1: \begin{abstract}

2:   This paper develops a framework for the estimation of the functional

3:   mean and the functional principal components when the functions form

4:   a random field. More specifically, the data we study consist of

5:   curves $X(\bs_k; t), \ t \in [0, T],$ observed at spatial points

6:   $\bs_1, \bs_2, \ldots, \bs_N$.  We establish conditions for the

7:   sample average (in space) of the $X(\bs_k)$ to be a consistent

8:   estimator of the population mean function, and for the usual

9:   empirical covariance operator to be a consistent estimator of the

10:   population covariance operator. These conditions involve an

11:   interplay of the assumptions on an appropriately defined dependence

12:   between the functions $X(\bs_k)$ and the assumptions on the spatial

13:   distribution of the points $\bs_k$. The rates of convergence may be

14:   the same as for iid functional samples, but generally depend on the

15:   strength of dependence and appropriately quantified distances

16:   between the points $\bs_k$.  We also formulate conditions for the

17:   lack of consistency. The general results are specialized to

18:   functional spatial models of practical interest. They are

19:   established using 

20:  an appropriate quadratic loss function which we can bound by terms that

21:   reflect the assumptions on the spatial dependence and the

22:   distribution of the points.  This technique is broadly applicable to

23:   all statistics obtained by simple averaging of functional data at

24:   spatial locations.

25: \end{abstract}