Upload
ilri
View
76
Download
1
Tags:
Embed Size (px)
Citation preview
Institute of Surveying, Remote Sensing and Land Information 1
Satellite-based drought monitoring
in Kenya in an operational setting
Clement AtzbergerUniversity of Natural Resources and Life Sciences, Vienna (BOKU),
Institute of Surveying, Remote Sensing and Land Information (IVFL)
Luigi LuminariNational Drought Management Authority (NDMA), Kenya
IBLI workshop, 9-11 June 2015, Nairobi
Institute of Surveying, Remote Sensing and Land Information
Traditional reaction to drought
The traditional reaction to drought and its effect has been to adopt a crisis management approach
This reactive approach is not good policy and should be replaced by a risk management approach which is anticipatory and preventive
Institute of Surveying, Remote Sensing and Land Information
WHY A CONTINGENCY FUND?
One of the main shortcomings in drought risk management remains the weak linkage between early warning and early response;
Inability of the Government and other relevant stakeholders to facilitate timely response is caused, to a large extent, by inadequate set-aside funds (contingency funds)
The availability of sufficient “set-aside contingency funds” can ensure timely measures to mitigate the impact of drought, protecting livelihoods and saving lives.
Institute of Surveying, Remote Sensing and Land Information
The criteria for the release of contingency funds must be systematic, evidence-based and transparent
Drought response activities are specific initiatives triggered by the stages of the drought cycle as signalled by the EWS
Multi-sectoral Contingency Plans are prepared and activated for rapid reaction to the early warning. They cover necessary interventions at each phase of drought
DISBURSEMENT OF DCF
Institute of Surveying, Remote Sensing and Land Information
EWS & DROUGHT PHASE CLASSIFICATION
The trigger points between warning stagesdetermined through four categories of drought indicators
ENVIRONMENTAL INDICATORS (impact on biophysical)
PRODUCTION INDICATORS (impact on livestock and crop production)
ACCESS INDICATORS(impact on market and access to food and water)
UTILISATION INDICATORS(impact on nutrition and coping strategy)
Institute of Surveying, Remote Sensing and Land Information
EN DI WEEE EI ???
Biomass measurements using reflected light in the visible (red) and near infrared (nIR) dnIR
dnIRNDVIRe
Re
Institute of Surveying, Remote Sensing and Land Information
Problem illustration: Clouds and aerosols are omni-present
Institute of Surveying, Remote Sensing and Land Information
NDVI time series (MODIS) for Kenya
Institute of Surveying, Remote Sensing and Land Information
Problem description: Anomaly indicators aggravatedata quality issues
Grassland z-score time profile
Institute of Surveying, Remote Sensing and Land Information
Problem description: Avoiding false alarms
Data quality matters:• Disaster contingency Funds (DCF)• Index-based insurance (IBLI)
Institute of Surveying, Remote Sensing and Land Information
VCI: Vegetation Condition Index
Institute of Surveying, Remote Sensing and Land Information 12
Sedano et al. (2014)
Smoothing applies in a post hoc sense, where there is a need to optimally interpolate past events in a time series.
Smoothing estimates a state based on data from both previous and later times.
Filtering is relevant in an online learning sense, in which current conditions are to be estimated by the currently available data.
Filtering involves calculating the estimate of a certain state based on a partial sequence of inputs.
Definitions
time
ND
VI
Institute of Surveying, Remote Sensing and Land Information
Existing filters … used in RS
Institute of Surveying, Remote Sensing and Land Information
Principle of Whittaker smoother (Eilers 2003)
Only one smoothing parameter
Interpolates automatically
No boundary effects
Inputs (MOD13 from Aqua & Terra):
NDVI
composite day of year
quality & cloud flags
Trade-off between fidelity to observations & smoothness of output
Institute of Surveying, Remote Sensing and Land Information
• Moving window of 175 days: all available MODIS observations are used
• Weighted filtering and interpolation with Whittaker smoother
• Constrained filtering: using „shape“ from statistics
• Filtered NDVI of last 5 weeks are saved (Mondays): 0 1 2 3 4• Smoothed NDVI of center week is saved
Constrained filtering using Whittaker smoother
Institute of Surveying, Remote Sensing and Land Information
Output: 1
Filtering: Consolidation periods (zero to fourteen weeks)
last 5 weeks are saved (Mondays)
Output: 0Output: 2Output: 3Output: 4Offline
Smoothing
Duration (in weeks) of consolidation period
Institute of Surveying, Remote Sensing and Land Information
“Uncertainty” modeling
Institute of Surveying, Remote Sensing and Land Information
Du
rati
on
(in
wee
ks)
of
con
soli
dat
ion
per
iod
Week of Year
4 weeks
2 weeks
0 weeks
Week 27
Uncertainty modelling used smoothed signal (“offline”) as reference &observation conditions as predictors
Institute of Surveying, Remote Sensing and Land Information
Filtering: Calculation of anomalies (VCI & ZVI)
100MinMax
MinVIVCI
SD
MeanVIZVI
(Kogan et al. 2003)
Institute of Surveying, Remote Sensing and Land Information
Downweighting of observations according to “uncertainty”
0
…
123
…
0
1
2
3
„Monday“Anomaly Uncertainties
monthlyaggregrated
Anomaly
4
Institute of Surveying, Remote Sensing and Land Information
wet
no drought
moderate drought
severe drought
extreme drought
Temporal aggregation to monthly VCI using uncertainties for weighting
Spatial and temporal aggregation of anomalies (e.g. VCI) incl. uncertainties
Vegetation condition index (VCI)
Spatial aggregation to zones e.g. counties & national livelihood zones
Institute of Surveying, Remote Sensing and Land Information
Comparison of anomalies with FEWS NET data
pentadal eMODIS NDVI provided by Famine Early Warning Systems Network (FEWS NET) of the USGS
VCI calculated for 2003-2014 from consolidated data temporally aggregated for 3 month interval spatially aggregated to arid and semi-arid land (ASAL)
counties of Kenya
General good agreement
RMSE = 6%R² = 0.89n = 3312
Intra-annualvariability
Inter-annual variabilitySpatial variability
Institute of Surveying, Remote Sensing and Land Information
Achievements
Efficient noise removal and gap-filling
Institute of Surveying, Remote Sensing and Land Information
Achievements
Efficient noise removal and gap-filling
Near real-time data processing & weekly updating cycle
Institute of Surveying, Remote Sensing and Land Information
Achievements
Efficient noise removal and gap-filling
Near real-time data processing & weekly updating cycle
Various consolidation phases
Strength of the consolidation
high …………………………..low
01234
Institute of Surveying, Remote Sensing and Land Information
Achievements
Efficient noise removal and gap-filling
Near real-time data processing & weekly updating cycle
Various consolidation phases
Consistent archive for the various consolidation phases
Current
Strength of the consolidation
high …………………………..low
01234
Archive (LTA, σ, min, max)
01234
Institute of Surveying, Remote Sensing and Land Information
Achievements
Efficient noise removal and gap-filling
Near real-time data processing & weekly updating cycle
Various consolidation phases
Consistent archive for the various consolidation phases
Modeling of uncertainties at pixel level & for all products
Institute of Surveying, Remote Sensing and Land Information
Achievements
Efficient noise removal and gap-filling
Near real-time data processing & weekly updating cycle
Various consolidation phases
Consistent archive for the various consolidation phases
Modeling of uncertainties at pixel level & for all products
Integration of uncertainty informationduring temporal (& spatial) aggregration
…
123
…
0
1
2
3
„Monday“Anomaly Uncertainties
monthlyaggregrated
Anomaly
4
Institute of Surveying, Remote Sensing and Land Information 29
Conclusions & Outlook
Data quality is of utmost importance…… errors propagate
Perfect filtering (in near-real-time) is unrealistic…. but uncertainty can be modeled
Filtering is necessary…… any filtering is better than none
User perception matters …. different products confuse users
Unified NDVI products for Kenya/HoA would be an asset for all parties
Institute of Surveying, Remote Sensing and Land Information 30
THANKS!
University of Natural Resources and Life Sciences, Vienna, Austria (BOKU)
Institute of Surveying, Remote Sensing and Land Information (IVFL)
Clement ATZBERGER
[email protected]://ivfl-info.boku.ac.at/
National Drought Management Authority (NDMA), Nairobi, Kenya
Luigi LUMINARI
[email protected]://www.ndma.go.ke/
Automated MODIS data download & data preparation (projection & mosaicking)
Offline smoothingof entire time series
Constrained NRT filteringusing „shape“ to constrain
Statistics of NRT filtered data &
quality indicators
NRT calculation of anomalies and associated
uncertainties
NRT calculation of temporally and spatially aggregated anomalies
Uncertainty
modelling