified. We generated Landsat-derived extents and productivity (seed yield or its proxy) of major MSS plants [watergrass (Echinochloa crusgalli) and smartweed (Polygonum spp.): WGSW, and swamp timothy (Crypsis schoenoides): ST] and their uncertainties in all Central Valley managed wetlands from 2007 - 2017. From these maps, we tested the effects of water year, land protection status and sub-region on plant area and productivity with a multifactor nested analysis of variance. Finally, we developed indicators to support management decisions to restore or enhance wetlands. MSS plant area maps were based on a support vector machine learning classification of Landsat phenology metrics trained with Valley-wide data from vegetation surveys and interpretation of four years of high-resolution imagery (2017 map overall accuracy: 89%). ST productivity maps were created with a regression model of seed yield (n=68, R² = 0.53, normalized RMSE = 10.5%). Annual maximum of the green chlorophyll index (GI) served as a proxy for WGSW productivity. The total Central Valley-wide estimated area for ST in 2017 was 32,369 ha ± 2,524 ha (C.I.), and 13,012 ha ± 1,384 ha for WGSW. Water year (drought years) was significantly (p<0.05) related to WGSW area, ST area, and ST seed yield and less related to WGSW GI (p<0.10). WGSW area and ST seed yield decreased while ST area increased in critical drought years compared to normal water years (Scheffe's test, p<0.05). Greenness anomalies, or z-scores of ST seed yield can serve to evaluate the performance of a wetland to determine need for habitat enhancement. Updated maps will support monitoring, water management and conservation planning in future years, which are likely to face greater uncertainty in water availability with climate change.