Browse
Abstract
A simple—yet powerful—indicator for monitoring agricultural drought is the water requirement satisfaction index (WRSI). In data-sparse, food-insecure areas, the WRSI is used to guide billions of dollars of aid every year. The WRSI uses precipitation (PPT) and reference evapotranspiration (RefET) data to estimate water availability relative to water demand experienced over the course of a growing season. If the season is in progress, to-date conditions can be combined with climatological averages to provide insight into potential end-of-season (EOS) crop performance. However, if the average is misrepresented, these forecasts can hinder early warning and delay precious humanitarian aid. While many agencies use arithmetic average climatologies as proxies for “average conditions,” little published research evaluates their effectiveness in crop-water balance models. Here, we use WRSI hindcasts of three African regions’ growing seasons, from 1981 to 2019, to assess the adequacy of the arithmetic mean climatological forecast—the Extended WRSI. We find that the Extended WRSI is positively biased, overestimating the actual EOS WRSI by 2%–23% in East, West, and southern Africa. The presented alternative combines to-date conditions with data from previous seasons to produce a series of historically realistic conclusions to the current season. The mean of these scenarios is the WRSI Outlook. In comparison with the Extended WRSI, which creates a single forecast scenario using average inputs that are not covarying, the WRSI Outlook employs an ensemble of scenarios, which more adequately capture the historical distribution of distribution of rainfall events along with the covariability between climate variables. More specifically, the impact of dry spells in individual years is included in the WRSI Outlook in a way that is smoothed over in the Extended WRSI. We find that the WRSI Outlook has a near-zero bias score and generally has a lower RMSE. In total, this paper highlights the inadequacies of the arithmetic mean climatological forecast and presents a less biased and more accurate scenario-based approach. To this end, the WRSI Outlook can improve our ability to identify agricultural drought and the concomitant need for humanitarian aid.
Abstract
A simple—yet powerful—indicator for monitoring agricultural drought is the water requirement satisfaction index (WRSI). In data-sparse, food-insecure areas, the WRSI is used to guide billions of dollars of aid every year. The WRSI uses precipitation (PPT) and reference evapotranspiration (RefET) data to estimate water availability relative to water demand experienced over the course of a growing season. If the season is in progress, to-date conditions can be combined with climatological averages to provide insight into potential end-of-season (EOS) crop performance. However, if the average is misrepresented, these forecasts can hinder early warning and delay precious humanitarian aid. While many agencies use arithmetic average climatologies as proxies for “average conditions,” little published research evaluates their effectiveness in crop-water balance models. Here, we use WRSI hindcasts of three African regions’ growing seasons, from 1981 to 2019, to assess the adequacy of the arithmetic mean climatological forecast—the Extended WRSI. We find that the Extended WRSI is positively biased, overestimating the actual EOS WRSI by 2%–23% in East, West, and southern Africa. The presented alternative combines to-date conditions with data from previous seasons to produce a series of historically realistic conclusions to the current season. The mean of these scenarios is the WRSI Outlook. In comparison with the Extended WRSI, which creates a single forecast scenario using average inputs that are not covarying, the WRSI Outlook employs an ensemble of scenarios, which more adequately capture the historical distribution of distribution of rainfall events along with the covariability between climate variables. More specifically, the impact of dry spells in individual years is included in the WRSI Outlook in a way that is smoothed over in the Extended WRSI. We find that the WRSI Outlook has a near-zero bias score and generally has a lower RMSE. In total, this paper highlights the inadequacies of the arithmetic mean climatological forecast and presents a less biased and more accurate scenario-based approach. To this end, the WRSI Outlook can improve our ability to identify agricultural drought and the concomitant need for humanitarian aid.
Abstract
We examine polarimetric rainfall estimates of extreme rainfall through intercomparisons of radar rainfall estimates with rainfall observations from a dense network of rain gauges in Kansas City. The setting provides unique capabilities for examining range dependence in polarimetric rainfall estimates due to the overlapping coverage of the Kansas City, Missouri, and Topeka, Kansas, WSR-88D radars. We focus on polarimetric measurements of specific differential phase shift, K DP, for estimating extreme rainfall. Gauge–radar intercomparisons from the “close-range” Kansas City radar and from the “far-range” Topeka radar show that K DP can provide major improvements in estimating extreme rainfall, but the advantages of K DP rainfall estimates diminish with range. Storm-to-storm variability of multiplicative bias remains an important issue for polarimetric rainfall estimates; variability in bias is comparable at both close and far range from the radar. “Conditional bias,” in which peak radar rainfall estimates are lower than rain gauge observations, is a systematic feature of polarimetric rainfall estimates, but is more severe at far range. The Kansas City region has experienced record flooding in urban watersheds since the polarimetric upgrade of the Kansas City and Topeka radars in 2012. Polarimetric rainfall estimates from the far-range Topeka radar provide useful quantitative information on basin-average rainfall, but the ability to resolve spatial variation of the most extreme rain rates diminishes significantly with range from the radar.
Abstract
We examine polarimetric rainfall estimates of extreme rainfall through intercomparisons of radar rainfall estimates with rainfall observations from a dense network of rain gauges in Kansas City. The setting provides unique capabilities for examining range dependence in polarimetric rainfall estimates due to the overlapping coverage of the Kansas City, Missouri, and Topeka, Kansas, WSR-88D radars. We focus on polarimetric measurements of specific differential phase shift, K DP, for estimating extreme rainfall. Gauge–radar intercomparisons from the “close-range” Kansas City radar and from the “far-range” Topeka radar show that K DP can provide major improvements in estimating extreme rainfall, but the advantages of K DP rainfall estimates diminish with range. Storm-to-storm variability of multiplicative bias remains an important issue for polarimetric rainfall estimates; variability in bias is comparable at both close and far range from the radar. “Conditional bias,” in which peak radar rainfall estimates are lower than rain gauge observations, is a systematic feature of polarimetric rainfall estimates, but is more severe at far range. The Kansas City region has experienced record flooding in urban watersheds since the polarimetric upgrade of the Kansas City and Topeka radars in 2012. Polarimetric rainfall estimates from the far-range Topeka radar provide useful quantitative information on basin-average rainfall, but the ability to resolve spatial variation of the most extreme rain rates diminishes significantly with range from the radar.
Abstract
A comprehensive ecohydrological analysis is designed to understand the formation and evolution of lake Lop Nur and the environmental change over the Tarim River basin. Three temporal scales from century-based climatological mean to decade-based quasi-steady state change and to annual-scale-based abrupt change test are included. Combining the Budyko and Tomer–Schilling framework, this research first analyzes hydroclimatic and ecohydrological resistance/resilience conditions, then attributes observed changes to external/climate impact or to internal/anthropogenic activities, and finally diagnoses the possible tipping point on ecohydrological dynamics. (i) The arid regions reveal less sensitivity in terms of low variabilities of excess water W and energy U and show high resilience, which will more likely stay the same pattern in the future. (ii) Present towns situated in the semiarid regions with a natural hydraulic linkage with the mainstream of the Tarim River show a higher sensitivity and likelihood to be affected if drier scenarios occurred in the future. (iii) The attribution from two subsequent quasi-steady states indicates increasing effects of the anthropogenic activities increase (1961–80 versus 1981–2000) and provides climatical evidence that the central Tarim River basin was getting wetter before 1960 and then kept drying afterward. (iv) In the Kongqi subcatchment, the excess water reveals a significant decrease-then-increase evolution, from which the years with abrupt changes are observed in the 1960s. Generally, both points iii and iv are in agreement with that the closed-basin lake Lop Nur desiccation until the 1970s and its connection with the eastern part of the Taklamakan Sand Sea.
Abstract
A comprehensive ecohydrological analysis is designed to understand the formation and evolution of lake Lop Nur and the environmental change over the Tarim River basin. Three temporal scales from century-based climatological mean to decade-based quasi-steady state change and to annual-scale-based abrupt change test are included. Combining the Budyko and Tomer–Schilling framework, this research first analyzes hydroclimatic and ecohydrological resistance/resilience conditions, then attributes observed changes to external/climate impact or to internal/anthropogenic activities, and finally diagnoses the possible tipping point on ecohydrological dynamics. (i) The arid regions reveal less sensitivity in terms of low variabilities of excess water W and energy U and show high resilience, which will more likely stay the same pattern in the future. (ii) Present towns situated in the semiarid regions with a natural hydraulic linkage with the mainstream of the Tarim River show a higher sensitivity and likelihood to be affected if drier scenarios occurred in the future. (iii) The attribution from two subsequent quasi-steady states indicates increasing effects of the anthropogenic activities increase (1961–80 versus 1981–2000) and provides climatical evidence that the central Tarim River basin was getting wetter before 1960 and then kept drying afterward. (iv) In the Kongqi subcatchment, the excess water reveals a significant decrease-then-increase evolution, from which the years with abrupt changes are observed in the 1960s. Generally, both points iii and iv are in agreement with that the closed-basin lake Lop Nur desiccation until the 1970s and its connection with the eastern part of the Taklamakan Sand Sea.
Abstract
The control of latent heat flux (LE) by soil moisture (SM) is a key process affecting the moisture and energy budgets at the land–atmosphere interface. SM–LE coupling relationships are conventionally examined using metrics involving temporal correlation. However, such a traditional linear approach, which fits a straight line across the full SM–LE space to evaluate the dependency, leaves out certain critical information: nonlinear SM–LE relationships and the long-recognized thresholds that lead to dramatically different behavior in different ranges of soil moisture, delineating a dry regime, a transitional regime of high sensitivity, and a wet (energy-limited) regime. Using data from climate models, reanalyses, and observationally constrained datasets, global patterns of SM–LE regimes are determined by segmented regression. Mutual information analysis is applied only for days when SM is in the transitional regime between critical points defining high sensitivity of LE to SM variations. Sensitivity is further decomposed into linear and nonlinear components. Results show discrepancies in the global patterns of existing SM regimes, but general consistencies among the linear and nonlinear components of SM–LE coupling. This implies that although models simulate differing surface hydroclimates, once SM is in the transitional regime, the locations where LE closely interacts with SM are well captured and resemble the conventional distribution of “hotspots” of land–atmosphere interactions. This indicates that only the transitional SM regime determines the strength of coupling, and attention should focused on when this regime occurs. This framework can also be applied to investigate extremes and the shifting surface hydroclimatology in a warming climate.
Significance Statement
Evaporation is sensitive to soil moisture only within a specific range that is neither too dry nor too wet. This transitional regime is examined to quantify how strongly soil moisture controls local evaporation. We identify the dry, transitional, and wet regimes across the globe, and the locations where each regime is experienced; the spatial patterns among climate models and observationally based datasets often show discrepancies. When we determine dependencies between soil moisture and evaporation only within the transitional regime, we find general consistency of locations having simple linear dependencies versus more complex nonlinear relationships. We conclude that although surface hydroclimates differ between climate models and observations, the locations where soil moisture can control evaporation are well captured. These results have potential application for improved forecasting and climate change assessment.
Abstract
The control of latent heat flux (LE) by soil moisture (SM) is a key process affecting the moisture and energy budgets at the land–atmosphere interface. SM–LE coupling relationships are conventionally examined using metrics involving temporal correlation. However, such a traditional linear approach, which fits a straight line across the full SM–LE space to evaluate the dependency, leaves out certain critical information: nonlinear SM–LE relationships and the long-recognized thresholds that lead to dramatically different behavior in different ranges of soil moisture, delineating a dry regime, a transitional regime of high sensitivity, and a wet (energy-limited) regime. Using data from climate models, reanalyses, and observationally constrained datasets, global patterns of SM–LE regimes are determined by segmented regression. Mutual information analysis is applied only for days when SM is in the transitional regime between critical points defining high sensitivity of LE to SM variations. Sensitivity is further decomposed into linear and nonlinear components. Results show discrepancies in the global patterns of existing SM regimes, but general consistencies among the linear and nonlinear components of SM–LE coupling. This implies that although models simulate differing surface hydroclimates, once SM is in the transitional regime, the locations where LE closely interacts with SM are well captured and resemble the conventional distribution of “hotspots” of land–atmosphere interactions. This indicates that only the transitional SM regime determines the strength of coupling, and attention should focused on when this regime occurs. This framework can also be applied to investigate extremes and the shifting surface hydroclimatology in a warming climate.
Significance Statement
Evaporation is sensitive to soil moisture only within a specific range that is neither too dry nor too wet. This transitional regime is examined to quantify how strongly soil moisture controls local evaporation. We identify the dry, transitional, and wet regimes across the globe, and the locations where each regime is experienced; the spatial patterns among climate models and observationally based datasets often show discrepancies. When we determine dependencies between soil moisture and evaporation only within the transitional regime, we find general consistency of locations having simple linear dependencies versus more complex nonlinear relationships. We conclude that although surface hydroclimates differ between climate models and observations, the locations where soil moisture can control evaporation are well captured. These results have potential application for improved forecasting and climate change assessment.
Abstract
Knowledge about precipitation generation remains limited in the tropical Andes due to the lack of water stable isotope (WSI) data. Therefore, we investigated the key factors controlling the isotopic composition of precipitation in the Páramo highlands of southern Ecuador using event-based (high frequency) WSI data collected between November 2017 and October 2018. Our results show that air masses reach the study site preferentially from the eastern flank of the Andes through the Amazon basin (73.2%), the Orinoco plains (11.2%), and the Mato Grosso Massif (2.7%), whereas only a small proportion stems from the Pacific Ocean (12.9%). A combination of local and regional factors influences the δ 18O isotopic composition of precipitation. Regional atmospheric features (Atlantic moisture, evapotranspiration over the Amazon rainforest, continental rain-out, and altitudinal lapse rates) are what largely control the meteoric δ 18O composition. Local precipitation, temperature, and the fraction of precipitation corresponding to moderate to heavy rainfalls are also key features influencing isotopic ratios, highlighting the importance of localized convective precipitation at the study site. Contrary to δ 18O, d-excess values showed little temporal variation and could not be statistically linked to regional or local hydrometeorological features. The latter reveals that large amounts of recycled moisture from the Amazon basin contribute to local precipitation regardless of season and predominant trajectories from the east. Our findings will help to improve isotope-based climatic models and enhance paleoclimate reconstructions in the southern Ecuador highlands.
Abstract
Knowledge about precipitation generation remains limited in the tropical Andes due to the lack of water stable isotope (WSI) data. Therefore, we investigated the key factors controlling the isotopic composition of precipitation in the Páramo highlands of southern Ecuador using event-based (high frequency) WSI data collected between November 2017 and October 2018. Our results show that air masses reach the study site preferentially from the eastern flank of the Andes through the Amazon basin (73.2%), the Orinoco plains (11.2%), and the Mato Grosso Massif (2.7%), whereas only a small proportion stems from the Pacific Ocean (12.9%). A combination of local and regional factors influences the δ 18O isotopic composition of precipitation. Regional atmospheric features (Atlantic moisture, evapotranspiration over the Amazon rainforest, continental rain-out, and altitudinal lapse rates) are what largely control the meteoric δ 18O composition. Local precipitation, temperature, and the fraction of precipitation corresponding to moderate to heavy rainfalls are also key features influencing isotopic ratios, highlighting the importance of localized convective precipitation at the study site. Contrary to δ 18O, d-excess values showed little temporal variation and could not be statistically linked to regional or local hydrometeorological features. The latter reveals that large amounts of recycled moisture from the Amazon basin contribute to local precipitation regardless of season and predominant trajectories from the east. Our findings will help to improve isotope-based climatic models and enhance paleoclimate reconstructions in the southern Ecuador highlands.
Abstract
Surface meteorological analyses serve a wide range of research and applications, including forcing inputs for hydrological and ecological models, climate analysis, and resource and emergency management. Quantifying uncertainty in such analyses would extend their utility for probabilistic hydrologic prediction and climate risk applications. With this motivation, we enhance and evaluate an approach for generating ensemble analyses of precipitation and temperature through the fusion of station observations, terrain information, and numerical weather prediction simulations of surface climate fields. In particular, we expand a spatial regression in which static terrain attributes serve as predictors for spatially distributed 1/16° daily surface precipitation and temperature by including forecast outputs from the High-Resolution Rapid Refresh (HRRR) numerical weather prediction model as additional predictors. We demonstrate the approach for a case study domain of California, focusing on the meteorological conditions leading to the 2017 flood and spillway failure event at Lake Oroville. The approach extends the spatial regression capability of the Gridded Meteorological Ensemble Tool (GMET) and also adds cross validation to the uncertainty estimation component, enabling the use of predictive rather than calibration uncertainty. In evaluation against out-of-sample station observations, the HRRR-based predictors alone are found to be skillful for the study setting, leading to overall improvements in the enhanced GMET meteorological analyses. The methodology and associated tool represent a promising method for generating meteorological surface analyses for both research-oriented and operational applications, as well as a general strategy for merging in situ and gridded observations.
Abstract
Surface meteorological analyses serve a wide range of research and applications, including forcing inputs for hydrological and ecological models, climate analysis, and resource and emergency management. Quantifying uncertainty in such analyses would extend their utility for probabilistic hydrologic prediction and climate risk applications. With this motivation, we enhance and evaluate an approach for generating ensemble analyses of precipitation and temperature through the fusion of station observations, terrain information, and numerical weather prediction simulations of surface climate fields. In particular, we expand a spatial regression in which static terrain attributes serve as predictors for spatially distributed 1/16° daily surface precipitation and temperature by including forecast outputs from the High-Resolution Rapid Refresh (HRRR) numerical weather prediction model as additional predictors. We demonstrate the approach for a case study domain of California, focusing on the meteorological conditions leading to the 2017 flood and spillway failure event at Lake Oroville. The approach extends the spatial regression capability of the Gridded Meteorological Ensemble Tool (GMET) and also adds cross validation to the uncertainty estimation component, enabling the use of predictive rather than calibration uncertainty. In evaluation against out-of-sample station observations, the HRRR-based predictors alone are found to be skillful for the study setting, leading to overall improvements in the enhanced GMET meteorological analyses. The methodology and associated tool represent a promising method for generating meteorological surface analyses for both research-oriented and operational applications, as well as a general strategy for merging in situ and gridded observations.
Abstract
The efforts to develop a hydrologic model-based operational streamflow forecast in India are limited. We evaluate the role of bias correction of meteorological forecasts and streamflow postprocessing on hydrological prediction skill in India. We use the Variable Infiltration Capacity (VIC) model to simulate runoff and root-zone soil moisture in the Narmada basin (drainage area: 97 410 km2), which was used as a testbed to examine the forecast skill along with the observed streamflow. We evaluated meteorological and hydrological forecasts during the monsoon (June–September) season for the 2000–18 period. The raw meteorological forecast displayed relatively low skill against the observed precipitation at 1–3-day lead time during the monsoon season. Similarly, the forecast skill was low with mean normalized root-mean-square error (NRMSE) more than 0.9 and mean absolute bias larger than 60% for extreme precipitation at the 1–3-day lead time. We used empirical quantile mapping (EQM) to bias-correct precipitation forecasts. The bias correction of precipitation forecasts resulted in significant improvement in the precipitation forecast skill. Runoff and root-zone soil moisture forecasts were also significantly improved due to bias correction of precipitation forecasts where the forecast evaluation is performed against the reference model run. However, bias correction of precipitation forecasts did not cause considerable improvement in the streamflow prediction. Bias correction of streamflow forecasts performs better than the streamflow forecasts simulated using the bias-corrected meteorological forecast. The combination of the bias correction of precipitation forecasts and postprocessing of streamflow resulted in a significant improvement in the streamflow prediction (reduction in bias from 40% to 5%).
Abstract
The efforts to develop a hydrologic model-based operational streamflow forecast in India are limited. We evaluate the role of bias correction of meteorological forecasts and streamflow postprocessing on hydrological prediction skill in India. We use the Variable Infiltration Capacity (VIC) model to simulate runoff and root-zone soil moisture in the Narmada basin (drainage area: 97 410 km2), which was used as a testbed to examine the forecast skill along with the observed streamflow. We evaluated meteorological and hydrological forecasts during the monsoon (June–September) season for the 2000–18 period. The raw meteorological forecast displayed relatively low skill against the observed precipitation at 1–3-day lead time during the monsoon season. Similarly, the forecast skill was low with mean normalized root-mean-square error (NRMSE) more than 0.9 and mean absolute bias larger than 60% for extreme precipitation at the 1–3-day lead time. We used empirical quantile mapping (EQM) to bias-correct precipitation forecasts. The bias correction of precipitation forecasts resulted in significant improvement in the precipitation forecast skill. Runoff and root-zone soil moisture forecasts were also significantly improved due to bias correction of precipitation forecasts where the forecast evaluation is performed against the reference model run. However, bias correction of precipitation forecasts did not cause considerable improvement in the streamflow prediction. Bias correction of streamflow forecasts performs better than the streamflow forecasts simulated using the bias-corrected meteorological forecast. The combination of the bias correction of precipitation forecasts and postprocessing of streamflow resulted in a significant improvement in the streamflow prediction (reduction in bias from 40% to 5%).
Abstract
In many parts of the world, humans rely on afternoon rainfall for their water supply. However, it is not fully understood how land surface properties influence afternoon precipitation. In fact, disagreement remains regarding the relative prevalence of “wet-soil advantage” regimes, in which wet soils receive more precipitation than do dry soils, and “dry-soil advantage” regimes, in which the opposite occurs. Recent studies have proposed that the permanent wilting point (PWP) soil moisture threshold influences the location and organization of convective clouds. Motivated by this work, we investigate how changes in soil moisture relative to the PWP affect the timing and amount of surface rainfall, as well as how this response depends on the presence or absence of vegetation. This investigation is carried out by conducting several series of high-resolution, idealized numerical experiments using a fully coupled, interactive soil–vegetation–atmosphere modeling system. From these experiments, a new soil moisture–precipitation relationship emerges: in the presence of vegetation, simulations with moderately dry soils, whose initial liquid water content slightly exceeds the PWP, generate significantly less surface precipitation than do those with the driest or wettest soils. This result suggests that simulated wet-soil advantage and dry-soil advantage regimes may not necessarily be mutually exclusive, insofar as extremely wet and extremely dry soils can both exhibit an advantage over moderately dry soils. This nonmonotonic soil moisture–precipitation relationship is found to result from the PWP’s modulation of transpiration of water vapor by plants. In the absence of vegetation, a wet-soil advantage occurs instead in these idealized simulations.
Significance Statement
This modeling study suggests a new type of rainfall response to soil moisture in which intermediate-moisture soils receive less rainfall than do the driest or wettest soils. Previous studies have suggested that afternoon rainfall, which impacts populations across the globe, consistently increases or decreases with increasing soil moisture; our results suggest that this relationship can instead be nonmonotonic under certain conditions. This nonmonotonic response only occurs in the presence of vegetation, suggesting that plants play a key role in determining the soil moisture dependence of afternoon precipitation. In examining the mechanisms behind these trends, we shed light on interactions between soil, vegetation, the boundary layer, and clouds that coarse-resolution models may fail to capture.
Abstract
In many parts of the world, humans rely on afternoon rainfall for their water supply. However, it is not fully understood how land surface properties influence afternoon precipitation. In fact, disagreement remains regarding the relative prevalence of “wet-soil advantage” regimes, in which wet soils receive more precipitation than do dry soils, and “dry-soil advantage” regimes, in which the opposite occurs. Recent studies have proposed that the permanent wilting point (PWP) soil moisture threshold influences the location and organization of convective clouds. Motivated by this work, we investigate how changes in soil moisture relative to the PWP affect the timing and amount of surface rainfall, as well as how this response depends on the presence or absence of vegetation. This investigation is carried out by conducting several series of high-resolution, idealized numerical experiments using a fully coupled, interactive soil–vegetation–atmosphere modeling system. From these experiments, a new soil moisture–precipitation relationship emerges: in the presence of vegetation, simulations with moderately dry soils, whose initial liquid water content slightly exceeds the PWP, generate significantly less surface precipitation than do those with the driest or wettest soils. This result suggests that simulated wet-soil advantage and dry-soil advantage regimes may not necessarily be mutually exclusive, insofar as extremely wet and extremely dry soils can both exhibit an advantage over moderately dry soils. This nonmonotonic soil moisture–precipitation relationship is found to result from the PWP’s modulation of transpiration of water vapor by plants. In the absence of vegetation, a wet-soil advantage occurs instead in these idealized simulations.
Significance Statement
This modeling study suggests a new type of rainfall response to soil moisture in which intermediate-moisture soils receive less rainfall than do the driest or wettest soils. Previous studies have suggested that afternoon rainfall, which impacts populations across the globe, consistently increases or decreases with increasing soil moisture; our results suggest that this relationship can instead be nonmonotonic under certain conditions. This nonmonotonic response only occurs in the presence of vegetation, suggesting that plants play a key role in determining the soil moisture dependence of afternoon precipitation. In examining the mechanisms behind these trends, we shed light on interactions between soil, vegetation, the boundary layer, and clouds that coarse-resolution models may fail to capture.
Abstract
Anthropogenic climate change–induced snowpack loss is affecting streamflow predictability, as it becomes less dependent on the initial snowpack conditions and more dependent on meteorological forecasts. We assess future changes to seasonal streamflow predictability over two large river basins, Liard and Athabasca in western Canada, by approximating streamflow response from the Variable Infiltration Capacity (VIC) hydrologic model with the Bayesian regularized neutral network (BRNN) machine learning emulator. We employ the BRNN emulator in a testbed ensemble streamflow prediction system by treating VIC-simulated snow water equivalent (SWE) as a known predictor and precipitation and temperature from GCMs as ensemble forecasts, thereby isolating the effect of SWE on streamflow predictability. We assess warm-season mean and maximum flow predictability over 2041–70 and 2071–2100 future periods against the1981–2010 historical period. The results indicate contrasting patterns of change, with the predictive skills for mean flow generally declining for the two basins, and marginally increasing or decreasing for the headwater subbasins. The predictive skill for maximum flow declines for the relatively warmer Athabasca basin and improves for the colder Liard basin and headwater subbasins. While the decreasing skill for the Athabasca is attributable to substantial loss in SWE, the improvement for the Liard and headwaters can be attributed to an earlier maximum flow timing that reduces the forecast horizon and offsets the effect of SWE loss. Overall, while the future change in SWE does affect the streamflow prediction skill, the loss of SWE alone is not a sufficient condition for the reduction in streamflow predictability.
Significance Statement
The purpose of this study is to evaluate potential changes in seasonal streamflow predictability in relation to snowpack change under future climate. This is highly relevant because snowpack storage provides a means of predicting available freshet water supply, as well as peak flow events in cold regions. We use a machine learning model as an emulator of a hydrologic model in a testbed ensemble prediction system. Our results provide insights on hydroclimatic controls and interactions that affect future streamflow predictability across two river basins in western Canada. We conclude that besides snowpack, predictability depends on a number of other factors (basin/subbasin characteristics, streamflow variables, and future periods), and the loss of snowpack alone is not a sufficient condition for the reduction in streamflow predictability.
Abstract
Anthropogenic climate change–induced snowpack loss is affecting streamflow predictability, as it becomes less dependent on the initial snowpack conditions and more dependent on meteorological forecasts. We assess future changes to seasonal streamflow predictability over two large river basins, Liard and Athabasca in western Canada, by approximating streamflow response from the Variable Infiltration Capacity (VIC) hydrologic model with the Bayesian regularized neutral network (BRNN) machine learning emulator. We employ the BRNN emulator in a testbed ensemble streamflow prediction system by treating VIC-simulated snow water equivalent (SWE) as a known predictor and precipitation and temperature from GCMs as ensemble forecasts, thereby isolating the effect of SWE on streamflow predictability. We assess warm-season mean and maximum flow predictability over 2041–70 and 2071–2100 future periods against the1981–2010 historical period. The results indicate contrasting patterns of change, with the predictive skills for mean flow generally declining for the two basins, and marginally increasing or decreasing for the headwater subbasins. The predictive skill for maximum flow declines for the relatively warmer Athabasca basin and improves for the colder Liard basin and headwater subbasins. While the decreasing skill for the Athabasca is attributable to substantial loss in SWE, the improvement for the Liard and headwaters can be attributed to an earlier maximum flow timing that reduces the forecast horizon and offsets the effect of SWE loss. Overall, while the future change in SWE does affect the streamflow prediction skill, the loss of SWE alone is not a sufficient condition for the reduction in streamflow predictability.
Significance Statement
The purpose of this study is to evaluate potential changes in seasonal streamflow predictability in relation to snowpack change under future climate. This is highly relevant because snowpack storage provides a means of predicting available freshet water supply, as well as peak flow events in cold regions. We use a machine learning model as an emulator of a hydrologic model in a testbed ensemble prediction system. Our results provide insights on hydroclimatic controls and interactions that affect future streamflow predictability across two river basins in western Canada. We conclude that besides snowpack, predictability depends on a number of other factors (basin/subbasin characteristics, streamflow variables, and future periods), and the loss of snowpack alone is not a sufficient condition for the reduction in streamflow predictability.
