A Bioclimate-Based Maximum Entropy Model for Comperiella calauanica Barrion, Almarinez & Amalin (Hymenoptera: Encyrtidae ), Parasitoid of Aspidiotus rigidus Reyne, in the Philippines

Background: Comperiella calauanica Barrion, Almarinez & Amalin (Hymenoptera: Encyrtidae ) is a host-specific endoparasitoid and effective biological control agent of Aspidiotus rigidus Reyne (Hemiptera: Diaspididae ), whose outbreak in 2010 to 2015 severely threatened the coconut industry in the Philippines. Using the maximum entropy (MaxEnt) algorithm, we developed a Species Distribution Model (SDM) for C. calauanica based on 19 bioclimatic variables, using occurrence data obtained mostly from field surveys conducted in A. rigidus-infested areas in Luzon Island from 2014 to 2016. Results: The calculated AUC values for the model were very high (0.966, standard deviation=0.005), indicating the model’s high predictive power. Precipitation seasonality was found to have the highest relative contribution to model development. Response curves produced by MaxEnt suggested the positive influence of mean temperature of the driest quarter, and negative influence of precipitation of the driest and coldest quarters on habitat suitability. Given that has been found to always occur with A. rigidus in Luzon Island due to high host-specificity, the SDM for the parasitoid may also be considered and used as a predictive model for its host. This was confirmed through field surveys conducted between late 2016 and early 2018, which found and confirmed the occurrence of A. rigidus in three areas predicted by the SDM to have moderate to high habitat suitability or probability of occurrence of Zamboanga City in Mindanao; Isabela City in Basilan Island; and Tablas Island in Romblon. This validation in the field demonstrated the utility of the bioclimate-based SDM for calauanica in predicting habitat suitability or probability of occurrence of A. rigidus in the Philippines.


Introduction
The Philippines is a primarily agricultural nation in Southeast Asia, despite rapid industrialization in many areas of the archipelago. Statistics in 2015 indicate that 29.15% of total employment in the Philippines is in agriculture [1]. The agricultural sector has provided the fourth highest contribution to the country's Gross Domestic Product (GDP), with the latest data summarized by the Philippine National Statistics Coordination Board indicating GDP from agriculture at 53.7 billion Philippine pesos (equivalent to about 1 billion US dollars). Coconut is one of the high value commercial crops of the country and has been recognized for years as a top agricultural export [2]. However, production of this crop was severely threatened by an outbreak of the destructive coconut scale, Aspidiotus rigidus Reyne (Hemiptera: Diaspididae), which devastated plantations in the Southern Tagalog region of Luzon Island from 2010 to 2015. Feeding of this diaspidid on the foliage of coconut palms has been found to impair photosynthesis, consequently affecting flowering, fruiting, and even compromising the survival of the infested tree [3]. A native parasitic wasp belonging to genus Comperiella Howard (Hymenoptera: Encyrtidae) was subsequently discovered and was found to be highly and very specifically parasitizing A. rigidus in the outbreak areas from 2014 onwards. Preliminary findings and observations from field and laboratory studies suggested the potential of the parasitoid for biological control [4]. Additionally, the encyrtid was not only the first native record in the Philippines for its genus, but was also described as a new species, C. calauanica Barrion, Almarinez & Amalin [5]. C. calauanica has been found to be very specific to A. rigidus, although mathematical modeling and simulations by [6] assumed that the parasitoid may exhibit Holling type III functional response in which parasitization on an alternate host is necessary for survival in the absence of the primary host. Predictive geographical modeling that is based on the dependence of species and community distributions on environmental factors has been viewed as an important means to assess the impact of natural and anthropogenic environmental change on the distribution of organisms [7]. Climate-based ecological models can help in conservation efforts by providing information for resource and habitat management [8]. Among the popular algorithms used in modeling species distributions is the maximum entropy (MaxEnt) approach, which requires presenceonly data as an indication of the species' occurrence.
Models produced using MaxEnt can be easily understood and interpreted and provide valuable insights on distribution and habitat suitability for a species [9,10]. MaxEnt modeling has been used to predict the current and potential distributions of invasive species [11], as well as those of a variety of forest and agricultural insect pests which include: the large pine weevil, Hylobius abietis L., and the horse-chestnut leaf miner, Cameraria ohridella Deschka and Dimič [12], and six tephritid fruit flies [13] [16,17]; and the cotton mealybug, Phenacoccus solenopsis Tinsley, in India [18] and worldwide [19]. The use of MaxEnt modeling as a tool in integrated pest management, particularly in forecasting potential areas of new pest invasion relative to climate, has not yet been explored very well in the Philippines. Hence, in view of the use of C. calauanica for biological control of A. rigidus, the MaxEnt approach was employed in this study to generate a bioclimate-based Species Distribution Model (SDM) for the prediction of either the presence of the parasitoid or suitability of areas for its occurrence. This study provides a peer into the potential of bioclimate-based SDMs as tools for integrated pest management, especially in view of climate change. The ability and utility of the distribution model of a highly specific parasitoid to predict the potential distribution or areas of new invasion by its host are likewise demonstrated.

Species presence, Bioclimatic variables, and Other data
Presence-only data pertaining to occurrence of C. calauanica were derived from GPS coordinates recorded from periodic field Island where sightings of C. calauanica were reported in January 2016 but was not actually covered by our surveys. The occurrence points were encoded in spreadsheet form (with three columns for species, longitude, and latitude in that order) using Microsoft Excel and saved as a Comma-Separated Values (CSV) file. Bioclimatic data sets were downloaded from the WorldClim Global Climate Database (accessible from http://worldclim.org/current). These bioclimatic data were derived from global climate data interpolated by [20] and represent current conditions. The downloaded raster data sets, in BIL format with 30 arc-seconds resolution, pertain to 19 variables   [20].

Precipitation seasonality bio15
Precipitation of the wettest quarter (mm) bio16 Precipitation of the driest quarter (mm) bio17

Precipitation of the warmest quarter (mm) bio18
Precipitation of the coldest quarter (mm) bio19

Analysis and assessment of the species distribution model
The

MaxEnt species distribution model for Comperiella calauanica
The generated bioclimate-based distribution model for C.
calauanica Areas with non-zero habitat suitability were nonetheless predicted in other parts of the Philippine archipelago that were outside the range of the survey points. Although the model in raw expression Figure 1B shows predictions of moderate to high habitat suitability throughout almost the entire archipelago, areas whose predicted probabilities may be considered substantial (between "lowmoderate" and "high") consistent with the logistic expression and Sulu. It should be noted that raw values tend to be significantly lower than their logistic equivalent. Given the spectral scale for qualitative interpretation of colors on the SDM, the predicted probability value for a given point could be considered "high" when expressed as raw, but only "moderate" when logistically expressed.
This would explain the apparent spectral discrepancy between the raw and logistic expressions of the same SDM. Nevertheless, areas with "low" to "moderate" predicted probability (or habitat suitability) should be treated as potential areas for population establishment, especially if preferable environmental conditions beyond bioclimate (e.g. presence of hosts) may occur in such areas. The SDM was developed for C. calauanica, which has been found to be very specific to its diaspidid host, A. rigidus. Although recent mathematical modeling with simulations assumed Holling type III functional response [6] which would require C. calauanica to parasitize an alternate host in the absence of A. rigidus, such alternate host has not been found and the parasitoid is so far known to parasitize only A. rigidus. Hence, it is reasonable to view MaxEnt SDM for C. calauanica as a predictive model that may also apply to

Model Performance and Contribution of Bioclimatic Variables
The training AUC value of the C. calauanica SDM was 0.996, and the test AUC value was 0.966 (standard deviation=0.005). Both values are higher than 0.8, the value above which the AUC must be for the predictive ability of the model to be considered "convincing [21]. The C. calauanica SDM, therefore, has very high predictive power based on these AUC values. Precipitation seasonality (bio15) was the variable found to have the highest relative contribution in the development of the model at 51.5%. This suggests that distribution of C. calauanica may be influenced more by precipitation than by temperature, especially considering that the Philippines is a tropical country, where temperatures throughout the year tend to vary less than in temperate regions. In comparison, variables pertaining to temperature or its variations were found to have significant influence on the predicted distributions of insect pest species in temperate regions, namely Dacus spp. [14] and Lobesia botrana [15] in China, Ricania shantungensis in Korea [17], Hylobius abietis and Cameraria ohridella in Europe [12], and six species of tephritid fruit flies in Europe [13]. Three bioclimatic variables were found to have clear unidirectional upward or downward trends, and therefore potentially having the greatest impact on occurrence

Curr Inves Agri Curr Res
1128 Furthermore, it could be noted in the set of response curves that the predicted probability of occurrence remained constant across changes in variables pertaining to temperature more than precipitation, namely annual mean temperature (bio01), maximum temperature of the warmest month (bio05), minimum temperature of the coldest month (bio06), and mean temperature of the warmest quarter (bio10). If these response curves provide an approximation of the actual ecophysiological responses of C. calauanica, then it is possible that habitat suitability for either insect may be influenced more by precipitation than by temperature.

Confirmation of Occurrences Predicted by the MaxEnt Model
The MaxEnt SDM predicted moderate to high habitat suitability in areas outside the previously known and reported A. rigidus- twice to more than thirty times as many occurrence records Table 3. Nevertheless, the SDM developed for C. calauanica using relatively few points was able to correctly predict the occurrence of A. rigidus in Zamboanga City and in Romblon, and together with the parasitoid in Basilan Island. MaxEnt has been recognized for being much less sensitive to sample size compared to other distribution modeling algorithms, capable of being able to produce useful, predictive models with as small as 5 occurrence points [22,23]. To date and to our knowledge, this is the first field-based validation of the occurrence or habitat suitability predicted by MaxEnt SDM for an insect species that is important to agriculture or forestry.