Combining sensor and forecast information to aid decision making: real-time determination of hydrological peat fire risk in Kalimantan

Transcription

1 Combining sensor and forecast information to aid decision making: real-time determination of hydrological peat fire risk in Kalimantan Jaap Schellekens Based on work by: Aljosja Hooijer, Ronald Vernimmen, Marnix vd Vat Peter Gijsbers, and Albrecht Weerts

2 Peat and Fire Overall research objective: Improve understanding of relations between SE Asian peatland hydrology (water management), peatland functioning (CO2 emissions, fire risk, production loss) and ecology (remaining forest values, rehabilitation options), in support of improved management. This presentation focusses on the possibilities for forecasting the hydrological peat fire risk using remotely sensed data and models

3 Large CO2 emissions by peat: Oxidation due to drainage Peat Fires

4 Why peat? Why Fires?

5 Why peat? Why Fires? Established (but poor) water management Previous peat level Leaning oil palms because of soft soils & subsidence

6 Why peat? Why Fires? Palm oil plantation Deep drainage (1-3m in canals) causing major CO2 emissions, fire risk and subsidence/production loss. Palm fruit (for oil) not sold because palm oil export as biofuel decreasing.

7 Processes Peat dome Clay / sand substrate 5 to 50 km 1 to 10 m Stream channel Natural situation: Water table close to surface Peat accumulation from vegetation over thousands of years Drainage: Water tables lowered Peat surface subsidence and CO 2 emission starts Continued drainage: Decomposition of dry peat: CO 2 emission High fire risk in dry peat: CO 2 emission Peat surface subsidence due to decomposition and shrinkage PEAT-CO2 / Delft Hydraulics End stage: Most peat carbon above drainage limit released to the atmosphere, unless conservation / mitigation measures are taken

8 Processes 120 CO2 emission (t ha-1 y-1) Ali and others 2006 Armentano and Menges 1986 Jauhiainen and others 2004 Melling and others 2005 Murayama and Bakar 1996 Wösten and Ritzema 2001 y=91x R2= Groundwater depth (m)

9 Relation water depth and fire risk Tentative relation established between peatland water depth and fire risk, by determining correlation between annual water depth and annual fire extent in a fire-prone area ( Block C in Central Kalimantan) as determined by timeseries analysis of burn scar extent (by University of Leicester). Number of days groundwater table below threshold value Area of Block C burnt (%) Days < -0.4m EMRP S Days <- 0.8m EMRP S Days < -1m EMRP S % Burnt = 0,17*GWD + 1,75 R2 = 0, Number of days GWD < -0.8 m (Block C South) 1995 Area of Block C burnt (%) % Burnt = 0.24*GWD R 2 = Number of days GWD < -1 m (Block C South) 2007

10 Peat fire forecasting systems Current fire forecasts are short-term and arguably of little use: no fire fighting response possible. Long-term forecast would allow some sort of fire prevention response..but this requires a far longer warning time than can be provided by current systems. At the same time global medium-term climate forecasts have improved a lot in recent years, and appear to now beat statistical methods GW Level as an indicator for peat fire risk?

11 Simple water budget groundwater level model groundwater depth below surface (m) tr.1 measured (avg. 22 dipwells) modelled (forest) modelled (noforest) 8-Jun-04 5-Dec-04 3-Jun Nov May Nov-06 Model written in pcraster environment embedded in delft-fews. Allows gridded operation over SE Asia (only grid cells with peat land-cover) groundwater depth below surface (m) tr.2 measured (avg. 22 dipwells) modelled (forest) modelled (noforest) Jun-04 5-Dec-04 3-Jun Nov May Nov-06

12 Refine the research questions (1) Can the TRMM precipitation estimates be used to reliably (as good as the simulations with the ground based data) simulate peatland groundwater levels in the selected areas in Kalimantan? (2) Can a freely available seasonal forecast system be used to drive the groundwater model and what is the maximum lead time for which the peat fire risk can be forecasted? (3) Can Delft-FEWS be used to set-up a near real-time hydrological peatland fire risk forecasting system based on the available data?

13 TRMM input peatland fire forecasting TRMM satellite rainfall data shown to be comparable to ground based data, but more reliable and with much better spatial coverage TRMM sliding observed 30 day monthly P (mm) precipitation (mm) Palangka Raya (all months) Palangka Raya Palangkaraya Banjarmasin Mar Apr Jun Aug Sep Nov Jan measured measured monthly TRMM precipitation (mm) y = x R 2 = Dry-season rainfall, Jul-Oct. The extreme drought & fire years of 02 and 06 come out well. Regional variability is clear.

14 Testing model with different inputs Model results with TRMM are as good as those with measured P We can estimate the current GW level (and thus fire risk) with some confidence groundwater depth below surface [m] Measured groundwater depth Modelled - TRMM precipitation -1.2 Modelled - Measured precipitation 01-Jan Jan Jan-2008

15 The fire warning process Detection Present state of the system (fire, dry soil..) Warning Dissemination to public and authorities Response Take action to reduce damage and loss of life

16 The fire warning process Increase lead time Detection Present state of the system (fire, dry soil..) Forecasting Future state of the system (fire, dry soil..) Warning Dissemination to public and authorities Response Take action to reduce damage and loss of life

17 Forecasting Now Historical run. Initial conditions updated regularly. Data source: TRMM Forecast, Run in ensemble mode source: CFS Monitoring Warning Response Forecasting groundwater depth below surface [m] Measured groundwater depth -1 Modelled - TRMM precipitation Modelled - Measured precipitation Jan Jan Jan-2008 Simulation

18 CFS Forecasts The NCEP Climate Forecast System (CFS) was developed at the Environmental Modeling Center at NCEP. It is a fully coupled model representing the interaction between the Earth's oceans, land and atmosphere. A description of the CFS is given in Saha et al, Four forecasts (ensemble members) per day available via ftp (grib format) Archive with retrospective forecasts (1981 to 2006) 1 forecast per month 15 ensemble members In general, the following steps must be taken to correct the raw forecast field and apply a (bias) correction: Retrieve the raw daily forecast Subtract the forecast climatological value (available for download) Add the observed climatological values (constructed from ERA 40 dataset)

19 Forecasting Evaluating forecast performance: retrospective forecasts for period with TRMM data ( ) Total of 26K forecasts processsing > 4TB of data in total Performed using Batch mode in Delft- 0 Fews Groundwater Level [m] TRMM 30 day 90 day 180 day 270 day Days since Sept

20 Results based on ensemble mean R2 (Nash/Sutcliff) Bias Lead Time All points Model Efficiency Bias MSE lead time (days) lead time (days) Mean sqr err all values <-0.1 < lead time (days) R2 (Nash/Sutcliff) lead time (days)

21 Making a system Setting up the Operational environment using Delft-Fews Reading TRMM, CFS, Climate data for bias correction Data processing Real time reports (web-based) CFS seasonal forecast grib files ftp pull Import directory for PFFS. Time lagged ensemble of 40 forecasts PFFS system Extract SE Asia from TRMM and CFS ftp pull Resample CFS and TRMM to daily TRMM data Merge TRMM and CFS to one continious series Run water budget model with forecast data (40 runs)

22 Forecasting System Delft-Fews Basic components of forecasting system Data Result Dissemination Model(s)

23 Delft FEWS: Open System (fewswiki.wldelft.nl) No implications when introducing new model concepts Maintain current models/investments Easily introduce new advances HD Model Model 1 Organisation Delft FEWS Model Current 2 University Model 3

24 Example National Flood Forecasting System (NFFS) in England & Wales Implementation Harmonisation of 8 regions with 8 different systems & procedures Complex methods and procedures Fluvial as well as coastal forecasting > 20 different model types > 2000 forecasting locations NWS (USA) Replacing current RFS by CHPS (now 2011)

25 WP4: RS input peatland fire forecasting

26 Reporting examples

27 Initial work done what now? Fully probabilistic determination of forecast performance Use other bias correction (climate), Bayesian statistics Use other and multi model forecast -> Contact with ECMWF Improve GW model (add ET?) Use remotely sensed soil moisture Improve relation Hydrological parameters to fire risk