Introduction Data & Methodology

Discussion Conclusions

Acknowledgments

Tenerife, the largest of the seven Canary Islands, has the highest elevation of Spain, the Teide volcano(3.718 m. – the third largest volcano in the world from its base). Its climate is hard influenced both for its tropical latitude and for its exposition to the Trade Winds (alisios), which blow from the northeast for almost the whole year. These two key facts (an island with such a high hill, which implies not only high peaks in a small area, but also strong slopes all over the territory; and a climate strongly influenced for the prevalent Trade Winds) interact together producing a very special local climate – the most specific phenomenon is the sea of clouds, which happens at mid-heights when the air moisture condenses for the Trade Winds pushing of the air against the hilltops.

This causes a special distribution ofthe rainfall all over the island (forthe land facing the Trade Winds, and for the lands isolated of thatwinds because of the hills), a special distribution which must be taken account in a project for thecreation of a set of monthly rainfallmaps – like ours.

As input data we have used the daily series ofprecipitation registered at the Spanish MeteorologicalAgency (AEMET) rain gauges located in the island, for the 1951-2010 period (where available). All thesedata must be aggregate in monthly data for tryingdifferent interpolation techniques. We have tried bothKriging (in its ordinary version) and Thin Plate Splines(in its 2-dimension version, 3-D version, and itsversion with dependence of a covariance); InverseDistance Weigthing (IDW) is used as a base methodfor comparing the results. For determing the best method, we have used a cross-validation leave-one-out (LOO) simulation and we have compared thesimulated values against the observed ones. For themeasure of the similitude, we have used not onlymeasures based in central measures (BIAS, MAE, correlation, Model Efficiency,... between observedand simulated data) but also goodness-of-fit tests(bootstrap Kolmogorov-Smirnov, Anderson-Darling) and the comparison between the transects (over theheighest point – Teide – and over the rainiest point –Matanza gauge) of the simulated data against the observed.

Fig. 2. a) Location ofobservatories in thethree area divisions andthe two transect lines. b) Temporal evolution ofthe number ofobservatories. Note thecut at 1985.

Results

Monthly rainfall maps for Tenerife (Canary Islands): comparing interpolation methods under local

climate conditionsF.J. Pórtoles, L. Torres, J. Ribalaygua, R. Monjo, and C. García Angulo

Climate Research Foundation (FIC), Spain, fic@ficlima.org

- Our results show that the use of the relationship between elevation and precipitation do not increase the accuracy of the interpolation (no matter what method is used), but the division of the island in climatic zones does. Therefore, we have divided the island in three climatic zones (the one facing the Trade Winds, in which the sea of clouds causes a special rainfall; the one isolated of the Trade Winds but exposed to the Sub-Tropical Cyclones; and the zone in the highest places, which has mountain climate - even snow) with geographical / climatic criteria.- The used statistics and transects showed that the result is acceptable, and served to choose the most appropriate interpolation method: TPS-2D in a three area division, with square root of rainfall and starting at 1985.- At the end, we have achieved 12 monthly maps of accumulated rainfall and one map of annual rainfall.

These results are part of the “Estudio del impacto del cambio climático sobre la diversidad y la composiciónde las cubiertas forestales en los Parques Nacionales españoles” study, funded by the Biodiversity Foundation, a public foundation of the Spanish Government. The meteorological data are collected by the Spanish Meteorological Agency (AEMET). We have used the R statistical environment (R Development CoreTeam) and the fields and gstat packages.

- Applying the square root to precipitation data for interpolating gives a better BIAS that if it is not applied (Fig. 3a).- LOO cross-validation shows no significant differences between starting to interpolate from 1950 or 1985 (Fig. 3a); we prefer to start on 1985 because since that date we have better spatial coverage (Fig. 2b).- The division of the island into three areas (North, Central and South) obtains a slight improvement in MAE over the other options (the whole island and two zones) (Fig. 3b).- TPS-3D obtained the best BIAS and MAE results (Fig. 4), but the transects show thatinterpolated precipitation is too much dependent on the height (Fig. 5) and is not realistic.- Ordinary kriging obtained similar errors than TPS-2D (Fig. 4).- We choose TPS-2D because is smoother than IDW (which has bull´s eye effects), and similar than ordinary kriging (but needs less computational effort).

Fig. 5. a) Transect of UTMX = 338.8 km (over Teide) for TPS-3D interpolation. b) Transect of UTMX = 338.8 km for TPS-2D interpolation. c) Transect ofUTMX = 360.0 km (over Matanza gauge) for TPS-2D interpolation.

Fig. 3. a) BIAS comparison between using SQRT and not using, for two starting dates (1950 and 1985). b) MAE comparison between using one, two and three division areas for Tenerife

Fig. 4. a) BIAS and b) MAE comparison between several interpolation methods

a)

b)

North area: 114Central area: 45South area: 185

Bootstrapped Kolmogorov-Smirnov test for all LOO cross validations shows that simulations are notsignificantly different of observations, except in summer months (Fig. 8a) (note the extremely dry summer, Fig. 5). In such cases, the median p-value is close to 0.05, so that, despite reduced summer rainfall, the simulations are very close to the observations. Anderson-Darling test (not showed) gave similar results to theKS test, but with a lower p-value because it is more restrictive for the extremes.Pearson correlations (Fig. 8b) are acceptable, even for summer, and are similar for all interpolation methods. Model Efficiency (Fig. 8c) is acceptable for all methods except for the summer period.

(covariance) (covariance)

Fig. 8. a) P-value of Kolmogorov-Smirnov test, b) Pearson correlation, and c) Model Efficency for all LOO cross validations.

Despite the apparently good relationship between rainfall and height (cross validation LOO of TPS-3D), the transects show that precipitation in the peaks of Tenerife would be too high (Fig. 5). For that reason, we finally chose the TPS-2D method. We have divided the island in three climatic zones, and we have interpolated the data corresponding to the rain gauges existing in that areas into a 200 x 200 m. grid, and we have merged the simulated data in the borders between the defined climatic zones (Fig. 6 and Fig. 7). As a monthly example, note that in December the southwest of Tenerife is affected by the Atlantic low pressures, which inject a very moist southwest flow.

Fig. 6. Annual precipitation interpolated with TPS-2D

Fig. 7. Precipitation of December, interpolated with TPS-2D

European Conference on Applications of MeteorologyEMS Annual MeetingBerlin | Germany |12-16 September 2011

Fig. 1. a) Sea of clouds over Tenerife (and other Canary islands). Source: MODIS Rapid Response Team, NASA-GSFC. b) Detail of Tenerife

a) b)

a) b) c)

TEIDE

TENERIFE

a) b)

a) b)

a) b) c)