Final Project 1. Point pattern: Analyze spatial distribution of the three tree species in Vancouver Island. Data: “victoria.dat” in your R workspace. (1) Detect spatial distributions of Douglas-fir and Hemlock using a quadrat-based index of your choice. Vary quadrat size from 22, 55, 1010, 1515 to 2020 m and then examine the change of the index with the quadrat size (test whether Douglas-fir and Hemlock are at csr). Answer: Quadrat based index: Standardized Morisita index of dispersion (Id) Null hypothesis is that distributions of tress are random. I use two different strategies to test csr: 1) randomly toss quadrats on the research area (Figure 1, left); 2) seamlessly pave quadrats over the area (Figure 2, right). The number of quadrat for all tests with randomly tossing is 2244, which is the number of 22 quadrat can seamlessly cover this area. Indexes are calculated by R function dispindmorisita() from package vegan. Quadrats are generated by quadrat.count.main and quadratcount from package spatstat. tree.ppp 13 12 1 3 36 29 27 24 0 10 11 5 1 12 7 17 23 22 1 4 5 3 6 3 0 0 20 12 15 0 2 17 21 14 10 4 6 9 47 1 7 16 4 9 7 0 0 1 10 3 5 9 5 11 16 7 5 0 4 0 1 3 8 25 22 9 25 4 5 0 1 6 15 38 19 39 30 5 9 0 4 9 11 47 27 18 5 1 2 0 20 40 y 60 80 22 0 20 40 60 80 100 x Figure 1. distribution of Douglas-fir(left) and hemlock(right) Standardized index ranges from -1 to 1, with 95% confidence limited at (-0.5, 0.5). Negative values indicate regularity, positive indicate aggregation, while around 0 is for randomness. Tab. 1 Standardized Morisita index of dispersion for Douglas-fir and Hemlock Quadrat Radom (size) 22 55 1010 1515 2020 Regular (number) 5144 2118 109 76 54 Random selection Douglas-fir Hemlock -0.40726 -0.52998 0.45016 0.50000 0.50000 0.50043 0.50026 0.50018 0.50013 0.50009 Regular selection Douglas-fir Hemlock -0.57113 -0.50163 0.07365 0.34249 0.50063 0.50039 0.50186 0.50495 0.50837 0.51310 The results show that distribution of hemlock is aggregation under all conditions. Distribution of douglas-fir varies according to quadrat sizes. Types of distribution change at size 1010 (Tab. 1). Quadrat number affect results. Small number of quadrat sometime shows a ‘correct’ answer for spatial distribution, for example 4 for 2020 quadrat and 8 for 15 quadrat. However, conclusions randomly change according to where quadrats tossed. (2) Detect spatial distribution of Douglas-fir and Hemlock using a distance-based index of your choice. Calculate the index up to the 100th nearest neighbor distance and test whether Douglas-fir and Hemlock are at csr. Answer: In this case, I choose Pielou’s index of non-randomness (α) to test if these trees distribute randomly or not. The index is defined as: 𝛼 = 𝜋𝜆𝜔 ̅ ,where λ is density, 𝜔 ̅ is expect distance. Tab. 2 show the parameters calculated by Pielou’s index with 200 points. ω is observed distances between points to trees. Tab. 3 show the results tested with different number of points. Results are calculated by R function distance.main from package spatstat. There are 652 douglas-fir and 982 hemlock in the 10383m area. The critical region for both trees is (0.866204, 1.143264). If index value outsides the interval, CSR will be rejected. Tab. 2 results of Pielou’s index test (200 points to trees) λ 0.07357 0.11244 Douglas-fir Hemlock ω 4.08036 10.2604 ̅ 𝝎 0.02342 0.03579 α CSR 0.94314 Yes 3.62434 No Tab. 3 Pielou’s index for different number of points to trees Douglas-fir Hemlock 50 0.84525 3.08498 100 0.96797 4.05171 150 0.93867 3.01503 200 1.00532 3.44778 300 0.98935 3.70481 600 800 0.94097 Null 3.55299 3.88273 The value of Pielou’s index varies when number of points changed. However, different tests have similar results for distribution. When set a small number of points for test, the index can be randomly inside or outside of the critical interval, for example, Douglas-fir in the test with 50 points. R function nndist is chosen to calculate the n-th nearest neighbor distance. Meanwhile, a curve for expect distances is drawn for comparing if the distribution is aggregation or random. 30 Observed distances of Douglas-fir are closed to CSR in 20. It is bigger than CSR in the rest distance. Hemlock aggregates under the 100th neighbor. DF CSR 2 3 log of Distance, E(m) 15 10 1 5 0 Distance, E(m) 20 25 4 DF CSR 0 50 100 n-th nearest neighbor 150 0 1 2 3 4 5 6 log of n-th nearest neighbor 7 30 HL CSR 3 0 0 5 1 2 log of Distance, E(m) 15 10 Distance, E(m) 20 25 4 HL CSR 0 50 100 150 n-th nearest neighbor 0 1 2 3 4 5 6 7 log of n-th nearest neighbor Figure 2. n-th nearest neighbor distance and logarithmic distance for Douglas-fir and Hemlock (3) Compute Ripley’s L function for Douglas-fir and Hemlock. Compare and analyze L functions with the quadrat-based and distance-based results. Answer: Distribution is aggregated distribution if index of Ripley’s L-function is bigger than 0, random when it is 0, and regular when less than 0. The results show (Figure 3) that Douglas-fir distribution is regular at low numbers, and then it becomes random and as numbers increased. The distribution becomes aggregated when h>5. Hemlock’ distribution is constantly in all the h value, which is aggregation. These three methods performance similarly. L(h) -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 DF 0 5 10 15 20 15 20 h L(h) 0.0 0.5 1.0 1.5 2.0 2.5 3.0 HL 0 5 10 h Figure 3 Ripley’s L function and simulation envelopes of Douglas-fir and hemlock (4) Compute bivariate L functions between Douglas-fir-Hemlock, Douglas-fir-Cedar, and Hemlock-Cedar. Interpret the results. Figure douglas-fir (live-death) 1.0 0.6 L function -0.2 0.2 1.0 0.6 L function 0.2 -0.2 0 5 10 15 20 0 5 15 20 h DF CD 0.6 0.2 -0.2 L function 1.0 h DF HL 10 0 5 10 15 20 h HL CD Figure 4 2. Geostatistic: Kriging surface PO4 for Gigante plot (PO4srf in “soil.dat”) and produce the maps of the kriged PO4 and its variance. Evaluate the kriging prediction use cross-validation. Answer: Step 1: EDA exploration, removing trend, checking for stationarity and isotropy PO4 (ppm) is sampled from 349 points, which scattered in 400*800 meters area. The proportion of PO4 ranges from 0.011 to 12.112 ppm (Tab 4). The sampling intervals are from 2 to 20 m (Figure 5). Tab 4. Statistics of PO4 Min. 0.011 1st Qu. 1.058 Median 1.826 Mean 2.178 3rd Qu. 2.66 Max. 12.11 Histogram of PO4 0 0 20 200 40 60 Frequency 400 Y (metre) 80 600 100 800 PO4 surface 0 100 200 300 400 0 X (metre) 2 4 6 8 10 12 PO4 (ppm) Figure 5. Distribution of phosphate (left) and histogram of phosphate (right) Second order 2.0 1.0 1.5 semivariance 2 0.0 0.5 1 0 semivariance 2.5 3 3.0 3.5 4 First order 0 100 200 300 400 500 distance 0 100 200 300 400 500 distance Figure 6. First and second order of empirical variogram Step 2: Computing the empirical variogram Empirical variogram is computed by directly considering trend (Figure 6.) Step 3: Fitting and selecting a theoretical variogram model to the empirical variogram. Logistic model has shorter range (100) than spheric model does (128). In this studying, I choose logistic model as the theoretical varigoram model. Spheric model: parameter estimates: tausq sigmasq phi 1.439 1.226 127.762 Practical Range with cor=0.05 for asymptotic range: 127.7620 variofit: minimised weighted sum of squares = 5932.714 logistic model: v ~ c0 + a * u^2/(1 + b * u^2) parent.frame() a b 0.0009485 0.0006551 sum-of-squares: 17.27 2.0 1.5 1.0 0.5 Spheric model Logistic model 0.0 semivariance 2.5 3.0 3.5 model: data: c0 1.1967563 residual 0 100 200 300 400 distance Figure 7. theoretical varigram model 500 Step 4: Computing the weight w using the fitted theoretical variogram, i.e., kriging. Step 5: Predicting the values at the locations of interest 800 R function krige.conv is use to predict the value on my research area (Figure 8). 5.5 5 6 5.5 6 5.5 6.5 5.5 5 6 600 6 5 5 5.5 5.5 5.5 400 5.5 6 5 6 200 5.5 5 5 6 5 6 5.5 5 5 0 Y Coord 6 5.5 6 6 5 6 6 5.5 6 -200 0 200 400 X Coord Figure 8. kriging surface for PO4 600 800 5.5 5 6 5.5 6 5.5 6.5 5.5 5 6 600 6 5 5 5.5 5.5 5.5 400 5.5 6 5 Y Coord 6 5.5 6 6 5 6 6 5.5 6 6 200 5.5 5 5 6 5 6 5.5 0 5 5 -200 0 200 400 600 X Coord Figure 9. Standard error surface of prediction Validation Cross-validation is chosen to validate the kringing surface. The function is xvalid in package geoR. Figure 10. cross-validation 3. Lattice data mapping and modeling: model the distribution of BCI species “ocotwh”. Data are included in the attached workspace, “ocotwh.RData”. The data “ocotwh.dat” contains the (x, y) locations of 1118 trees. To convert the point pattern into lattice data, implement: >occupancy.main(ocotwh.dat,10) # 10 m is the cell size. (1) Map the distribution of saplings at 2020 m cell size. Here, “sapling” is defined by the stems with dbh<2 cm, not include 2 cm. Note the measurement unit of the dbh in the file is mm not cm. There are 434 saplings in the data set. 500 400 300 200 100 0 200 400 600 800 1000 0 200 400 600 800 1000 0 100 200 300 400 500 0 (2) Use Poisson model to identify those cells which has the “relative risk” >1 (see Chapter 15). Suppose pixels contained saplings have high mortality than pixels have big tree. Therefore , P+=1118 O+=434 r+= O+/ P+= 0.3881932 (3) Use the CAR model to model the relative risk () in terms of the three topographic covariates (slope, elevation and convexity). The topographic data at 2020 m cell size are in the R workspace used throughout the course. The CAR model has the form (see the last slide of Chapter 15): log( ) 0 1 x ,