1. Introduction
Subseasonal-to-seasonal (S2S) hydrological forecast systems (S2S-HFS) are a fundamental component of early warning systems because they provide hydrometeorological predictions in time scales of extended weather (2 weeks) to 2 years (Vitart and Robertson 2018), allowing stakeholders reasonable timeframes for disaster preparedness and management of water resources and agriculture activities. Western tropical South America (WTSA; Fig. 1), a region that includes Peru and Ecuador, might benefit from the development of an S2S-HFS given the preponderance of livelihood and socioeconomic activities that rely on rainfed agriculture systems and the value of associated reservoir management decisions to the region’s hydropower sector.
Western tropical South America (WTSA) and rainfall regimes from WTSA-LDAS for each ecoregion within the region. Numbers in the map indicate ecoregions. Ecoregions are based on the ecological classifications for Central and South America and the Caribbean defined by the U.S. EPA. The time series for rainfall season variability is based on precipitation averaged across a given ecoregion.
Citation: Journal of Hydrometeorology 25, 5; 10.1175/JHM-D-23-0064.1
In the last few years, S2S-HFS based on coupled atmospheric–ocean–land general circulation models (CGCMs) have gained popularity due to a better understanding of the physical basis of global seasonal climate predictions, which have been attributed in part to the presence of strong and persistent teleconnections associated with El Niño–Southern Oscillation (ENSO; Bjerknes 1969; Yuan et al. 2015a; Mo and Lyon 2015; Vitart and Robertson 2018). Such improvement in the understanding of S2S meteorological predictions is thanks to international efforts of climate institutions in the development of multiple intermodal comparison projects such as the North American Multi-Model Ensemble (NMME; Kirtman et al. 2014). The NMME project is supported by the Global Earth Observation System (GEOS) S2S forecast ensemble (Molod et al. 2012), which the National Aeronautics and Space Administration (NASA) developed. While most of the global NMME meteorological evaluations have focused on the ensemble mean of the system (e.g., Yuan et al. 2015b; Becker and van den Dool 2016), few others have assessed the performance of the ensembles of each model in the system (e.g., Krakauer 2019; Slater et al. 2016; Kirtman et al. 2014; Thober et al. 2015). The latter studies showed that GEOS skill varies geographically, by season, and depends on model initialization times. For example, Krakauer (2019) showed that GEOS was among the models that yielded the highest mean correlation for global warming rate at different longer lags, while Slater et al. (2016) showed that GEOS displays better skills in predicting extreme precipitation and temperature in diverse geographic areas across the United States. Additionally, Borovikov et al. (2017) demonstrated the capability of GEOS to predict precipitation and temperature over tropical regions around the globe, and particularly, they found an overall good correlation anomaly (larger than 0.4) for both temperature and precipitation in the Amazon region throughout the seasonal forecast. While these studies have focused primarily on problems of meteorological prediction, it is also recognized that they offer a valuable platform for studies of S2S-HFS. Indeed, S2S-HFS simulations often yield higher skill than the meteorological S2S used to drive them because the S2S-HFS can draw hydrological predictive skill from a combination of skill in the meteorological forecast and accuracy of initial land surface conditions (Koster and Suarez 2003). The relative contribution of meteorological fields and land surface initial conditions to S2S-HFS skill varies by region and by season (DeChant and Moradkhani 2014; Shukla et al. 2013; Wood and Schaake 2008; Yossef et al. 2013).
One of the challenges of using S2S meteorological forecast to construct the S2S-HFS lies in the inherent difficulty of propagating a skillful representation of atmospheric dynamics at S2S time scales. To alleviate this challenge, previous studies have proposed the importance of downscaling the atmospheric forecast of global systems before applying them to hydrological prediction (Gutmann et al. 2014; Zamora et al. 2021; Zhou et al. 2021). For that reason, we implement the generalized analog regression downscaling (GARD) algorithm (Gutmann et al. 2022) as it has been demonstrated to improve the skill of regional drought forecasting systems across various climate ecoregions in the United States (Zamora et al. 2021) and Asia (Zhou et al. 2021).
Further, S2S-HFS can benefit from recent improvements in land data assimilation systems (LDAS, e.g., Rodell et al. 2004). An effective LDAS can provide skillful initial conditions for an S2S-HFS forecast, offering another source of skill. In South America, there have been some efforts to implement advanced LDAS (de Goncalves et al. 2006a,b; Maertens et al. 2021; Recalde et al. 2021; Recalde-Coronel et al. 2022). De Goncalves et al. (2006a) used LDAS to initialize a regional climate model, concluding that LDAS increases the performance of precipitation forecast due to a slightly better prediction of the location of the South Atlantic convergence zone (SACZ). Meanwhile, Recalde et al. (2021) found that overall LDAS hydrological results were temporally and geographically consistent with ground-based and satellite observations. In a further study, Recalde-Coronel et al. (2022) constructed a set of retrospective hydrological simulations and showed that hydrological simulation results were dominated by meteorological forcing.
For the reasons mentioned above, there are grounds to believe that skillful subseasonal to seasonal hydrological predictions are possible across WTSA. This study investigates this potential, using an S2S-HFS that employs the NASA GEOS-S2S-V1 meteorological forecast system and an LDAS that uses the Noah multiparameterization model (Noah-MP; Niu et al. 2011; Yang et al. 2011) land surface models (LSMs). Although a few global and regional studies on S2S-HFS have cast light on S2S performance in South America (e.g., Rodriguez and Cavalcanti 2006; Yossef et al. 2013), to the best of our knowledge, none of them have focused on WTSA. Given that WTSA has unique climate and geography characteristics and its own socioeconomic challenges, an analysis of an S2S-HFS for the region is needed to contribute to the understanding of sources of S2S predictability and the reliability that S2S forecast would have on operational applications.
2. Data and methods
a. Study area
Here, we present an S2S-HFS developed for WTSA, a region that includes Peru, Ecuador, Colombia, and portions of Venezuela, Brazil, Bolivia, and Chile. We focus specifically on Peru and Ecuador (Fig. 1). These countries have a population of approximately 50 million inhabitants whose livelihoods and socioeconomic development rely heavily on water resources for agro-industry, hydropower generation, and water consumption (Recalde-Coronel et al. 2022). WTSA has three major subgeographic regions: the Coast, the Andes Mountains, and the Amazon rain forest. Each of these subregions has its unique set of landforms, vegetation, climate, and hydrology characteristics; for this reason, we used eight representative ecoregion zones (Fig. 1) based on levels II and III of the ecological classifications defined by the U.S. Environmental Protection Agency (USEPA 2011) to evaluate the S2S-HFS.
The seasonality of rainfall in the region is mainly controlled by the migration of the intertropical convergence zone (ITCZ) and the South America monsoon system (SAMS). On the Coast and Andes, the rainy season usually takes place in February–April months, while a portion of the rainy season of the Amazon occurs in March–May (MAM) months (Fig. 1). During March and April, rainfall maximums are observed on the Ecuadorian coast and across the eastern Andes slopes, while in May larger amounts of daily rainfall are only confined to the north of Ecuador (Fig. 2). This sequence of rainfall distribution across the MAM season is a product of the meridional migration of the ITCZ, as well as the northward migration of SAMS toward the equator, which supports part of the rainfall from April through June in the tropical Amazon (Rao and Hada 1990). One of the most important components of the SAMS is the SACZ (Carvalho and Cavalcanti 2016) which particularly influences the Amazon climate (Nobre et al. 2009). The variability of the SACZ is largely modulated by oscillations on subseasonal time scales that depend on tropical–extratropical interactions (Nogués-Paegle and Mo 1997; Liebmann et al. 1999; Carvalho et al. 2011) such as the Antarctic Oscillation (AAO) also known as the Southern Annular Mode (SAM; Carvalho et al. 2005). The AAO is the dominant pattern of atmospheric variability in the Southern Hemisphere (Rogers and van Loon 1982; Thompson and Wallace 2000), and its wave trains modulate the SACZ convection influencing precipitation variability in South America (Liebmann et al. 1999; Carvalho et al. 2004, 2005). At interannual time scales, the region exhibits significant hydroclimatic anomalies during both phases of ENSO (Aceituno 1988, 1989; Garreaud et al. 2009; Poveda et al. 2011). The coastal and Andean regions generally experience positive rainfall anomalies during the warm phase of ENSO (El Niño) and negative anomalies during the cold phase (La Niña), while the opposite is true for the Amazon. However, there is significant scatter in these relationships, as the regional precipitation response to ENSO is sensitive to ENSO flavor, to season, and to interactions between ENSO teleconnections and other drivers of climate variability.
Spatial distribution of the long-term mean from 2002 to 2017 for rainfall (mm day−1), evapotranspiration (ET; mm day−1), and soil moisture content (SMC) expressed as fraction [0, 1] during March–May. Rainfall and SMC come from WTSA-LDAS, while ET corresponds to ALEXI.
Citation: Journal of Hydrometeorology 25, 5; 10.1175/JHM-D-23-0064.1
Both evapotranspiration (ET; Fig. 2 second row) and soil moisture content (SMC; Fig. 2 third row) show lower month-to-month variability during the MAM season compared to rainfall. In the case of evapotranspiration, low intraseasonal variability in the Amazon might be due to the dense canopy forest and humid conditions in these ecoregions, which make for relatively high evapotranspiration throughout the year. Clay–loam and clay soil types found over most coastal and Amazon ecoregions also contribute to low intraseasonal variability in evapotranspiration and soil moisture, as these soil types have hydraulic properties that allow them to retain more water than sandy soils. For this reason, hydrological conditions in clay soils generally respond to forcings over longer time scales rather than short time scales (Miguez-Macho and Fan 2012).
b. Data
1) Observational dataset
We use two distinct dataset types to perform the evaluation of the hindcast system: one corresponds to satellite-derived products, and the other is a retrospective LDAS simulation available in the region. The satellite-derived datasets are the Atmosphere–Land Exchange Inverse (ALEXI; Anderson et al. 1997, 2007) diagnostic evapotranspiration product and the European Space Agency’s Climate Change Initiative for Soil Moisture (ESA CCI SM; Gruber et al. 2019). Meanwhile, the retrospective dataset corresponds to a land data surface model simulation of WTSA using Noah-MP (Recalde et al. 2021).
(i) ALEXI
The ALEXI evapotranspiration (ET) dataset is employed to assess the evapotranspiration retrospective forecast. ALEXI ET is a satellite-derived inverse model that calculates ET based on the relationship between time-differential land surface temperature observations from geostationary satellites (e.g., GOES) and the time-integrated energy balance within the surface–atmospheric boundary layer system (Anderson et al. 1997, 2007). Daily ALEXI ET estimates from 2002 to 2019 are available at a 0.05° resolution spanning 70°N–60°S. In assessment studies that compare ALEXI ET against flux tower measurements, Anderson et al. (2011) found favorable results over diverse climate zones across North America, while Salazar-Martínez et al. (2022) found low correlations but bias close to zero in wet tropical climates. Further, ALEXI ET has been used both to examine surface hydrological processes (Paca et al. 2019) and as a successful indicator of agricultural drought (Anderson et al. 2016) in the Brazilian Amazon basin.
(ii) ESA CCI SM
We use ESA CCI SM (v04.7) to evaluate soil moisture in the hindcast simulations. ESA CCI SM is a global combined active–passive satellite-based product available at a daily resolution on a 0.25° regular grid from 2012 through 2019 (Gruber et al. 2019). The ESA CCI SM product appears extensively in scientific publications and dataset applications worldwide. Evaluations indicate that ESA CCI SM generally captures soil moisture spatial and temporal patterns across climates, land cover, and soil types (e.g., Dorigo et al. 2015, 2017, and references therein).
(iii) WTSA-LDAS
This dataset is used for two purposes: 1) as initial conditions for the hydrological forecast system and 2) to validate the hydrological hindcast (i.e., precipitation, streamflow, soil moisture, and evapotranspiration) across WTSA. This dataset contains meteorological, surface energy balance, and surface hydrological simulations from 2002 to 2019 performed at 10-km resolution with Noah-MP LSM and the Hydrological Modeling and Analysis Platform (HyMAP; Getirana et al. 2012) for a region that encompasses WTSA from the Pacific Coast through the western Amazon (1°N–18°S, 82°–67°W). Outputs were applied at daily temporal resolution. WTSA-LDAS is derived from a combination of models and observations, in which Noah-MP is driven with Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2) and the Climate Hazards Infrared Precipitation with Stations, version 2 (CHIRPS) meteorological fields. Recalde et al. (2021) performed a regional assessment of WTSA-LDAS retrospective simulations of rainfall, streamflow, soil moisture, and evapotranspiration using both observed and satellite-derived datasets for evaluation, and they found that, in general, Noah-MP/HyMAP captures the diverse hydrological seasonal patterns across the region; however, some biases exist for all variables, particularly in complex terrains.
2) Forecasting dataset
In this study, we implement the GARD framework to downscale the GEOS-S2S-V1 dataset to obtain the surface meteorological fields for the region of interest (i.e., WTSA). GARD requires three input datasets for downscaling: a prediction dataset, an observational meteorological dataset taken to be “truth,” and a training dataset. The description of each dataset is as follows:
(i) Prediction dataset
GEOS subseasonal to seasonal forecast ensemble, version 1 (GEOS-S2S-V1; Molod et al. 2012; Borovikov et al. 2017) is utilized as the prediction dataset. Although an updated version of GEOS exists, the earliest version (V1) was preferred because it has a long hindcast record to evaluate its performance. GEOS-S2S-V1 is a global meteorological forecast hindcast dataset that consists of 10 ensemble members with a horizontal gridded resolution of 1° × 1.25° and is available from 2000 to 2017. Each of the ensemble members of GEOS-S2S-V1 is initialized (executed) every 5 days and valid for 9 months. In this analysis, we select all 10 ensemble members from GEOS-S2S-V1 that are initialized on 1 March each year during the hindcast period. Our analysis focuses on the 3-month lead period because previous analysis has found that after 3 months, GEOS-S2S-V1 precipitation skill is very limited (Zhou et al. 2021).
Using the 10 GEOS-S2S-V1 ensemble members, we selected the following eight GEOS-S2S-V1 near-surface meteorological fields to be spatially downscaled with GARD using the observational dataset described below: air temperature (Tair), specific humidity (Qair), air pressure (Psurf), zonal (Zwind) and meridional (Mwind) wind speed, downward shortwave radiation (SWdown), downward longwave radiation (Lwdown), and precipitation (Rain). We used specifically this set of meteorological forcing fields because they are the same ones that were used in the development of the WTSA-LDAS dataset (Recalde et al. 2021), and thus allows for a consistent comparison between the hindcast and the WTSA-LDAS dataset.
(ii) Observational dataset
The observational meteorological forcings used as a reference to downscale the atmospheric predictor’s dataset in the hydrological forecast system come from MERRA-2 and CHIRPS (Funk et al. 2015). They are used as the observed meteorological forcings to downscale the predictor dataset in the hydrological forecast system. MERRA-2 is used as a baseline for the meteorological forcings fields, while CHIRPS is used as a reference for the precipitation field. MERRA-2 is the latest atmospheric reanalysis of the modern satellite era produced by NASA and GMAO. MERRA-2 is derived from a combination of the GEOS model which provides the dynamical basis, and satellite observations which are assimilated into the model to provide an optimal reanalysis. MERRA-2 has a spatial resolution of about 50 km in the meridional direction and 67 km in the zonal direction, and it provides data from 1980 to the present (Gelaro et al. 2017). Draper et al. (2018) suggested that for global land annual averages, MERRA-2 appears to overestimate the land surface latent heating and sensible heating and the downwelling shortwave radiation while underestimating the downwelling and upwelling longwave radiation. Furthermore, CHIRPS is a quasi-global daily rainfall dataset that spans from 50°S to 50°N (and all longitudes) with a high resolution of 0.05°. CHIRPS blends infrared geostationary satellite data with in situ station rainfall observations, and it is available from 1981 to date. Several studies have assessed CHIRPS performance using local rain gauges at different sites over South America. These studies demonstrate that overall, CHIRPS adequately represents the seasonal and interannual spatial and temporal precipitation variability; however, large biases were found, particularly in complex geographic regions (e.g., Ceccherini et al. 2015; Paredes-Trejo et al. 2017; Rivera et al. 2018; López-Bermeo et al. 2022).
(iii) Training dataset
The dataset to train GARD was developed by Zhou et al. (2021). This training dataset pairs MERRA-2 and CHIRPS observations with the first 5 days of GEOS-S2S-V1 forecasts throughout the entire period. Only the first 5 days of the forecast are used in order to minimize the impact of forecast divergence on regression fit, and this dataset is solely used to downscale the meteorological fields using GARD.
c. Hydrological subseasonal reforecast (hindcast)
Our S2S-HFS applies NASA’s Land Information System framework (LIS; Kumar et al. 2006) to perform the hydrological simulations using the downscaled GEOS-S2S-V1 forecast. For the present study, we focus exclusively on the austral autumn season (MAM), because it captures both a part of the rainy season in the Coast and Andes that occurs in FMA months and a portion of the rainy season of the Amazon that occurs in MAM months (Fig. 1). The S2S-HFS can be considered as a two-step system approach: 1) downscaling the GEOS-S2S-V1 meteorological fields using GARD and 2) applying the downscaled GEOS-S2S-V1 forecast to drive hydrological simulation in LIS. The approach of the hydrological forecast system is described below and summarized in Fig. 3.
Schematic representation of the methodology used in this study.
Citation: Journal of Hydrometeorology 25, 5; 10.1175/JHM-D-23-0064.1
1) Forecast downscaling
The daily GEOS-S2S-V1 meteorological forcing data are spatially downloaded and bias-corrected using GARD. One of the advantages of the GARD downscaling algorithm is that it offers flexible analog and regression options that allow the user to set the best configurations to improve downscaling accuracy for any given application. For example, in a study for the United States, Zamora et al. (2021) showed that GARD downscaling led to improved representation of temperature across the territory, and although results for precipitation were mixed, GARD increased the skill in predicting drought relative to using GEOS-S2S-V1 data without downscaling. Similarly, Zhou et al. (2021) found that the root-mean-square error (RMSE) was reduced for most of the meteorological fields after applying GARD, which improved the S2S-HFS over many subregions in Asia. Given that those configurations benefit the hydrological S2S forecast over diverse geographic regions, in this study we apply similar GARD configurations used in Zamora et al. (2021) and Zhou et al. (2021). This includes the use of the analog–regression configuration, using univariate regression between the observation and forecast variable. In GARD, we set the number of analogs to 100 for rainfall (∼1.5% of the training dataset) and 250 for the other atmospheric variables (∼3.8% of the training dataset). In addition, we fix the option to weigh each analog by its inverse-square distance from the current predictand and apply a cube root transform to the input rainfall data to overcome the zero inflation in records of 6-hourly precipitation (Zamora et al. 2021). Using these configurations, each of the 10 ensemble members of the GEOS-S2S-V1 is downscaled to the resolution of the observation dataset. Recalling that MERRA-2 and CHIRPS are used as the observational dataset for atmospheric fields and precipitation field, respectively. The downscaled GEOS-S2S-V1 dataset has a resolution of around 5 km for the precipitation field and around 50 km for the other meteorological fields resolution.
Before using the downscaled GEOS-S2S-V1 dataset as a forcing on LIS, we need to disaggregate GEOS-S2S-V1 meteorological fields from daily to 6-hourly estimates to capture subdaily variations. To achieve this, we first calculated a scale factor from the subdaily observed fields from MERRA-2 and CHIRPS. The scale factor for solar radiation and precipitation forecast is the ratio between the 6-hourly long-term mean and the daily long-term mean using MERRA-2 and CHIRPS, respectively, while the scale factor for the additional atmospheric forcings is calculated as the difference between the daily long-term mean and the 6-hourly long-term mean using MERRA-2. Then, each respective scale factor is applied using an inverse operation to the corresponding GEOS-S2S-V1 meteorological variables to be transformed from daily to subdaily frequency. Once the daily GEOS-S2S-V1 dataset has been downscaled to a 6-hourly subdaily cycle, it is used as the meteorological forcing for LIS simulations with Noah-MP LSM to produce the 3-month hydrological forecast initialized on 1 March for all 10 ensemble members. Hereafter we refer to the spatially and temporally downscaled GEOS-S2S-V1 meteorological dataset as DF_GEOSV1.
To examine the spatial performance of the downscaling method, Fig. 4 shows an example of the long-term mean for March of total precipitation (top panel) and 2-m temperature (bottom panel) for the observed values (WTSA-LDAS), raw GEOS-S2S-V1, and the downscaled dataset (DF_GEOSV1). As expected, GEOS-S2S-V1 shows a coarser resolution than DF_GEOSV1 over the region for both precipitation and temperature. In terms of rainfall, DF_GEOSV1 captures the spatial distribution of precipitation much better than GEOS-S2S-V1; particularly, the bias-correction method benefits the distribution of precipitation over the Andes Mountains. For temperature, however, DF_GEOSV1 has large differences with respect to WTSA-LDAS across the region; for example, along the coastal areas and the Ecuadorian Andes temperatures are cooler and warmer than observed, respectively. Overall, GARD resolves properly the complex Andes topography, and captures better the spatial distribution of precipitation than temperature across the ecoregions. This is important in regions such as WTSA, where it has been shown that hydrological simulations depend highly on the precipitation forcings field (Recalde-Coronel et al. 2022).
March long-term averaged (2002–17) (top) total precipitation (mm day−1) and (bottom) 2-m temperature (K) for WTSA-LDAS (observed dataset), raw GEOS-S2S-V1, and downscaled precipitation using GARD (DF_GEOSV1).
Citation: Journal of Hydrometeorology 25, 5; 10.1175/JHM-D-23-0064.1
2) Hydrological hindcast
We conducted the S2S-HFS hindcast simulations from 2000 to 2017 using Noah-MP in offline mode within the LIS framework. The S2S-HFS hydrological simulations (Recalde-Coronel and Zaitchik 2024) consist of daily outputs at 10-km spatial resolution over WTSA (1°N–18°S, 82°–67°W). The S2S-HFS is run using the DF_GEOSV1 dataset as a meteorological forcing, and the WTSA-LDAS retrospective simulation dataset as initial conditions (Fig. 3). We note that on the initialization date (i.e., 1 March) all the hydrological states (e.g., soils moisture, water level) are the same for both the hydrological hindcast system and WTSA-LDAS retrospective simulations. One advantage of using WTSA-LDAS as initial conditions is that it was constructed with the same meteorological forcing fields (e.g., Tair, Qair, Rain, etc.) and Noah-MP land surface model parameterizations that the S2S-HFS simulations. It is important to mention that DF_GEOSV1 is further interpolated to 10-km horizontal resolution within LIS using both the static environmental lapse-rate and the slope-aspect correction methods (Arsenault et al. 2018). In this matter, the spatial and temporal comparisons between S2S-HFS and WTSA-LDAS are consistent.
d. Verification methods (diagnostics and metrics)
The ensemble mean of the S2S-HFS is evaluated at a monthly time scale for the 3-month lead time hindcast across diverse ecoregions in the study domain. S2S-HFS evaluation is performed using satellite-derived products and WTSA-LDAS retrospective simulations. Hindcast comparisons with satellite-derived hydrological estimates offer valuable independent evaluation. These comparisons are complicated, however, by uncertainty in the satellite data products, contrasting climatologies between model and satellite datasets, and in some cases by limitations in spatial or temporal coverage. For this reason, we also apply WTSA-LDAS dataset for hindcast evaluation. Using these evaluation datasets, both deterministic and probabilistic verification approaches are applied to evaluate the S2S-HFS, which is initialized on 1 March to forecast the austral autumn (MAM) for the period 2000–17.
The accuracy of the ensemble mean is assessed using mean error, linear correlation, and RMSE maps. For probabilistic evaluation, we measure the skill of the hindcasts using the Brier score (BS; Brier 1950) and the ranked probability score (RPS; Epstein 1969; Murphy 1970). The BS score can only be used for binary outcomes (e.g., it was below normal, or it was not below normal), while RPS score can be used to evaluate a multicategory probabilistic forecast. Using these scores, we compare the hindcast values to the WTSA-LDAS values for a specific category. We used BS to evaluate below-average and above-average categories, and RPS to evaluate three event categories: below-average, near-average, and above-average categories. Each category is calculated as the corresponding tercile value (0.33, 0.67) from the mean and the spread of each dataset. Therefore, below average corresponds to data in the lower tercile (<0.33), near average corresponds to the data within the 0.33 and 0.67 tercile, and above average corresponds to data that is larger than the 0.67 tercile.
In addition, we used the Brier skill score (BSS) and the ranked probability skill score (RPSS) to explore the relative skill of the probabilistic hindcast over that of a reference forecast to predict whether an event occurred or not (Wilks 2011). A reference forecast usually consists of a climatology forecast, a random forecast, or a persistence forecast. Here, we use the climatology of the WTSA-LDAS dataset as the control forecast because we assume that in the absence of a forecast system, the climatological values would be used instead.
Additionally, we examine impacts on precipitation hindcast performance for the first lead month associated to sea surface temperature (SST) anomalies in the Pacific Ocean and atmospheric teleconnection patterns linked with the Antarctic Oscillation (AAO; Thompson and Wallace 2000). For this purpose, we use the Niño-3.4 SST monthly anomaly based on the Extended Reconstructed Sea Surface Temperature (ERSST.v5; Huang et al. 2017) averaged over the Niño-3.4 region (5°N–5°S, 120°–170°W). For AAO we use the AAO index from the National Oceanic and Atmospheric Administration Climate Prediction Center (NOAA/CPC). The AAO index is constructed as the leading mode from empirical orthogonal function analysis of the monthly mean 700-mb (1 mb = 1 hPa) height anomalies south of 20°S.
3. Results and discussion
a. Deterministic verification scores
1) Rainfall
Rainfall correlation maps (first column of Fig. 5) illustrate statistically significant correlations between hindcasts and WTSA-LDAS across the region for 1-month lead (March). For April and May, however, localized positive values within the 90% confidence level are observed only over northern Napo-Putumayo, Yungas, and southern Amazonia forest. The correlation skills observed in Fig. 5 in part come from downscaling the meteorological forcings using CHIRPS in the training data in GARD, which influences the variability of DF_GEOS-V1 precipitation. These findings are consistent with previous studies that found meaningful subseasonal forecasting skills over tropical regions at 1-month lead time (Yuan et al. 2011; Wanders and Wood 2016; Bombardi et al. 2018). Notably, Bombardi et al. (2020) reported subseasonal prediction skills for the onset and demise dates of the monsoon system over South America at around 30 days lead times. Moreover, additional precipitation skills for 1-month lead forecast might come from the initial soil conditions, as in a weather precipitation forecast using LDAS, de Goncalves et al. (2006a) found that a climate model estimated properly surface precipitation and attributed these findings to an accurate prediction of SACZ position in South America.
Rainfall spatial distribution of correlation, percentage mean error (bias), and the root-mean-square error (RMSE; mm) between hindcast and WTSA-LDAS for 1-month lead (March), 2-month lead (April), and 3-month lead (May). Statistical significance was assessed with a t test: stippling indicates significance at 90% confidence.
Citation: Journal of Hydrometeorology 25, 5; 10.1175/JHM-D-23-0064.1
The spatial character of the sign and magnitude of the percentage of error mean varies across the MAM season (second column of Fig. 5). In March, a wet (dry) bias is observed toward the west (east) of the region. Conversely, a dry (west) bias is observed in the eastern (western) during May. However, through the MAM season, mean errors are lower than 18% for both under- and overestimation in most of the territory. This reflects the general success of GARD for bias correction of the raw GEOS-S2S-V1 precipitation (Fig. 4). We note that biases are not zero because GARD is trained using only the first 5 days of a forecast—that is, it is not a conventional post hoc bias correction method that perfectly corrects against climatological bias at all time steps.
Spatial distributions of RMSE (Fig. 5, bottom) illustrate better hindcast skills over the Andes and Peruvian coastal ecoregions. In contrast, RMSE larger than 2.2 mm day−1 is found over the Ecuadorian coastal and Amazon ecoregions. A larger RMSE in these wetter areas (see Fig. 3) is not surprising, given the larger magnitude of the precipitation signal in both observations and models that are characterized to have more climatological seasonal rainfall (Fig. 2).
2) Evapotranspiration
Comparing evapotranspiration (Fig. 6) and precipitation (Fig. 5) forecast skills, we observe that for many portions of WTSA, evapotranspiration has larger positive correlation coefficients, lower bias, and lower RMSE values than precipitation throughout the MAM season, particularly at longer lead times. Regarding evaporation correlation maps (first column of Fig. 6), we can see that although negative correlations are observed as the lead forecast months increase, positive correlations higher than 0.6 are maintained throughout the MAM season in coastal ecoregions, eastern Napo-Putumayo Forest, and southern Irregular Forest. Since the WTSA-LDAS and hindcast simulations use the same land surface model, high positive correlation scores (higher than 0.6) on these ecoregions suggest that DF_GEOSV1 precipitation meteorological forcing is skillful enough (Fig. 4) to provide reliable initial conditions to the land surface model to reproduce the temporal and spatial variability of evapotranspiration through the season. Meanwhile, negative correlations at longer leads might indicate that the impact of the initial conditions has been reduced after a few months of the initialization date.
As in Fig. 5, but for evapotranspiration.
Citation: Journal of Hydrometeorology 25, 5; 10.1175/JHM-D-23-0064.1
In terms of magnitude, the hindcast tends to overestimate evapotranspiration over the Ecuadorian Andes, Napo-Putumayo, and southern Peruvian ecoregions, while underestimating evapotranspiration over the Pacific Plains, central areas of both Peruvian Andes and Yungas Forest, and Ucayali-Irregular Forest (second column of Fig. 6). The spatial distribution of low values of RMSE (third column of Fig. 6) agrees with low percentages of mean error distributions, reflecting the model’s ability to capture the interannual variability of the mean magnitudes in most parts of the region. There appear to be some spatial disconnects between percentage bias between rainfall (second column of Fig. 5) and evaporation (second column of Fig. 6), but these differences are lower than 10%.
Larger RMSE is observed in certain areas within ecoregions, such as Dry Hills and Yungas, which show evapotranspiration bias larger than 20%. As with percent bias, there are spatial inconsistencies in RMSE performance between rainfall and evapotranspiration. Hence, Fig. 5 illustrates that areas with wet (dry) evapotranspiration bias coincide with dry (wet) rainfall bias shown in Fig. 5 (e.g., Dry Hills, Yungas). The spatial differences in the forecast skills reflect the contribution of initial conditions on the simulation of evapotranspiration.
Figure 7 compares hindcast to WTSA-LDAS, which offers a controlled view of hindcast performance but is not an independent evaluation of the system. In situ observations of evapotranspiration are extremely limited in this region (and over most of the world), so we rely on comparisons with the ALEXI satellite-derived ET dataset for an independent evaluation of hindcast evapotranspiration. Not surprisingly, this comparison yields substantially lower estimates of skill than seen in Fig. 6. Correlations are of mixed sign throughout the season, and percent error and RMSE are large, reflecting the different climatologies and magnitudes of variabilities between these products. ALEXI comparisons suggest that DF_GEOSV1 tends to underestimate evapotranspiration over Dry Hills and the center of the Andes and Ucayali-Irregular Forest while overestimating most areas of the Andes ecoregions. However, since there might be uncertainties in the ALEXI ET dataset related to the poor coverage of ground ET observation, evaluation of the magnitude of the bias has to be taken with caution (Mueller et al. 2011). While keeping this caveat in mind, it is intriguing that correlations tend to be negative over the humid Amazonian ecoregions, given that there are debates regarding physical controls on interannual variability of evapotranspiration in these environments: in these humid areas, energy limitation might dominate over water limitation. Negative correlations, then, might be an indication that Noah-MP does not simulate these climate sensitivities appropriately.
Evapotranspiration spatial distribution of correlation, percentage mean error (bias), and RMSE (mm) between hindcast and ALEXI satellite dataset for 1-month lead (March), 2-month lead (April), and 3-month lead (May). Statistical significance was assessed with a t test: stippling indicates significance at 90% confidence.
Citation: Journal of Hydrometeorology 25, 5; 10.1175/JHM-D-23-0064.1
3) Soil moisture content
Figure 8 shows that soil moisture content (top 10 cm of the soil) hindcast generally has greater skill than was seen for rainfall (Fig. 5) or evaporation (Fig. 6). There are positive correlations between hindcast and WTSA-LDAS that are higher than 0.7 across the region, and strong positive correlations persist throughout the MAM season (first column of Fig. 8). Comparing rainfall correlation (Fig. 5) to SMC (Fig. 8), we can see that rainfall correlation skill after 1 month is not large in ecoregions (e.g., northern Ucayali-Irregular Forest, Yungas Forest, and southern Peruvian Andes) where SMC has much higher correlations; therefore, we can implied that soil moisture skill is derived from the initial conditions (i.e., soil moisture at the beginning of the forecast period). These findings agree with previous studies that have found that the soil moisture forecast skills at a lead time of 1–3 months come from initial conditions (Mo and Lyon 2015).
As in Fig. 5, but for soil moisture content.
Citation: Journal of Hydrometeorology 25, 5; 10.1175/JHM-D-23-0064.1
In general, the percentage of mean error for soil moisture content is in the range of ±5% (second column of Fig. 8), and RSME values stay low across ecoregions, demonstrating the ability of the hindcast to replicate soil moisture magnitudes (third column of Fig. 8). Interestingly, even though soil moisture content has a minimal wet bias across the study region, the bias appears to be large enough to trigger wet bias in evapotranspiration over semiarid ecoregions (e.g., Dry Hills, Garua-Loma desert, small area of the Andes; second column of Fig. 6). The high performance of hindcast in most parts of the region might be due to the slow state evolution of soil moisture (Fig. 4), which maintains anomalies in larger time scales than the atmosphere (Koster and Suarez 2003) and thus preserves skill inherited from initial conditions for a longer time period.
Comparisons between soil moisture hindcast and the ESA CCI SM satellite dataset are restricted by the availability of ESA CCI SM data, which is extremely limited in mountainous and forest regions. Figure 9 illustrates that correlation coefficients are lower, percentage bias mean errors are larger, and RMSE has higher scores than those observed in Fig. 8. Climatologically, the hindcasts are drier than ESA CCI SM (Fig. 9, middle column) and RMSE is substantial. Again, significant uncertainty in the satellite-derived product prevents us from making firm conclusions about the skill of the hindcasts, but the inconsistency between satellite-informed soil moisture estimates and those produced by our modeling system emphasize the challenge of evaluating soil moisture forecasts for the region.
Soil moisture content (SMC) spatial distribution of correlation, percentage mean error (bias), and RMSE (mm) between hindcast and ESA CCI SM satellite dataset for 1-month lead (March), 2-month lead (April), and 3-month lead (May). Statistical significance was assessed with a t test: stippling indicates significance at 90% confidence.
Citation: Journal of Hydrometeorology 25, 5; 10.1175/JHM-D-23-0064.1
To further evaluate the hindcast simulations, Fig. 10 compares the climatology for each month of the hindcast (dashed lines) to WTSA-LDAS (solid line) for precipitation, evapotranspiration, and soil moisture content averaged over each ecoregion. As can be seen, the hindcasts replicate the observed seasonality for all hydrological variables in each ecoregion. This suggests that the implementation of GARD to bias-correct and download the meteorological forcing from FD_GEOSV1 (Fig. 4) has contributed to the successful prediction of the hydrological variables across WTSA. Thus far, the results indicate that S2S-HFS could be useful in an early warning system approach by providing hydrological estimates up to 3 months in advance. However, limitations on the hindcast system still exist as the temporal time series shows a large bias in some ecoregions that was also observed in the previous bias maps. For example, the substantial hindcast overestimation of rainfall (Fig. 5) and soil moisture (Fig. 8) over the Dry Hills ecoregion is observed in Fig. 9b. Therefore, additional S2S-HFS studies may focus on examining the use of recently developed capabilities that have implemented NMME precipitation forecast forcing within LIS, which has led to improvements in hydrologic forecast estimates than using GEOS-only-based forecast (Hazra et al. 2023).
Time series for the annual cycle comparing hindcast simulation (dashed line) and WTSA-LDAS retrospective simulation (solid line) for evapotranspiration (ET), rainfall, and soil moisture content (SMC) for 1-month lead (March), 2-month lead (April), and 3-month lead (May).
Citation: Journal of Hydrometeorology 25, 5; 10.1175/JHM-D-23-0064.1
b. Probabilistic verification scores
To perform the probabilistic verification, we used the BS and the RPS as well as the corresponding skill scores, BSS and RPSS, respectively, to evaluate the performance of a specific category of the hindcasts against WTSA-LDAS dataset. Figure 11 shows the BS for below-average (top-left panel) and above-average (top-right panel) conditions for rainfall, evapotranspiration, and soil moisture content for all ecoregions. Results show that hindcasts have better skills for below-average conditions than for above-average conditions for all hydrological variables. For the below-average category, better BS are observed for rainfall and evapotranspiration in the Amazon Forest and the northern coastal ecoregions (Garua-Loma desert and Dry Hills). In particular, BS for evapotranspiration indicates positive hindcast skills throughout the season for the Dry Hills and Pacific Plains. Regarding above-average rainfall and evapotranspiration scores, Fig. 11 (top-right panel) shows stronger hindcast performance throughout MAM season over the Ecuadorian ecoregions: Dry Hills, Garua-Loma desert, Ecuadorian Andes, and Napo-Putumayo Forest. Moreover, BS results show that hindcast performance for soil moisture content is generally better than rainfall and evapotranspiration in both categories. Hindcast skill for soil moisture persists for the entire 3-month period in various ecoregions.
Evapotranspiration (ET), rainfall (RR), and soil moisture content (SMC) heatmap of (top) BS for below average and above average and (bottom) RPS for three categories (below average, near average, and above average) for 1-month lead (March), 2-month lead (April), and 3-month lead (May) hindcasts. BS and RPS compare a specific category of the hindcast against the same specific category of WTSA-LDAS. For BS and RPS, a score of 0 means perfect accuracy.
Citation: Journal of Hydrometeorology 25, 5; 10.1175/JHM-D-23-0064.1
RPS results (Fig. 11, bottom panel) generally agree with BS findings, showing better hindcast skills in similar ecoregions to BS. Thus, in March, lower RPS values for rainfall are observed in the Amazon. Further, good scores are observed during MAM season over the Pacific Plains, Dry Hills, and Garua-Loma desert ecoregions for evapotranspiration. However, unlike BS, evapotranspiration RPS values decay in April or May in some of these ecoregions. Like BS, RPS reports the best scores for soil moisture across the 3-month forecast period, reflecting the long memory of that variable.
Figure 12 presents the BSS and RPSS scores for all ecoregions to analyze the hindcast’s performance relative to a reference forecast (climatology). Overall, both scores illustrate similar results in the sense that BSS and RPSS outperform the reference forecast in the same months for the same ecoregions. Rainfall hindcast is generally low skill, but some positive scores are seen in March, with larger scores observed over Napo-Putumayo and Ucayali-Irregular Forest ecoregions. The evapotranspiration hindcast is relatively better than the reference forecast beyond March for Ecuadorian coastal ecoregions (Plains Hills, Dry Hills, and Garua-Loma desert). Soil moisture content again shows the best scores for most ecoregions across all months. Last, we noticed that, even though BS and RPS showed lower skills in some ecoregions in Fig. 10, BSS and RPSS suggest that the S2S-HFS still brings a certain level of forecast skill to those ecoregions when it is compared to the climatological forecast.
As in Fig. 11, but for skill scores. BSS and RPSS measure the relative skill of the hindcast compared to the climatology reference from WTSA-LDAS for a specific category. For BSS and RPSS, a score above 0 indicates that the hindcast benefits with respect to the reference.
Citation: Journal of Hydrometeorology 25, 5; 10.1175/JHM-D-23-0064.1
c. SST anomalies and regional atmospheric features
As the occurrence of precipitation variability in South America has been associated with SST anomalies in the Pacific Ocean (Grimm and Tedeschi 2009), and with the AAO pattern (Carvalho et al. 2005), this section explores the ability of the hindcast to capture the influence of both SST anomalies and AAO teleconnections on MAM precipitation anomalies in WTSA. Although there are interactions between ENSO and AAO (e.g., Carvalho et al. 2005) that may explain some of the similarities on the univariate evaluations presented here, we note that ENSO and AAO exhibit low correlation in our season of interest (correlation coefficient between the two is 0.28 in March). Thus, it is worth examining associations between rainfall patterns and each index independently.
1) SST anomalies over the Pacific Ocean
Figure 13a illustrates the monthly correlation between WTSA-LDAS precipitation anomalies and SST monthly anomalies averaged across the Niño-3.4 Pacific Ocean region (hereinafter, the Niño-3.4 index). As can be seen, during March and April, positive correlations are observed mainly in ecoregions located between 0° and 8°S (e.g., Dry Hills, Napo-Putumayo Forest, Ucayali-Irregular Forest). These positive correlations can be attributed to the intensification of the ascending branch of the Walker circulation and the southern displacement of ITCZ (Tapley and Waylen 1990) associated with warmer than normal SST observed in the Niño-3.4 Pacific region. Accordingly, both atmospheric features enhance moisture flux and convective activity, driving positive rainfall anomalies closer to the equator. On the contrary, negative correlations observed over the Andes ecoregions agree with previous studies (Vuille et al. 2000a,b; Francou et al. 2004) that have shown that when there are positive SST anomalies over the central Pacific, the Andes Mountains tend to experience below-normal precipitation as a result of an anomalous Hadley cell that inhibits convection over the high terrain.
Monthly correlation maps between the Niño-3.4 index and (a) WTSA-LDAS rainfall anomalies and (b) hindcasts.
Citation: Journal of Hydrometeorology 25, 5; 10.1175/JHM-D-23-0064.1
In May, positive correlations between rainfall and SST intensify along the coastal region and over the southern Peruvian ecoregions, while negative correlations are observed over the region’s northeast (Fig. 13a). These positive correlations are more likely because, usually, when positive SST anomalies are observed in the Niño-3.4 Pacific region, trade wind intensity is reduced and warmer than normal SSTs spread toward the eastern Pacific. Thus, during May, positive temperature anomalies are observed across the western coast of South America, bringing more moisture flux, and supporting local precipitation. At the same time, negative rainfall anomalies in the region’s northeast might be due to the meridional displacement of the ITCZ (Sulca et al. 2018) or anomalies in the Hadley cell circulation (Poveda et al. 2006) which weaken convergence. Both of these mechanisms have been associated with positive SST anomalies in central Pacific.
Further, comparing the correlation maps of WTSA-LDAS precipitation and Niño-3.4 (Fig. 13a) to the correlation maps of the hindcast precipitation and Niño-3.4 (Fig. 13b), we see that hindcast agreement with WTSA-LDAS decays with prediction lead time. In March, hindcast and WTSA-LDAS exhibit similar patterns of correlation with Niño-3.4 across broad areas of the study region, with exceptions in the Pacific Plains and the Ecuadorian Andes, as shown in Fig. 13a. During April, rainfall hindcast correlation patterns match observed patterns well in the Ecuadorian ecoregions, the center of Ucayali-Irregular Forest, and the south of the Peruvian ecoregions, but there are large discrepancies in other areas. Similarly, in May we see significant differences in the correlation patterns, though there is some similarity between WTSA-LDAS and hindcasts in the Pacific Plains, Dry Hills, northern Ecuadorian Andes, Peruvian Andes, and Peruvian Amazonia. The predictability observed in May likely comes from the eastern SST anomalies, as they play an important role in rainfall impacts in these ecoregions (Recalde-Coronel et al. 2014).
Interestingly, there is a systematic improvement in hindcast skill in years with a distinct positive SST anomaly signal. In Table 1, we calculate the percentage of years in which there is a positive RPSS (i.e., hindcast better than climatology) for all years and if we only consider years categorized as El Niño or La Niña. Results are shown only for March, the month in which hindcasts best captured the spatial structure of ENSO correlations with rainfall (Fig. 12). Considering each of the eight ecoregions, forecast performance evaluated as the percent of years with positive RPSS is better in most ecoregions when we consider only El Niño or only La Niña years (Table 1, rows 2 and 3) than when considering the full study period (Table 1, row 1). The percentage of months with positive RPSS is enhanced in five of the eight ecoregions during a positive phase of the Niño-3.4 index, but RPSS declined in six of eight regions for the negative phase.
Percentage of months (March only) with positive RPSS, considering the full 2000–17 hindcast record, and percentages of positive RPSS during the positive and negative anomalies of the Niño-3.4 index and AAO index. Numbers in bold are cases where RPSS is higher for the given ENSO or AAO condition than it is for the full study period.
This result reflects the fact that S2S-HFS captures some of the El Niño precipitation impacts on the region, while La Niña teleconnection is not particularly strong in the region. Interestingly, S2S-HFS might properly simulate rainfall events linked with SST anomalies that have not followed the established rainfall impacts on WTSA in recent decades. For example, during El Niño 2005 (La Niña 2016), negative (positive) rainfall was observed on the Ecuadorian coast (not shown) when traditionally, ENSO is more likely to drive floods toward the west of the Andes during El Niño events and droughts over the Amazon during La Niña events (e.g., Poveda et al. 2011). The ability of the physically based S2S-HFS to capture nonstationary ENSO relationships that can degrade performance of statistical forecast systems will be tracked in future applications of the system.
2) AAO index
Figure 14a shows the monthly correlation between CHIRPS precipitation anomalies and an AAO scalar index. The spatial patterns indicate positive associations between the AAO index and rainfall anomalies in the Garua-Loma desert and over most parts of the northern territory during March, including the Ecuadorian Andes, Napo-Putumayo Forest, and Ucayali-Irregular Forest ecoregions. Similarly, positive associations are observed during April, except negative anomalies from the Peruvian Andes have shifted to the north. Meanwhile, lower correlations are observed during May. These observed associations are likely related to alterations on the SAMS components (e.g., SACZ, Bolivian high, and low-level winds), as previous studies have shown that AAO atmospheric forcing affects the intensity and position of SAMS (e.g., Silvestri and Vera 2003; Carvalho et al. 2005). Therefore, positive correlations during May and April are likely related to easterly moisture inflow from the Atlantic Ocean that are modulated during SAMS. In such a way, the low-level winds that enter northern Brazil bring warm moist air from the Atlantic toward northern South America (Grimm 2003). Meanwhile, lower correlations during May might be because the peak of the monsoon generally occurs during the austral summer season (DJF), and although the precipitation across the tropics is still observed during the austral autumn season (MAM), typically SAMS intensity is reduced by May (Vera et al. 2006).
Monthly correlation maps for AAO index against (a) WTSA-LDAS rainfall anomalies and (b) hindcast.
Citation: Journal of Hydrometeorology 25, 5; 10.1175/JHM-D-23-0064.1
Figure 14b shows the monthly correlation between hindcast precipitation anomalies and an AAO index. This figure illustrates clear similarities between precipitation anomalies from WTSA-LDAS and the hindcast model during March. Thus, positive correlations are observed in Garua-Loma desert, Ecuadorian Andes, Napo-Putumayo Forest, and Ucayali-Irregular Forest ecoregions, while negative anomalies are observed over the Peruvian Andes and Yungas Forest. However, as can be seen, similarities fade as the lag time increases. As longer lead times for this March initialized set of forecasts align with a seasonal weakening of the SAMS, reduced hindcast performance in April and May could be related both to long lead time and to weakening teleconnection at this time of year.
There is some evidence that hindcasts perform better for years with AAO−, as six out of eight ecoregions have enhanced RPSS for AAO− (Table 1). Carvalho et al. (2005) found that in intraseasonal time scales, the negative (positive) phase of the AAO is dominant when SST and convection anomalies resemble El Niño (La Niña) phases of ENSO. Therefore, it might be possible that better scores observed during the negative AAO phase on WTSA are a product of the interaction between AAO and ENSO. Results for AAO+ are weak, with five of eight regions showing worse RPSS performance for AAO+ than for the all-years average.
4. Summary and conclusions
The objective of this study was to implement and evaluate the skill of an S2S-HFS for WTSA. We look specifically at predictions of monthly precipitation, evapotranspiration, and soil moisture during the autumn rainy season. The system’s predictive skill has been examined by comparing hindcast outputs to WTSA-LDAS retrospective simulations and satellite data at lead times from 1 to 3 months within the 2002–17 period. We quantified hindcast skills using deterministic and probabilistic forecast verification metrics. In addition, we assessed the ability of the hindcast model to represent rainfall impacts associated with SST anomalies over the Pacific Ocean and to the mid- to high-latitude atmospheric Antarctic mode.
We find reasonably widespread skill in the S2S-HFS for all variables in the first month of the simulation. After that time the skill of the precipitation forecast drops to near zero, but the skill of evapotranspiration and, especially, soil moisture forecasts are maintained for a longer period (as evaluated against a reference retrospective simulation). The memory effect on these hydrological variables is particularly evident in the dry coastal lowlands—the Pacific Plains, Dry Hills, and Garua-Loma desert, as skill inherited from initial conditions can persist for longer in these regions for the March–May season. These patterns were difficult to confirm when forecasts were evaluated against satellite-derived observations instead of WTSA-LDAS retrospective reference simulation. This difficulty is in large part attributable to uncertainty in the satellite-derived estimates themselves, as well as to limitations in data coverage for soil moisture in the densely vegetated portions of the domain.
The fact that hindcast performance was somewhat stronger under ENSO+ and AAO− conditions is consistent with general understanding of drivers of climate variability in this region. It is encouraging that the modeling system was able to capture some of this influence, albeit only for short time horizons. The fact that this evaluation was performed for March–May made it a challenging case for teleconnection skill, since the South American monsoon systems weaken during this forecast period and ENSO predictability is particularly difficult in boreal spring. In future work we will test the potential of these teleconnections to explain and, perhaps, to contribute to S2S forecast skill through forecast conditioning techniques, in other seasons. Future work will also include expanding the S2S-HFS to include multiple meteorological forecast models, as this study was limited by the fact that GEOS-S2S-V1 was the only source of large-scale meteorological information.
Notwithstanding these uncertainties and areas for future research, it is encouraging to see some S2S forecast skills for hydrological variables in WTSA. The results of this study contribute to efforts to characterize sources of actionable hydrological information for the region and can be considered when developing forecast-informed management decisions for systems that are sensitive to climate variability on S2S time scales.
Acknowledgments.
This work was supported in part by NASA Applied Sciences Health and Air Quality project NNH13ZDA001N.
Data availability statement.
MERRA2 data used in this study are available from NASA at (https://gmao.gsfc.nasa.gov/reanalysis/MERRA-2/data_access/). CHIRPSv2 precipitation data are available from the Climate Hazards Center (https://www.chc.ucsb.edu/data/chirps). ESA-CCI data are available through the ESA Climate Office data archive (https://climate.esa.int/en/projects/soil-moisture/news-and-events/news/esa-cci-soil-moisture-product-new-version-release-v047/). Retrospective LDAS simulations applied in this study are available through the Johns Hopkins University Data Archive (https://archive.data.jhu.edu/dataset.xhtml?persistentId=doi:10.7281/T1/YQDI0F), as are the S2S-HFS hindcasts (https://doi.org/10.7281/T1/SQJ7G0).
REFERENCES
Aceituno, P., 1988: On the functioning of the Southern Oscillation in the South American sector. Part I: Surface climate. Mon. Wea. Rev., 116, 505–524, https://doi.org/10.1175/1520-0493(1988)116<0505:OTFOTS>2.0.CO;2.
Aceituno, P., 1989: On the functioning of the Southern Oscillation in the South American sector. Part II: Upper-air circulation. J. Climate, 2, 341–355, https://doi.org/10.1175/1520-0442(1989)002<0341:OTFOTS>2.0.CO;2.
Anderson, M. C., J. M. Norman, G. R. Diak, W. P. Kustas, and J. R. Mecikalski, 1997: A two-source time-integrated model for estimating surface fluxes using thermal infrared remote sensing. Remote Sens. Environ., 60, 195–216, https://doi.org/10.1016/S0034-4257(96)00215-5.
Anderson, M. C., J. M. Norman, J. R. Mecikalski, J. A. Otkin, and W. P. Kustas, 2007: A climatological study of evapotranspiration and moisture stress across the continental United States based on thermal remote sensing: 1. Model formulation. J. Geophys. Res., 112, D10117, https://doi.org/10.1029/2006JD007506.
Anderson, M. C., and Coauthors, 2011: Mapping daily evapotranspiration at field to continental scales using geostationary and polar orbiting satellite imagery. Hydrol. Earth Syst. Sci., 15, 223–239, https://doi.org/10.5194/hess-15-223-2011.
Anderson, M. C., and Coauthors, 2016: The evaporative stress index as an indicator of agricultural drought in Brazil: An assessment based on crop yield impacts. Remote Sens. Environ., 174, 82–99, https://doi.org/10.1016/j.rse.2015.11.034.
Arsenault, K. R., and Coauthors, 2018: The Land surface Data Toolkit (LDT v7.2)—A data fusion environment for land data assimilation systems. Geosci. Model Dev., 11, 3605–3621, https://doi.org/10.5194/gmd-11-3605-2018.
Becker, E., and H. van den Dool, 2016: Probabilistic seasonal forecasts in the North American multimodel ensemble: A baseline skill assessment. J. Climate, 29, 3015–3026, https://doi.org/10.1175/JCLI-D-14-00862.1.
Bjerknes, J., 1969: Atmospheric teleconnections from the equatorial Pacific. Mon. Wea. Rev., 97, 163–172, https://doi.org/10.1175/1520-0493(1969)097<0163:ATFTEP>2.3.CO;2.
Bombardi, R. J., L. Trenary, K. Pegion, B. Cash, T. Delsole, and J. L. Kinter III, 2018: Seasonal predictability of summer rainfall over South America. J. Climate, 31, 8181–8195, https://doi.org/10.1175/JCLI-D-18-0191.1.
Bombardi, R. J., V. Moron, and J. S. Goodnight, 2020: Detection, variability, and predictability of monsoon onset and withdrawal dates: A review. Int. J. Climatol., 40, 641–667, https://doi.org/10.1002/joc.6264.
Borovikov, A., R. Cullather, R. Kovach, J. Marshak, G. Vernieres, Y. Vikhliaev, B. Zhao, and Z. Li, 2017: GEOS-5 seasonal forecast system. Climate Dyn., 53, 7335–7361, https://doi.org/10.1007/s00382-017-3835-2.
Brier, G. W., 1950: Verification of forecasts expressed in terms of probability. Mon. Wea. Rev., 78, 1–3, https://doi.org/10.1175/1520-0493(1950)078<0001:VOFEIT>2.0.CO;2.
Carvalho, L. M. V., and I. F. A. Cavalcanti, 2016: The South American monsoon system (SAMS). The Monsoons and Climate Change: Observations and Modeling, L. M. V. de Carvalho and C. Jones, Eds., Springer, 121–148, https://doi.org/10.1007/978-3-319-21650-8_6.
Carvalho, L. M. V., C. Jones, and B. Liebmann, 2004: The South Atlantic convergence zone: Persistence, intensity, form, extreme precipitation, and relationships with intraseasonal to interannual activity and extreme rainfall. J. Climate, 17, 88–108, https://doi.org/10.1175/1520-0442(2004)017<0088:TSACZI>2.0.CO;2.
Carvalho, L. M. V., C. Jones, and T. Ambrizzi, 2005: Opposite phases of the Antarctic oscillation and relationships with intraseasonal to interannual activity in the tropics during the austral summer. J. Climate, 18, 702–718, https://doi.org/10.1175/JCLI-3284.1.
Carvalho, L. M. V., A. E. Silva, C. Jones, B. Liebmann, P. L. Silva-Dias, and H. R. Rocha, 2011: Moisture transport and intraseasonal variability in the South America monsoon system. Climate Dyn., 36, 1865–1880, https://doi.org/10.1007/s00382-010-0806-2.
Ceccherini, G., I. Ameztoy, C. P. R. Hernández, and C. C. Moreno, 2015: High-resolution precipitation datasets in South America and West Africa based on satellite-derived rainfall, enhanced vegetation index and digital elevation model. Remote Sens., 7, 6454–6488, https://doi.org/10.3390/rs70506454.
DeChant, C. M., and H. Moradkhani, 2014: Toward a reliable prediction of seasonal forecast uncertainty: Addressing model and initial condition uncertainty with ensemble data assimilation and sequential Bayesian combination. J. Hydrol., 519, 2967–2977, https://doi.org/10.1016/j.jhydrol.2014.05.045.
de Goncalves, L. G. G., W. J. Shuttleworth, S. C. Chou, Y. Xue, P. R. Houser, D. L. Toll, J. Marengo, and M. Rodell, 2006a: Impact of different initial soil moisture fields on Eta model weather forecasts for South America. J. Geophys. Res., 111, D17102, https://doi.org/10.1029/2005JD006309.
de Goncalves, L. G. G., W. J. Shuttleworth, E. J. Burke, P. Houser, D. L. Toll, M. Rodell, and K. Arsenault, 2006b: Toward a South America land data assimilation system: Aspects of land surface model spin-up using the simplified simple Biosphere. J. Geophys. Res., 111, D17110, https://doi.org/10.1029/2005JD006297.
Dorigo, W. A., and Coauthors, 2015: Evaluation of the ESA CCI soil moisture product using ground-based observations remote. Sens. Environ., 162, 380–395, https://doi.org/10.1016/j.rse.2014.07.023.
Dorigo, W. A., and Coauthors, 2017: ESA CCI soil moisture for improved Earth system understanding: State-of-the art and future directions. Remote Sens. Environ., 203, 185–215, https://doi.org/10.1016/j.rse.2017.07.001.
Draper, C. S., R. H. Reichle, and R. D. Koster, 2018: Assessment of MERRA-2 land surface energy flux estimates. J. Climate, 31, 671–691, https://doi.org/10.1175/JCLI-D-17-0121.1.
Epstein, E. S., 1969: A scoring system for probability forecasts of ranked categories. J. Appl. Meteor., 8, 985–987, https://doi.org/10.1175/1520-0450(1969)008<0985:ASSFPF>2.0.CO;2.
Francou, B., M. Vuille, V. Favier, and B. Cáceres, 2004: New evidence for an ENSO impact on low latitude glaciers: Antizana 15, Andes of Ecuador, 0°28′S. J. Geophys. Res., 109, D18106, https://doi.org/10.1029/2003JD004484.
Funk, C., and Coauthors, 2015: The climate hazards infrared precipitation with stations—A new environmental record for monitoring extremes. Sci. Data, 2, 150066, https://doi.org/10.1038/sdata.2015.66.
Garreaud, R. D., M. Vuille, R. Compagnucci, and J. Marengo, 2009: Present-day South American climate. Palaeogeogr. Palaeoclimatol. Palaeoecol., 281, 180–195, https://doi.org/10.1016/j.palaeo.2007.10.032.
Gelaro, R., and Coauthors, 2017: The Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2). J. Climate, 30, 5419–5454, https://doi.org/10.1175/JCLI-D-16-0758.1.
Getirana, A. C. V., A. Boone, D. Yamazaki, B. Decharme, F. Papa, and N. Mognard, 2012: The Hydrological Modeling and Analysis Platform (HyMAP): Evaluation in the Amazon basin. J. Hydrometeor., 13, 1641–1665, https://doi.org/10.1175/JHM-D-12-021.1.
Grimm, A. M., 2003: The El Niño impact on the summer monsoon in Brazil: Regional processes versus remote influences. J. Climate, 16, 263–280, https://doi.org/10.1175/1520-0442(2003)016<0263:TENIOT>2.0.CO;2.
Grimm, A. M., and T. Ambrizzi, 2009: Teleconnections into South America from the tropics and extratropics on interannual and intraseasonal timescales. Past Climate Variability in South America and Surrounding Regions: From the Last Glacial Maximum to the Holocene, F. Vimeux, F. Sylvestre, and M. Khodri, Eds., Springer, 159–191, https://doi.org/10.1007/978-90-481-2672-9_7.
Grimm, A. M., and R. G. Tedeschi, 2009: ENSO and extreme rainfall events in South America. J. Climate, 22, 1589–1609, https://doi.org/10.1175/2008JCLI2429.1.
Gruber, A., T. Scanlon, R. van der Schalie, W. Wagner, and W. Dorigo, 2019: Evolution of the ESA CCI soil moisture climate data records and their underlying merging methodology. Earth Syst. Sci. Data, 11, 717–739, https://doi.org/10.5194/essd-11-717-2019.
Gutmann, E., T. Pruitt, M. P. Clark, L. Brekke, J. R. Arnold, D. A. Raff, and R. M. Rasmussen, 2014: An intercomparison of statistical downscaling methods used for water resource assessments in the United States. Water Resour. Res., 50, 7167–7186, https://doi.org/10.1002/2014WR015559.
Gutmann, E. D., J. J. Hamman, M. P. Clark, T. Eidhammer, A. W. Wood, and J. R. Arnold, 2022: En-GARD: A statistical downscaling framework to produce and test large ensembles of climate projections. J. Hydrometeor., 23, 1545–1561, https://doi.org/10.1175/JHM-D-21-0142.1.
Hazra, A., and Coauthors, 2023: NASA’s NMME-based S2S hydrologic forecast system for food insecurity early warning in southern Africa. J. Hydrol., 617, 129005, https://doi.org/10.1016/j.jhydrol.2022.129005.
Huang, B., and Coauthors, 2017: Extended Reconstructed Sea Surface Temperature, version 5 (ERSSTv5): Upgrades, validations, and intercomparisons. J. Climate, 30, 8179–8205, https://doi.org/10.1175/JCLI-D-16-0836.1.
Kirtman, B. P., and Coauthors, 2014: The North American Multi-Model Ensemble (NMME): Phase-1 seasonal to interannual prediction; phase-2 toward developing intra-seasonal prediction. Bull. Amer. Meteor. Soc., 95, 585–601, https://doi.org/10.1175/BAMS-D-12-00050.1.
Koster, R. D., and M. J. Suarez, 2003: Impact of land surface initialization on seasonal precipitation and temperature prediction. J. Hydrometeor., 4, 408–423, https://doi.org/10.1175/1525-7541(2003)4<408:IOLSIO>2.0.CO;2.
Krakauer, N. Y., 2019: Temperature trends and prediction skill in NMME seasonal forecasts. Climate Dyn., 53, 7201–7213, https://doi.org/10.1007/s00382-017-3657-2.
Kumar, S. V., and Coauthors, 2006: Land information system: An interoperable framework for high resolution land surface modeling. Environ. Modell. Software, 21, 1402–1415, https://doi.org/10.1016/j.envsoft.2005.07.004.
Liebmann, B., G. N. Kiladis, J. A. Marengo, T. Ambrizzi, and J. D. Glick, 1999: Submonthly convective variability over South America and the South Atlantic convergence zone. J. Climate, 12, 1877–1891, https://doi.org/10.1175/1520-0442(1999)012<1877:SCVOSA>2.0.CO;2.
López-Bermeo, C., R. D. Montoya, F. J. Caro-Lopera, and J. A. Díaz-García, 2022: Validation of the accuracy of the CHIRPS precipitation dataset at representing climate variability in a tropical mountainous region of South America. Phys. Chem. Earth, 127, 103184, https://doi.org/10.1016/j.pce.2022.103184.
Maertens, M., G. J. M. De Lannoy, S. Apers, S. V. Kumar, and S. P. P. Mahanama, 2021: Land surface modeling over the Dry Chaco: The impact of model structures, and soil, vegetation and land cover parameters. Hydrol. Earth Syst. Sci., 25, 4099–4125, https://doi.org/10.5194/hess-25-4099-2021.
Miguez-Macho, G., and Y. Fan, 2012: The role of groundwater in the Amazon water cycle: 1. Influence on seasonal streamflow, flooding and wetlands. J. Geophys. Res., 117, D15113, https://doi.org/10.1029/2012JD017539.
Mo, K. C., and B. Lyon, 2015: Global meteorological drought prediction using the North American multi-model ensemble. J. Hydrometeor., 16, 1409–1424, https://doi.org/10.1175/JHM-D-14-0192.1.
Molod, A., L. Takacs, M. Suarez, J. Bacmeister, I. S. Song, and A. Eichmann, 2012: The GEOS-5 atmospheric general circulation model: Mean climate and development from MERRA to Fortuna. NASA Tech. Rep. NASA/TM-2012-104606, Vol. 28, 124 pp., https://ntrs.nasa.gov/api/citations/20120011790/downloads/20120011790.pdf.
Mueller, B., and Coauthors, 2011: Evaluation of global observations-based evapotranspiration datasets and IPCC AR4 simulations. Geophys. Res. Lett., 38, L06402, https://doi.org/10.1029/2010GL046230.
Murphy, A. H., 1970: The ranked probability score and the probability score: A comparison. Mon. Wea. Rev., 98, 917–924, https://doi.org/10.1175/1520-0493(1970)098<0917:TRPSAT>2.3.CO;2.
Niu, G.-Y., and Coauthors, 2011: The community Noah land surface model with multiparameterization options (Noah-MP): 1. Model description and evaluation with local-scale measurements. J. Geophys. Res., 116, D12109, https://doi.org/10.1029/2010JD015139.
Nobre, C. A., G. O. Obregón, J. A. Marengo, R. Fu, and G. Poveda, 2009: Characteristics of Amazonian climate: Main features. Amazonia and Global Change, Geophys. Monogr., Vol. 186, Amer. Geophys. Union, 149–162, https://doi.org/10.1029/2009GM000903.
Nogués-Paegle, J., and K. C. Mo, 1997: Alternating wet and dry conditions over South America during summer. Mon. Wea. Rev., 125, 279–291, https://doi.org/10.1175/1520-0493(1997)125<0279:AWADCO>2.0.CO;2.
Paca, V. H. M., G. E. Espinoza-Davalos, T. M. Hessels, D. M. Moreira, G. F. Comair, and W. G. M. Bastiaanssen, 2019: The spatial variability of actual evapotranspiration across the Amazon River Basin based on remote sensing products validated with flux towers. Ecol. Processes, 8, 6, https://doi.org/10.1186/s13717-019-0158-8.
Paredes-Trejo, F. J., H. A. Barbosa, and T. V. L. Kumar, 2017: Validating CHIRPS-based satellite precipitation estimates in Northeast Brazil. J. Arid Environ., 139, 26–40, https://doi.org/10.1016/j.jaridenv.2016.12.009.
Poveda, G., P. R. Waylen, and R. S. Pulwarty, 2006: Annual and inter-annual variability of the present climate in northern South America and southern Mesoamerica. Palaeogeogr. Palaeoclimatol. Palaeoecol., 234, 3–27, https://doi.org/10.1016/j.palaeo.2005.10.031.
Poveda, G., D. M. Álvarez, and Ó. A. Rueda, 2011: Hydro-climatic variability over the Andes of Colombia associated with ENSO: A review of climatic processes and their impact on one of the Earth’s most important biodiversity hotspots. Climate Dyn., 36, 2233–2249, https://doi.org/10.1007/s00382-010-0931-y.
Rao, V. B., and K. Hada, 1990: Characteristics of rainfall over Brazil: Annual variations and connections with the Southern Oscillation. Theor. Appl. Climatol., 42, 81–91, https://doi.org/10.1007/BF00868215.
Recalde, C., B. F. Zaitchik, and W. K. Pan, 2021: Data associated with: Retrospective land surface model simulations of western tropical South America. Johns Hopkins University Data Archive, accessed 26 September 2021, https://doi.org/10.7281/T1/YQDI0F.
Recalde-Coronel, C., and B. F. Zaitchick, 2024: Data associated with the publication: Recalde-Coronel G. C. et al. (2024) Contributions of initial conditions and meteorological forecast to subseasonal-to-seasonal hydrological forecast skill in western tropical South America. Journal of Hydrometeorology. Johns Hopkins Research Data Repository, accessed 20 April 2024, https://doi.org/10.7281/T1/SQJ7G0.
Recalde-Coronel, C., A. G. Barnston, and A. G. Muñoz, 2014: Predictability of December–April rainfall in coastal and Andean Ecuador. J. Appl. Meteor. Climatol., 53, 1471–1493, https://doi.org/10.1175/JAMC-D-13-0133.1.
Recalde-Coronel, C., B. Zaitchik, W. Pan, and A. Getirana, 2022: Influence of vegetation on simulation of the water balance and hydrological response to El Niño–Southern Oscillation in western tropical South America. J. Hydrol., 23, 1737–1757, https://doi.org/10.1175/JHM-D-21-0081.1.
Rivera, J. A., G. Marianetti, and S. Hinrichs, 2018: Validation of CHIRPS precipitation dataset along the central Andes of Argentina. Atmos. Res., 213, 437–449, https://doi.org/10.1016/j.atmosres.2018.06.023.
Rodell, M., and Coauthors, 2004: The Global Land Data Assimilation System. Bull. Amer. Meteor. Soc., 85, 381–394, https://doi.org/10.1175/BAMS-85-3-381.
Rodriguez, D. A., and I. F. A. Cavalcanti, 2006: Simulations of the hydrologic cycle over southern South America using the CPTEC/COLA AGCM. J. Hydrometeor., 7, 916–936, https://doi.org/10.1175/JHM534.1.
Rogers, J. C., and H. van Loon, 1982: Spatial variability of sea level pressure and 500 mb height anomalies over the Southern Hemisphere. Mon. Wea. Rev., 110, 1375–1392, https://doi.org/10.1175/1520-0493(1982)110<1375:SVOSLP>2.0.CO;2.
Salazar-Martínez, D., and Coauthors, 2022: Evaluation of remote sensing-based evapotranspiration products at low-latitude eddy covariance sites. J. Hydrol., 610, 127786, https://doi.org/10.1016/j.jhydrol.2022.127786.
Shukla, S., J. Sheffield, E. F. Wood, and D. P. Lettenmaier, 2013: On the sources of global land surface hydrologic predictability. Hydrol. Earth Syst. Sci., 17, 2781–2796, https://doi.org/10.5194/hess-17-2781-2013.
Silvestri, G. E., and C. S. Vera, 2003: Antarctic Oscillation signal on precipitation anomalies over southeastern South America. Geophys. Res. Lett., 30, 2115, https://doi.org/10.1029/2003GL018277.
Slater, L. J., G. Villarini, and A. A. Bradley, 2016: Evaluation of the skill of North American Multi‐Model Ensemble (NMME) global climate models in predicting average and extreme precipitation and temperature over the continental USA. Climate Dyn., 53, 7381–7396, https://doi.org/10.1007/s00382-016-3286-1.
Sulca, J., K. Takahashi, J.-C. Espinoza, M. Vuille, and W. Lavado-Casimiro, 2018: Impacts of different ENSO flavors and tropical Pacific convection variability (ITCZ, SPCZ) on austral summer rainfall in South America, with a focus on Peru. Int. J. Climatol., 38, 420–435, https://doi.org/10.1002/joc.5185.
Tapley, T. D., and P. R. Waylen, 1990: Spatial variability of annual precipitation and ENSO events in western Peru. Hydrol. Sci. J., 35, 429–446, https://doi.org/10.1080/02626669009492444.
Thober, S., R. Kumar, J. Sheffield, J. Mai, D. Schaefer, and L. Samaniego, 2015: Seasonal soil moisture drought prediction over Europe using the North American Multi-Model Ensemble (NMME). J. Hydrometeor., 16, 2329–2344, https://doi.org/10.1175/JHM-D-15-0053.1.
Thompson, D. W. J., and J. M. Wallace, 2000: Annular modes in the extratropical circulation. Part I: Month-to-month variability. J. Climate, 13, 1000–1016, https://doi.org/10.1175/1520-0442(2000)013<1000:AMITEC>2.0.CO;2.
USEPA, 2011: Level III ecoregions of Central and South America. U.S. EPA Office of Research and Development (ORD) and National Health and Environmental Effects Research Laboratory (NHEERL), accessed 15 December 2023, http://ecologicalregions.info/data/sa/sa_eco_l3.htm.
Vera, C., and Coauthors, 2006: Towards a unified view of the American monsoon system. J. Climate, 19, 4977–5000, https://doi.org/10.1175/JCLI3896.1.
Vitart, F., and A. W. Robertson, 2018: The Sub-Seasonal to Seasonal Prediction project (S2S) and the prediction of extreme events. npj Climate Atmos. Sci., 1, 3, https://doi.org/10.1038/s41612-018-0013-0.
Vuille, M., R. S. Bradley, and F. Keimig, 2000a: Climate variability in the Andes of Ecuador and its relation to tropical Pacific and Atlantic sea surface temperature anomalies. J. Climate, 13, 2520–2535, https://doi.org/10.1175/1520-0442(2000)013<2520:CVITAO>2.0.CO;2.
Vuille, M., R. S. Bradley, and F. Keimig, 2000b: Interannual climate variability in the central Andes and its relation to tropical Pacific and Atlantic forcing. J. Geophys. Res., 105, 12 447–12 460, https://doi.org/10.1029/2000JD900134.
Wanders, N., and E. F. Wood, 2016: Improved sub-seasonal meteorological forecast skill using weighted multi-model ensemble simulations. Environ. Res. Lett., 11, 094007, https://doi.org/10.1088/1748-9326/11/9/094007.
Wilks, D. S., 2011: Statistical Methods in the Atmospheric Sciences. Academic Press, 676 pp.
Wood, A. W., and J. C. Schaake, 2008: Correcting errors in streamflow forecast ensemble mean and spread. J. Hydrometeor., 9, 132–148, https://doi.org/10.1175/2007JHM862.1.
Yang, Z.-L., and Coauthors, 2011: The community Noah land surface model with multiparameterization options (Noah-MP): 2. Evaluation over global river basins. J. Geophys. Res., 116, D12110, https://doi.org/10.1029/2010JD015140.
Yossef, N. C., H. Winsemius, A. Weerts, R. van Beek, and M. F. P. Bierkens, 2013: Skill of a global seasonal streamflow forecasting system, relative roles of initial conditions and meteorological forcing. Water Resour. Res., 49, 4687–4699, https://doi.org/10.1002/wrcr.20350.
Yuan, X., E. F. Wood, L. Luo, and M. Pan, 2011: A first look at Climate Forecast System version 2 (CFSv2) for hydrological seasonal prediction. Geophys. Res. Lett., 38, L13402, https://doi.org/10.1029/2011GL047792.
Yuan, X., E. F. Wood, and Z. Ma, 2015a: A review on climate-model based seasonal hydrologic forecasting: Physical understanding and system development. Wiley Interdiscip. Rev.: Water, 2, 523–536, https://doi.org/10.1002/wat2.1088.
Yuan, X., J. K. Roundy, E. F. Wood, and J. Sheffield, 2015b: Seasonal forecasting of global hydrologic extremes: System development and evaluation over GEWEX basins. Bull. Amer. Meteor. Soc., 96, 1895–1912, https://doi.org/10.1175/BAMS-D-14-00003.1.
Zamora, R. A., B. F. Zaitchik, M. Rodell, A. Getirana, S. Kumar, K. Arsenault, and E. Gutmann, 2021: Contribution of meteorological downscaling to skill and precision of seasonal drought forecasts. J. Hydrometeor., 22, 2009–2031, https://doi.org/10.1175/JHM-D-20-0259.1.
Zhou, Y., B. F. Zaitchik, S. V. Kumar, K. R. Arsenault, M. A. Matin, F. M. Qamer, R. A. Zamora, and K. Shakya, 2021: Developing a hydrological monitoring and sub-seasonal to seasonal forecasting system for South and Southeast Asian river basins. Hydrol. Earth Syst. Sci., 25, 41–61, https://doi.org/10.5194/hess-25-41-2021.
