Challenges for the operational assimilation of satellite image data in agrometeorological models

Transcription

1 Challenges for the operational assimilation of satellite image data in agrometeorological models Mark Danson Centre for Environmental Systems Research, University of Salford, UK 0

2 Objectives The main objective of this presentation is to examine the challenges for the operational assimilation of satellite image data in agrometeorological models The starting point is an assumption that remotely sensed data have been shown (by our invited experts) to be useful in such models and that it is meaningful to examine the factors that affect the availability of appropriate data and observations What sort of observations? Previous published research has shown that the key variables required for assimilation in agrometeorological models are either satellite radiance (or reflectance) or a range of inferred variables Key inferred variables include leaf area index, FAPAR, surface temperature and various vegetation indices..are there others?

3 What sort of resolution? Many surveys of user requirements for agricultural applications of remote sensing have been undertaken over the last 40 years and there is some consensus that, with the exception of precision agriculture applications, that the following characteristics are desirable in a European agricultural context: Spatial resolution: 10-30m Temporal resolution: weekly or better Spectral resolution: Landsat-like wavebands including the SWIR Angular resolution: observations at different view angles may be desirable in some circumstances Lets discuss this briefly..

4 Observations with a cloud-free atmosphere! If we start by assuming a cloud free atmosphere, then three things determine whether we can observe a given location at a given time: 1. Is the satellite in a position to make the observation? This is determined by the satellite orbit and the pointing capability of the platform; satellite orbits for environmental remote sensing are generally very similar; some platforms are highly agile while some have a fixed view angle. This is the revisit time. 2. Is the instrument activated? This is determined by the operational mode of the instrument; some instruments record data continuously (more or less), whilst others are only activated following user requests 3. Is the instrument functioning? Instruments have a finite lifetime after which they no longer function; some instruments have specific problems like Landsat ETM scan line corrector error; others like Landsat 5 TM can miraculously come back to life!

5 Landsat ETM+ Scan Line Corrector Error No missing data

6 Observations with a cloud-free atmosphere! What determines temporal resolution? Length of line of latitude 45 degrees (which goes through Bordeaux) is about 28,000km Satellites orbit every 100 minutes or so, or 14 orbits/day We therefore need a swath width of about 2000km for daily coverage at 45 degrees north (for example Novi Sad, Serbia!) (or 2800km at the equator) This is why Modis Terra/Aqua achieves near-daily global coverage (swath width 2300km)

7 Observations with a cloud-free atmosphere! Revisit time (how frequently could a given point be imaged Repeat cycle (how frequently does ground track repeat) With no pointing capability: Revisit time for Modis Terra is 1 day, repeat cycle is 16 days With pointing capability: Revisit time for Ikonos is 1-5 days, repeat cycle is 114 days We know there is a trade-off between spatial and temporal resolution but there is another approach

8 The Disaster Monitoring Constellation (DMC) First Generation AlSAT-1 (Algeria), launched November BilSAT (Turkey), completed mission in August 2006 due to failed battery cells [1] NigeriaSAT-1 (Nigeria), launched September 2003 UK-DMC (United Kingdom), launched September Second Generation Beijing-1 (China), launched October UK-DMC 2 (United Kingdom[2]), launched July Deimos-1 (Spanish commercial[3]), launched July NigeriaSAT-2, planned to launch in 2010 Constellation revisit time : Four satellites in same orbit at 0, 90, 180 and 270 degrees provide potentially global coverage at 32m on a daily basis

9 DMC s SLIM-6 sensor gives 2x300km swath With 4 satellites we get 4 x 600km swaths on each orbit imaging 2400km swath in total this is why we can image any selected point on a daily basis

10 DMC European mosaic 2007

11 Observations with a cloudy atmosphere Cloud cover prevents surface observations so that the probability of recording useful data is the product of observation probability, as previously discussed, and cloud free imaging probability One question here is whether we need observations for specific points, or observations over given areas since the probability of seeing a given point will always be higher than that of observing a prescribed area Which scenario is most likely for data assimilation problems? Let s discuss.

12 Measuring cloud free frequencies Modis Terra NOAA AVHRR

13 Cloud free frequency in the UK Landsat ETM+ 16 day revisit/repeat cycle means 22 opportunities for observation per year With a cloud free frequency of 20% we might expect only 4 5 cloud free images per year With daily revisit and 20% cloud free we would expect 73 cloud free image per year

14 Another approach multi-sensor, multi-platform, multi-system data assimilation To increase the frequency of observations of a given location it is possible to incorporate data from several sensors. Many of the newer satellite sensors have similar broad visible and near infrared wavebands (very few have a SWIR waveband) If the spectral response functions of two different satellite sensors are different it is possible to derive cross-calibration equations for vegetation indices computed from different sensors.this may not be necessary when variables are determined through a physical-based model inversion approach Accurate radiometric and atmospheric correction are critical for multidate, multi-sensor applications.. Let s discuss.

15 New sensors for agrometeorological applications Based on a snapshot of the CEOS database there currently 40 satellites currently flying that have land-based observations (including ice) as their primary target. Discounting very high resolution cartographic and SAR instruments, and the established series like Landsat, SPOT and IRS, there is a small number which may be suitable for agrometeorological observations ALOS AVNIR-2 (Advanced Visible and Near Infrared Radiometer) is a visible and near infrared bands with 10 meters spatial resolution in four spectral bands. Swath 70 km at nadir and the pointing angle across track in a +/ 44º range to allow a target revisiting time of up to one day. Japanese Aerospace Exploration Agency JAXA CBERS-2B CCD camera five bands, 113km swath, 20m resolution and +/-32 degree pointing capability Brazil/China THEOS (THailand Earth Observation Satellite) the first Earth observation satellite of Thailand, was successfully launched by Dnepr launcher from Yasny, Russian Federation, on Wednesday, October 1, Four bands, 90km swath, 15m spatial resolution up to 4000km imaged strip length SumbandilaSat, RapidEye, Formosat-2, IMS-1, Kompsat-2, Pleiades 1(?)

16 DMC times series for LAI monitoring in UK

17 Results summary Root mean square error = 0.47 LAI Range of errors LAI

18 Calibration/validation using ground measurements There are still large uncertainties in the precision and accuracy of ground measurements of vegetation biophysical properties in the previous example the error in LAI measured by a Delta T SunScan LAI meter were about the same size as the error in the LAI estimates from the remotely sensed data Salford Advanced Laser Canopy Analyzer (Salca)

19 Final thoughts and discussion points Operational assimilation of data in agrometerological models will require new approaches to collection and analysis There are probably enough platforms in space to collect daily 30m spatial resolution data for the whole of Europe The effects of cloud cover need to be quantified some areas may be too cloudy (<14% cloud free frequency) to allow weekly observations cloud free frequency may be seasonally dependent does anyone know? The most promising solution is constellation of identical satellite in same orbit. Small satellite technology opens the possibility of a European AgriSatellite constellation Current systems lack waveband sin the SWIR, and many have limited onboard storage and data transmission capabilities Cal/Val activities are critical and operational atmospheric correction for local areas is a pre-requisite