Inter-annual variability in Prosopis caldenia pod production in the Argentinean semiarid pampas: A modelling approach

Inter-annual variability in Prosopis caldenia pod production in the Argentinean semiarid Pampas: A modelling approach Lucia Risio1, 2,*, Rafael Calama2, 3 Stella M. Bogino4 and Felipe Bravo1, 2 1 Sustainable Forest Management Research Institute University of Valladolid-INIA, Av. Madrid 44, 34004 Palencia, Spain. 2 Departamento de Producción Vegetal y Recursos Forestales, E.T.S. de Ingenierías Agrarias, Universidad de Valladolid, Palencia, Spain. 3 Departamento de selvicultura y Gestión Forestal, CIFOR-INIA. Madrid, Spain. 4 Departamento de Ciencias Agropecuarias, Universidad Nacional de San Luis, Argentina. * Corresponding author: Tel: +34 979108427, Fax: +34 979108440, E-mail address:[email protected]. 1 Abstract The driest part of the Argentinean pampas is occupied by semiarid woodlands dominates by Prosopis caldenia Burkart (Calden). Calden pods are a highly valuable fodder supplement for livestock but its production is highly variable. Our objective was to analyze and model the temporal pattern in inter-annual variability of Calden pod production. Our key hypothesis is that weather conditions are the main determinant of the pod masting behavior. Tree size and climatic variables were evaluated as explanatory covariates using a zero-inflated log-normal modelling approach. The proposed final model structure incorporated 25 parameters, including four variance components, two intercepts for both the logistic and the log-normal parts of the model, and nineteen parameters associated with fixed effects. Climate had a strong influence on the flowering-fruiting Calden process and on the inter-annual variability of the final pod production at the tree level. Temperatures during bud breaking, flowering and fruit shedding, together with the precipitation from the final month of fruit shedding and the total amount of the prior vegetative cycle, were the main weather covariates that affect the processes. Keywords Zero-inflated, fruit, non-wood forest product, calden. Abbreviation list g per crown area square metre grams per square metre of crown area; ZILN zero inflated log normal This is a post-copyedited, author-produced PDF of an article published in 2016 in Journal of Arid Environments. The final publication: Volume 131, Issue 4, pp 59-66 is available at ScienceDirect via: http://www.sciencedirect.com/science/article/pii/S0140196316300349 2 1. Introduction Calden forests (Caldenales) are semiarid woodlands covering about 170,000 km2 of central Argentina. This xerophytic open woodland is a transitional ecosystem between the Pampas grasslands and the dry Monte shrublands, dominated by calden trees (Prosopis caldenia Burkart) an endemic species of Argentina. Two opposing processes simultaneously operate in the Caldenales. Firstly, there is a high rate of deforestation across the area of its distribution (0.86% per year), leaving only 18% of the original area (SAyDS, 2007). Secondly, as the original area shrinks, adjacent grasslands not occupied by crops are increasingly encroached upon by calden, leading to very high density secondary woodland (Dussart et al., 1998). The invasion of pristine grasslands by calden and the increased densities of this species in savannas are well-known vegetation changes in the semiarid region of central Argentina (Dussart et al., 1998). Woody-plant encroachment has long been of concern to a broad range of stakeholders, from pastoral farmers to ranchers, because of the subsequent negative impact on livestock production (Anadon et al., 2014), the main economic activity in these woodlands after beekeeping. The non-wood forest products are a major source of income for the Caldenales woodland owners since wood extraction, mostly firewood production, is only a marginal activity (SAyDS, 2007). Fruits of Prosopis spp. have served as a food source for humans and domestic cattle in rural communities of arid and semiarid environments around the world since ancient times (Burkart, 1952). Calden pods are a highly valuable fodder supplement for livestock due to the pod’s nutritional characteristics: 15% raw protein, 2.2 Mcal of metabolizable energy per kg of dry matter and a 52% of dry matter in situ digestibility (Privitello et al., 2001; Menvielle, 1985), during its ripening period (June-July). Due to its availability, the pods can provide the sole source of food at certain times of the year or supplement grass for cattle in the winter months (Privitello and Gabutti, 1988). Calden flowers are also recognised as a source of nectar (Genise et al., 1990) and pollen for honey bees (Andrada et al., 2005). Also, the spatial and temporal dynamics of flowering, fruiting and seeding can be considered as a key to controlling natural recruitment of plant populations (Calama et al., 2011). 3 Most Prosopis species produce abundant flowers at a predictable time of the year (spring and/or summer depending on the species) (Simpson et al., 1977). Despite the predictability of the Prosopis blooming, high variability in fruit production has been observed (Salvo et al., 1988). Calden pod production is highly variable from year to year and from tree to tree, even with trees from the same stand (Peinetti et al., 1991). In years of high fruit production, it is common to find trees that allocate a high proportion of photoassimilates into pod production growing close to neighboring trees of a similar size that do not bear fruits (Peinetti et al., 1991). Many long-lived plant species exhibit strong synchronized annual variability in fruit production, this phenomenon is known as masting or mast fruiting (Ostfel and Keesing, 2000). The interval between consecutive masting and the degree of periodicity is species specific and varies depending on endogenous control factors, weather conditions and resource availability (Thomas and Packham, 2007; Han et al., 2008). The mast seeding has important ecological consequences, not only on the recruitment of the species that exhibit this reproductive behaviour (e.g. seedling establishment may be limited to mast years, (Negi et al., 1996)), but also on a myriad of organisms in other trophic levels, whether directly or indirectly related to seed and fruit availability: e.g. direct consumers of seeds (insects, birds, small and large mammals), and predators of seed consumers and parasites (Espelta et al., 2008; Kelly et al., 2008). Edible fruits and seeds from some forest species can represent an important non-timber forest product (Scarascia-Mugnozza et al., 2000). Additionally, when fruit production is one of the main objectives in forest management planning, adequate estimates of fruit production on spatial and temporal scales are often required. Our objective is to describe, analyze and model the temporal pattern in interannualvariability of Prosopis caldenia pod production in woodlands located in the northern limit of its natural distribution area. We shall identify the factors controlling the temporal variability in order to develop models that allow us to predict the annual pod production at tree level. Our key hypothesis is that weather conditions are the main determinant of the masting behaviour. Given the data structure, a Zero-Inflated LogNormal mixture distribution will be evaluated and fitted as the modelling approach for our data. 4 2. Material and Methods 2.1.1 Study area Prosopis caldenia woodlands thrive at the edge of the driest area of the Argentinean Pampas, across 34-36ºS and 64-66ºW (Fig. 1) (Anderson et al., 1970). Across its natural distribution area, the total annual precipitation varies from 450 to 620 mm, and it is concentrated in the spring and summer months (78%, from October to March). Temperature ranges from the annual isotherms of 16 to 18 ºC. The area is a well-drained plain with moderate slopes produced by wind and fluvial processes (SAyDS, 2007). Soil types are mainly poorly developed and well drained, with scarce horizon differentiation and low water holding capacity and with 1.5 to 3% organic matter. The area is severely affected by wind and water erosion due to its poor soil structure (Peña Zubiate et al., 1998). Five sampling sites were located on privately-owned properties in the San Luis province, near the northern edge of the natural distribution area of calden woodlands. These lands are used for cattle ranching, and there is no active silvicultural management practices on them. 2.1.2 Prosopis phenology, flowering and fruiting. Prosopis species sprout at the beginning of spring (average temperature around 16ºC) and stay in leaf until autumn. The initiation of leaf production and cambium activity appears to be rather independent of rainfall (Mooney et al., 1977; Villalba, 1985). The flowers are hermaphrodite, entomophilous and they depend on pollinating insects for seed setting (Fig. 2). Most Prosopis species produce abundant flowers at a predictable time of the year, since they bloom regardless the yearly rainfall fluctuation (Simpson et al., 1977), responding rather to photoperiod as well as the length of the growing season (Solbrig and Cantino 1975). High variability in Prosopis fruit production has been observed (Salvo et al., 1988); close observation of native stands of Prosopis indicated that only between 0.05 and 0.25% of the flowers buds initiated fruits, and only 20–45% of the initiated fruits reached full size (Cariaga et al., 2005). Similar patterns have been observed in P. flexuosa and P. chilensis (Mooney et al., 1977). Flower mortality takes into accounts either inflorescence abortion and/or intra-inflorescence flower abortion 5 Cariaga et al., 2005). Peinetti (1991) reported that for P. caldenia only 5x10-4% of flowers reached fruit maturity, following the flower and fruit abortion patterns of the genus. 2.1.3 Data Tree data We sampled P. caldenia pod production in five different sites during the months of June and July (when the ripe pods dehiscence occurred ) of seven non-consecutive years (1982, 2000, 2004, 2011, 2012, 2013, 2014). A total of 400 trees were sampled, and the number of sampled trees varied among years and sites (Table 1). Only the trees of site 1 (64) were measured twice (1982 and 2000). Different trees were measured at the remaining sampling sites. At each tree, diameter at breast height (at 1.3 m) in cm, diameter in the base of the stem (at 0.3 m) in cm, tree height (m), and crown area (m) were recorded (Table 2). Once a year when the dehiscence of the annual pod production occurred , four samples per tree of 1 m2 in each cardinal point below the tree crown, (in the middle of their projection above the ground), were collected and taken to the laboratory, where they were oven dried at 80º C until they reached a constant weight. Wild animal predation or pod redistribution before sample was not taken in account. In order to evaluate the morphological variability (weight and length) a random subsample (n=50) of the 2014 pod production were measured. Climate data The climate data of the EEA INTA Villa Mercedes meteorological station was used due to its proximity to the samples sites (40 kilometers from the farthest site). Data was organized according to the southern hemisphere vegetative cycle, from May to April. Climate data is available at: http://siga2.inta.gov.ar/en/datoshistoricos/ 2.2 Methods 2.2.1 Response and explanatory variables Annual pod-yield per tree can be expressed either by the number of pods or by their dry weight. In our case we decided to use the weight as the response variable since its takes into account both the phenomena related to the initial processes of floral induction and pollination (determining pod numbers) and the pod growth. Furthermore, 6 weight is a better indicator of the total amount of resources allocated to the reproductive effort than the number of pods, since it reflects the observed variability in pod length and weight. Tree pod productivity was expressed in grams per square metre of crown area (g per crown area square metre) because production per crown unit area is the most objective way to measure productivity and compare between different stands and locations (Gea-Izquierdo et al., 2006). For both variables (pod number and weight), the distribution of frequencies did not fulfill the standard normality assumption, displaying: - Asymmetry: empirical distribution is significantly skewed towards the higher values of the variable, with a massive number of observations showing smaller values of pod production, and only a small number of trees in a few years giving very large crops and forming a long tail to the right (Fig. 2). - Zero inflation: the distribution displays a strong mode at zero (corresponding to null production by sampled trees), comprising 45% of the observations in the fitting data set (Table 1). - Truncation: given the nature of the response variable, negative values are not possible. Furthermore, the hierarchical structure of the data (repeated observations from trees nested in sample plots within natural units) implies a lack of independence among observations, which prevented us from using estimation methods based on ordinary least squares minimization. The tree size group variables; diameter at breast height, crown radio, basal areas at breast height and crown width were evaluated to explain spatial variability in pod production. The temporal variability in pod production was explained by evaluating different characteristics of the weather over the course of the study period: monthly rainfall (mm), mean, maximum and minimum temperatures (ºC), monthly sum of chill hours, monthly sum of effective sunlight hours, frost free period (days), monthly sum of the days with precipitation and with a mean wind velocity over 18 km.h-1. The last two variables were included due to the possible effect they can have on insect pollination activity. Because flower bud induction is produced the year prior to the floweringfruiting year, weather variables from the induction year were also evaluated. Finally, due to the southern hemisphere location of the study area, the climate variables were considered according to the vegetative year from May to April, involving two different calendar years. 7 2.2.2 Modeling approach. We used zero-inflated models for our modelling approach. The explanatory covariates selection was carried out by first independently fitting a binomial regression model for the dichotomized data for fruiting occurrence and, after, fitting a log-normal model using only the non-null intensity data (weight of pods), as proposed by Heilbron (1994) and Woollons (1998). The independent fitting of these generalized linear models can be accomplished using maximum likelihood estimation methods. Information criteria such as −2LL and AIC were used to define the best independent model for each component. In a subsequent step, simultaneous fitting using ZILN was carried out with these pre-selected covariates, testing the significance level of the parameters and removing those that were non-significant. The explanatory covariates may or may not be common to both the occurrence and intensity models. We compared the three possible alternatives of additional level of random variability (site, year and tree), and then selected the best according to Akaike´s Information Criterion (AIC), Bayesian Information Criterion (BIC), and minus two likelihood (-2LL). Site per year, site per tree and tree per year iteration terms were also evaluated as random sources of variability, but problems in the model convergence were detected and, thus, it was discarded. The accuracy of the selected model was checked using the fitting data set, since not more data was available. Two alternatives were evaluated to predict the pod production from a tree using the fitted model: a) a cut-off value of 0.55 was set (proportion of fruiting trees in the data set). If the value predicted by the occurrence part of the model was greater than the cut-off value, the pod production was predicted using the intensity part of the model, otherwise the predicted production was zero; b) The pod production from a tree was equal to the product of the expected probability of occurrence, as estimated using the logistic part, and the expected value of pod production estimated by the log-normal part. In this case, no zero values are predicted. Approaches (a) and (b) were compared using the fitting data set, considering the mean error (E), root mean squared error (RMSE) and modelling efficiency (EF) (see the Electronic Appendix for details). In the case of approach (a), specifity (rate of correctly 8 modeling allows a simultaneous and correlated estimation of the parameters explaining both processes (Calama et al., 2011). It also prevents us from violating basic statistical assumptions derived from zero abundance, non-normality and inherent correlation among observations, factors which have been identified as the main impediments to modelling annual fruit production (Calama et al., 2008). Both of our modelling approaches (Electronic Appendix, Table 2) allowed us to accurately predict pod production at the tree level. Weather is not the only controlling factor over the masting behavior of P. caldenia. A deeper analysis is needed that includes; a) physiological factors (e.g., hormonal inhibitions caused by the ripening seeds (Lee, 1979)), b) intraspecific genetic variability. Extrapolation of Prosopis pod production data from one “natural stand” to another would be extremely difficult, even under similar climatic and moisture conditions because of the genetic variability in Prosopis; Felker et al. (1984), c) dendrochronological analysis for evaluating the resource depletion theory whereby the demands on resources in a bumper crop leave the tree with insufficient resources to support a normal crop the following year, d) the incorporation of stand attributes to take into account spatial variability such as stem density, stand age and site index, and e) predation by bruchid beetles (Vir, 1996). 5. Conclusions Climate has strong influence on the flowering-fruiting of P. caldenia and on the interannual variability of the final pod production at the tree level on the northern limit of its natural distribution. Temperatures in the months of bud break, flowering and fruit shedding, together with precipitation from the final month of fruit shedding and the total amount of the prior vegetative cycle, are the main weather covariates that affect the processes. Zero-inflated models allow us to take into consideration the idiosyncrasies of data from plant flowering-fruiting studies without violating standard assumptions or using data transformation. Our statistical approach allows an accurate prediction of P. caldenia annual pod production, which will enable forest managers to carry out annual planning activities such as predicting the amount of pods to include in their livestock 15 management plan or estimating the crops which can be expected under different climatic scenarios. 6. Acknowledgments We would like to thank Liliana Privitello, Elba Gabutti and Guillermo Cozzarin for allowing us to use part of their data base. Thanks to Elena Scapinni, Silvina Mercado and Sergio Chiofalo who participated in the lab work. Finally, we gratefully acknowledge the funding from the ERASMUS MUNDUS ECW 2009 1655/001-001 European Union mobility program, a fellowship awarded to the corresponding author. 7. References Affleck, D.L.R., 2006. Poisson mixture models for regression analysis of stand level mortality. Canadian Journal of Forest Research 36, 2994–3006. Aitchinson, J., 1955. On the distribution of a positive random variable having a discrete probability mass at the origin. Journal of American Statistical Association 50, 901–908. Anderson, D.L., Del Aguila, J.A., Bernardon A.E., 1970. Las formaciones vegetales en la provincia de San Luis. RIA, Serie 2, Vol. VII, Nº3, 153-183. Anadón, J.D., Sala, O.E., Turner, B.L., Bennett, E.M., 2014. Effect of woody plant encroachment on livestock production in North and South America. Proceedings of the National Academy of Sciences of the United States of America 111 DOI: 10.1073/pnas.1320585111 Andrada, A., Tellería, M.C., 2007. Pollen collected by honey bees (Apis mellifera L.) from south of Caldén district (Argentina): botanical origin and protein content. Grana, 44, 115-122 Belasco, E., Ghosh, S.K., 2008. Modeling censored data using zero-inflated regressions with an application to cattle production yields. American Agricultural Economics Association Annual Meeting, Florida, July 2008. Available on line at http://ageconsearch.umn.edu/bitstream/6341/2/456273.pdf. Burkart, A., 1952. Las leguminosas Argentinas silvestres y cultivadas. Acme Agency. 16 Calama, R., Mutke, S., Gordo, J., Montero, G., 2008. An empirical ecological-type model for predicting Stone pine (Pinus pinea L.) cone production in the Northern Plateau (Spain). Forest Ecology and Management 255, 660-673 Calama, R., Mutke, S., Tomé, J., Gordo, J., Montero, G., Tomé, M., 2011. Modelling spatial and temporal variability in a zero-inflated variables: The case of stone pine (Pinus pinea L.) cone production. Ecological Modelling 222, 606-618. Cariaga, R.E., Aguero, P.R., Ravett, D.A., Vilela, A.E., 2005. Differences in production and mortality of reproductive structures in two Prosopis L. (Mimosaceae) shrub species from Patagonia, Argentina. Journal of Arid Environments 63, 696–705. Contreras, S., Santoni, C., Jobbágy, E., 2013. Abrupt watercourse formation in a semiarid sedimentary landscape of central Argentina: the roles of forest clearing, rainfallvariability and seismic activity. Ecohydrology 6, 794-805. Dussart, E., Lerner, P., Peinetti, R., 1998. Long-term dynamics of 2 populations of Prosopis caldenia Burkart. Journal of Rangeland Management 51, 985-991. Espelta, J.M., Cortés, P., Molowny-Horas, R., Sánchez-Humanes, B., Retana, J., 2008. Masting mediated by summer drought reduces acorn predation in mediterranean oak forest. Ecology 89, 805-817. Felker, P., Clark, P., Osborn, J., Cannel, G.H., 1984. Prosopis pod productioncomparison of North American, South American, Hawaiian and African germplasm in young plantations. Economic Botany 38, 36-51 Fortin, M., DeBlois, J., 2007. Modelling tree recruitment with zero-inflated models: the example of hardwood stands in Southern Québec. Forest science 53 , 529-539. Gea-Izquierdo, G., Cañellas, I., Montero, G., 2006. Acorn production in Spanish holm oak woodlands. Investigaciones Agrarias: Sistemas de Recursos Forestales 15, 339-354. Genise, J., Palacios, R.A., Hoc, P., Carrizo, R., Mofffat, L., Mom, M. P., Agullo, M.A., Picca, P., Torregrosa, S., 1990. Observaciones sobre la biología floral de Prosopis (Leguminosae, Mimosoidae) II. Fases florales y visitantes en el Distrito Chaqueño Serrano. Darwiniana 30, 71-85 17 Han, Q., Kabeya, D. Iio, A., Kakubari,Y., 2008. Masting in Fagus crenata and its influence on the nitrogen content and dry mass of Winter buds. Tree Physiology 28, 1269-1276. Heilbron, D., 1994. Zero-altered and other regression models for count data with added zeros. Biometrical Journal 36, 531–547. Jobbágy, E., Nosetto, M., Villagra, P., Jackson, R., 2011. Water subsides from mountains to desert: Their role in sustaining groundwater-fed oases in a Sandy landscape. Ecological applications 21, 678-694. Kelly, D., Koenig, W.D., Liebhold, A.M., 2008. An intercontinental comparison of the dynamic behavior of mast seeding communities. Population ecology 50, 329-342. Lambert, D., 1992. Zero-inflated Poisson regression, with an application to defects in manufacturing. Technometrics 34, 1–14. Lee, K.J., 1979. Factors affecting cone initiation in pines:a review. Research Report 15. Institute of Forest Genetics, Soweon (Korea). Ostfeld, R. Keesing, F., 2000. Pulsed resources and community dynamics of consumers in terrestrial ecosystems. Tree 15 , 232-237. Peinetti, R., Martinez, O., Balboa O., 1991. Intraespecific variability in vegetative and reproductive growth of a Prosopis caldenia Burkart population in Argentina. Journal of Arid Environment 21, 37-44. Peña Zubiate C. A., Anderson, D. L., Demmi, M.A., Saenz, J. L., D´ Hiriart, A., 1998. Carta de suelos y vegetación de la provincia de San Luis. Secretaría de Agricultura, Ganadería, Pesca y Alimentación, INTA and Gob. de la prov. de San Luis. 115 pp. Privitello, L. Gabutti, E., 1988. Producción de vainas de caldén (Prosopis caldenia BURKART) y análisis de la calidad forrajera. VI congreso forestal argentino. Tomo I: 169-171. Santiago del Estero, Argentina. Privitello, L., Gabutti, E., Leporati, J., 2001. Chauchas de caldén. Factores ambientales que afectan su producción. Actas en la XVII reunión de la Asociación Latinoamericana de producción animal. La Habana. Cuba. Mayer, D., Roy, D., Robins, J., Halliday, I., Sellin, M., 2005. Modelling zero-inflated fish counts in estuaries – a comparison of alternate statistical distributions. In: Zerger, A., Argent, R.M. (Eds.), MODSIM 2005 International Congress on Modelling and 18 Simulation. Modelling and Simulation Society of Australia and New Zealand, December 2005, pp. 2581–2587, ISBN: 05-9758400-2-9. http://www.mssanz.org.au/modsim05/papers/mayer.pdf. Menvielle, E.E, Hernandez, O.A., 1985. El valor nutritivo de las vainas de caldén (Prosopis caldenia Burk.) RevArg. Prod. Animal. Vol. 5 Nº 7-8, 433-439. Mutke, S., Gordo, J., Gil, L., 2005. Variability of Mediterranean Stone pine cone production: yield loss as response to climatic change. Agricultural and forest meteorology 132, 263-272. Mooney, H.A., Simpson, B.B., Solbrig, O.T., 1977. Phenology, morphology, physiology. In: Simpson BB (ed), Mesquite. Its biology in two desert scrub ecosystems. U.S./ibp synthesis series 4. Hutchinson & Ross, Dowden, pp 26–43 Negi, A.S., Negi, G.C.S., Singh, S.P., 1996. Establishment and growth of Quercus floribunda seedlings after a mast year. Journal of vegetation science 7, 559-799. Nilsen, E.T, Sharifi, M.R., Rundel, P.W., 1984. Comparative water relations of phreatophytes in the Sonoran Desert of California. Ecology 65, 767-776. Sarker, A., Erskine, W., Singh, M., 2003. Regression models for lentil seed and straw yields in Near East. Agricultural and forest meteorology 116, 61-72. Sharifi, M.R, Nilsen, E.T., Rundel, P.W., 1982. Biomass and net primary production of Prosopis glandulosa (Fabaceae) in the Sonoran Desert of California. American journal of botany 69, 760-767. Sawal, R.K, Ratan, R, Yadav, S., 2004. Mesquite (Prosopis juliflora) pods as a Feed Resource for Livestock –A reviewAsian Australasian Journal of Animal Science Vol 17, 719-725. Scarascia-Mugnozza, G., Oswald, H., Piussi, P., Radoglou, K., 2000. Forests of the Mediterranean region: gaps in knowledge and research needs. Forest ecolology and Management 132, 97–109. Secretaría de ambiente y desarrollo sustentable de la Nación., 2007. Informe Regional Espinal. Segunda Etapa. ANEXO I. Estado de Conservación del Distrito Caldén. Salvo, B., Botti, C., Pinto, M., 1988. Flower induction and differentiation in Prosopis chilensis (Mol.) Stuntz and their relationship with alternate fruit bearing. In: Habitat MA (ed), The current state of knowledge of Prosopis juliflora. FAO, Rome, pp 269– 275. 19 Simpson, B.B., Neff, J.L., Moldenke, A.R., 1977. Prosopis flowers as a resource. In: Simpson BB (ed), Mesquite. Its biology in two desert scrub ecosystems. US/IBP Synthesis Series 4. Hutchinson & Ross, Dowden, pp 84–105 Solbrig, O.T., Cantino, P.D., 1975. Reproductive adaptations in Prosopis (Leguminosae, Mimosoideae). Journal of the Arnold arboretum 56,185–210 Thomas, P.A., Packham, J.R., 2007. Ecology of woodlands and forest. Cambridge University Press, Cambridge. Tooze, J.A., Grunwald, G.K., Jones, R.H., 2002. Analysis of repeated measures data with clumping at zero. Statistical Methods in Medical Research 11, 341–355. Toro, H., Chiappa, E., Covarrubias, R., Villaseñor, R., 1993. Interrelaciones de polinización en zonas áridas de Chile. Acta Entomologica Chilena 18, 20–29 Tu, W., 2002. Zero-inflated data. In: El-Shaarawi, A.H., Piegorsch, W.W. (Eds.), Encyclopedia of Environmetrics. John Wiley and Sons, Chichester, pp. 2387–2391. Villalba, R., 1985. Xylem structure and cambial activity in Prosopis flexuosa D.C. IAWA Bulletin n.s. 6, 119–130. Vir, S., 1996. Bruchid infestation of leguminous trees in the Thar desert. Tropical Science 36, 11-13. Welsh, A.H., Cunningham, R.B., Donnelly, C.F., Lindenmayer, D.B., 1996. Modelling the abundance of rare species: statistical models for counts with extra zeros. Ecological modelling 88, 297–308. Woollons, R.C., 1998. Even-aged stand mortality estimation through a two-step regression process. Forest Ecology and Management 105, 189–195. 20 Highlights (for review) Highlights: We model the Prosopis caldenia pod production at tree level. Zero-Inflated Log-Normal mixture distribution with random components was fitted. Climate has strong influence on the flowering-fruiting Prosopis caldenia process. Temperatures and precipitation are the main covariates that affect the processes. Figure Figures. Figure 1. Figure 2. Figure 3. Figure 4. Table 1. Fixed parameters and variance component estimates for the selected ZILN mixed model including correlated random parameters at plot level. Covariate Logistic part (α) Log normal part (β) Intercept -1.5123 -1.9832 Tree covariates d 1.9401 0.9524 CA - 0.0897 Climate covariates mTO 0.0021 0.0073 mMinTS -0.0725 -0.1876 mMinTN -1.1045 - mMaxTD 0.3875 - maxMaxTJ -0.0540 -0.7854 mMinTD -0.4532 - mTA 1.1201 - mMinTO - -0.4875 mTD - 0.7589 mTF - -0.0037 mTM - 0.8751 Pp-1 0.1701 - PpApr - 0.7986 Random components σ2 u (year) 0.5107 - σ2 v (year) - 0.3574 σ2(residual) - 1.5019 ρ (correlation term) 0.5612 Table 2. Ability of the model for predicting annual production of P. caldenia pod at tree level over fitting data set. The ZILN approaches used were “a” (fixed cut-off=0.55 for defining fruiting-non fruiting) and “b” (the production of pod equals the product of the expected probability of occurrence and the expected value of pod production). Approach (a) (b) E (pp.CA m2 year-1) 9.52 1.67 p-value (t-test) 0.0006 0.0012 pw_pred (pp.CA m2 year-1) 51.02 58.87 pw_obs (pp.CA m2 year-1) 60.54 - RMSE 6.987 6.782 EF (%) 29.84% 31.02% Sensitivity (%) 59.32% - Specificity (%) 67.54% - Events/no events obs 220/180 - Events/no events pred 130/270 - E: mean error; pw_pred and pw_obs: predicted and observed mean pod production (pp.CA m2, pod production per m2 of crown area (g)); RMSE: root mean squared error; EF: model efficiency; sensitivity: rate of correctly classified events; specificity: rate of correctly classified non-events.