Papers

(i) How did the concept of Water Footprint develop in the 10–15 years in which it started being used worldwide?

(ii) Is it a reliable indicator? Are there any limits to its application?

(iii) What are the limits of studies carried out till present?

(iv) How can its application and reliability be improved?

1. Introduction

2. Methods

2.1. Water Footprint definition

2.1.1 Green, blue, and grey water

The concepts of green, blue, and grey water, how they are calculated, and the context of agricultural v.s. domestic and industrial sectors.

2.2. Methodology adopted

Reviewed 96 scientific papers (until Nov 2015), and found that most of the studies until 2008 focuses on the global agricultural sector. So the authors focused on the crops in their literature analysis. 

Water footprint consists of three components: (1) consumptive use of rain water (green water); (2) consumptive use of surface water and ground water (blue water); (3) pollution of water (grey water). It can be defined for a product, a group of people, or an economic sector (Hoekstra 2008).

Summary:

This paper focuses on water footprint assessment methods in the agricultural and industrial sectors. A literature synthesis is performed in order to identify emerging issues on freshwater resources’ management for agrifood products, and a hierachical decision-making framework for water footprint management in agrifood supply chains is proposed. 

Literature synthesis - water footprint assessment for agricultural and/or industrial products (operations):

- Scope: studies that focuses on freshwater consumption and pollution in various stages of a product’s life cycle, i.e. a supply chain perspective

Water footprint management for agrifood supply chains:

- Where & How does the heaviest water footprint occur during the supply chain and how to mitigate them?

Ackerman & Fisher 2013 Is there a water–energy nexus in electricity generation? Long-term scenarios for the western United States

The paper investigated four different scenarios of long-term electricity generation for the western United States:  (1) No new constraints (2) CO2 emission limit (3) Water use limit (4) Both CO2 emission and water use limits Then, CO2 prices and water consumption prices were imposed to find which prices make each scenario cost-effective. Cost-effectiveness in this paper means how much must the saved CO2 and water worth, to make each scenario the least-cost one. It does not mean that the extra cost is unnecessary - given environmental constraints. But in the context of this paper (western United States), the saved amount of water was indeed too small (<1% of the state’s total consumption) to be worth it or necessary. 

2. Model design and 3. Electricity data and assumptions The model simulates the state-level demand and supply of the Western Electric Coordinating Council (WECC) region during 2008-2100, and was used to examine the electricity cost, CO2 emissions, and water use during this period. - Electricity demand is driven by population, temperature changes, and assumed increase or no-increase in  energy efficiency. - Electricity supply is modeled: ---- Scenario assumptions on fuel mix are specified by users instead of being determined by optimization. ---- The demand, transmission & distribution losses, and fuel mix are used to estimate the required annual generation for the WECC region. ---- The required generation is distributed into three sub-regions, the appropriate resource types, and individual states in proportion to the assumed fuel mix, current generation in 2008, and current distribution of the fuel types. ---- Existing power plants' generation, fuel use, water consumption, and economic performance are tracked over time (using 2008 data). Old power plants are retired or de-rated if the existing supply exceeds the required generation for that fuel type, starting with the oldest. New power plants of each fuel type are built to meet demand as needed. (I think) their water consumption rate uses the average of the current rate of the same cooling system type. Their fuel use is calculated by assuming the same capacity factors as the average capacity factor of same-type existing power plants in 2010 (with CCS imposing a 35% reduction in net generation). Cooling system types are gradually retrofitted, and CCS are retrofitted in some scenarios. The CO2 emissions of the power plants are also tracked using the current average emission rates. - Resource costs are modeled: ---- The fuel use of the power plants are calculated above. The fuel costs are at the cost of these fuels in 2008. ---- The maintenance and generation costs of current power plants, the costs of building new resources, and the avoided costs due to retiring old power plants are tracked. The retrofit costs of cooling systems and CCS are tracked. See paper for specific values. ---- Energy efficiency initiatives are assumed to cost $0.045/kWh. - Outputs are saved by annual time step.

4. Scenario definitions (Each scenario was modeled with efficiency and without efficiency, because the effect of efficiency tends to dominate over other factors)

Scenario A/AE. No new constraints - The fuel mix (by percentage_ of the Annual Energy Outlook 2010 was maintained until 2035, after which the same percentages were used until 2100. - No specific meausres were taken by the electricity sector to reduced CO2 or water consumption. - Temperature trajectory follows IPCC SRES A2.

Scenario B/BE. Water use limit - Fuel mix are the same as Scenario A/AE. - "All once-through cooling systems were retrofitted by cooling towers by 2030. Half the coal and biomass units were retrofitted with dry cooling by 2050, and the rest by 2075. Half the gas units were retrofitted with dry cooling by 2030, and the rest by 2050. Nuclear was shifted towards hybrid wet/dry cooling by 2050." - Temperature trajectory follows IPCC SRES A2.

Scenario C/CE. CO2 emission limit - Fuel mix in C has reduced coal generation, increased nuclear generation and non-hydro renewable generation (see paper for specific values), and CCS technology was retrofitted to 65% of coal generation by 2050 and 100% by 2100. - Fuel mix in CE has high penetration of renewable energy and natural gas. All coal power plants were retired by 2050. Natural gas consumption was reduced by 50% between 2050 and 2100. Non-hydro renewable penetration was increased to 46% of generation by 2100. - No specific move was made to reduce water consumption. - Temperature trajectory follows IPCC SRES B1.

Scenario D/DE. Both water and CO2 emission limit - Fuel mix and CCS adoption in D follows Scenario C. Fuel mix in DE follows Scenario CE. - Cooling system retrofits follow Scenario B. - (I guess) temperature trajectory follows IPCC SRES B1.

5. Results Table 1 shows the electricity generation, CO2 emissions, and water consumptions of all eight scenarios. The efficiency scenarios cut all values by nearly half compared to the non-efficiency scenarios. The CO2 emissions of C and D were about 1/3 of that of A and B, the water consumption of C was a little higher than A (due to the CCS), the water consumption of B was less than 1/5 of that of A, and the water consumption of D was about 1/3 of that of A. In order to investigate the economic feasibility of these different scenarios, the total capital and operating costs of the electricity system during 2008-2100 were calculated using a discount rate of 5.4%. The lowest-cost scenarios were A and AE, which might be expected. Then, the "shadow prices" of water and CO2 were calculated. Different externality prices of water consumption and CO2 emission were added to the costs of each scenario, and the lowest-cost scenario was identified for each combination of the two externality prices. The results showed that scenarios C/CE and D/DE had the least cost when the externality price of CO2 was above $50-$70 / ton CO2, and scenarios B/BE and D/DE had the least cost when the externality price of water was above $4500-$13000 / acre-foot (Fig. 2 and Fig. 3). The cost of retrofitting all the cooling systems was clearly not worth it given that ocean desalination costs less than $3000 / acre-foot. Besides, the electricity sector's water consumption was small, the saved water between the least and most water-intensive scenarios being <1% of the region's total consumption.

Moore et al. 2015 A High Spatiotemporal Assessment of Consumptive Water Use and Water Scarcity in the Conterminous United States

 The study developed a temporal-spatial disaggregation scheme for the USGS water-consumption data (1985, 1990, 1995, and 2000) in the irrigation/agriculture, domestic, industrial, thermoelectric, livestock and mining sectors. The disaggregation was from county-scale to 1/8o resolution.

The disaggregation method was preliminary but all the water-use categories were covered. It was a good study demonstrating that more detailed spatial-temporal assessment of water scarcity can reveal better information to water management. But the calculated water scarcity at 1/8o resolution was probably better not taken at literal value.

 Methodology

1. Spatial disaggregation of the water consumption

(1) Domestic: proportional to population; population was assumed to be evenly distributed within each census block

(2) Thermoelectric: proportional to the number of thermoelectric power plants in each grid cell (using EIA 2010 data); all power plants were assumed to have equal consumptive water use

(3) Industrial, irrigation, livestock, mining: proportional to the remotely-sensed land use/land cover data in each grid cell

2. Temporal disaggregation of the water consumption

(1) Thermoelectric: based on cooling and heating degree days of the apparent temperature

(2) Domestic and livestock: based on cooling degree days of the apparent temperature

(3) Irrigation: First, the growing season was determined using either the growing season length or the growing season index method. Then, the weights within the growing season was calculated using the degree day, the inverse aridity index, or the water deficit proportion methods

(4) Industrial and mining: No seasonal variation assumed

3. The water scarcity index

Defined as consumptive water use divided by runoff. A review of water scarcity indices is available in Brown and Matlock 2011

Results

To increase the spatial resolution from HUC-2 watersheds to HUC-4 watersheds and 1/8o degrees increases the proportion of  units under water scares from 0% to 11.1% to 13.7%.

Dube et al. 2016 Location Theories and Business Location Decision: A Micro-Spatial Investigation of a Nonmetropolitan Area in Canada

Region of interest

individual business establishments @ Lower St Lawrence, Canada

Objective

We provide an empirical analysis of the determinants of individual establishments' location decisions in relation to their main economic activity within a random utility model framework.

The objective of this paper:

"First, it sets out to develop a conceptual framework within which to analyze individual location decisions at a micro-geographic scale." "Second, it aims at testing (i) whether the classical factors underlined by location theories, such as agglomeration economies, are statistically related to individual establishments’ location choices, and (ii) whether the conclusions drawn from such analyses are sensitive to the choice of geographical scale used to define local individual spatial indicators."

Methodology

i - establishment s - sector v - main centers

A linear, multinomial regression model under the random utility maximization framework

Short Summary

Newmark et al. 2010 Water Challenges for Geologic Carbon Capture and Sequestration

This paper touches three issues on the interface between CCS and water - increased water use by CCS - potential leakage issue that may contaminate groundwater - brine displacement and the re-use of brine

Carbon capture and sequestration (CCS)

For geologic sequestration, the carbon dioxide is captured from large point sources (e.g. power plants or other industrial sources), transported to the injection site, and injected into deep geological formations for storage. This will produce new water challenges, such as the amount of water used in energy resources development and utilization and the "capture penalty" for water use. 

- At depth, brine displacement within formations, storage reservoir pressure increases resulting from injection, and leakage are potential concerns

- Potential impacts range from increasing water demand for capture to contamination of groundwater through leakage or brine displacement

Quote: “there is a ‘‘capture penalty’’ of increased water use for currently available capture and compression operations (DOE-NETL 2007c).”

---- From DOE-NETL 2007c: "The carbon capture section of this analysis uses a 1st order approach derived from a recent NETL study of the cost and performance impacts associated with CCS technologies on coal-based power plants."(citation: 15). "Net outputs from retrofitted PC plants are de-rated by approximately 30%, resulting in 72.7 GW of required replacement power." (citation: 14)

-------- The citation 15 is a 2007 study on the design and performance of retrofitting existing coal-fired power plants with CO2 capture and compression systems. It mainly investigates a MEA system for CO2 capture, compression, and liquefaction: 

------------ The MEA system is a post-combustion CO2 capture system that uses monoethanolamine (MEA) in its amine-based scrubber. In brief, the system modifieds the flue-gas desulfurization system so that part (Case 1-4: 90%, 70%, 50%, 30%; Case 5: 96%) of the CO2 is removed, compressed, and liquefied using (Case 1-4:) an advanced "state of the art" amine scrubbing system, which was designed and cost-estimated in 2006, or (Case 5:) the Kerr-McGee/ABB Lummus Global's commercial MEA process, which was designed and cost-estimated in 2000. Also, the exhaust steam from the intermediate pressure turbine is re-routed for solvent regeneration, resulting power loss at the low pressure turbine. The 2006 MEA system requires less energy for solvent regeneration than the 2000 system (1550 Btu/lbm-CO2 versus 2350 Btu/lbm-CO2 ). The energy requirement of solvent regeneration has considerable influence on the power loss. 

------------ For result, the changes in electricity output, cooling water requirement, and cost depend on the cases (amount of CO2 captured and technology). The original net generation 434 MW, and the Case 5, 1-4 outputs were, respectively, 252 MW, 303 MW, 333 MW, 363 MW, 392 MW, which amounted to between 10%-42% reduction on the original output. The less energy-intensive solvent regeneration gave about 26% improvement (Case 1 corrected to 96% CO2 capture) compared to the old solvent regeneration (Case 5). In Cases 1-4, the cooling water requirements for the cooling tower at the power plant were, respectively, 696944, 54217, 38693, 22991 gallons per minute. The increases in the cost of electricity were, for Case 5, 1-4, respectively, 6.92, 5.32, 3.64, 2.31, and 12.54 cents per kWh. 

------------ The study also discussed a few other post combustion CO2 capture technologies, but did not investigate them. Those include: aqueous ammonia, amine-enhanced sorbents, ionic liquids, enzymatic CO2 sorbents, dry regenerable sorbents, chilled ammonia, KS(R) Solvents (KS1, KS2, KS3). If those could be commercially implemented, further decrease in power generation efficiency penalty is expected. 

------------ The parasitic efficiency loss for the post-combustion technique in citation 15 seems to be standard (see this paper)?

CO2 capture technologies include: 

1. Post-combustion technique: basically amine; least costly but most water intensive because water is used in solvent-based capture capture and additional cooling water is needed because of decreased generation efficiency; amine + NGCC increases cooling water use by ~81%, and amine + PC increases cooling water use by ~95%. 

2. Pre-combustion technique: the fuel is first converted to syngas, then checmially shifted in a water-shift reaction, and CO2 is separated using the Selexol process. The process needs more cooling water becauese syngas needs to be cooled, and steam is needed in the water-shift reactor; Selexol + IGCC increases cooling water use by ~37%. 

3. Oxyfiring separates oxygen from air, resulting almost pure CO2 and steam output that can be separated by compression. Additional water use is incurred by the air separation and flue gas condensation. The additional water use comes from the make-up water, and no capture penalty exists.

"It displaces formation fluids wherever a pressure gradient develops in response to injection." CO2 rises towards the caprock, forming a plume that depend on various factors.

Recent studies have focused on large-scale changes in the storage formation and neighboring units due to CO2 injection. As injection progresses, pressure increases with the injection formation over a relatively large region. The pressure pulse travels much faster than the mass of the CO2 plume, which as the potential to displace reservoir fluids swiftly, far from the CO2 plume itself. Outside the injection zone, the pressure increases are low. Pressure perturbations may reach shallow aquifers, causing fluid displacement. Lateral brine flow veolocities induced by CO2 injection are relatively small. Vertical velocity of brine displacement through sealing units is negligibly small except where localized high-permeability flow paths occur.

A number of techniques exist today to monitor and verify CO2 plumes.

Groundwater Concerns Leakage Interaction with groundwater "Groundwater hazards stem from different leakage scenarios: well leakage, fault leakage or cap rock leakage". Industrial experience suggests that "most leaks occurred via either unsealed fault and fracture zones or through improperly constructed or abandoned wells."Good technologies exist in preventing well leakage and sealing wells, but the number of wells are very large in the United States. Another risk is related to the potential mobilization of toxic species from drinking water aquifers by the CO2 leakage.

Energy-Water Nexus Opportunity - "A coarse estimate has been made that, for a modern 1 GW IGCC plant generating 7.5 million m3 of CO2/year, treating displaced brine would provide half of the plant's operating fresh water, including cooling requirements." The challenge lies in dealing with potentially wide-varying water composition.

Sanders 2014 Critical Review: Uncharted Waters? The Future of the Electricity-Water Nexus

This review contains a huge amount of information.

It starts with background, where the water use of different fuel, prime mover, and cooling system is summarized. The content is not limited to the general argument based on the thermal efficiency of generation technology and known ranking among once-through, recirculating, and dry cooling systems due to the inclusion of rich details (numbers and citations). It touches upon the potential impact of various policies and environmental regulations, and comprehensively summarized the existing consensus on how the thermoelectric generation sector is going to change. It concludes with some suggestions on how research can move forward. 

Background

EIA Form 2012 shows that about 87% of U.S. generation needed water for cooling. About 62% of the electricity generation were from steam turbine facilities, and about 24% from combined cycle facilities.

1. Fuel and prime mover considerations

Fuel: Coal; Natural Gas; Nuclear; Biomass, CSP and geothermal systems; Hydroelectric; Wind and Solar PV

Prime mover: Steam; Combined cycle; Combustion turbine

---- [All Cons] Coal are relatively inefficient and water-inefficient compared to natural gas because coal contains moisture and contaminants, and needs more pollution controls. According to literature review, the most water-efficient IGCC power plants consumes more than 300 Gal/MWh, while average Natural Gas Combined Cycle (NGCC) consumes 205 Gal/MWh. 

---- [All Pros] NGCC power plants has low water requirement per unit electricity generation because of their high efficiency. They also have "efficiency, cost, and operational benefits over other forms of thermoelectric generation". From a life-cycle perspective, replacing coal-fired generation by NGCC (using shale gas) in Texas result in 60% reduction in water consumption per unit electricity generation, though the detailed spatial distribution of the consumption is not always decrease (citation: Grubert et al. 2012; Pasci et al. 2014). 

---- [Only peak load] Natural gas combustion turbines has little water requirement, but they are usually peaking facilities because they are smaller and more expensive to build, but ramp fast. Some natural gas combustion turbine units pre-chill the air with small amount of water to increase efficiency. the majority of the natural gas combustion turbine and NGCC units in California are equipped with this technology. 

---- [Pro: CO2; Con: water] Nuclear facilities represent ~19% of the electricity generation in the United States. They generally operate with lower thermal efficiency than coal and natural gas power plants, and no heat is brought away in flue gas, resulting in higher water use. They also cannot use dry cooling due to safety reasons, unless the experimental small modular reactors (SMRs) become commercially widely deployed. 

---- [Neutral] Biomass, CSP, and geothermal power plants require water for cooling. 

---- [Difficult] Hydroelectric power plants consume water via evaporation, but the consumption varies several magnitudes (from 10.6 Gal/MWh to 55200 Gal/MWh) depending on climate conditions and design. They also have water quality and ecosystem impacts. 

---- [All Pros] Wind turbines and solar photovoltaics (PV) use no water and thus are good for arid regions. Example study on water savings conducted on Germany - 80% renewable penetration by 2050 translates to 70% decrease in annual water consumption (citation: Johst et al. 2014)

2. Cooling system considerations

Cooling system: Once-through, Recirculating, Dry, Hybrid

---- Based on the USGS model-estimated water use for 2010 (citation: Maupin et. al. 2014) -------- Once-through cooling represents 30% of the water-cooled generation capacity, 70% of the water withdrawal, with 75% freshwater and 25% saline, and 20% of the freshwater consumption. -------- Recirculating pond and tower cooled systems represented 50% of the water-cooled generation capacity, 2% of the water withdrawal, and 56% of the freshwater consumption. -------- Hybrid power plants (multiple cooling system types, generation tech., or fuels) represent 20% of the water-cooled generation capacity, 28% of the water withdrawal, and 24% of the freshwater consumption.

---- Once-through cooling systems have declined significantly in the past. The water withdrawal per unit electricity generation decreased from 63 Gal/KWh in 1950 to 21 Gal/KWh in 2000, after the Clean Water Act requirements on thermal discharge in the 1970s (citation: Abrams & Hall 2010). 

---- The shift from once-through will result in increase in consumption, which is not likely to negatively impact water supply in the water-abundant east, but may impact the west (citation: Ackerman & Fischer 2013; USGS 2005; Stewart et al. 2013; Tidwell et al. 2012). The shift can increase the power plants' resilience to drought (citation: Stillwell et al. 2011; Stillwell et al. 2013). (Note to myself: these studies are relevant; how have they quantified the consumption change? need follow-up)

---- Dry cooling systems are limited in deployment because the are more expensive, may reduce the electricity generation efficiency by 8%-25% during hot summer days and by 2% on annual average (citation: Badr et al. 2012). An evaluation suggested that large-scale conversion to dry-cooling is not a cost-effective way to reduce water consumption in the western United States (citation: Ackerman & Fischer 2013) (Note to myself: what is the water consumption reduction estimated by Ackerman & Fischer? need follow-up)

----- To use alternative cooling supplies from wastewater is another option to reduce water use, apart from dry cooling. This can be cost-effective when the price difference between freshwater and treated municipal wastewater is more than $0.14 per kL and resources are widely available, though additional water treatment is needed for the cooling water, and the equipment can be negatively impacted by scaling, stress cracking, and biological growth (citation: Zemlick et al. 2013; Walker et al. 2013; Tidwell et al. 2014; Hightower & Pierce 2008). Wastewater reuse in the United States is growing by about 15% per year (citation: Hightower & Pierce 2008).

3. Environmental control considerations (mainly on environmental regulations)

---- There is sometimes synergy and sometimes trade-off between water use reduction and air emission (CCS (CO2), SOx) reduction. 

---- The federal Endangered Species Act also limit the water use by thermoelectric generation and limit the development of wind power, solar PV, hydroelectric, and transmission lines. 

---- The CWA 316(B) provision's implementation, and the phase-out of once-through cooling systems in California will decrease water withdrawal though the exact impact is not clear. 

---- Other environmental regulations that may impact the thermoelectric sector and its water use include the 2011 Mercury and Air Toxics Standards for Utilities, the 2011 CrossState Air Pollution Rule, and the 2010 proposed Coal Combustion Residuals rule. The collective impact of the three, plus CWA 316(B), is that about 234-258 gigawatts of power generation capacity will be retrofitted by 2015, and another 36-59 gigawatts of capacity will be retired or derated by 2018 (citation: NERC 2011). The 2014 Clean Power Plan is expected to result in the retirement of about 30-49 gigawatts by 2020 (citation: NERC 2011). 

4. Climate considerations

---- Increase in air and water temperatures can decrease the efficiency of power generation and some research has been conducted, but not enough research has been done at regional level (citation: Maulbetsch & DiFillippo 2006; Santhaye et al. 2013; Li et al. 2014; Miara et al. 2013).

-------- Combustion turbine and NGCC power efficiency loss is because air density is decreased with increase in air temperature. The thermoelectric output is reduced by about 0.3%-0.7% per 1 degree Celsius temperature increase (citation: Li et al. 2014). 

-------- Hydropower will be affected by increasing temperature and drought. The hydroelectric power generation decreases by about 3% per 1% decrease in streamflow (citation: Mellilo et al. 2009). 

-------- The efficiency of solar PV is decreased by about 0.65% per 1 degree Celsius temperature increase (citation: Li et al. 2014). 

---- Other climate change concerns include increasing precipitation and sea-level rise. For example, 25-30 coastal power plants in California are at risk (citation: Sathaye et al. 2013; Heberger et al. 2009). 

---- Various anecdotes of how water concerns have impacted electricity generation/resulted in shutdown of power plants (citation: e.g. CAISO 2014)

---- Some system-scale studies on the impacts of water temperature impacts on power generation have been conducted (citation: Flores-López & Yates 2013 (southeastern United States); Forster et al. 2009 (a nuclear power plant in Central Europe); Koch & Voogele 2013 and Koch et al. 2014 (nuclear power plants in in Germany))

---- Increase in air temperature and wild fire frequency can negatively affect electricity transmission and distribution (T&D) infrastructure, and electricity demand increases, adding stress.

5. Grid-scale considerations ---- Future "smart grid", integration of the transportation sector into the grid electricity exchange system, penetration of more intermittent renewable electricity sources, integration of energy storage, ... may have potential impact on water resources.

Future Shifts in the Power-Water Nexus 1. This section first briefly reviewed the status of current projection studies: 

---- While the studies agree that future water withdrawal will decline, and that wide deployment of carbon capture and storage and nuclear would increase water consumption, the overall projected trend in consumption varies across studies. Since the result of projection studies depend on the scenarios and modeling techniques, a review table (Table 3) was provided. All except three studies used water use factors from literature, and the different studies covered a range of scenarios (environmental regulation, climate change, mitigation, and business as usual) and electricity projection model. 

2. This section then goes on to summarize some consensus on various aspects of the electricity-water nexus: 

---- On fuel preferences: coal is expected to decrease, natural gas and renewable energy generation will increase, and the nuclear regulatory environment is unclear

---- On cooling system preferences: once-through cooling system will continue to decline, dry cooling will increase despite the disadvantage in efficiency and cost, and alternative water supply will be increasingly used for cooling. 

---- On environmental regulations: increase is expected in greenhouse gas reduction and other pollution control regulations, and the Endangered Species Act is expected to pose increasing constraint on energy development

---- On the impact of climate change: more climate-related disruptions are expected; evaporative water losses will increase, the efficiency of electricity generation will decrease, and hydropower generation will decrease

---- On the expected grid transformation: expanding smart grid, electrification of the transporation sector, increasing renewable energy penetration and electricity storage

Jones & Carvalho 2014 Sensitivity to Madden–Julian Oscillation variations on heavy precipitation over the contiguous United States

 Takeaway: MJO amplitude is positively related to precipitation in the western and central US (from lower Mississipi to the Ohio River Valley), while the picture in eastern US is more ambiguous. The intensified precipitation is related to overlaying low sea-level pressure centre, 500 hPa trough (negative anomaly in geopotential height), high moisture flux transport from the adjacent tropical Pacific or Gulf of Mexico, and enhanced 500 hPa positive absolute vorticity advection and 500 hPa positive advection of temperature.

This is a case study on the Dec 18, 2004 - Jan 20, 2005 MJO event using the WRF model. Heavy precipitation occurred in the western and mid-eastern United States during that period (Fig. 5). The WRF run began on Nov 1, 2004 00UTC and ended on Mar 31, 2005 18UTC. That winter had weak warm ENSO (tropical Pacific SST anomalies ~0.6 oC), weak PNA (0.26 standardized index), and strong NAO (1.4 standardized index). The magnitude of the study MJO event was not large (index = 1.34), being below the median (index = 1.67).

Definition of MJO was the same as in Jones & Carvalho 2012.

Evaluation of WRF: WRF precipitation had high bias in California (due to the relatively coarse 30km resolution), and in Great Lakes region and Maine.

Understand the convolution between MJO and other large-scale signals like NAO, ENSO, and PNA by analysis on the large-scale pattern of 200 hPa zonal wind (U200) standard deviation anomaly: I am not sure about the significance of using the standard deviation of U200 as the indicator of intraseasonal variability. Maybe it is because U200 is related to MJO, and that if low wavenumber accounted for large part of the variability in U200, it means MJO explains well the intraseasonal variability in U200? But why the authors mentions NAO?

Perturbation experiments: The purposes of the perturbation experiments were to understand the sensitivity of heavy precipitation over the CONUS. The driving CFSR reanalysis was perturbed by altering the large-scale intraseasonal signal: - "First, a 20–100 day band-pass Murakami filter (Murakami, 1979) is applied to the time series of geopotential height, temperature, zonal, meridional wind components and specific humidity (all model levels)." - "Next, a spatial filter is applied to these variables by retaining only zonal wavenumbers 1–5 which is typical of the large-scale nature of the MJO (Hendon and Salby, 1994; Jones, 2009)." - Finally, the extracted MJO signal in the driving CFSR reanalysis was modified in 7 different ways, resulting in 7 perturbation experiments: NOMJO (no MJO signal), reduced to 25% and 50%, and increased by 25%, 50%, and 75%. The modification was done for each day during the MJO event (i.e. Dec 18, 2004 - Jan 20, 2005).

Result Fig. 12: For the Western and Central U.S., stronger MJO gave higher precipitation; For the eastern U.S., the trend was less clear - the control run gave the highest precipitation.

- mechanistic explanations:

Fig 13: Sea-level pressure, 500 hPa geopotential height, and the magnitude and direction of vertically integrated moisture flux were shown. On Jan 9, the Pacific Northwest heavy precipitation was explained by a low sea-level pressure center directly overlaying; the California heavy precipitation was explained by an 500 hPa trough and intense southwesterly moisture flux that drops moisture against the mountain ranges. On Jan 13, the Midwest heavy precipitation was also explained by a low-pressure system and 500 hPa trough, and intense moisture flux carried from the Gulf of Mexico.

Fig. 14 gives further comparison on 500 hPa geopotential height differences - NOMJO minus control, MJO decreased by 25% minus control, and MJO increased by 75% minus control. The 500 hPa trough over northwestern US on Jan 9 and over central northern US on Jan 13 was basically removed under NOMJO experiment, but strengthened under increased MJO magnitude.

Fig. 15 gives further comparison on vertically integrated moisture flux intensity [kg m-1 s-1]. The result is that in western and central US, the moisture flux increased with growing MJO amplitude, while in eastern US, only the upper quartiles of the moisture flux and above increased with growing MJO amplitude.

Jones & Carvalho 2012 Spatial–Intensity Variations in Extreme Precipitation in the Contiguous United States and the Madden–Julian Oscillation

Overall, not a very informative paper because they used joint probability to characterize the influence of MJO on the spatial extent and intensity of extreme precipitation regions, instead of the conditional probability, and because of the lack of demonstration on the mechanisms (for example, why ENSO modulates the tail of precipitation in the US?). The information on ENSO-MJO interaction is good, but again not clear enough due to the use of joint probability. Their aggregation of CREP area and intensity over large regions is good for regional natural hazard prevention but not good when I want to know exactly how the regional precipitation looks like.

For a summary of their conclusions: more intense precipitation when MJO is active[*], and when ENSO is in neutral or warm SST phases. Phase 3 and 7 of MJO seem to correspond to more occurrence of extreme precipitation but not necessarily larger area or higher intensity. Amplitude of MJO did not seem to have any effect.

*: This contradicts the previous paper which finds drying signal in Florida. This can be explained: the previous paper's central south wetting region partially straddles this paper's central south, southeastern, and in MJO Phase 7, northeastern regions. Also, the drying signal in Florida was mainly due to frequency decrease and only slightly due to intensity decrease. So the drying signal in the previous paper could have been clouded out by the wetting signal due to this paper's delineation of region and focus on extreme precipitation which has nothing to do with frequency.

Method 1. Division of regions

The U.S. were divided into 6 regions (northwest, central north, northeast, southwest, central south, southeast) with somewhat arbitrary boundaries. "The spatial-intensity characteristics of extreme precipitation, their probabilities of occurrence, and the importance of the MJO are aggregated in each sector over the CONUS."

2. Definition of MJO MJO events were identified by performing combined EOF analysis on 15oS-15oN averaged U200 and U850 band-pass filtered anomalies, that is, the 850 hPa and 200 hPa zonal winds were de-seasonalized, de-trended, and filtered to retain the 20-200 days frequency domain (intraseasonal variations). Data source is NCEP-NCAR reanalysis I. Their definition was similar to but slightly different from the Wheeler and Hendon 2004 definition because the bandpass-filtered definition was less noisy and better estimates intraseasonal signals. Some care was taken to correctly identify eastward propagating MJO events. This whole definition procedure differed from the previous paper which used the Wheel and Henden 2004 definition of MJO index.

Fig. 2 shows a composite of the general life cycle of a MJO event. Fig. 3 further shows how MJO relates to 500 hPa geopotential height (H500) anomalies: when enhanced convection was over the Indian Ocean (Phase 1-3), negative H500 anomaly was seen in eastern North America; when the enhanced convection passed to the western Pacific (Phases 4-6), small positive H500 anomaly occurred over the eastern coast of North America; in Phases 7-8, negative H500 anomaliescame go dominate eastern North America again. But this section did not relate the H500 anomalies to precipitation yet.

3. Characterization of the spatial-intensity variability of extreme precipitation

First, Gamma-distribution was fitted to each winter month's precipitation series separately. Then, the 75th and 90th percentile of the gamma distribution were used as threshold for "extremes". Next, spatially connected grids of precipitation were identified. Finally, the quantities of interest are: type 1, intensity > 75th percentile and the connected area >= 5 grids; type 2, intensity > 90th percentile and the connected area >= 25 grids. (abbr. CREPs)

To characterize the spatial extents of the two CREPs, the authors calculated the percentage area of each of the six climate region occupied by the two CREPs when MJO is active/inactive.

To characterize the intensity of the two CREPs, the authors defined the intensity as "the total precipitation associated with CREPs falling in each sector".

Result

1) "What is the probability of extreme precipitation over the CONUS when the MJO is active?"

Fig. 7 and Fig. 8 plot the joint probabilities of MJO being active/inactive and the CREPs occupying more than a certain percentage area of each of the six climate regions. The ratios of the probabilities when MJO was active:inactive were close to 2:1. This kind of plot is simply not informative because we know the MJO being active:inactive ratio were close to 2:1 and we need the conditional, instead of joint, probabilities to actually make the percentage areas of CREPs comparable between MJO active and inactive periods.

Fig. 9 and Fig. 10 show the joint probabilities, but with the intensities. There were again nice 2:1 ratios between MJO active:inactive days.

2) "Does the spatial–intensity probability of extreme precipitation associated with the MJO significantly vary with ENSO? "

Linkage between MJO and ENSO is that MJO tends to propagate farther (less) eastward when tropical Pacific SST anomaly is positive (negative) during ENSO events, which can have implications on precipitation (more details on ENSO regulation of MJO in this paper, this paper, and related). Thus, the authors further investigated how the joint probabilities of MJO active/inactive and CREP area exceedence depended on ENSO. This is shown in Fig. 11, which compares the joint probabilities of MJO active and 75th percentile CREP area exceedence during warm, cold, and neutral ENSO. Fig. 12 is the same but the joint probabilities are for 75th percentile CREP intensity.

(for Fig. 11 and Fig. 12: solid lines = neutral ENSO solid lines with squares = warm ENSO solid lines with triangles = cold ENSO)

In both Fig. 11 and Fig. 12, the joint probabilities were neutral ENSO >(=) warm ENSO >(=) cold ENSO in all six climate regions. Again, due to their use of joint probabilities, we do not know whether the higher joint probability was due to more active MJO or larger area/intensity of CREP.

3) "Do probabilities of spatial–intensity characteristics of extreme precipitation over the CONUS vary with MJO phases?" The joint probabilities of CREP area exceedence/intensity and MJO phases were not investigated in detail because the study period was too short to allow definitive conclusion. But the general distribution of CREP across MJO phases indicates a little bit more CREPs occurrence in Phases 3 and 7 (which would be consistent with the previous paper). Phases 3 and 7 are when the MJO "convective signals are over the Indian Ocean or western Pacific". Break-down into the six climate regions further suggest that northwest, southwest, and southeast had the most prominent response to Phase 3 and Phase 7. Note their definition of southeast involved a large portion of the lower Mississippi area.

Finally, Fig. 14 and 15 shows some conditional probabilities! But the conditional probabilities were not comparing MJO active and inactive periods. They only compared the conditional probabilities of CREP area exceedance and intensity exceedance when MJO is active, but in different phases. The conditional probabilities did not seem to have preferential phases, which contradicts that the CREPs' occurrence preferred Phases 3 and 7. But after all, the exceedance area and intensity were more detailed measures than simple probabilities of occurrence. They attributed this to that other physical factors might be at play on the area and intensity of CREPs.

Becker et al. 2011 Modulation of Cold-Season U.S. Daily Precipitation by the Madden–Julian Oscillation

The influence of MJO events on cold-season (November-March) precipitation in the United States was investigated. Clear conclusion and demonstration of the mechanisms using composite analysis. But this study only touched upon the interaction between MJO and ENSO, which has effect on circulation and thus maybe precipitation (see this paper)

Methodology First, MJO events were identified using an EOF-based MJO index (Wheeler and Hendon 2004) and the Climate Prediction Center (CPC)'s MJO event classification for the years 1979-2005. Fig. 1 shows the November-March composite outgoing longwave radiation (OLR) to show their progress through the phases of a MJO. Areas of enhanced convection (which develops over the Indian Ocean in phases 1-3 and then propagates into the Pacific Ocean in phases 5-8) were characterized by negative OLR. The identified MJO events and the corresponding MJO index during the cold seasons of 1979-1987 are displayed in Fig. 2. The MJO phases 5-7 were of interest to this study.

Results: MJO modulation of US precipitation during the cold season a. Precipitation patterns (data from CPC) - Composite analysis suggests that different precipitation anomalies occur during different MJO phases. Phase 1, 5-8 was related to precipitation response in the eastern half of the United States. In Phase 5, positive anomalies occur in the lower Mississippi River basin, and negative anomalies occur in the Southeast. The positive anomalies in the lower Mississippi River basin and negative anomalies in Florida persist in Phase 6-7. In Phase 8, the negative anomalies moved up to the Mid-Atlantic coastal states. In Phase 1, positive anomalies exists in the Southeastern states. - Further break-down into intensity and frequency component suggests that these two anomalies during Phase 5-7 were of the same sign but different magnitude. The positive anomaly in Mississippi River basin was mainly due to intensity, while the negative anomaly in Florida was mainly due to frequency.

b. Mechanisms (data from NARR) - Vertically integrated moisture flux convergence anomaly during cold season MJO Phases 5-7 was positive above the positive anomaly over the lower Mississippi River basin. But no consistent negative moisture flux convergence anomaly existed over Florida or the Southeastern US (Fig. 7). Understandably, this was because in Florida, the decreased precipitation was mainly related to frequency rather than intensity (good explanation! So, moisture availability would be mainly related to precipitation intensity, while frequency was explained by some other forcing). - 200 hPa zonal wind was used to indicate the North American jet. - The decrease in precipitation in Florida during Phases 5-7 was related to northward shift of the North American jet, exhibited as negative anomaly of the 200 hPa zonal wind over Florida. - Storm track could be identified as the standard deviation of the un-filtered 200 hPa meridional wind. During Phase 5-7, more active storm track (that is, postive standard deviation in the 200 hPa standard deviation of meridional wind) was found in the Midwest and less active storm track in the Gulf.

c. MJO-related precipitation anomalies compared to ENSO - Like MJO, ENSO also "modulates the characteristics of daily precipitation over the United States through tropical-extratropical teleconnections". ENSO-related enhanced convection in tropical Pacific is on the seasonal to interannual time scales, while the MJO-related enhanced convection is on the intraseasonal time scales. - Two 2x2 latlon boxes were drawa to better understand the precipitation anomalies during phases of MJO and ENSO. One box is in the lower Mississippi River basin, and the other is over central Florida.

(box 1)

(box 2)

Dhanya & Villarini 2017 An investigation of predictability dynamics of temperature and precipitation in reanalysis datasets over the continental United States

 Explanation of the spatial pattern of the predictability indices by topography and their inter-relationship is good.

 The Chaotic System Indices

1. Maximum predictability

The maximal Lyapunov exponent is a measure of the predictability of a chaotic system. For a highly complex nonlinear system like the atmosphere, the nonlinear finite time Lyapunov exponent (FTLE) was appropriate. So FTLE was used to evaluate the predictability of temperature and precipitation over the continental US.

2. Predicative error (the mean predictive error around saturation)

3. Predictive instability (the random fluctuation of predictive error around a common mean)

Data

Four reanalysis products are evaluated: NCEP-NCAR Reanalysis I, MERRA, ERA-Interim, JRA-55

Results 4.1 General daily precipitation features - Climatological mean daily precipitation is better captured by ERA-Interim, followed by MERRA and JRA-55, and finally Reanalysis I - All the reanalysis datasets underestimate the standard deviation especially over the southeastern United States. This means Reanalysis I's bias is mainly systematic and can be easily corrected. - The signal-to-noise ratio of precipitation is somewhat slightly overestimated by ERA-Interim, MERRA, and JRA-55 in the southeastern US, and severely overestimated by Reanalysis I, which is not surprising given Reanalysis I's overestimation of the climatological mean and underestimation of the standard deviation.

<seasonal performance is missing from this study>

4.2 Predictability features of daily precipitation - All reanalysis products reproduced the overall predictability limit pretty well - Biases by the reanalysis products in simulating predictive error can be attributed to inability to simulate non-rainy days, which are underestimated in the eastern US and overestimated in some parts of the western US - For predicative instability, the biases in reanalysis products exhibit an inverse relationship to the predictive errors, which is expected because predicative instability = standard deviation of fluctuations / predicative error

4.3 and 4.4 are skipped here because they concern temperature and the results are less relevant & clear to me. But MERRA is said to perform the best on the chaotic dynamics indices.

Chen & Liu 2016 Global water vapor variability and trend from the latest 36 year (1979 to 2014) data of ECMWF and NCEP reanalyses, radiosonde, GPS, and microwave satellite

This article looked at global trends in precipitable water vapor in two reanalysis datasets (ECMWF and NCEP-DOE) and three observational datasets (Radiosonde, GPS, microwave satellite). It also looked at the relationship between temperature and PWV, and compared and explained the difference between them and Clausius-Clayperon relationship by moisture divergence & convergence. 

PWV has the larger standard deviation in spring and autumn, and low standard deviation in summer and winter. It's average value is high in summer and low in winter. The standard deviation of PWV is also larger over land than over ocean, and larger in the monsoonal regions. Globally, the trend in PWV is positive. ECMWF's PWV agrees better to the observations than NCEP. The trend in ECMWF is weaker than in NCEP, but the two agreed well after 1992. This is because ECMWF ingested a different dataset after 1992: it overestimated the PWV before then. Temperature is positively correlated with PWV where there is no strong moisture convergence or divergence. Strong moisture divergence remove the moisture, creating a negative relationship. Strong convergence zones, such as the ITCZ, have weak temperature trend, so the correlation also easily became negative. 

The papers conclusion is good!

Zhang et al. 2016 Scale-dependent Regional Climate Predictability over North America Inferred from CMIP3 and CMIP5 Ensemble Simulations

The title seems a little misleading to me. It is less about "predictability" than about that the signal-to-noise ratio of the current generation of GCMs is too low at below 1000 km for temperature and below 2000 km for precipitation, so that robust climate decision-making cannot be based on them. But, low signal-to-noise ratio does not mean that the signal is false, right? (see this)

 Figure 1 is good:

- Ensemble average mean bias is < Ensemble spread, but the largest spread does not always coincide with the largest error, implying systematic biases. Systematic overestimation of temperature by 3-5 oC exists near the western coast and adjacent sea, systematic underestimation of temperature by -2 oC exists off the Gulf coast (larger than the average ensemble spread of merely 1 oC). Precipitation has severe dry bias (-2.5 mm) in Florida and the Gulf coast, and wet bias (0.5-2 mm) in the lee of the Rockies.

 Confidence in regional projections of surface temperature and precipitation is low:

- Whetton, P., I. Macadam, J. Bathols, and J. O’Grady, 2007: Assessment of the use of current climate patterns to evaluate regional enhanced greenhouse response patterns of climate models. Geophys. Res. Lett., 34, L14701.

- Separovic, L., R. de El´ıa, and R. Laprise, 2008: Reproducible and irreproducible components in ensemble simulations with a Regional Climate Model. Mon. Wea. Rev. 136, 4942–4961.

- Watterson, I. G., and P. H. Whetton, 2011: Distributions of decadal means of temperature and precipitation change under global warming. J. Geophys. Res., 116, D07101.

- Deser, C., R. Knutti, S. Solomon, and A. S. Phillips, 2012: Communication of the role of natural variability in future North American climate. Nature Clim. Change, 2, 775–779.

Gianotti et al. 2014 The potential predictability of precipitation occurrence, intensity, and seasonal totals over the continental United States

 Potential predictability (PP) is defined as the low frequency variation of the climate system as opposed to the day-to-day variation due to weather. This metric is relevant to weather forecast on the seasonal and year-ahead scale, but is not relevant to statistical precipitation downscaling, because statistical downscaling uses the relationship between circulation predictors and local precipitation. The downscaled precipitation therefore contains low-frequency variability and day-to-day variability from the circulation predictors.

 Informative graphs!

Is interested in the PP of seasonal precipitation.

Total PP is found lower than the PP of occurrence and higher than the PP of intensity.

The Atlantic coast generally has low PP (moderate in JJN and SON) except Florida.

 Methods to assess predictability

------------------ Process models

Koster et al. 2004

Moron et al. 2006

Jia and DelSole 2011

------------------ Observational approaches

Madden et al. 1999

Anderson et al. 2009

Robertson et al. 2009

Gianotti et al. 2013 What do rain gauges tell us about the limits of precipitation predictability?

 Metrics for PP

------------------ Signal to noise ratios

Ferguson et al. 2011

------------------ Variance partitioning

Koster et al. 2004

Jia and DelSole 2011

Boer 2009

------------------ Through the divergence of states from initial model conditions

key points:

study period 1976-2000 GHCN land-based weather station