J-functions for inhomogeneous point processes in space and...
Transcript of J-functions for inhomogeneous point processes in space and...
-
◭ � ◮
J-functions for inhomogeneous point
processes in space and time
Marie-Colette van Lieshout
CWI Amsterdam
The Netherlands
J-functions for inhomogeneous point processes in space and time – p. 1/20
-
◭ � ◮
J-functions for inhomogeneous point processes in space and time – p. 2/20
-
◭ � ◮
Exploratory analysis for spatial point processes
cells
0.00 0.05 0.10 0.15
0.0
0.2
0.4
0.6
0.8
1.0
envelope(cells, Gest, simulate = expression(runifpoint(42)))
r
G(r)
obsmmeanhilo
plot(envelope(cells, Gest, simulate=expression(runifpoint(42))))
Generating 99 simulations by evaluating expression ...
lty col key label meaning
obs 1 1 obs obs(r) observed value of G(r) for data pattern
mmean 2 2 mmean mean(r) sample mean of G(r) from simulations
hi 1 8 hi hi(r) upper pointwise envelope of G(r) from simulations
lo 1 8 lo lo(r) lower pointwise envelope of G(r) from simulations
J-functions for inhomogeneous point processes in space and time – p. 3/20
-
◭ � ◮
Summary statistics for stationary point processes
Let X be a stationary point process on Rd with intensity ρ > 0.
Popular statistics include
F (t) = P(X ∩B(0, t) 6= ∅)
G(t) = P(X \ {0} ∩B(0, t) 6= ∅ | 0 ∈ X)
K(t) = 1ρE
[
∑
x∈X\{0} 1{x ∈ B(0, t)} | 0 ∈ X]
J(t) = (1−G(t)) / (1− F (t))
where B(0, t) is the closed ball of radius t ≥ 0 centred at the origin.
Write P!0(X ∈ A) = P(X \ {0} ∈ A | 0 ∈ X) for the reduced Palm
distribution of X.
J-functions for inhomogeneous point processes in space and time – p. 4/20
-
◭ � ◮
Summary statistics for inhomogeneous patterns
Define
Kinhom(t) =1
|B|E
[
∑ 6=
x,y∈X
1{x ∈ B} 1 {y ∈ B(x, t)}
ρ(x) ρ(y)
]
regardless of the choice of bounded Borel set B ⊂ Rd with strictly
positive volume |B|, and using the convention a/0 = 0 for a ≥ 0.
Baddeley, Møller and Waagepetersen, SN 2000.
Remark: the definition can be extended to space-time point processes.
Gabriel and Diggle, SN 2009.
Goal: define analogues of F , G, J for inhomogeneous point processes.
J-functions for inhomogeneous point processes in space and time – p. 5/20
-
◭ � ◮
Product densities and correlation functions
The product densities ρ(n) of a simple point process X satisfy
E
[
∑ 6=
x1,...,xn∈Xf(x1, . . . , xn)
]
=
∫
· · ·
∫
f(x1, . . . , xn) ρ(n)(x1, . . . , xn) dx1 · · · dxn
for all measurable f ≥ 0 (6= indicates a sum over n-tuples of distinct
points).
In words: ρ(n)(x1, . . . , xn) dx1 · · · dxn is the infinitesimal probability of
finding points of X at each of dx1, . . ., dxn.
N-point correlation functions are defined recursively by ξ1 ≡ 1 and
ρ(n)(x1, . . . , xn)
ρ(x1) · · · ρ(xn)=
n∑
k=1
∑
D1,...,Dk
ξn(D1)(xD1) · · · ξn(Dk)(xDk )
where {D1, . . . , Dk} is a partition of {1, . . . , n}, xDj = {xi : i ∈ Dj}.
J-functions for inhomogeneous point processes in space and time – p. 6/20
-
◭ � ◮
Summary statistics and product densities
Let X be a stationary point process with intensity ρ > 0. Then
K(t) =∫
B(0,t)
ρ(2)(0,x)
ρ2dx
F (t) = −∑∞
n=1(−1)n
n!
∫
B(0,t)· · ·
∫
B(0,t)ρ(n)(x1, . . . , xn) dx1 · · · dxn
G(t) = −∑∞
n=1(−1)n
n!
∫
B(0,t)· · ·
∫
B(0,t)
ρ(n+1)(0,x1,...,xn)ρ
dx1 · · · dxn
J(t) = 1 +∑∞
n=1(−ρ)n
n!Jn(t)
for
Jn(t) =
∫
B(0,t)
· · ·
∫
B(0,t)
ξn+1(0, x1, . . . , xn) dx1 · · · dxn.
Hence
J(t)− 1 ≈ −ρ (K(t)− |B(0, t)|) .
J-functions for inhomogeneous point processes in space and time – p. 7/20
-
◭ � ◮
Intensity-reweighted moment stationarity
Let X be a simple point process on Rd for which product densities of all
orders exist and infx ρ(x) = ρ̄ > 0. If the ξn are translation invariant, that
is,
ξn(x1 + a, . . . , xn + a) = ξn(x1, . . . , xn)
for almost all x1, . . . , xn ∈ Rd and all a ∈ Rd, X is intensity-reweighted
moment stationary.
Examples
• Poisson point processes;
• location dependent thinning of a stationary point process;
• log Gaussian Cox processes driven by a Gaussian random field with
translation invariant covariance function.
J-functions for inhomogeneous point processes in space and time – p. 8/20
-
◭ � ◮
Inhomogeneous J-function
Let X be an intensity-reweighted moment stationary point process. Set
Jn(t) =
∫
B(0,t)
· · ·
∫
B(0,t)
ξn+1(0, x1, . . . , xn) dx1 · · · dxn
and define
Jinhom(t) = 1 +
∞∑
n=1
(−ρ̄)n
n!Jn(t).
Remarks
• Jinhom(t) > 1 indicates inhibition at range t;
• Jinhom(t) < 1 suggests clustering;
• Jinhom(t)− 1 ≈ −ρ̄ (Kinhom(t)− |B(0, t)|);
• the definition does not depend on the choice of origin.
J-functions for inhomogeneous point processes in space and time – p. 9/20
-
◭ � ◮
The generating functional
For measurable v : Rd → [0, 1] that are identically 1 except on some
bounded subset of Rd,
G(v) = E
[
∏
x∈X
v(x)
]
where by convention an empty product is taken to be 1.
Properties
• the distribution of X is determined uniquely by G;
• suppose product densities of all orders exist and let u : Rd → [0, 1] be
measurable with bounded support. Then
G(1−u) = 1+∞∑
n=1
(−1)n
n!
∫
· · ·
∫
u(x1) · · ·u(xn) ρ(n)(x1, . . . , xn) dx1 · · · dxn
(provided the series converges).
J-functions for inhomogeneous point processes in space and time – p. 10/20
-
◭ � ◮
Jinhom in terms of generating functionals
Write, for t ≥ 0 and a ∈ Rd,
uat (x) =ρ̄ 1{x ∈ B(a, t)}
ρ(x), x ∈ Rd.
Then
Jinhom(t) =G!a (1− uat )
G (1− uat )
where G!a is the generating functional of the reduced Palm distribution
P!a at a, G that of P itself.
Note: G!a (1− uat ) and G (1− uat ) do not depend on the choice of a and
lend themselves to estimation.
J-functions for inhomogeneous point processes in space and time – p. 11/20
-
◭ � ◮
Remarks
For stationary X, uat (x) = 1{x ∈ B(a, t)}, hence
G (1− uat ) = P(X ∩B(a, t) = ∅) = 1− F (t)
G!a (1− uat ) = P!a(X ∩B(a, t) = ∅) = 1−G(t)
Consequently, one retrieves the classic definition of the J-function.
For intensity-reweighted moment stationary X, we get counterparts of
the F - and G-functions:
Finhom(t) = 1−G (1− uat ) ;
Ginhom(t) = 1−G!a (1− uat )
which do not depend on the choice of origin a and are easy to use in
practice.
J-functions for inhomogeneous point processes in space and time – p. 12/20
-
◭ � ◮
Minus sampling estimators
In practice, X is observed in some bounded window W .
Let L ⊆ W be a finite point grid. Set
̂1− Finhom(t) =
∑
lk∈L∩W⊖t
∏
x∈X∩B(lk,t)
[
1− ρ̄ρ(x)
]
#L ∩W⊖t,
where W⊖t is the eroded set. Similarly, set
̂1−Ginhom(t) =
∑
xk∈X∩W⊖t
∏
x∈X\{xk}∩B(xk,t)
[
1− ρ̄ρ(x)
]
#X ∩W⊖t.
Proposition: ̂Finhom(t) is unbiased, ̂Ginhom(t) is ratio-unbiased.
J-functions for inhomogeneous point processes in space and time – p. 13/20
-
◭ � ◮
Pakistani earthquakes
Pakistan is regularly affected by earthquakes due to the subduction of
the Indo-Australian continental plate under the Eurasian plate.
There are two convergence zones: One crosses the country from its
South-West border with the Arabic Sea to Kashmir in the North-East, the
other crosses the northern part of the country in the East-West direction.
Two major earthquakes were recorded during 1973–2008
• 1997: magnitude 7.3 along the SW to NE zone; about seventy
casualties;
• 2005: magnitude 7.6 in Kashmir; devastating with at least 86, 000
casualties.
Restrict to shallow quakes (depth less than 70 km) of magnitude at least
4.5 as deeper or weaker ones may not be felt.
J-functions for inhomogeneous point processes in space and time – p. 14/20
-
◭ � ◮
Pakistani earthquakes: intensity function
• To avoid edge effects, include earthquakes within a distance of
about one degree from the Pakistan border;
• aggregate into a single pattern, but exclude the major earthquake
years;
• calculate the kernel estimator of intensity (isotropic Gaussian kernel
with standard deviation 0.5, approximately 50 km).
Pooled earthquake intensity
010
2030
4050
60
J-functions for inhomogeneous point processes in space and time – p. 15/20
-
◭ � ◮
Earthquake data 2005–2007
Note: a KPSS test indicates the intensity pattern persists over the years.
2005 2006 2007
Question: can the data be explained by a series of inhomogeneous
Poisson processes?
J-functions for inhomogeneous point processes in space and time – p. 16/20
-
◭ � ◮
Inhomogeneous J-functions 2005–2007
0.0 0.5 1.0 1.5
0.85
0.90
0.95
1.00
2005
0.0 0.5 1.0 1.50.
980.
991.
001.
011.
02
2006
0.0 0.5 1.0 1.5
0.98
0.99
1.00
1.01
1.02
2007
Ĵinhom for the locations of shallow earthquakes of magnitude at least 4.5
in 2005–2007 with upper and lower envelopes based on 19 independent
realisations of an inhomogeneous Poisson process.
Conclusion: evidence of clustering over and beyond that explained by
the spatial variation.
J-functions for inhomogeneous point processes in space and time – p. 17/20
-
◭ � ◮
Space-time point processes
Let Y be a simple point process on Rd × R equipped with the supremum
distance
d((x, s), (y, r)) = max{||x− y||, |s− r|}.
One may define Jn and JSTinhom as before. However, it is more natural to
scale space and time differently. Then
Jn(t1, t2) =
∫
B(0,t1)
∫ t2
−t2
. . .
∫
B(0,t1)
∫ t2
−t2
ξn+1(0, y1, . . . , yn)n∏
i=1
dyi
and
JSTinhom(t1, t2)− 1 ≈ −ρ̄
∫
B(0,t1)
∫ t2
−t2
ξ2((0, 0), (x, s)) dx ds,
which corresponds to the K∗ST -approach of Gabriel and Diggle (2009).
J-functions for inhomogeneous point processes in space and time – p. 18/20
-
◭ � ◮
Marked point processes
Let X be a point process with continuous marks in a bounded set
M ⊂ R. Assume that the ξn are translation invariant in the spatial
component and set, for Borel sets A,B ⊂ M ,
JABn (t) =1
|A|
∫
A
∫
(B(0,t)×B)⊗nξn+1((0, b), y1, . . . , yn) db
n∏
i=1
dxi dmi
where yi = (xi,mi) are marked points.
Write XA for the (unmarked) point process of locations in X which are
marked by a point in A. Then
• XA and XB independent ⇒ JABinhom ≡ 1;
• XA and XAc independent ⇒ JAMinhom(t) = J
XAinhom(t);
• If the points in X are randomly labelled with probability density ν(m),
m ∈ M , then JABinhom(t) = Jthinhom(t), the inhomogeneous J-function of
the thinning of XM with retention probability p =∫
Bν(m) dm.
J-functions for inhomogeneous point processes in space and time – p. 19/20
-
◭ � ◮
References
1. M.N.M. van Lieshout. A J-function for inhomogeneous point
processes. Statistica Neerlandica, 65:183–201, 2011.
2. M.N.M. van Lieshout and A. Stein. Earthquake modelling at the
country level using aggregated spatio-temporal point processes.
Mathematical Geosciences, 44:309–326, 2012.
3. K. Türkyilmaz, M.N.M. van Lieshout and A. Stein. Comparing the
Hawkes and trigger process models for aftershock sequences
following the 2005 Kashmir earthquake. Mathematical Geosciences,
45:149–164, 2013.
4. O. Cronie and M.N.M. van Lieshout. A J-function for inhomogeneous
spatio-temporal point processes. Submitted.
5. O. Cronie and M.N.M. van Lieshout. Summary statistics for
inhomogeneous marked point processes. In preparation.
J-functions for inhomogeneous point processes in space and time – p. 20/20
Exploratory analysis for spatial point processesSummary statistics for stationary point processesSummary statistics for inhomogeneous patternsProduct densities and correlation functionsSummary statistics and product densitiesIntensity-reweighted moment stationarityInhomogeneous $J$-functionThe generating functional$J_{m {inhom}}$ in terms of generating functionalsRemarksMinus sampling estimatorsPakistani earthquakesPakistani earthquakes: intensity functionEarthquake data 2005--2007Inhomogeneous $J$-functions 2005--2007Space-time point processesMarked point processesReferences