Droughts are one of the costliest of natural disasters, posing a significant threat to both man-made and natural systems. Hundreds of drought indices are currently available for the monitoring of drought magnitude, severity, and extent; however, most of these indices were primarily designed for the analysis of drought’s impact on human concerns, such as crop production and freshwater supplies, and do not consider greater environmental aspects such as stream health. To the best of my knowledge, no universal drought index has been developed with the ability to comprehensively quantify different aspects of drought (e.g. meteorological, agricultural, hydrological, and stream health). In addition, there is no general agreement for drought definition even within each drought category. This means that different drought indices, even in the same category, can report contradictory results. In order to address these issues, we designed a study based on the following research objectives: 1) development of an index capable of determining the impact of drought on aquatic ecosystems and stream health; 2) creation of a universal drought index for the measurement of multiple impacts of drought (e.g. meteorological, hydrological, agricultural, and stream health); and 3) determination of a predictive drought model that is able to capture both the categorical and overall impacts of drought. To address the first objective, we coupled a soil and water assessment tool (SWAT) with a regional-scale habitat suitability model to investigate drought conditions in the Saginaw River Watershed. Using the ReliefF algorithm as our variable selection method along with partial least squared regression, six predictive stream health drought models were developed to monitor stream health drought conditions. Of these models, the version with five flow-related variables was determined to be the best tool for predicting both stream health and drought severity. For objective two, thirteen commonly used drought indices from the following categories were integrated to devise a definition of drought that is both categorical and universal: meteorological (4 indices), hydrological (4 indices), agricultural (4 indices), and stream health (1 index). The three closest indices to each other in each category were selected and then averaged to obtain the categorical drought scores; next, the simple average method was used to aggregate the categorical scores, which then provided the universal drought score. For objective three, the ReliefF algorithm was used to select the best variable set for each of the categorical drought scores as well as for the universal drought score. The highest ranked variables were then used in the development of the various predictive drought models via the adaptive network-based fuzzy inference system. The adaptive network-based fuzzy inference system successfully produced four predictive drought models, including the three categorical models (meteorological, agricultural, and hydrological) and the universal drought model.
Read
