Pests of mustard_Identification_Management_Dr.UPR.pdf
Loustau, Denis: Early detection of the effects of a changing environment on ecosystems
1. Early detection of the effects of a changing environment
on ecosystems
• Lesson from the fluxnet 2015 data set
• Application to the ICOS ecossytem network .
Moreaux V., Panthou G., Josse B., Lamy K., Bert G., Papale D., Loustau D.
2. Can we detect the CO2 signal in historical time series the flux data?
The CO2 atmospheric increase
CO2 increase is worldwide
No « control » site
Analysis on temporal trend
CO2 effects assumed similar within PFT s
Process / variables to be targeted
Target variables
• Annual sum : normalized GPP
• Maximal ½ hourly value : normalized GPPmax
= upper quartile of light saturated values in
June, between 10 et 12 am, wet soil, max LAI,
VPD < 1500 Pa
Analysis of the
Fluxnet 2015 data
3. Can we detect a CO2 signal in Flux data ?
The Fluxnet 2015
dataset
𝜀 𝐺𝑃𝑃, 𝐺𝑃𝑃𝑚𝑎𝑥 = 𝜀 𝑁𝐸𝐸𝑑𝑎𝑦 2 + 𝜀 𝑁𝐸𝐸𝑛𝑖𝑔ℎ𝑡 2
Annual GPP:
𝛿𝐺𝑃𝑃
𝛿𝑡
GPPmax:
𝛿𝐺𝑃𝑃max
𝛿𝑡
𝜎(𝐺𝑃𝑃) = 𝑓 𝑐𝑙𝑖𝑚𝑎𝑡𝑒, 𝑁𝑥, 𝑂𝑥, 𝑆𝑥, 𝑚𝑎𝑛𝑎𝑔𝑒𝑚𝑒𝑛𝑡, 𝑎𝑔𝑒
𝜎(𝐺𝑃𝑃max) = 𝑓( 𝑁𝑥, 𝑂𝑥, 𝑆𝑥, 𝑚𝑎𝑛𝑎𝑔𝑒𝑚𝑒𝑛𝑡, 𝑎𝑔𝑒)
Uncertainty:
Temporal trend:
Detrended variability:
• Constant u* threshold (CUT)
• Gap filled (MDS )
• GPP – RECO (Lasslop et al. (DT))
• Selection of 10yr long time series :
• 55 stations in 9 PFTs
• Pastorello et al. 2020
Number of stations PFT
FLUXNET time series(Pastorelllo et al. 2020)
4. • A mathematical model of ecosystem continuous measurements:
X t, n = 1 + a × t + N (0, α ) a = f (e, σ )
Data uncertainty Natural variabilityX normalised variable (ex: GPP)
n network size / PFT, eco domain
t time
a step change or linear drift in time
Can we detect a CO2 signal in Flux data ?
5. Temporal variability s (GPPmax)0.75
0.50
0.25
0.00
GPPmax: Forest < Grassland, Shrubs < Crops
( GPP annual : same)
Can we detect a CO2 signal in Flux data ?
Uncertainty, e (GPPmax)
0.30
0.20
0.10
0.00
GPP max: median = 0.02 to 0.05
(GPP annual: ENF, GRA and SH > 0.05)
6. Annual maximum, GPPmax
GPP max : ENF is the only PFT with a significant,
positive trend.
a = + 0.028 yr-1 p < 0.001
e = 0.033
s = 0.17
(GPP annual : same conclusion)
a = + 0.015 yr-1 p < 0.001
e = 0.08
s = 0.10
Can we detect a CO2 signal in Flux data ?
GPPmax shows higher temporal variability and trend than GPP annual.
A significant positive trend in GPPmax and GPP annual is observed only in the
coniferous forests PFT (ENF)
The trend is steeper than the expected CO2 effect (0.028 vs ~ 0.005 yr-1) and
might be confounded with other drivers (temperature ?).
7. Application to the ICOS Ecosystem network: a virtuous cycle
Scenarios :
Chemistry-transport
atmospheric model
• T, ew, SW, LW, precipitations, CO2,
Ndry , Nwet and Ox at 50 km resolution
• Ecodomain delimitation : Control vs
Impacted areas
Impacts:
Canopy model + literature
based expertise
• Effects on Ecosystem
• Definition of Target Variables
𝜕𝑋
𝜕𝑡
across ICOS stations
or
comparison analysis
X(control) vs X(exposed)
Impact detected p=0,05
N Y
Attribution can be proven
N Y
Network optimisation :
• Station number ?
• Variable accuracy ?
• Add new variable ?
• Control stations ?
Success !!
Detectability :
Part 2.
8. • The surface detection concept.
gives the network size, accuracy and duration
detecting a given temporal drift with p=0.95
Temporal drift (a, yr-1)
2.
X t, n = 1 + a × t + N (0, α )
Application to the ICOS Ecosystem network:
Extraction of (n, t, a) detecting a with p=0,05
(Monte Carlo, 5000 runs)
• Calibration: test with the fluxnet
2015 dataset
EX. ENF stations :
ε = 0.09
σ = 0.08
a = 0.12
9. a = 0.45
a = 0.25
a = 0.12
Optimisation of the ICOS Ecosystem network
• A loss in accuracy
degrades the network
sensitivity
• Increasing the station
number, n, is efficient for a
small network size
• Reducing data uncertainty,
e, is difficult…
• …but reducing the
detrended variability, s,
can be achieved through
synchronic comparison
between control and
exposed areas
2.
Temporal drift (a, yr-1)
10. Preliminary conclusion (more to come in the RINGO final report)
Optimising in situ networks for the early detection of environmental changes
a (impact) maximisation
- target variables are key
- long time series are requested for accumulative effects detection
- PFTs with continuous exposition to the atmosphere, covering large geographic areas
e uncertainty reduction
- measurements harmonisation
- high QC
- minimal network size / PFT
a minimising
- Importance of PFTs with low management intensity (ENF and forest PFTs, grasslands, mires)
- Distribute stations among control and exposed areas (suppl. slide)
Attributability
Target variables whose drivers are well understood (high attributability)
11. • Detection of future impacts (climate, N deposition, Ox)
across Europe:
Analysis of expected impacts on the ICOS Ecosystem
sites
Model simulation of climate-chemistry scenarios 2020-
2050 at 50km resolution
2035 anomaly of Ox deposition over Europe (B. Josse et al.
2019).
GPP (CO2 Hist) –
GPP (CO2 = 350)
(gC.m-2 y-2)
3.
𝛿𝐺𝑃𝑃
𝛿𝑡
= 6,8 𝑔𝐶𝑚−2
𝑦−2