| dc.description.abstract | Regional analysis of hydrometeorological variables (e.g., evapotranspiration, precipitation), hydrological processes (e.g., runoff), and their extremes (e.g., drought, floods) is essential for various applications. The analysis can be effective when based on homogeneous regions (groups of sites). Conventionally, hard regions are formed using a regionalization approach considering lumped (time-invariant) statistics of different hydrometeorological variables as attributes. The information on the temporal dynamics of attributes and the degree of resemblance between different sites, which could prove useful in delineating effective regions, is often discarded. To address this, a new regionalization methodology referred to as fuzzy dynamic clustering (FDC) is presented in Chapter 2 for the formation of regions in the fuzzy framework. It accounts for temporal dynamics of various attributes influencing the predictand. Application of the methodology to regionalization of potential evapotranspiration (PET) considering predictor climate variables as attributes, yielded eighteen regions in India, which are validated for homogeneity using PET-related statistics. The regions are shown to be statistically more homogenous than existing agro-ecological zones and regions formed using the global fuzzy c-means clustering method.
Various applications of the homogeneous PET regions formed using the proposed FDC approach are demonstrated in Chapter 3. For each region, relevance vector regression (RVR) relationships are developed between FAO-PM estimate of PET and subsets of its predictor climate variables for estimation of PET from fewer variables at data-sparse locations in the regions. The performance of the RVR relationships in arriving at PET estimates at data-sparse locations is shown to be better than multiple linear regression-based relationships and three widely used empirical equations (Hargreaves, Mcguinnes-Bordne, Priestly-Taylor).
Regional trend analysis was carried out on each of the 18 homogeneous PET regions. A significantly decreasing trend in annual PET was evident for nine regions in north India, whereas a contrasting increasing trend in the same was noted for three regions in south India during the period 1951-2013. Furthermore, change points were identified for PET and its predictor climate variables in each of the regions.
The sensitivity of PET and surface runoff to changes in their predictor climate variables was assessed for each region by considering a newly developed third-order Taylor series approximation of their functional relationships. PET and surface runoff from different regions were found to be sensitive to relative humidity, net solar radiation, Tmax, wind speed, and Tmin (in that order). In the regions located in south and northeast India, net solar radiation and relative humidity (RH) were found to be the key climate variables that govern the PET changes. On the other hand, wind speed followed by RH appeared to have the most influence on PET for regions in rest of the India.
The evaporation paradox (i.e., decrease in PET despite an increase in temperature) was evident in four regions. The paradox in two north-east regions could be attributed to global stilling (i.e., decrease in wind speed), global dimming (i.e., decrease in net solar radiation), and an increase in relative humidity, whereas that in two north-west regions could be attributed to global stilling.
Bouchet-Morton's hypothesis on the complementary relationship (CR) between PET and actual evapotranspiration (AET) was found to be valid in eleven of the eighteen homogeneous PET regions that are located in central, west, and north India. Significant divergence in the trend of PET and AET was evident for two regions in north-east India.
In Chapter 4, future regional changes in PET and its effect on freshwater availability (FWA) were assessed for the 18 homogeneous PET regions. For this purpose, climate simulations were considered for the period 2015-2100 from three CMIP-6 GCMs corresponding to four socio-economic pathways (SSPs 126, 245, 370, and 585) and the ensemble average of those GCMs for the four scenarios. The regional trend of annual PET was projected to increase significantly (at a 5% significance level) in all the regions for all four SSPs. An increase in net solar radiation and increasing/decreasing changes in relative humidity were found to be the main factors (in that order) causing future PET changes. Future projections of PET and temperature did not indicate possibility of evaporation paradox in any of the 18 regions, although the paradox was evident in the north-west and north-eastern parts of India in analysis with historical data. Despite projected increase in PET, significant increase in future FWA was projected for all the regions corresponding to higher SSPs (245, 370, and 585) due to a significant increase in projected precipitation. Consequently, future FWA per capita is projected to significantly increase for SSPs 245 and 585, as expected. However, for SSPs 126 and 370, projected trends in future FWA were significantly increasing and decreasing, respectively. It could be attributed to the higher projected population for SSP-370 and a lower projected population for SSP-126 relative to other SSPs.
To account for nonstationarities of climate variables in drought analysis, nonstationary versions of two multivariate drought indices (namely NSPEI, and NSDDI) were proposed in Chapter 5. The indices were quantified by fitting time-varying probability distribution to the drought reference variable considering the distribution’s location parameter as a function of large-scale climate indices. Regional analysis is appropriate for analyzing drought characteristics/phenomena. To facilitate this, the fuzzy dynamic clustering methodology presented in Chapter 2 is extended for use in regionalization for drought analysis. It aids in delineating homogeneous drought regions by accounting for temporal dynamics of the drought reference variable. Further, a multi-site block bootstrap-based modified seasonal Mann-Kendall test was proposed for testing nonstationarity in regional trend of drought reference variable. It is helpful to a modeler in deciding whether the drought analysis for the target region is appropriate to be performed with nonstationary or stationary drought indices.
In addition, for the first time, the performances and agreement of different pairs of stationary and nonstationary versions of three multivariate standardized drought indices, namely SPEI, SDDI, RDI (Reconnaissance Drought Index), were analyzed at at-site and regional scales through study on different climate regions in Karnataka state, India. Drought Severity-Area-Frequency (SAF) curves indicated that estimates of regional drought severity based on nonstationary versions of drought indices are generally lower than those based on their stationary counterparts for 6-12 months accumulation periods. Similarly, drought Intensity-Area-Duration (IAD) curves indicated that estimates of regional drought intensity are lower with nonstationary drought indices than with their stationary counterparts for various chosen durations and areal extents.
In Chapter 6, a partial duration series-based approach was presented for at-site and regional frequency analysis of extreme precipitation events resulting from different mechanisms. Its effectiveness in arriving at extreme precipitation quantile estimates (EPQEs) at ungauged and sparsely gauged locations is illustrated through a study on central India. The monsoon trough passes through this region over which extreme precipitation is known to be affected by severe low pressure systems and other weather systems. Errors in EPQEs obtained for ungauged locations were within ±20% when derived using regional frequency analysis (RFA) considering causal mechanism-based samples and regional transfer of information from surrounding grids and kriging procedures. The errors were relatively higher when RFA was performed, considering ROI-based regions. In analysis with partial duration series, the EPQEs obtained based on a single sample were found to be generally higher than those obtained using causal mechanism-based samples at the majority of grids. Consequently, the peaks of flood hydrographs were considerably lower when EPQEs were based on the proposed approach than those based on the conventional procedure. This was illustrated through a study on the Chiddgaon Ganges catchment in the Narmada basin, India. Overall the results indicated that the use of the proposed approach for estimating EPQEs could lead to economical designs of flood control/conveyance infrastructure. | en_US |
