We limited our review of climate change assessments for agriculture and hydrology to those studies that used process models (e.g., CERES-Maize, SWAT) requiring climate information (historical or future) at a daily time step. Google Scholar was used to identify peer-reviewed assessments meeting this criterion that were published during the 10-year period, 2005-2014. A total of 39 publications for agriculture and 89 publications for hydrology were identified. Because of the vast literature on ecological impacts of climate change, we confined our literature review to those studies focusing on habitat suitability and that used one or both of two popular species distribution models (i.e. Maxent and GARP). A systematic literature search was performed using Scopus. A total of 135 studies published during the period 2006 to 2015 are included in the analysis. Differences between the applications were summarized including: 1) the typical number and type of GCM simulations and greenhouse gas emissions scenarios from which downscaled projections were developed, 2) the approaches used to downscale the GCM simulations, and 3) the methods used to evaluate, incorporate, and communicate the uncertainty surrounding the future climate.
An important finding is that the availability of user-friendly datasets developed for specific sectors has greatly facilitated climate change assessment research. This is particularly true for habit suitability assessments. WorldClim, a global fine-resolution (~1 km) dataset of long-term (1950-2000) averages of 19 bioclimatic variables (http://www.worldclim.org), was used in over 63 percent of the habitat suitability assessments in our sample as either the climate baseline and/or as input to the development of species distribution models. Moreover, over 57 percent of the habitat suitability assessments used climate projections available from WorldClim that were created by applying change factors between GCM future and control simulations to the baseline gridded fields of the bioclimatic variables. Another 20 percent used datasets of future climate conditions that were partially developed from the WorldClim historical baseline, i.e., the CliMond (https://www.climond.org/ClimateData.aspx) and Climate Change, Agriculture and Food Security (CCAFS; http://ccafs-climate.org/) datasets. Only a small proportion of the habitat suitability assessments obtained historical climate data directly from national centers or GCM simulations from the Climate Model Intercomparison Project (CMIP, http://pcmdi-cmip.llnl.gov/cmip5/).
A broader range of approaches for downscaling GCM simulations to finer resolutions was used in the agricultural and hydrological climate change assessments that we reviewed. The delta method is also popular for these two sectors, with 26 percent of the agricultural assessments and 18 percent of hydrological assessments adjusting historical climate fields by change factors calculated from future and control GCM simulations. A few assessments directly incorporated GCM simulations into their models with no downscaling, and the use of statistical downscaling methods beyond the delta method, particularly the use of weather generators, was popular. The Bias Corrected Spatially Downscaled (BCSD) method ( http://gdo-dcp.ucllnl.org/downscaled_cmip_projections/dcpInterface.html), developed initially for hydrological assessments and based on the statistical downscaling approach of quantile mapping, was used in approximately 10 percent of the hydrological assessments and 3 percent of the agricultural assessments. Dynamically-downscaled climate projections were much more popular for the agricultural and hydrological assessments than for habitat suitability assessments. Twenty percent of the agricultural assessments and 33 percent of the hydrological assessments employed simulations from regional climate models (RCMs). Our review suggests that climate impact assessments often do not take advantage of the large suite of GCMs simulations that are available to evaluate the uncertainty associated with the climate projections. For all three sectors (agriculture, hydrology, habitat suitability) over 35 percent of the assessments relied on the simulations from a single GCM. An additional 20 percent included simulations from only 2-3 GCMs.
The findings reveal that the availability of user-friendly, fine resolution gridded datasets of climate variables and future projections has helped to facilitate climate change impact assessments, as seen by the wide use of a small number of popular datasets. On the other hand, the extensive use of a few sources of climate projections raises concerns regarding the possible influence of potential biases on the aggregate interpretation of climate change impacts and the characterization of the uncertainty surrounding the climate projections. Numerous opportunities exist for new approaches for climate change assessments for agriculture, hydrology, and habitat suitability.