Preparing for and responding to drought requires integrating scientific information into complex decision making processes. In recognition of this challenge, regional drought early warning systems (DEWS) and related drought-information tools have been developed under the National Integrated Drought Information System (NIDIS). Despite the existence of many tools and information sources, however, the factors that influence if a tool(s) is (are) used, which tools are used, and how much benefit those tools provide remain poorly understood. Using the Upper Colorado River DEWS as a case study, this study investigated how water, land, and fire managers select from among many available tools. The Upper Colorado River Basin (UCRB) was one of the first pilot areas, beginning in 2009, for implementation of a regional drought early warning system (DEWS) under the NIDIS program, which now supports eight regional DEWS. (In 2016, the UCRB DEWS was expanded and reconfigured into the Intermountain West DEWS). The selection of the UCRB for a pilot DEWS reflects the regional importance of drought monitoring for managing water supply for agriculture and other uses, and the need for effective decision support related to drought. New drought-information tools were developed specifically for the UCRB DEWS, and a number of others have been created since 2009, adding to the pre-existing toolkit for drought decision making. The various tools that are now available in the Upper Colorado River Basin region can be expected to be more or less suitable for different decision makers’ needs. As a result, the broad decision context of this case study (managing drought) was fixed, but the information needs of users varied. This provided the opportunity to examine the varied choices decision makers make about which of the available tools to use or not use. The research identified four broad categories of tool use that map to particular decision contexts. Water supply managers, land managers with rangeland management responsibilities, land managers focused on ecological health, and fire managers each use a suite of indicators and tools that match their particular decision context and timeframe at which they make decisions. Important differences also emerged in how respondents find out about tools, with water managers reporting strong inter-agency connections while land managers tend to rely on information from others within their agencies. Fire managers also play a key role in keeping others in the land management agencies informed about drought.

The purpose of this project was to estimate and map the probability that grassland converts to cropland in the northern plains and prairie region given potential climate change. This region provides critical breeding and migratory habitat for waterfowl and other wetland-dependent species, and is also a highly productive agricultural region. Generally, the regional effects projected by climate models are increasing temperatures and more variable precipitation, which could provide incentives for private landowners to convert native and managed grassland to intensive cropland. Conversion of grassland to cropland can result in habitat loss for dependent species and the degradation of a range of ecosystem services. If climate change alters the spatial distribution of both agricultural land use and suitable habitat, land managers and conservationists may need to alter efforts to offset the negative consequences of combined climate and land-use change on habitats and dependent species. The land-use change projections associated with this report provide information for such management efforts.

Projected suitable habitat models were constructed in randomForest (R package, version 4.6-10) using a set of presence points for the species derived from element occurrence and herbarium records, together with temperature, precipitation, and soil variables. The current distribution used modeled historic period (1970-2000) climate variables from the appropriate matching GCM model run. These model parameters were then used with projected climate data to get future (2020-2050) modeled suitable habitat for each scenario. Modeled past suitable habitat and modeled future suitable habitat are combined to show areas of change, using various thresholds to distinguish change categories, as well as current mapped J. osteosperma habitats from LANDFIRE existing vegetation (version 1.3.0). Current JUOS habitat is represented as areas with probability greater than the all-scenario average model-reported threshold (sensitivity = specificity) AND currently mapped as JUOS. These probability threshold levels were also applied to projected future habitat (since we have no “future” mapping), with the final model was classified as: Value Habt Class Current 2035 1 Lost >= 0.90 < 0.55 2 Threatened >= 0.90 >= 0.55 and < 0.90 3 Persistent >= 0.90 >= 0.90 4 Emergent < 0.90 >= 0.90 0 none of the above where: 0.90 is the average probability of occurrence value from the 3 scenarios, current timeframe, where JUOS is known to occur (using LANDFIRE vegetation). 0.55 is the average probability of occurrence value from the 3 scenarios, current timeframe, where the model specificity = the model sensitivity.

Projected suitable habitat models were constructed in randomForest (R package, version 4.6-10) using a set of presence points for the species derived from element occurrence and herbarium records, together with temperature, precipitation, and soil variables. The current distribution used modeled historic period (1970-2000) climate variables from the appropriate matching GCM model run. These model parameters were then used with projected climate data to get future (2020-2050) modeled suitable habitat for each scenario. Modeled past suitable habitat and modeled future suitable habitat are combined to show areas of change, using various thresholds to distinguish change categories, as well as current mapped J. osteosperma habitats from LANDFIRE existing vegetation (version 1.3.0). Current JUOS habitat is represented as areas with probability greater than the all-scenario average model-reported threshold (sensitivity = specificity) AND currently mapped as JUOS. These probability threshold levels were also applied to projected future habitat (since we have no “future” mapping), with the final model was classified as: Value Habt Class Current 2035 1 Lost >= 0.90 < 0.55 2 Threatened >= 0.90 >= 0.55 and < 0.90 3 Persistent >= 0.90 >= 0.90 4 Emergent < 0.90 >= 0.90 0 none of the above where: 0.90 is the average probability of occurrence value from the 3 scenarios, current timeframe, where JUOS is known to occur (using LANDFIRE vegetation). 0.55 is the average probability of occurrence value from the 3 scenarios, current timeframe, where the model specificity = the model sensitivity.

Projected suitable habitat models were constructed in randomForest (R package, version 4.6-10) using a set of presence points for the species derived from element occurrence and herbarium records, together with temperature, precipitation, and soil variables. The current distribution used modeled historic period (1970-2000) climate variables from the appropriate matching GCM model run. These model parameters were then used with projected climate data to get future (2020-2050) modeled suitable habitat for each scenario. Modeled past suitable habitat and modeled future suitable habitat are combined to show areas of change, using various thresholds to distinguish change categories, as well as current mapped J. osteosperma habitats from LANDFIRE existing vegetation (version 1.3.0). Current JUOS habitat is represented as areas with probability greater than the all-scenario average model-reported threshold (sensitivity = specificity) AND currently mapped as JUOS. These probability threshold levels were also applied to projected future habitat (since we have no “future” mapping), with the final model was classified as: Value Habt Class Current 2035 1 Lost >= 0.90 < 0.55 2 Threatened >= 0.90 >= 0.55 and < 0.90 3 Persistent >= 0.90 >= 0.90 4 Emergent < 0.90 >= 0.90 0 none of the above where: 0.90 is the average probability of occurrence value from the 3 scenarios, current timeframe, where JUOS is known to occur (using LANDFIRE vegetation). 0.55 is the average probability of occurrence value from the 3 scenarios, current timeframe, where the model specificity = the model sensitivity.

Projected suitable habitat models were constructed in Maxent (version 3.3; Phillips et al. 2004, 2006) using a set of presence points for the species derived from element occurrence and herbarium records, together with temperature, precipitation, and soil variables. The current distribution used modeled historic period (1970-2000) climate variables from the appropriate matching GCM model run. These model parameters were then used with projected climate data to get future (2020-2050) modeled suitable habitat for each scenario. Modeled past suitable habitat and modeled future suitable habitat are combined to show areas of change, using various thresholds to distinguish change categories, as well as current mapped sagebrush-occupied habitats from SWReGAP landcover (USGS 2004). Current occupied habitat is represented as areas with probability greater than the all-scenario average model-reported threshold (sensitivity = specificity) AND currently mapped as the appropriate sagebrush type. These probability threshold levels were also applied to projected future habitat (since we have no “future” mapping), with the final model was classified as: Value Habt Class Current 2035 1 Lost >= 0.56 < 0.34 2 Threatened >= 0.56 >= 0.34 and < 0.56 3 Persistent >= 0.56 >= 0.56 4 Emergent < 0.56 >= 0.56 0 none of the above where: 0.56 is the average probability of occurrence value from the 3 scenarios, current timeframe, where vaseyana is known to occur (using SWReGAP landcover). 0.34 is the average probability of occurrence value from the 3 scenarios, current timeframe, where the model specificity = the model sensitivity.

Projected suitable habitat models were constructed in Maxent (version 3.3; Phillips et al. 2004, 2006) using a set of presence points for the species derived from element occurrence and herbarium records, together with temperature, precipitation, and soil variables. The current distribution used modeled historic period (1970-2000) climate variables from the appropriate matching GCM model run. These model parameters were then used with projected climate data to get future (2020-2050) modeled suitable habitat for each scenario. Modeled past suitable habitat and modeled future suitable habitat are combined to show areas of change, using various thresholds to distinguish change categories, as well as current mapped sagebrush-occupied habitats from SWReGAP landcover (USGS 2004). Current occupied habitat is represented as areas with probability greater than the all-scenario average model-reported threshold (sensitivity = specificity) AND currently mapped as the appropriate sagebrush type. These probability threshold levels were also applied to projected future habitat (since we have no “future” mapping), with the final model was classified as: Value Habt Class Current 2035 1 Lost >= 0.56 < 0.34 2 Threatened >= 0.56 >= 0.34 and < 0.56 3 Persistent >= 0.56 >= 0.56 4 Emergent < 0.56 >= 0.56 0 none of the above where: 0.56 is the average probability of occurrence value from the 3 scenarios, current timeframe, where vaseyana is known to occur (using SWReGAP landcover). 0.34 is the average probability of occurrence value from the 3 scenarios, current timeframe, where the model specificity = the model sensitivity.

Projected suitable habitat models were constructed in Maxent (version 3.3; Phillips et al. 2004, 2006) using a set of presence points for the species derived from element occurrence and herbarium records, together with temperature, precipitation, and soil variables. The current distribution used modeled historic period (1970-2000) climate variables from the appropriate matching GCM model run. These model parameters were then used with projected climate data to get future (2020-2050) modeled suitable habitat for each scenario. Modeled past suitable habitat and modeled future suitable habitat are combined to show areas of change, using various thresholds to distinguish change categories, as well as current mapped sagebrush-occupied habitats from SWReGAP landcover (USGS 2004). Current occupied habitat is represented as areas with probability greater than the all-scenario average model-reported threshold (sensitivity = specificity) AND currently mapped as the appropriate sagebrush type. These probability threshold levels were also applied to projected future habitat (since we have no “future” mapping), with the final model was classified as: Value Habt Class Current 2035 1 Lost >= 0.56 < 0.34 2 Threatened >= 0.56 >= 0.34 and < 0.56 3 Persistent >= 0.56 >= 0.56 4 Emergent < 0.56 >= 0.56 0 none of the above where: 0.56 is the average probability of occurrence value from the 3 scenarios, current timeframe, where vaseyana is known to occur (using SWReGAP landcover). 0.34 is the average probability of occurrence value from the 3 scenarios, current timeframe, where the model specificity = the model sensitivity.

Projected suitable habitat models were constructed in Maxent (version 3.3; Phillips et al. 2004, 2006) using a set of presence points for the species derived from element occurrence and herbarium records, together with temperature, precipitation, and soil variables. The current distribution used modeled historic period (1970-2000) climate variables from the appropriate matching GCM model run. These model parameters were then used with projected climate data to get future (2020-2050) modeled suitable habitat for each scenario. Modeled past suitable habitat and modeled future suitable habitat are combined to show areas of change, using various thresholds to distinguish change categories, as well as current mapped sagebrush-occupied habitats from SWReGAP landcover (USGS 2004). Current occupied habitat is represented as areas with probability greater than the all-scenario average model-reported threshold (sensitivity = specificity) AND currently mapped as the appropriate sagebrush type. These probability threshold levels were also applied to projected future habitat (since we have no “future” mapping), with the final model was classified as: Value Habt Class Current 2035 1 Lost >= 0.46 < 0.21 2 Threatened >= 0.46 >= 0.21 and < 0.46 3 Persistent >= 0.46 >= 0.46 4 Emergent < 0.46 >= 0.46 0 none of the above where: 0.46 is the average probability of occurrence value from the 3 scenarios, current timeframe, where wyomingensis is known to occur (using SWReGAP landcover). 0.21 is the average probability of occurrence value from the 3 scenarios, current timeframe, where the model specificity = the model sensitivity.