Statistical climate model downscaling for impact projections in the Midwest United States

Abstract

For future climate projections to be useful they must be actionable at the local level. In this study, we develop daily temperature and precipitation climate scenarios suitable for use in projections of drought, energy use, water use, and crop production. We investigate the magnitude of future changes to air temperature and precipitation in the Midwest United States in response to three future climate change scenarios. Results are used to assess changes to incidence of precipitation extremes and human comfort (using heat index) associated with the anticipated climate changes in the region.We use self-organizing maps and random forest based techniques to generate daily realizations of temperature and precipitation for 279 weather stations in a region centred on Illinois. We determine that the random forest model performs best for maximum and minimum temperatures, while the self-organizing map performs best for precipitation. Using nine models from the Coupled Model Inter-Comparison Project Phase 5, downscaled daily temperature and precipitation values are generated for low, moderate, and high greenhouse gas emissions scenarios for historical and future periods. Based on recent trends, we focus our results on the high emissions scenario, and show an average increase of 4.3 C in maximum daily air temperature across the region for the 2071-2100 period. Precipitation decreases by up to 15% in the southern half of the study region, with a similar percentage increase in the northern half of the region. The regional environmental changes result in an increase of 5.8 in average summer heat index, and increase of 48% in the number of days likely to produce extreme heat, and a decrease in the average value of the standardized precipitation and evapotranspiration index of 1.9 (indicating increased drought) across the region by 2100.

Maize performance in a kura clover living mulch under drought conditions

“…Fertilizer N rate treatments did not affect maize performance at either site. This result adds to previous conflicting findings regarding N requirements for maize in KCLM (Alexander et al., 2019a; Sawyer et al., 2010). Although Alexander et al.…”

“…(2019a) found that the magnitude of N benefits from the KCLM was dependent on previous stand history, Sawyer et al. (2010) concluded that the system supplied no more N than is credited to maize following soybean [ Glycine max (L.) Merr.]. The present results add to the discussion of N fertility management in KCLM systems, but further research is needed to identify the timing and magnitude of N contributions from the living mulch as well as factors that influence N fertilizer requirements for maize.…”

“…This research aimed to determine the influence of row-establishment technique and fertilizer N rate on maize yield and N uptake for maize grown in a KCLM. Because of the unique growing conditions experienced during the 2021 growing season and projected increases in the frequency and severity of drought (Polasky et al, 2021), results will be presented with emphasis on drought intensity and its influence on cropping system performance under row-establishment and fertilizer N treatments. Examining agronomic performance under these extreme environmental conditions will aid in the identification of potential management interventions to reduce risk under uncertainty.…”

Statistical downscaling for precipitation projections in West Africa

“…Using that most similar pattern, we take the set of days in the training period matching the pattern, and sample from the distribution of precipitation values from those days to get a downscaled value for that day. A more complete description of the SOM method can be found in Polasky et al (2021).…”

Assessing statistical downscaling in Argentina: Daily maximum and minimum temperatures

“…A variety of statistical methods—with different complexity—have been employed for statistical downscaling, varying from traditional and well known techniques such as linear regression, analogs, weather generators, weather typing and bias adjustment methods (D'onofrio et al ., 2010; Asong et al ., 2016; Bettolli and Penalba, 2018; Gutiérrez et al ., 2019; Casanueva et al ., 2020; Fan et al ., 2021; Olmo and Bettolli, 2021) to more complex machine learning methods including artificial neural networks, support vector machines, random forest, wavelet‐based methods, among others (Sachindra et al ., 2013; MoradiKhaneghahi et al ., 2019; Kumar et al ., 2021; Polasky et al ., 2021; Quesada‐Chacón et al ., 2021; Sun and Lan, 2021; Hernanz et al ., 2021a). In the case of temperatures, promising new machine learning techniques can be found in the recent literature.…”