Study area

The study area is located between geographical coordinates 18.6° and 19.8° N and 97.4° and 96.0° W. It covers an area of 20 328 km² in the central mountainous region of the state of Veracruz, Mexico. Twelve sugar mills and four alcohol plants are located in the area (Figure 1). The four alcohol plants produce ethanol for sale as fuel in the Veracruz-Boca del Río urban area (^{Agüero-Rodríguez, Tepetla-Montes, & Torres-Beristaín, 2015}). Based on the above, it is deduced that the study area has a potential demand for sugarcane, jatropha and moringa production for bioenergy purposes, which in turn could also encourage the production of biodiesel, as has been proposed by ^{SAGARPA (2009)}.

RegCM4

Mexico currently has no information related to the implementation of RegCM4 for assessing land-use changes; however, at the international level, there are studies on future scenarios in Africa, in the region known as the Greater Horn of Africa (^{Otieno & Anyah, 2012}), in three Himalayan glaciers (^{Zaman, Waqas, & Rasul, 2011}) and in China (^{Hua, Chen, & Li, 2015}).

RegCM4 is a regional climate model developed and distributed by the International Centre for Theoretical Physics (ICTP). RegCM4 was the first limited area model for studying long-term climate, which is why this methodological approach was used in this study.

The model is mainly composed of three stages: pre-processing, numerical integration and post-processing. The land-surface characteristics and processes are represented by a biosphere-atmosphere transfer scheme (BATS), which consists of describing the vegetation-soil interaction in the energy, momentum and water transfers between the surface and the atmosphere (^{Dickinson, Henderson-Sellers, & Kennedy, 1993}). With the BATS scheme, a land cover type is assigned to each point of the model mesh. Currently, RegCM4 considers 22 land-use and vegetation classes (Table 1), taken from BATS modifications and updates, although two types have also been added, which are urban and suburban environments. This is because such environments modify not only the albedo and the surface energy balance, but also create impermeable surfaces with large effects on runoff and evapotranspiration (^{Elguindi et al., 2013}).

Sequence for land-use allocation and simulations

Step 3

The first approach, known as the mother domain (Figure 1), consists of a domain that covers a large part of Mexico and simulates the atmospheric conditions of the selected period (2000-2010); the spatial resolution was 27 km and the temporal resolution every three hours. Initial and boundary conditions were provided through ERA-Interim reanalysis data with a 1.5° x 1.5° spatial resolution and a six-hourly temporal resolution. These data were selected because they present a considerable improvement in the hydrological cycle description, atmospheric circulation quality and meteorological fields (^{Dee et al., 2011}).

Figure 1 Orography of the areas of interest. A) Mother domain (resolution 27 km) covering a large part of the national territory; domain with a 9.0 km resolution (black line box). B) Area of interest (red line box) for the introduction of crops.  

Step 4

After the simulation of the mother domain, the dynamic scale was downscaled to eliminate potential errors due to abrupt orographic changes and to allow the model to represent these changes more accurately (^{Önol, 2012}). The downscaling technique consists of using the information generated by the mother domain to establish the initial and boundary conditions required by the simulations with sugarcane, jatropha and moringa crops, plus current land-use conditions, identified as the control. The control simulation was made with the land-use and vegetation that INEGI's mapping (²⁰¹⁵) provides (step 1). All four new simulations had a spatial resolution of 9.0 km and the same simulated period as that of the mother domain. The process flow chart is shown in Figure 2.

Figure 2 Process flow chart for assessing land-use change with the introduction of three bioenergy crops, using RegCM4.  

Data analysis

Four regions were considered in the data analysis. These regions had to have the greatest contrast with respect to the change in land-use and vegetation, and climatological information close to the time of the substitution of the three crops, in order to determine the behavior of the environmental temperature and rainfall. According to INEGI's cartographic description (²⁰¹⁵), the zones were the following: region 1, which is characterized by being allocated for agricultural use; region 2, which represents zones with perennial needleleaf trees; and regions 3 and 4, where diverse forests and irrigated agriculture are located, respectively. Subsequently, the monthly climatology of each simulation was obtained to make the climatological variability comparison derived from the substitution of each of the three crops in the four regions.

Study region model calibration and validation

For the best model performance, the bias of the environmental temperature and rainfall variables was corrected. For temperature, the ^{Piani et al. (2010)} methodology was used by means of the following equation:

xcor=a+bx where x and x _cor represent the simulated and corrected variables, respectively. Values a and b, which are parameters of the model, were estimated by

a=Tobs--Tsim- b=xobs-xsim-

where Tobs- and Tsim- represent the observed and simulated monthly climatic temperatures; and xobs- and xsim- are the average observed and simulated daily variables observed and simulated.

Rainfall was fitted with the ^{Ines and Hasen (2006)} technique:

xi'=xixobs-xsim- where xi' and xi represent the simulation correction with the model on day i. In turn xobs- and xsim- represent the observed and simulated monthly climatic rainfall.

In order to evaluate the model’s results, the control simulation was compared to the data observed at four weather stations located within the regions analyzed: Villa Tejeda (region 1), Maltrata (region 2), Naranjal (region 3) and El Copital (region 4) (Figure 3). These stations belong to the National Meteorological Service (^{Comisión Nacional del Agua [CONAGUA], 2010}) and were selected for their proximity to the study areas (less than 5.0 km) and because they have the largest number of data recorded during the analysis period. The differences between the two data sets were determined by the mean square error (MSE), the formula for which is as follows:

MSE=1n∑i=1nSi-Oi2

where S _i and O _i represent the simulated and observed data, respectively, and n indicates the total number of data in each set. This technique is used in climatology because it allows evaluating the performance of numerical models; large MSE values indicate greater differences between data sets (^{Bonilla-Ovallos & Mesa, 2017}). Additionally, the root-mean-square error was used to determine value thresholds that allow evaluating the good performance of a model (^{Andrade & Blacutt, 2010}).

Figure 3 Selected weather stations (black triangles) in the state of Veracruz study region. 

Validation of the RegCM4

Figure 4 illustrates the results of the environmental temperature simulations. In region 1, RegCM4 obtained data similar to those observed, as the largest differences were of 1 °C. For region 4, the model was able to adequately simulate temperature for almost the entire year, having a small overestimate of only 0.3 °C in October and November. In contrast, for region 3, the MSE increased by almost three units (Table 2), resulting in a maximum difference of 2.0 °C between actual and simulated data. For region 2, the MSE increased more than 20-fold, resulting in differences of up to 6 °C between observed and simulated temperatures.

Figure 4 Comparison graphs between the information of the selected stations in the state of Veracruz and the results of the control simulation (current situation: INEGI’s series V land-use and vegetation map ^[2015]) with RegCM4, with respect to environmental temperature: a) region 1, b) region 2, c) region 3 and d) region 4. 

According to Figure 5, in regions 1, 2 and 4, the model overestimated the monthly rainfall accumulated throughout the annual cycle; otherwise it was recorded in region 3, where the model underestimated the rainfall for the summer months (June, July and August). It should be noted that the model detected a heatwave at the Maltrata, El Naranjal and El Copital (regions 2, 3 and 4, respectively) stations; however, the control data only showed a moderate heatwave in Maltrata (region 2). Additionally, in regions 1, 3 and 4, the model simulated rainfall peaks in September and October, which resulted in a lag in relation to the control data.

Figure 5 Comparison of the information from the selected stations in the state of Veracruz and the results from the control simulation (current situation: INEGI’s series V land-use and vegetation map ^[2015]) with RegCM4, regarding rainfall: a) region 1, b) region 2, c) region 3 and d) region 4.  

In the four regions, the MSE of the rainfall variable had values greater than 7 000 units, while the root-mean-square error (RMSE) was greater than 60 units (Table 2); according to ^{Andrade and Blacutt (2010)}, the value 60 is considered as a reference threshold to determine that the model fails to adequately simulate this variable. The opposite resulted for temperature, where the RMSE was below five units. In this regard, research by ^{Fuentes-Franco, Coppola, Giorgi, Graef, and Pavia (2014)}, ^{Pal et al. (2007)} and ^{Ali, Li, Congbin, and Yang (2015)} conducted in regions of Mexico, South Africa, South America, South Asia, and China, respectively, indicated that the RegCM shows greater biases in complex orography areas, such as those evaluated in this research. Based on the above, the RegCM is considered to present problems when simulating the rainfall behavior in regions with very marked elevation gradients. In this sense, it should be emphasized that in the study zone, elevations vary from sea level to almost 2 000 m; the El Copital and Villa Tejeda weather stations are located below 350 m (14 and 348 m, respectively), whereas the Maltrata and Naranjal stations are above 900 m (1 713 and 967 m, respectively). According to the results and references cited, it is considered that the model did not adequately simulate the conditions of the study sites for the rainfall variable.

This condition is not unique to the RegCM, since ^{Andrade and Blacutt (2010)}, in their study conducted in Bolivia, found that the PRECIS regional climate model performs better when simulating temperature and rainfall in low areas (<500 m) than in intermediate (500 to 3 500 m) and high areas (>3 500 m). Therefore, it is considered that, in sites with high variations in elevation, the estimation of temperature and, especially, of rainfall generates problems of bias in these climate models.

In this study, temperature was the variable with the greatest capacity to be adequately simulated with RegCM4, while rainfall showed strong intensity and magnitude biases at all sites. This coincides with the results of ^{Sanabria, Marengo, and Valverde (2009)}, ^{Camargo-Bravo and García-Cueto (2012)}, ^{Fuentes-Franco et al. (2014)}, ^{Hassan et al. (2014)}, and ^{Ji and Kang (2015)}, who concluded that temperature can be simulated more adequately. In contrast, ^{Tiwari et al. (2015)} indicate that the best simulated variable is rainfall, although this may be due to orographic traits at the study sites.

Temperature and rainfall simulation from crop introduction

Figure 6 shows the environmental temperature simulations with RegCM4 for each crop. Regions 1 and 4 show the largest temperature variation with respect to current land-use. In this sense, the sugarcane crop (rainfed agriculture), for January to May, causes cooling of the area (0.7 °C); subsequently, from June to December its behavior is similar to the control (with smaller than 0.3 °C variations). Such cooling coincides with what was found by ^{Lobell, Bala, and Duffy (2006)}; ^{Kueppers, Snyder, and Sloan (2007)}; ^{Sacks, Cook, Buenning, Levis, and Helkoski (2009)}; and ^{Marcella and Eltahir (2014)}, who state that cooling is caused by increased soil moisture and albedo, caused by the irrigated crop, which generates changes in the latent and sensitive heat fluxes (^{Mahmood et al., 2014}; ^{Pielke et al., 2007}). For jatropha (deciduous shrub) and moringa (perennial needleleaf tree) crops, the behavior in relation to sugarcane is reversed from June to December with a maximum cooling of 0.5 °C. This cooling is due to the contrast with bare areas allocated for crops, compared to vegetated areas that have denser foliage and greater root depth, which causes an increase in evapotranspiration and a decrease in sensible heat (^{Pielke Sr. et al., 2007}). For regions 2 (perennial needleleaf trees) and 3 (mixed forest), as they are areas with tree vegetation, changes in temperature would be between 0.1 and 0.2 °C because jatropha and moringa crops are also tree species, which in case of substituting the original vegetation would have a similar behavior.

Figure 6 Comparison of the temperature obtained in the simulations with RegCM4 for each crop with respect to the control (current situation: INEGI’s series V land-use and vegetation map ^[2015]) in the four selected regions of the state of Veracruz. Current control situation: a) region 1 = agriculture/mixed crop (Villa Tejeda), b) region 2 = Perennial needleleaf trees (Maltrata), c) region 3 = mixed forest (Naranjal) and d) region 4 = irrigated agriculture (El Copital).  

Figure 7 shows precipitation comparisons among crops. Even though the model did not adequately simulate this variable, the results were congruent with the vegetation types evaluated, since, in region 1, the model with sugarcane (irrigated agriculture) produced the highest rainfall from January to June with an average increase of 21 mm over the control. This result directly correlates with the increase in soil moisture, which causes greater evapotranspiration and upward movement of the low troposphere, leading to higher rainfall (^{Zhao, Fang, Cui, & Huang, 2012}). For the other regions (2, 3 and 4), rainfall would have a maximum variation of 6.4 mm with respect to the control. In the case of moringa and jatropha, there would be a 7.0 mm increase in rainfall caused by the greater vegetation cover and soil moisture that can be evapotranspirated, which generates greater humidity in the atmosphere, causes greater convective activity and, therefore, greater precipitation (^{Mahmood et al., 2014}; ^{Pielke Sr. et al., 2007}). This coincides with what is indicted by several authors, that the greater the amount of vegetation cover, the greater the precipitation (^{Lee & Berbery, 2015}; ^{Malhi et al., 2008}; ^{Pitman, Narisma, Pielke Sr., & Holbrook, 2004}; ^{Salazar et al., 2016}).

Figure 7 Comparison of the rainfall obtained in the simulations with RegCM4 for each crop with respect to the control (current situation: INEGI’s series V land-use and vegetation map ^[2015] in the four selected regions of the state of Veracruz. Current control situation: a) region 1 = agriculture/mixed crop (Villa Tejeda), b) region 2 = Perennial needleleaf trees (Maltrata), c) region 3 = mixed forest (Naranjal) and d) region 4 = irrigated agriculture (El Copital). 

Based on Table 3, when comparing each region and each crop for the winter and summer seasons, it was determined that the greatest variation in temperature and rainfall would occur in regions 1 and 4. Of the three species, jatropha turned out to be the crop that would cause the least climatic variation, since it is the plant with the greatest leaf area and land cover, traits that help to conserve surface relative humidity (^{Salazar et al., 2016}). The opposite is shown for the sugarcane crop, which causes the greatest variation in temperature and rainfall. On the other hand, moringa has a smaller leaf area and causes more radiation to reach the surface, because it is a tree with small, thin leaves, which results in greater variations in temperature and rainfall than with jatropha. It is important to mention that although the temperature variation is less than 1 °C, authors such as ^{Georgescu et al. (2013)}, ^{Hallgren et al. (2013)} and ^{Khanal et al. (2013)} state that even these small climatic variations could cause adverse effects at both, the local and mesoscale level.