Climate diagnostics and forecasting of hydro-climate variables in the Gunnison river basin, Colorado
Abstract:
Spring streamflows are the major contribution of the annual flows in the Gunnison River Basin (GRB). Statistical methods in collaboration with climate signal are used to forecast spring flows at various locations in the GRB. In this process, this study looked at the GRB characteristics and found key variables as snow and snow driven spring flows. Principal Component Analysis (PCA) of the snow and spring flows suggested basin as homogeneous and unique modulating climate signal for the entire basin. Diagnostics of the basin key variables revealed the influence of local variables such as Geopotential Height, Surface Air Temperature, Zonal and Meridional wind in addition to Sea Surface Temperatures. Using above mentioned climate variables as predictors, spring flows are forecasted at different lead times i.e., on January 1st and on April 2nd. Results are quite promising.
Introduction:
Water is an important commodity and it plays a vital role in the economy of the western USA as it entangled with many other purposes than elsewhere. In this regard, many rivers in the western USA dammed by dams and reservoirs, mainly, to provide reliable supply of water to the community and additionally to serve many other purposes. Large portions of annual runoff in these rivers come from the spring flows i.e., spring snowmelt. In the western USA, snowpack accumulations are an important source of runoff and water supply. So, snowpack accumulations used to model and forecast peak and annual flows (Gray and Male, 1981). Being snowmelt-dominated basins, spring flows could be directly forecasted from the predictors winter snowpack predictors. In this regard, literature that relates predictors of the winter snowpack is presented below. Both dynamical and statistical methods are being used to forecast streamflows. One pathway is the use of General Circulations models (GCMs) of the Ocean and atmosphere, followed by “downscaling” using Regional Climate Methods or statistical approaches, followed by lumped or distributed rainfall-runoff models. Theoretically it sounds well. However, GCMs are able to reproduce general spatial and temporal distributions of hydro-climate variables only on global scale but on regional scales it vary markedly from the observed values (Grotch and MacCraken, 1990); and it still needs evaluation in the context of resource management at relevant spatial and temporal scales of interest (Francisco Assis Souza Filho and Upmanu Lall, 2003). However, GCM simulations of synoptic scale weather patterns are more reliable than temperature and precipitation, because weather patterns are of a spatial scale that is compatible with the spatial resolution of most GCMs (McCabe and Legates, 1992, Hay et al., 1992). In fact, before the advent of computers and the recent inventory of large-scale climate patters (e.g., ENSO, PDO, NAO, and PNA) and their teleconnections, meteorologists related large-scale precipitation regimes to the upper air-circulation, which cited as “developing synoptic patterns for precipitation” or “synoptic climatology of precipitation”. Synoptic pressure-contour patterns were related with precipitation over the western USA and adjacent areas (Smith, 1942, Klein, 1948, Stidd, 1954, and Malone et al. 1956); physical mechanism and associated moisture flow patterns are well explained by the calculation of the correlation fields between precipitation and synoptic patterns (Stidd, 1954, Klein, 1963,and Klein and Bloom 1987). In late 20th century, research focused on a much more regional scale level; many regions of the western USA’s precipitation are snow and snow accumulation on April1 represented as winter precipitation to investigate the impact of large-scale climate features and synoptic patterns.
Changnon et al. (1993) analyzed 1 April1 snowpack (for 35 winters) across the Rockies in association with 500-mb synoptic patterns to investigate the influence of climate variability in the spatial and temporal winter snowpack patterns. Seven different synoptic patterns observed in the winter and these explained nearly 90% of the non-average annual SN patterns. Each synoptic pattern represents a different airflow that provided precipitation opportunities for certain areas across the selected region. Trends in daily synoptic patterns suggested increasing trends that favor heavier precipitation in the Southwest and decreasing trends that favor lesser precipitation in the Northwest.
McCabe and Legates (1995) observed negative relationship between atmosphere circulation (winter mean 700-hPa height anomalies) and the temporal patterns of April1 snowpack in the western United States i.e., above average April1 snowpack observed with negative 700-hPa height anomalies over the eastern north pacific Ocean and the western United States. Negative 700-hPa height anomalies over the western USA are indicative of anomalous cyclonic circulation and a weakening of the ridge of high pressure that normally exit over the western US. The weakened ridge is indicative of enhanced zonal flow from the eastern North Pacific Ocean into the western USA. The increased zonal flow and anomalous cyclonic circulation cause anomalous westerly flows of moist air and storms from the eastern North pacific Ocean into the western USA; consequently increases winter precipitation and increase April 1 Snowpack. In contrast, below average April 1 Snowpack related to positive 700-hPa height anomalies over the western USA. Positive anomalies reflect anomalous anticyclonic circulation and strengthened ridge, which prevent intrusion of moisture air from the east North Pacific Ocean into the western USA. In addition, the increased atmospheric pressures over the western USA indicate anomalous subsidence of air, which leads to a drying and warming of the air and thereby by decreases winter precipitation. McCabe and Legates classified winter mean anomaly 700-hPa into five patterns and related with spatial patterns of April1 snowpack. These anomaly circulation patterns are in consistency with the mechanism that derived from the correlation approach. McCabe and Legates calculated long-term trends in the winter mean 700-hPa heights anomalies. Increased trends observed over the western Canada and northwestern USA whereas eastern North Pacific Ocean, south central USA and the Gulf of Mexico experienced decreasing trend in winter mean 700-hPa height anomalies; consequently, decreased and increased April1 snowpack observed in north western USA and in south-eastern part of the western USA respectively. Qualitatively similar trends observed by Changnon et al (1993). Cayan and Peterson (1989) found similar results between winter mean 700-hPa height anomalies and winter precipitation/December-August streamflows.
McCabe (1996) observed relationships between winter atmospheric circulation and streamflows. Negative correlations observed between winter mean 700-hpa height anomalies and average annual streamflow of the selected clusters, which are in consistent with relation between winter mean 700-hpa height anomalies and April1 snowpack (McCabe and Legates, 1995). East-central western USA streamflows exhibited weaker correlations. It is consistent with the results of the Cayan and Peterson (1989), Klein et al. (1965) and Weare and Hoeschele (1983) where they observed different behavior in the Rocky Mountains and the Great basin, unlike to the west coast, which is less explained by upper level geopotential height anomalies. Winter precipitation in these areas is significantly affected by other factors such as topography, surface heating, and surface friction (Cayan and Roads, 1984).
Above-mentioned literature described the relation between precipitation and local features of the circulation. Trends in the local circulation features are modulated by natural climate/atmosphere oscillation or (and) by large-scale climate signals’ teleconnections. In terms of Wallace and Gutzler (1987), teleconnections refer simultaneous correlations between temporal fluctuations in meteorological parameters at widely separated points on the earth.
Cayan (1996) observed influence of PNA and ENSO patterns in having anomalously low/high April1 SWEs in all five regions of the western USA, which were classified by Rotated Principal Component Analysis. However, regions in the interior west affected by several different winter storms and consequently exhibited weaker relations with seasonal mean atmospheric circulations. Clark et al. (2001) analyzed the El Nino and La Nina effects on seasonal snow pack evolution in the major sub basins of Columbia and Colorado River systems using Snow-water equivalent (SWE) data. The Influence of the ENSO events clearly appeared in the SWE by analyzing the seasonal SWE with respect to El Nino and La Nina years. In the Columbia River basin, El Nino is associated with a decrease in SWE and La Nina is associated with an increment of SWE. During El Nino, the Colorado River basin shows a transition between drier than average conditions in the north and wetter than average conditions in the southwest and associations during La Nina are generally opposite to those in El Nino years. McCabe and Dettinger (2002) analysis revealed PDO, not ENSO, as the primary driving force of variability in 1 April snowpack in the western United States. Many studies extended the investigation of climatic influences on hydro-climate variables by examining the temperature, precipitation, and streamflows in the western USA and found linear relations between SOI and winter precipitation (Redmond and Coach, 1991), and between winter mean sea level pressure and December-August streamflows totals in the western USA. McCabe and Dettinger (1999) observed variability among the above relations on a decadal scale
In this study, we analyzed Gunnison River Basin’s (GRB) winter snowpack and streamflows in association with synoptic and large-scale climate patterns. GRB is situated in the interior west where climate signal is not characterized by well known phenomenon (Cayan 1996, Cayan 1999, Clark et al. 2001, Klein 1963, Klein and Bloom 1987, McCabe and Legates 1995). Though past literature suggested linear relations between winter precipitation and winter mean 700-hPa height anomalies, regression fit of the same did not explain more than 50% variance of the precipitation (Klein 1963, McCabe 1994, McCabe and Legates, 1995). In this regard it is wise to look into other predictors which will increase the variance of the regression and by revealing the associated physical mechanism.
Data and Methodology:
The study Area
The Gunnison River Basin (Figure 1) situated in the southwest of the Colorado. U.S Geological Survey categorized GRB as a sub region, which resides in the Upper Colorado region and it includes 6 cataloging units i.e., East-Taylor (760 sq.mi), Upper Gunnison (2380 sq.mi), Tomichi (1090 sq.mi), North Fork (959 sq.mi), Lower Gunnison (1630 sq.mi), and Uncompahange (1110 sq.mi). The Gunnison River Basin (GRB) has a drainage area of approximately 20,534 km2 and basin elevations are extremely variable, ranging from 1387 to 4359 m (McCabe, 1994). The GRB has similar attributes like other basins in the western USA and it is considered as a major tributary of the Colorado River. The Gunnison River contributes approximately 42% of the streamflow of the Colorado River at the Colorado-Utah Stateline (Ugland et al, 1990). Large volumes of the GRB flows have great influence in various purposes i.e., municipal supply of water, recreational purpose, release of flows for endangered spices. Being a major contribution of the annual flows, spring flows’ forecasting couple of seasons ahead makes reservoirs’ operation easier in relation with various compacts and agreements e.g., Colorado River Compact and endangered spices.
This study analyzed streamflows, snow course locations’ data and climate variables.
Streamflow:
In this analysis, six streamflow locations are selected from the Hydro Climate Data Network (HCDN). This network, HCDN, developed by USGS to analyze the climate impacts on the rivers and it has more than 1000 streamflow stations across the conterminous USA that are not much affected by any kind of human activities (Slack and Landwehr 1992). Selected streamflow locations situated in five of the six cataloging units such that will represent the Gunnison River Basin. Selected locations’ have a continuous flow records from 1938 onwards.
Snow:
Snow Water Equivalent (SWE) data obtained from snow-course surveys conducted by the Natural Resourced Conservation Service (NRCS). SWE measurements are generally taken on or about the beginning of each month and these are most frequently taken at the beginning of the April, which is representative of peak SWE in many regions. In this study, 13 April 1st SWE locations considered and SWE measurements recorded at least 80% of the years during 1949-1999.
Climate Variables:
This study considered constant pressure patterns (Geo potential Height), Surface Air Temperature, and Wind patterns (zonal and meridional at 700mb height) to grab local scale climate variables and Sea Surface Temperatures to grab large-scale climate patterns. These variables recorded from 1948 onwards and available at Climate Diagnostics Center (www.cdc.noaa.gov).
Methodology:
Principal Component Analysis (PCA) is implemented to reduce the dimensionality and to obtain the dominant modes of the spring flows and SWE of the GRB. Then selected principal components (PCs) of the spring flows are correlated with the local and large-scale climate patterns to identify the predictors that are modulating spring flows. Regions that are highly correlated identified and averaged climate variables of the corresponding region are downloaded. In this way, several predictors of the spring flows are determined; and few of them could interrelate. Inter dependence among the predictors causes over fitting of the regression. In this regard, it is necessary to find the best optimized combination of the predictor set that explains much of the variance of the predictand and reduces the possibility of over fitting. This can be done in two ways: 1) by calculating statistical parameters (Cp, Adjusted R2, and GCV) for each combination and selecting the combination of the predictors that has best of the above-mentioned statistics 2) by doing a PCA analysis on the set of selected predictor set and selecting dominant principal components that explain much of the variance. Since PCA yields uncorrelated independent principal components one can use first few PCs as predictors or can select the best combination of the same by repeating the step 1. Generally, PCA approach is applicable if predictors are interrelated or if number of predictors are more. Finally, optimized predictor set will be obtained. PC1-flows simulated with selected predictors through semiparametric algorithm (Francisco Assis Souza Filho and Upmanu Lall, 2003) in association with bootstrap KNN (Lall and Sharma, 1996) and modified KNN methods (Prairie et al., 2004). Semiparametric method is a combination of parametric and non-parametric methods. Initially, it develops a best linear regression between the predictand and predictors and then utilizes the regressed information in the calculation of the neighbors i.e., using of the regressed coefficients as weights while implementing non-parametric methods. Two methods, KNN and modified KNN, simulate the possible flows by bootstrapping from the nearest neighbors. Unlike parametric method, regular KNN makes use of limited data that resembles the current state in a constructed embedding phase space. Modified KNN performs similar to KNN but it produces the values that have not seen historically by bootstrapping the residuals of the associated selected neighbors and adding then to the forecasted value through local polynomial regression.