Toward a categorical drought prediction system based on U.S. Drought Monitor (USDM) and climate forecast

Toward a categorical drought prediction system based on U.S. Drought Monitor (USDM) and climate forecast

Accepted Manuscript Research papers Toward a categorical drought prediction system based on U.S. Drought Monitor (USDM) and climate forecast Zengchao ...

532KB Sizes 3 Downloads 126 Views

Accepted Manuscript Research papers Toward a categorical drought prediction system based on U.S. Drought Monitor (USDM) and climate forecast Zengchao Hao, Youlong Xia, Lifeng Luo, Vijay P. Singh, Wei Ouyang, Fanghua Hao PII: DOI: Reference:

S0022-1694(17)30404-3 http://dx.doi.org/10.1016/j.jhydrol.2017.06.005 HYDROL 22058

To appear in:

Journal of Hydrology

Received Date: Revised Date: Accepted Date:

19 December 2016 30 March 2017 3 June 2017

Please cite this article as: Hao, Z., Xia, Y., Luo, L., Singh, V.P., Ouyang, W., Hao, F., Toward a categorical drought prediction system based on U.S. Drought Monitor (USDM) and climate forecast, Journal of Hydrology (2017), doi: http://dx.doi.org/10.1016/j.jhydrol.2017.06.005

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Toward a categorical drought prediction system based on U.S. Drought Monitor (USDM) and climate forecast

Zengchao Hao1, Youlong Xia2, Lifeng Luo3,Vijay P. Singh4, Wei Ouyang1, Fanghua Hao1 1

Green Development Institute, School of Environment, Beijing Normal University, Beijing,

100875, China 2

Environmental Modeling Center (EMC) of the National Centers for Environmental Prediction

(NCEP) and I. M. Systems Group, College Park, Maryland, USA 3

Department of Geography, Environment, and Spatial Sciences, Michigan State University, MI,

48824,USA 4

Department of Biological and Agricultural Engineering and Zachry Department of Civil

Engineering, Texas A&M University, College Station, TX, USA

Submit to Journal of Hydrology Corresponding author: Zengchao Hao ([email protected])

1

Abstract Disastrous impacts of recent drought events around the world have led to extensive efforts in drought monitoring and prediction. Various drought information systems have been developed with different indicators to provide early drought warning. The climate forecast from North American Multimodel Ensemble (NMME) has been among the most salient progress in climate prediction and its application for drought prediction has been considerably growing. Since its development in 1999, the U.S. Drought Monitor (USDM) has played a critical role in drought monitoring with different drought categories to characterize drought severity, which has been employed to aid decision making by a wealth of users such as natural resource managers and authorities. Due to wide applications of USDM, the development of drought prediction with USDM drought categories would greatly aid decision making. This study presented a categorical drought prediction system for predicting USDM drought categories in the U.S., based on the initial conditions from USDM and seasonal climate forecasts from NMME. Results of USDM drought categories predictions in the U.S. demonstrate the potential of the prediction system, which is expected to contribute to operational early drought warning in the U.S..

Key words: drought prediction; U.S. Drought Monitor; climate forecast; NMME

2

1

Introduction

A series of severe droughts in the U.S. in the past decade, such as 2012 central U.S. drought (Hoerling et al., 2014), have highlighted the necessity of early drought warning to reduce the potential impacts (Mariotti et al., 2013; Pozzi et al., 2013; Schubert et al., 2007). Substantial advances have been achieved in the past decade with different drought prediction methods from statistical and dynamical perspectives (Mishra and Singh, 2011; Mishra et al., 2015). A wide array of regional and global drought information systems has been developed to aid drought early warning (Hao et al., 2014; Luo and Wood, 2007; Nijssen et al., 2014; Sheffield et al., 2014). The availability of season climate forecast from state-of-the-art general circulation models (GCMs) has been among the most important advances in drought prediction in recent decades (Hao et al., 2017; Wood et al., 2015; Yuan et al., 2013). From the perspective of operational drought management, it is important to produce drought predictions in a probabilistic manner for informed decision making, which can be generally achieved through the multimodel ensembles approaches. The seasonal climate forecast through the North American Multimodel Ensemble (NMME) (Kirtman et al., 2014) plays an important role in the probabilistic prediction of drought (Becker et al., 2014; Shukla et al., 2014). Recently, the NMME forecast have been widely used for meteorological drought prediction or coupled with hydrological models to provide agricultural and hydrological drought prediction at different regions (Mo and Lyon, 2015; Thober et al., 2015; Yuan et al., 2015).

3

The U. S. Drought Monitor (USDM) (Svoboda et al., 2002) has been commonly used by a range of users, including policy makers and water resource managers, for drought monitoring in the U.S. which employs multiple drought categories, including abnormally dry (D0), moderate drought (D1), severe drought (D2), extreme drought (D3), and exceptional drought (D4), to indicate the degree of drought severity. The key feature of the USDM for drought characterization is the incorporation of multiple drought sources, including gauged observation, land surface simulations, remote sensing products, and inputs of impacts from local experts. This has inspired the development of several drought information systems to characterize drought with different drought categories consolidating multiple drought indicators (Hao et al., 2014; Lawrimore et al., 2002; Matthias et al., 2016). Development of objective approaches to integrate multiple sources of information for drought monitoring and prediction without losing the advantage of the USDM is a critical need as well as a challenge (Wood et al., 2015). Even though the USDM has been commonly used to evaluate drought indicators in different regions in the U.S. (Anderson et al., 2013; Anderson et al., 2011; McEvoy et al., 2012), few efforts have been devoted to incorporating USDM category in drought characterization and prediction (Hao et al., 2016a; Xia et al., 2014). With profound advances of the NMME in drought prediction and wide application of USDM, it would be of critical importance to predict USDM drought categories incorporating climate forecast from NMME. The objective of this study therefore is to develop a categorical drought prediction system based on USDM drought category and seasonal climate forecast from NMME. With the USDM providing the initial drought condition, the categorical drought prediction is achieved through the ordinal regression model with seasonal climate forecast from NMME as 4

predictors. Results of categorical drought prediction based on the categorical framework demonstrate its potential to aid the operational early drought warning in the U.S. 2

Methods

In this study, the USDM drought category was used as the “observed” drought condition. Multiple variables related to drought condition were used as predictors (denoted as a random vector X) for modeling the drought category. The drought condition labeled by n drought categories is denoted as a random variable Y for each month (Yt=1, … ,n). The relationship between drought category Y and random variable X can be modeled with the ordinal regression model (Hao et al., 2016a). The lagged USDM drought category was used as the initial condition and only one month lag USDM drought category was used mainly for parsimonious purposes. Thus, the statistical model for the non-exceedance probability P(Yt≤j) of the drought category Y for the time period t can be expressed as (Fokianos and Kedem, 2003; Hao et al., 2016a):  P(Yt ≤ j )   = α j + β1 X 1t + ... + β m X mt + γYt −1 log 1 P ( Y ≤ j ) t  

(1)

where Yt and Yt-1 are USDM drought categories for the period t and t-1, respectively; X=(X1, X2,…Xm) is the vector of m indices or variables that are related to drought condition or occurrence; P(Yt≤j) is the probability of the drought category less severe than a category j (j=1,2,…,n); αj is the intercept parameter; β =(β1, β2,…βm) is regression coefficients associated with m indices or variables. The model parameters were estimated separately for each lead time L. The intercept and regression coefficients were estimated with the maximum likelihood estimation (MLE) method 5

(Fokianos and Kedem, 2003; Hao et al., 2016a). For each grid, the model with different predictors for predicting USDM drought categories was selected by Akaike’s Information Criterion (AIC) (Akaike, 1974). By estimating P(Yt≤j), the probability of drought condition falling into each drought category P(Yt=j) was obtained, providing the uncertainty of drought category predictions. Note that the sum of probabilities of n drought categories for a specific month is equal to one, i.e., P(Yt=1)+… +P(Yt=n)=1.The drought category with the highest probability was regarded as the predicted category for each month (Hao et al., 2016a). 3 3.1

Data and evaluation metrics Data

USDM The USDM drought category is available from 2000 on a weekly basis (http://droughtmonitor.unl.edu/). The weekly USDM products for each month for the period 2000-2014 are averaged to obtain monthly USDM products. All the USDM data are digitized to the 1 degree resolution to match the resolution of NMME models. To obtain a relatively large sample for each drought category to facilitate the statistical modeling, USDM categories are reorganized into three drought categories (DC), including DC1 (no drought and D0 category), DC2 (D1 category) and DC3 (D2-D4 category). NMME The seasonal forecast of monthly precipitation and temperature from NMME of 1- and 3- month lead for the period 1982-2014 at a 1° spatial resolution in the U.S. were obtained from http://iridl.ldeo.columbia.edu/SOURCES/.Models/.NMME/(Kirtman et al., 2014). Forecast of six 6

models from the NMME project were used in this study, including (total 62 ensemble members): (1) COLA-RSMAS-CCSM3 (6 members), (2) COLA-RSMAS-CCSM4 (10 members) (3) GFDL-CM2p1-aer04(10 members), (4) GFDL-CM2p5-FLOR-A06 (12 members),(5) GFDLCM2p5-FLOR-B01 (12 members) and (6) NASA-GMAO-062012 (12 members). For all these models, the precipitation and temperature forecast for the whole period from 1982 to 2014 are available. 3.2

Evaluation metrics

To evaluate the performance of the proposed framework in predicting USDM drought categories, the persistence of the USDM drought category (i.e., Yt-1 in equation (1)) was defined as the reference forecast. The Rank Probability Skill Score (RPSS) was used as the probabilistic measure of prediction performance (Wilks, 2011), which is defined as.

RPSS = 1 −

< RPS > < RPS ref >

(2)

where the angle bracket <·> denotes the average of the scores over certain numbers of prediction and observation pairs; RPS is the Ranked Probability Score (RPS). For the observation with n categories, the RPS for a single observation and prediction pair is defined as:

RPS =

1 n (Pi − Oi )2 ∑ n − 1 i=1

(3)

where Pi and Oi are elements of cumulative prediction and observation vectors. The RPSS ranges from -∞ to 1. Postive values of RPSS indicate better performance of the proposed model than the reference forecast. 7

Meanwhile, the deterministic measure of the prediction performance is also used, which is defined based on the mean absolute error (MAE) of drought categories that can be expressed as:

MAE =

1 N

N



| Pc − Oc |

(4)

i =1

where Pc and Oc are observed and predicted categories. 4

Results

4.1

Model setting

The model in equation (1) is employed for the categorical drought prediction in the U.S., which is capable of modeling the USDM drought category with respect to a suite of drought related variables. The lagged USDM drought category is selected as a predictor to provide the initial drought condition. Since precipitation and temperature have a decisive influence on local hydrological cycle and are main variables provided by climate forecast, the precipitation and temperature anomaly from NMME climate forecast are used as predictors. The overall prediction framework is shown in Figure 1. For the L-month lead prediction of the USDM drought category, the prediction equation based on the initial USDM drought category, precipitation and temperature forecast can then be expressed from equation (1) as:  P(Yt + L ≤ j )   = α j + β1 Pa t + L + β 2Ta t + L + γYt + L −1 log 1 P ( Y ≤ j ) t+L  

8

(5)

where Yt+L is the L-month lead USDM drought category to be predicted; Pa t+L and Ta t+L are the anomaly of precipitation and temperature forecast for the period t+L from NMME; Yt+L -1 is the lag-1 U.S. drought category. 4.2

Model verification

The ensemble mean of precipitation and temperature forecast from NMME were obtained by averaging the forecast of all members with equal weight, from which the anomaly can be obtained. The quantile mapping method (Wood et al., 2002) was then used for the bias correction of the anomaly of ensemble mean of raw precipitation and temperature forecast from NMME. The anomaly of monthly mean precipitation and temperature from North America Land Data Assimilation System Phase 2 (NLDAS-2 (Xia et al., 2012b; Xia et al., 2012a)) were used as surrogates of observations for the period from 1982 to 2014 (Mo and Lettenmaier, 2014; Tian et al., 2014; Yoon et al., 2012). The leave one out cross validation (LOOCV) was used for assessing the predictive performance of the categorical prediction framework for the period from 2000 to 2014. Following (Tian et al., 2014), the target season was left out in creating the cumulative distribution function (CDF) of the precipitation and temperature anomaly of NLDAS-2 for bias correction. For the USDM of target month, the model was first estimated using the remaining (or training) datasets of corrected precipitation/temperature anomaly from NMME and initial USDM condition to estimate parameters, which was then used to predict the USDM of the target period (not used for parameter estimation) to assess its predictive performance. By repeating the same procedure, the predictions of USDM categories for each month during the period from 2000 to 2014 were obtained.

9

The RPSS values of 1- and 3-month lead prediction of USDM drought categories based on NMME for the period from 2000 to 2014 for the cross validation are shown in Figure 2 (a,b). In almost all regions, RPSS values were positive, implying that the proposed model outperformed the reference forecast (or the initial condition). With the increase of lead time, the RPSS was higher (and positive), indicating the degradation of persistence forecast and the added prediction skill from the climate forecast. The difference of MAE of the proposed method and the reference forecast is shown in Figure 2(c, d). For the 1-month lead prediction, the MAE difference was relatively small for most regions in the U.S.. The main reason likely resided in the high persistence of drought, for which the initial condition (or persistence) played an important role. For the 3-month lead prediction, the difference of MAE was relatively small in western regions and is relatively large (and negative) in eastern regions of the central U.S.. The negative MAE difference in eastern regions implied the proposed model performed better than the persistence forecast in these regions. These results above show that the categorical framework effectively integrated the initial condition from USDM and climate forecast from NMME to provide useful drought prediction information beyond the USDM initial condition (or NMME forecast) alone. 4.3

Case study of 2012 central U.S. drought

The application of the categorical prediction framework was illustrated with the case study of 2012 summer drought in central U.S.. This drought event occurred without an early warning (often referred to as “flash drought”) and was among the most severe droughts in recent years in U.S. resulting in huge losses to many sectors (Hoerling et al., 2014). The observed USDM drought categories for August 2012 is shown in Figure 3(a), along with those for July and May 2012 in figure 3(b, c), which are the initial condition for the 1- and 3- month lead prediction, 10

respectively. From the USDM observations, the drought condition mainly resided in southwestern region in May, and then expanded to the central and Midwest U.S. by August, 2012. The 1 and 3-month lead predictions of USDM for August 2012 is show in Figure 3 (d,e). Overall drought categories from the 1-month lead prediction were close to observations for this period in large regions. For example, the predicted drought severity for August 2012 initiated in July 2012 in the central U.S. resembled the observations relatively well. This is understandable, since the initial condition generally provides useful information for the prediction in the near future after the onset due to the drought persistence, as shown though comparisons with the initial condition of July 2012. The prediction generally degraded with the increase of lead time, as expected. For the 3-month lead prediction of August 2012 initiated in May 2012, the drought condition in the central region was not predicted well. It has been shown previously that the rapid development of dry condition during summer in central U.S. in 2012 largely resulted from the internal atmospheric process (or natural variations in weather) and the climate forecast generally did not predicted this drought well (Hoerling et al., 2014; Wang et al., 2014; Wood et al., 2015). The U.S. Seasonal Drought Outlook (SDO, http://www.cpc.ncep.noaa.gov/products/expert_assessment/sdo_summary.php) initiated in May 2012 showed that the drought condition mainly persisted in southwestern and southeastern regions while did not predict intensified drought condition in central Great Plains (Hoerling et al., 2014; Wood et al., 2015). Due to the lack of the forecast skill from the climate forecast, overall the 3-month lead prediction did not perform well and the predicted drought category largely resembled the initial condition from NMME, as shown from Figure 3(c, e). In certain regions 11

(e.g., central-west Texas), the 3-month lead prediction of USDM categories showed certain improvements over the initial condition (or reference forecast), which likely resulted from the climate forecast. These results indicate that integrating initial conditions from USDM and reliable climate forecast may aid USDM drought category prediction in the U.S.. Drought prediction in a probabilistic manner is important to convey the uncertainty of prediction for informed decision making by users. The categorical prediction framework facilitates the probabilistic drought prediction by estimating the probability of each drought category (with PDC1+PDC2+PDC3=1). The high probability (e.g., >0.70) indicates that the predicted drought category is very likely to occur in the target period, while a low probability indicates that there is low confidence in the predicted drought category. The probability of predicted drought condition falling in each drought category in August 2012 for 1-month and 3-month lead times is shown in Figure 4 (the probability DC1 “no drought” is not shown), which indicates the uncertainty of forecasted drought category. From Figure 4, the probability of drought category was high for short lead prediction (i.e., 1-month) in most regions and became relatively low in certain regions for the long lead prediction (i.e., 3-month). For the 3-month lead prediction, the probability of predicted drought category in the central U.S. was relatively low, indicating large uncertainty (or low confidence) in this region. However, for part of the southwestern regions, the probability of the estimated category was still high for the 3-month lead prediction, implying the high confidence of predicted drought category. This type of information would be useful to provide the uncertainty of prediction to aid decision making.

12

5

Discussion

The drought prediction skills of the proposed framework are mainly determined by two factors. First, it relies on the performance of using multiple drought indicators to model the U.S. drought category. It should be noted that the USDM composes a suite of indicators and thus ideally more drought indicators are needed to establish the relationship between USDM and drought indicators for prediction purposes. The potential limitation of this study is that only the precipitation and temperature are used to produce the drought category prediction. Since more variables have been used for developing the USDM, it is required to integrate more information from other sources (Nijssen et al., 2014; Sheffield et al., 2014; Thober et al., 2015), such as soil moisture and runoff from the land surface simulation or remote sensing with finer spatial resolution, to improve model performance. Second, the drought prediction performance also relies on the prediction skill of precipitation and temperature from climate forecast, which always depends on regions and seasons (Becker et al., 2014; Becker and Van Den Dool, 2016; Infanti and Kirtman, 2014; Mo and Lyon, 2015). The improvement of dynamical forecast of precipitation and temperature from NMME, which is critical for predicting land surface variables through hydrologic models, is needed for improving USDM prediction presented in this study. Instead of using the ensemble mean with equal weighting of NMME, recent developments of ensemble member selection or weighting methods may be used to improve the climatic forecast (Doblas-Reyes et al., 2013; Niko and Eric, 2016; Shukla et al., 2014; Thober and Samaniego, 2014; Wanders and Wood, 2016) and the USDM category prediction through the proposed framework. In addition, large-scale tele-connection patterns, such as El Niño–Southern Oscillation (ENSO) have been shown to affect drought 13

occurrence (Bonaccorso et al., 2015; Dutra et al., 2013; Hoerling et al., 2014; Mo, 2011; Schubert et al., 2016; Wang and Kumar, 2015), which may also be incorporated in the proposed framework for improving the USDM prediction in certain regions and seasons. These potential extensions of the proposed framework with improved ensemble selection and additional indictors (or predictors) will be addressed in the future. 6

Conclusions

A prototype of drought prediction system is presented in this study for predicting the USDM drought category in the U.S., based on the state-of-art climate forecast from NMME with the initial condition from USDM. It facilitates the probabilistic prediction of drought categories and would be useful to aid informed decision making. The cross validation based on USDM and climate forecast for the period from 2000 to 2014 showed that the proposed framework generally outperformed the reference forecast (or initial condition) in large regions in U.S.. Considering the wide application of USDM and advances of climate forecast from NMME, it is expected that the categorical drought prediction system would be useful to aid operational drought early warning in the U.S.. This study demonstrates the potential of objectively integrating USDM initial condition and climate forecast from NMME into a categorical drought prediction system. Since the USDM data record for model validation is relatively short, rigorous testing of the model performance in different regions or seasons is needed in the future. This study mainly provides a test bed for the development of categorical drought prediction system by using the precipitation and temperature forecast from NMME, while an improved drought information system incorporating other

14

variables based on land surface simulations with forcing from climate forecast (e.g., Climate Forecast System Version 2 (CFSv2)) is under development. The U.S. Seasonal Drought Outlook (SDO) produced by NOAA's Climate Prediction Center (CPC) with a subjective and manual process relies on forecaster expertise/judgment to consolidate drought information from sources such as CPC temperature and precipitation outlooks, long-lead forecasts from CFS, short-term forecasts from Global Forecast System (GFS) and ECMWF forecasts, along with current drought conditions from USDM (Wood et al., 2015). The categorical drought prediction framework presented in this study is expected to be useful in this regard for producing operational drought outlooks in the U.S. by objectively combining multiple forecast sources with an automated and reproducible approach.

15

7

Acknowledgments

This work is supported by the National Natural Science Foundation of China (NSFC, No. 41601014) and Youth Scholars Program of Beijing Normal University (Grant No. 2015NT02). Dr. Luo's effort is supported by NOAA Modeling, Analysis, Predictions and Projections (MAPP) program through grant NA12OAR4310081. We acknowledge the International Research Institute for Climate and Society (IRI) for making the NMME products available.

16

8

References

Akaike H. (1974). A new look at the statistical model identification. IEEE Trans. Automat. Contr., 19(6): 716-723. Anderson M. C., Hain C., Otkin J., et al. (2013). An intercomparison of drought Indicators based on thermal remote sensing and NLDAS-2 simulations with US Drought Monitor classifications. Journal of Hydrometeorology, 14(4): 1035-1056. Anderson M. C., Hain C., Wardlow B., et al. (2011). Evaluation of drought indices based on thermal remote sensing of evapotranspiration over the continental United States. Journal of climate, 24(8): 2025-2044. Becker E., den Dool H. v. and Zhang Q. (2014). Predictability and Forecast Skill in NMME. Journal of climate, 27(15): 5891-5906. Becker E. and Van Den Dool H. (2016). Probabilistic Seasonal Forecasts in the North American Multimodel Ensemble: A Baseline Skill Assessment. Journal of climate, 29(8): 3015-3026. Bonaccorso B., Cancelliere A. and Rossi G. (2015). Probabilistic forecasting of drought class transitions in Sicily (Italy) using standardized precipitation index and North Atlantic oscillation index. Journal of Hydrology, 526: 136150. Doblas-Reyes F. J., García-Serrano J., Lienert F., et al. (2013). Seasonal climate predictability and forecasting: status and prospects. Wiley Interdisciplinary Reviews: Climate Change, 4(4): 245-268. Dutra E., Giuseppe F. D., Wetterhall F., et al. (2013). Seasonal forecasts of droughts in African basins using the Standardized Precipitation Index. Hydrology and Earth System Sciences, 17(6): 2359-2373. Fokianos K. and Kedem B. (2003). Regression theory for categorical time series. Stat. Sci., 18 (3): 357-376. Hao Z., AghaKouchak A., Nakhjiri N., et al. (2014). Global integrated drought monitoring and prediction system. Scientific Data, 1: 140001. Hao Z., Hao F., Xia Y., et al. (2016a). A statistical method for categorical drought prediction based on NLDAS-2. J. Appl. Meteor. Climatol., 55(4): 1049-1061. Hao Z., Yuan X., Xia Y., et al. (2017). An overview of drought monitoring and prediction systems at regional and global scales. Bulletin of the American Meteorological Society, In press. Hoerling M., Eischeid J., Kumar A., et al. (2014). Causes and predictability of the 2012 Great Plains drought. Bulletin of the American Meteorological Society, 95(2): 269-282. Infanti J. M. and Kirtman B. P. (2014). Southeastern U.S. Rainfall Prediction in the North American Multi-Model Ensemble. Journal of Hydrometeorology, 15(2): 529-550. Kirtman B., Min D., Infanti J., et al. (2014). The North American Multimodel Ensemble: Phase-1 Seasonal-toInterannual Prediction; Phase-2 toward Developing Intraseasonal Prediction. Bulletin of the American Meteorological Society, 95(4): 585–601. Lawrimore J., Heim Jr R. R., Svoboda M., et al. (2002). Beginning a new era of drought monitoring across North America. Bulletin of the American Meteorological Society, 83(8): 1191-1192. Luo L. and Wood E. F. (2007). Monitoring and predicting the 2007 US drought. Geophysical Research Letters, 34(22): L22702. Mariotti A., Schubert S., Mo K., et al. (2013). Advancing Drought Understanding, Monitoring, and Prediction. Bulletin of the American Meteorological Society, 94(12): ES186-ES188. Matthias Z., Luis S., Rohini K., et al. (2016). The German drought monitor. Environmental Research Letters, 11(7): 074002.

17

McEvoy D. J., Huntington J. L., Abatzoglou J. T., et al. (2012). An evaluation of multiscalar drought indices in Nevada and Eastern California. Earth Interactions, 16(18): 1-18. Mishra A. K. and Singh V. P. (2011). Drought modeling-A review. Journal of Hydrology, 403(1-2): 157–175. Mishra A. K., Sivakumar B. and Singh V. P. (2015). Drought processes, modeling, and mitigation. Journal of Hydrology, 526: 1-2. Mo K. C. (2011). Drought onset and recovery over the United States. Journal of Geophysical Research (Atmospheres), 116(D15): 20106. Mo K. C. and Lettenmaier D. P. (2014). Hydrologic prediction over Conterminous US using the National Multi Model ensemble. Journal of Hydrometeorology, 15(4): 1457–1472. Mo K. C. and Lyon B. (2015). Global meteorological drought prediction using the North American Multi-Model Ensemble. Journal of Hydrometeorology, 16(3): 1409-1424. Nijssen B., Shukla S., Lin C.-Y., et al. (2014). A prototype Global Drought Information System based on multiple land surface models. Journal of Hydrometeorolgy, 15: 1661–1676. Niko W. and Eric F. W. (2016). Improved sub-seasonal meteorological forecast skill using weighted multi-model ensemble simulations. Environmental Research Letters, 11(9): 094007. Pozzi W., Sheffield J., Stefanski R., et al. (2013). Towards global drought early warning capability: expanding international cooperation for the development of a framework for global drought monitoring and forecasting. Bulletin of the American Meteorological Society, 94(6): 776–785. Schubert S., Koster R., Hoerling M., et al. (2007). Predicting drought on seasonal-to-decadal time scales. Bulletin of the American Meteorological Society, 88(10): 1625-1630. Schubert S. D., Stewart R. E., Wang H., et al. (2016). Global Meteorological Drought: A Synthesis of Current Understanding with a Focus on SST Drivers of Precipitation Deficits. Journal of climate, 29(11): 3989-4019. Sheffield J., Wood E. F., Chaney N., et al. (2014). A Drought Monitoring and Forecasting System for Sub-Sahara African Water Resources and Food Security. Bulletin of the American Meteorological Society, 95: 861-882. Shukla S., Funk C. and Hoell A. (2014). Using constructed analogs to improve the skill of National Multi-Model Ensemble March–April–May precipitation forecasts in equatorial East Africa. Environmental Research Letters, 9(9): 094009. Svoboda M., LeComte D., Hayes M., et al. (2002). The drought monitor. Bulletin of the American Meteorological Society, 83(8): 1181-1190. Thober S., Kumar R., Sheffield J., et al. (2015). Seasonal Soil Moisture Drought Prediction over Europe Using the North American Multi-Model Ensemble (NMME). Journal of Hydrometeorology, 16(6): 2329-2344. Thober S. and Samaniego L. (2014). Robust ensemble selection by multivariate evaluation of extreme precipitation and temperature characteristics. Journal of Geophysical Research: Atmospheres, 119(2): 594-613. Tian D., Martinez C. J., Graham W. D., et al. (2014). Statistical downscaling multimodel forecasts for seasonal precipitation and surface temperature over the southeastern United States. Journal of climate, 27(22): 8384-8411. Wanders N. and Wood E. F. (2016). Improved sub-seasonal meteorological forecast skill using weighted multimodel ensemble simulations. Environmental Research Letters, 11(9): 094007. Wang H. and Kumar A. (2015). Assessing the impact of ENSO on drought in the US Southwest with NCEP climate model simulations. Journal of Hydrology, 526: 30-41. Wang H., Schubert S., Koster R., et al. (2014). On the Role of SST Forcing in the 2011 and 2012 Extreme U.S. Heat and Drought: A Study in Contrasts. Journal of Hydrometeorology, 15(3): 1255-1273. Wilks D. S. (2011). Statistical Methods in the Atmospheric Sciences. Academic Press, San Diego, CA.

18

Wood A. W., Maurer E. P., Kumar A., et al. (2002). Long-range experimental hydrologic forecasting for the eastern United States. Journal of Geophysical Research: Atmospheres, 107(D20): 4429. Wood E. F., Schubert S. D., Wood A. W., et al. (2015). Prospects for Advancing Drought Understanding, Monitoring and Prediction. Journal of Hydrometeorology, 16(4): 1636–1657. Xia Y., Ek M. B., Peters-Lidard C. D., et al. (2014). Application of USDM Statistics in NLDAS-2: Optimal Blended NLDAS Drought Index Over the Continental United States. Journal of Geophysical Research: Atmospheres, 119(6): 2947–2965. Xia Y., Mitchell K., Ek M., et al. (2012b). Continental‐scale water and energy flux analysis and validation for North American Land Data Assimilation System project phase 2 (NLDAS‐2): 2. Validation of model‐simulated streamflow. Journal of Geophysical Research: Atmospheres (1984–2012), 117(D3). Xia Y., Mitchell K., Ek M., et al. (2012a). Continental‐scale water and energy flux analysis and validation for the North American Land Data Assimilation System project phase 2 (NLDAS‐2): 1. Intercomparison and application of model products. Journal of Geophysical Research: Atmospheres (1984–2012), 117(D3). Yoon J. H., Mo K. and Wood E. F. (2012). Dynamic-Model-Based Seasonal Prediction of Meteorological Drought over the Contiguous United States. Journal of Hydrometeorology, 13(2): 463-482. Yuan X., Roundy J. K., Wood E. F., et al. (2015). Seasonal forecasting of global hydrologic extremes: system development and evaluation over GEWEX basins. Bulletin of the American Meteorological Society, 96: 1895-1912. Yuan X., Wood E. F., Roundy J. K., et al. (2013). CFSv2-Based Seasonal Hydroclimatic Forecasts over the Conterminous United States. Journal of climate, 26(13): 4828–4847.

19

9

Figure

Figure 1 Flow chart of the categorical drought prediction of USDM drought categories based on USDM and climate forecast from NMME.

20

Figure 2 The rank probability skill score (RPSS) and mean absolute error (MAE) difference for the model validation based on NMME for the period from 2000 to 2014 for 1 and 3month lead prediction.

21

Figure 3 The observed USDM drought category for August 2012 (a) and initial condition in July (b) and May (c) for the 1 and 3-month lead prediction (d,e). ( 1, 2 and 3 in the colorbar represent the drought categories DC1, DC2 and DC3).

22

Figure 4 The probability of the 1-month (left) and 3-month (right) lead prediction of USDM drought category for August 2012 in the U.S.

23

Highlights



Predict the drought category incorporating the NMME seasonal forecast and USDM.



Assess the prediction performance in comparison to the initial USDM condition



Facilitate the probabilistic drought prediction for operational early warning.

24