Global Drought Monitoring and Forecasting based on Satellite Data and Land Surface Modeling Justin Sheffield, Eric F. Wood, Princeton University; David Lobell, Stanford University Introduction Monitoring drought globally is challenging because of the lack of dense in-situ hydrologic data in many regions. This is especially problematic for developing regions such as Africa where water information is arguably most needed, but virtually non-existent on the ground. With the emergence of remote sensing estimates of all components of the water cycle there is now the potential to monitor the full terrestrial water cycle from space to give global coverage and provide the basis for drought monitoring. However, many challenges remain in using these data, especially due to biases in individual satellite retrieved components, their incomplete sampling in time and space, and their failure to provide budget closure in concert.
Global Water Cycle and Drought Monitoring
A potential way forward is to use modeling to provide a framework to merge these disparate sources of information to give physically consistent and spatially and temporally continuous estimates of the water cycle and drought. Here we present results from our experimental global water cycle monitor and its African drought monitor counterpart. The system relies heavily on satellite data to drive the Variable Infiltration Capacity (VIC) land surface model to provide near real-time estimates. Ongoing work is adding a drought forecast component based on a successful implementation over the U.S. and agricultural productivity estimates based on output from crop yield models.
The Monitoring System
Historic Drought 1950-2000
The Princeton U.S. drought monitor has been extended to an experimental global water cycle and drought monitoring system, with a focus on Africa. The VIC model is run in near real time forced by a mixture of observations and modeled/remotely sensed meteorology to produce updates of water cycle variables (soil moisture, runoff, evapotranspiration, snow). Drought is monitored in terms of low soil moisture percentiles.
An index of drought is developed based on soil moisture percentiles relative to the local and seasonal climatology. This allows us to compare conditions historically and spatially. A drought is defined as a run of soil moisture values below a threshold (20th percentile) and a number of drought characteristics can be defined, including the duration, severity and intensity as well as the spatial area in drought (contiguous or not).
http://hydrology.princeton.edu/monitor
Sahel Drought
The system comprises three parts: (1) a retrospective reconstruction of the global terrestrial water cycle forced by merged reanalysis/ observational meteorological forcing. This forms the climatology against which current and future conditions are compared; (2) a real-time monitoring system driven by remote sensing precipitation and atmospheric analysis data that tracks drought conditions in near realtime; (3) a seasonal forecast system based on downscaled climate forecasts to provide seasonal hydrologic, drought and agricultural data.
1950-2000
One Year of Real-Time Monitoring: Aug2008 – July2009
Ongoing drought in the east, especially in Ethiopia/Somalia
Increased water availability in the West but flooding in Mozambique
Nov 1991 Forecast
Experimental Seasonal Forecasting
Worsening drought conditions in the east
Feb 1992 Forecast
Severe drought in Kenya, Ethiopia, Somalia
May 1992 Forecast
Recently a seasonal forecast component has been implemented based on our existing U.S. system (http:hydrology.princeton.edu/forecast). The system uses seasonal global climate forecasts from the NCEP Climate Forecast System (CFS) and merges them with observed climatology in a Bayesian framework to produce ensemble atmospheric forcings that better capture the uncertainties. At the same time, the system bias corrects and downscales the monthly CFS data to scales more appropriate for hydrologic modeling. We show some initial African seasonal (up to 6-month lead) hydrologic forecast results for the 1991/92 drought. The forecasts are compared to those derived using traditional ESP (random sampling of historic record) and a multi-model prediction based on merging of CFS forecasts with those from the DEMETER multi-model database
Nov DEMETER forecast not available
The southern African drought of 1991/92 was ENSO forced. Both CFS and ESP show no skill in predicting the onset of drought
Challenges of Global Drought Monitoring In Africa, where real time (and historic) observations of key hydrologic variables such as precipitation are generally lacking, we resort to alternative sources of data. These include remote-sensing based precipitation that can be used as input to the hydrologic modeling. Remotely sensed evapotranspiration, soil moisture can be used directly or through assimilation into the model. Below shows comparisons of observation-based precipitation versus three satellite MW-IR merged products that are available in near real-time.
Once the drought has started, only the multimodel forecast shows any skill in predicting the continuation of drought.
During the following dry season, all forecasts are skillful only because of the persistence of observed initial conditions.
Development of Agricultural Productivity Component Initial development of an experimental agricultural monitoring component has focused on using existing global agricultural datasets and statistical yield models. Regressions with climate data (Schlenker and Lobell, 2010) are used to predict changes in yield, which are anchored to yield estimates for the year 2000 (Monfreda et al., 2008) and adjusted based on cropping area and harvested area. Future plans include prediction on crop yields based on these regressions, process-based models and CFS climate forecasts.
Annual
P
Annual
Rain
Days
P(R|R)
Figures above and right show comparisons of the VIC retrospective (for 2002) and real-time (for 2008) simulations with soil moisture retrievals from AMSRE brightness temperatures globally using the Princeton Land Surface Microwave Emission Model (LSMEM) and evapotranspiration derived using a Penman-Monteith formulation based on remotesensing radiation (ISCCP, SRB) and vegetation characteristics (AVHRR)
P(D|D)
Acknowledgments: this work was funded by UNESCO grant 4500041171 and NOAA grant NA10OAR4310130 Contact: Justin Sheffield (
[email protected])
Crop yield monitoring for 2008 for 5 major crops