Operational Drought Monitoring and Forecasting at the USDANRCS

Operational Drought Monitoring and Forecasting at the USDA-NRCS Tom Pagano Tom. Pagano@por. usda. gov 503 414 3010

Monitoring networks Data products Seasonal forecasts Soil moisture Challenges Frontiers

Monitoring networks

1906 2005 Manual Snow Surveys Metal tube inserted into snow and weighed to measure water content. +300, 000 snow course measurements as of June 2008

Snotel (SNOw TELemetry) network Automated, remote stations Primary variables: Snow water Precipitation Temperature Also: Snow depth Soil moisture SNOTEL and Snow course records often spliced together

Number of sites Snowcourse (solid) and SNOTEL (hashed) active station installation dates Active year

Soil climate analysis network (SCAN) Soil moisture/energy balance emphasis Short period of record (some from 1990 s) Data available but few products

Manual snow-course SNOTEL SCAN

Data products

Time series charts

CSV flat files Google Earth

Forecast products

Location Forecasts are coordinated with the National Weather Service (NWS). Both agencies publish identical numbers.

Location Historical Average Time Period Forecasts are coordinated with the National Weather Service (NWS). Both agencies publish identical numbers.

Historical Average Location Time Period Error Bounds “The” Forecast Water Volume Forecasts are coordinated with the National Weather Service (NWS). Both agencies publish identical numbers.

Seasonal water supply volume forecasts (available in a variety of formats) NRCS formats:

Seasonal water supply volume forecasts (available in a variety of formats) NRCS formats:

Basic Forecasting Methods Statistical regression Apr-Jul streamflow % avg S Fork Rio Grande, Colo May 1 snowpack % avg

Basic Forecasting Methods Statistical regression Simulation modeling S Fork Rio Grande, Colo Apr-Jul streamflow % avg Snow Rainfall Heat Snow pack Soil water May 1 snowpack % avg Runoff

Principal Components Regression (Garen 1992) Prevents compensating variables. Filters “noise”.

Principal Components Regression (Garen 1992) Prevents compensating variables. Filters “noise”. Z-Score Regression (Pagano 2004) Prevents compensating variables. Aggregates like predictors, emphasizing best ones. Does not require serial completeness. Relative contribution of predictors

Daily forecast updates Existing seasonal forecasts issues once per month Why not develop 365 forecast equations/year and automate the guidance? We currently do Apr-Jul Streamflow = a * April 1 Snowpack + b Why not something like Apr-Jul Streamflow = a * April 8 Snowpack + b

Period of record range (10, 30, 70, 90 percentile) 1971 -2000 avg Period of record median

Period of record range (10, 30, 70, 90 percentile) 1971 -2000 avg Period of record median Official coordinated outlooks

Daily Update Forecasts Period of record range (10, 30, 70, 90 percentile) 1971 -2000 avg Period of record median Official coordinated outlooks

Official forecasts

Expected skill Daily forecast 50% exceedence Official forecasts

SWSI Methodology varies by state Available 8 Western states Rescaled percentile of [reservoir + streamflow] Calibrated on observed, forced with streamflow forecasts (real-time variance too low) No consistent calibration period

Soil moisture and runoff efficiency

Expansion of soil moisture to SNOTEL network (data starts ~2003)

Blue Mesa Basin, Colorado Soil Moisture 2001 -2008 (According to the Univ Washington Model- top 2 layers)

Blue Mesa Basin, Colorado Soil Moisture 2001 -2008 (According to the Univ Washington Model- top 2 layers) (According to Park Cone Snotel- ~0 -30” depth) Snotel does poorly in frozen soils, so that has been censored Model resembles snotel, but also remember we’re comparing basin average with point measurement

What influence humans? Does it matter? Blue Mesa For each site, all measurements Jan-Jun, Jul-Dec are averaged by year. Station half-year data then converted into standardized anomaly (o-avg(o))/std(o) vs period of record for the half year. Multiple stations are then averaged.

Spring precipitation, especially the sequencing with snowmelt is also important Rainfall Snowmelt Rainfall mixed with snowmelt “normal” Runoff April July

Spring precipitation, especially the sequencing with snowmelt is also important Rainfall Snowmelt Rainfall mixed with snowmelt “normal” Runoff Rainfall boosting snowmelt Larger volumes Snowmelt and rainfall separate Not enough “momentum” to produce big volumes April July All these interactions are tough to “cartoonize”; Simulation models can handle this… but still it’s tough to predict beyond 1 -2 weeks.

Spring precipitation, especially the sequencing with snowmelt is also important Rainfall Snowmelt Rainfall mixed with snowmelt “normal” Runoff Rainfall boosting snowmelt Larger volumes Snowmelt and rainfall separate Not enough “momentum” to produce big volumes Even then, however, high heat and no rain can lead to “pouring sunshine” April July All these complex interactions are tough to “cartoonize”; Simulation models can handle this… but still it’s tough to predict beyond 1 -2 weeks.

Challenges and frontiers

Seasonality/lag of drought in snowmelt regions Precipitation and impacts can be separated by months. Highly managed systems How to separate drought from poor planning or overbuilding? Also: Humans react to forecasts e. g. evacuating reservoirs Regional/local vulnerability Whose drought? Stickiness of drought When is the drought over? Never… (also risk of “Drought fatigue”)

Seasonality/lag of drought in snowmelt regions Precipitation and impacts can be separated by months. Highly managed systems How to separate drought from poor planning or overbuilding? Also: Humans react to forecasts e. g. evacuating reservoirs Regional/local vulnerability Whose drought? Stickiness of drought When is the drought over? Never… (also risk of “Drought fatigue”) Incomplete understanding of natural system (esp soil moist, sublim) Can we even close the water balance? Institutional and infrastructure barriers Limited agency resources, increasing restrictions Non-stationarity Could climate change be the new normal?

The future may have more and better: Products from and understanding of soil moisture data Automation and “smart” objectification of forecast process Quantification and use of anecdotal evidence Forecast transparency (i. e. access to raw guidance)

The future may have more and better: Products from and understanding of soil moisture data Automation and “smart” objectification of forecast process Quantification and use of anecdotal evidence Forecast transparency (i. e. access to raw guidance) Communication of uncertainty, especially graphically Understanding of local user vulnerabilities Consolidation of data from multiple networks: universal, uniform access and multi-agency products Understanding of the “long view”: how relevant is data from 10, 50, 100, 500 years ago?

Variable Snow “Significance” 60 -90 Fall precip Winter precip Spring precip 5 -20 30 -60 10 -25 Baseflow Soil Moisture 5 -15 5 -10 Temperature Wind Radiation Relative humidity 10 -25 5 -20 5 -15 5 -10 Source: 1972 Engineering Handbook

January 1 Daily forecast Skill: (Correlation)2 Variance Explained

April 1 Daily forecast Skill: (Correlation)2 Variance Explained

NWS formats:

NWS formats: