*3.2. Data*

We used a nationwide rice farming survey in Indonesia administered by the Indonesian Bureau of Statistics (BPS). The survey was conducted from May 2014 through June 2016 and covered a sample of 87,330 rice farm households (RFHs). The sample selection involved two-stage stratified random sampling. In the first stage, the national BPS o ffice randomly selected census blocks from a total of 844,946 blocks and obtained 8933 sample blocks. Only census blocks with 10 or more RFHs were considered eligible for sample selection. In the second stage, the district BPS o ffice stratified RFHs by land size. Only RFHs with a land size no less than 550 m<sup>2</sup> for lowland rice and 100 m<sup>2</sup> for upland rice were eligible for sample selection. Figure 2 shows, for each province in Indonesia, the distribution of sample RFHs and the percentage of farmers who experienced the e ffects of CC.

**Figure 2.** The distribution of the sample and the percentage of farmers who experienced effects of climate change (CC) for each province in Indonesia.

Table 1 provides summary statistics, the expected signs of predictors, and the definition of each variable.The outcome variable is the perceived impact severity of CC, which was measured by the question: "Were you impacted by climate change? If yes, how much yield loss was caused by it?" The responses consisted of 4 choices: <25% yield loss, 26–50% yield loss, 51–75% yield loss, and 75% yield loss. The response was then converted into three categories: "no impact" for those who answered 'no' to the first question, "low impact" for those who reported less than 50% yield loss, and "high impact" for those who reported greater than 50% yield loss. The purpose of this further recoding was to clearly distinguish between low and high impact. The original impact category (consisting of 4 groups) yielded a biased estimation between higher-low and lower-low impact and between higher-high and lower-high impact. The data show that 66.2% (57,771 farmers) reported no impact, 29% (25,306 farmers) reported a low impact, and 4.9% (4253 farmers) reported a high impact.

The predictors included 17 variables, which were described in Section 3. The average age of Indonesian rice farmers is 49.5 years, having only elementary schooling (5.81 years, six years of elementary school), and most are male. The majority of farmers (70.7%) cultivate their land and have an average landholding of 0.467 ha. Of rice farmers, 90% financed their farming, leaving only 9% who financed their farm with a loan. The irrigation infrastructure covered 45.3% of the rice farmers in Indonesia, whereas the rest depended on non-technical (46.4%) and rain-fed irrigation (7.9%). Ninety-six percent of farmers only cultivated rice and only 3.9% of farmers cultivated mixed species. Chemical fertilizer is the primary input in rice farming; 91% of farmers apply it, while 8.7% of farmers do not. The average cultivation frequency is twice annually. As with the institutional factors, 52% of farmers participate in a farmer group, 25% have access to extension services, and only 11.2% have participated in a farmer field school. Finally, the climatic variables showed that 7.6% of farmers experienced a flood, 18.9% of farmers experienced a drought, and 5.8% and 1.5% reported experiencing heavy rain and other climate-induced hazards, respectively.


**Table 1.** Summary statistics, the expected signs of predictors, and the definition of variables.

Note: − and + denote decreasing and increasing effect to farmers' P-I, respectively.

#### *3.3. Empirical Model*

To estimate the effect of each predictor on the ordinal response variable, we used an ordered probit regression. Ordered probit estimate can be used to estimate how each predictor determines the probability that farmers perceive (whether high or low) an impact of CC on their rice farming. The use of ordered-probit regression to analyze the effect of independent variables on the ordinal response was favored to avoid false alarm (detecting a non-existent effect) and loss of power (failure to detect an effect) problems [83]. Equation (1) specifies the model:

$$y\_i^\* = \sum\_{i=1}^{17} \beta x\_i + \varepsilon\_i, i = 1, 2, \dots, N \tag{1}$$

where *y*∗*i* is the response variable that represents the perceived impact of CC, β is the parameter to be estimated, *xi* is the vector of predictors, ε*i* is the error term, and *N* is the number of observations.

#### **4. Result and Discussion**

#### *4.1. Estimation Results*

Of the 17 predictors analyzed in the ordered probit regression, 13 had a statistically significant effect on the perceived impact of CC, 10 of these had the expected signs, 6 variables had a positive sign, and 7 had a negative sign. However, since the climatic variables logically increase the degree of the P-I, nine variables practically represent farmer characteristics that have a statistically significant e ffect. The likelihood test ratio and the pseudo *R*<sup>2</sup> indicated that the model is robust.

The estimation results of the social variables revealed that education and gender have a statistically significant e ffect on P-I, having negative and positive e ffects, respectively. In the economic category, only land tenure had a statistically significant adverse e ffect. The technical variables seem to primarily affect the degree of P-I since all variables in this category have a statistically significant e ffect. Farmers with access to technical irrigation, practicing monoculture farming, and applying chemical fertilizer reported a lower degree of P-I. However, farmers with a higher cultivation frequency seem to have experienced a greater impact of CC. The results of institutional variables sugges<sup>t</sup> that participation in a farmer group and having access to extension services reduce the degree of P-I. However, participation in a farmer's field school is not likely to have a significant e ffect in terms of decreasing the degree of P-I.

The purpose of incorporating climatic variables in the estimation was to identify which type of climate-related hazard is perceived as causing the most severe damage. The estimation results revealed that all variables in this category have a statistically significant positive e ffect on the degree of P-I. The obtained coe fficients indicated that farmers perceive flood as causing the most damage, followed by drought, heavy rain, and other hazards. The estimation result for each variable is provided in Table 2.



**Table 2.** *Cont.*


Note: \*\*\*, \*\*, and \* denote significant at 1%, 5%, and 10% levels, respectively. ns denotes a statistically not significant effect.

## *4.2. Discussion*

The identification of factors affecting the degree of a farmer's P-I of CC was the primary purpose of this study. The estimation results in Table 2 show the effect of each variable. The variables were further classified into weakening and reinforcing factors. A reinforcing factor decreases the degree of P-I, whereas a weakening factor increases it. Figure 3 summarizes the reinforcing and weakening factors.

**Figure 3.** The reinforcing and weakening factors of the climate change perceived impact.

#### 4.2.1. Social Variables

The estimation results show that two out of three variables had a statistically significant effect. Education and gender had significantly negative and positive effects, respectively, whereas age did not have any statistically significant effect. The result suggests that farmers with higher levels of education better adapt to CC and thus perceive a lower degree of CC impact. Many studies have reported the role of education in moderating the adverse e ffects of CC. A study on farmer use of CC adaptation practices in Pakistan revealed that farmers with higher education are better at adjusting sowing time, using a drought-tolerant crop, and practicing crop-shifting, and these practices have resulted in higher food security [39]. Another study in Pakistan [67] and one in Kenya [84] that assessed farmer awareness of CC also found a positive e ffect of education on adaptation ability. Similarly, a study in Ethiopia found that education positively a ffects the probability of farmers adapting to CC via soil conservation and changing planting dates [65]. These findings show that farmers with higher education perceive a lower degree of CC impact because they are better at adapting to and have a higher awareness of CC.

The positive regression coe fficient of gender indicates that male farmers perceived a higher degree of impact than female farmers. The gender issue in the CC discussion has gained considerable attention. In some studies, male farmers were reported to be better at adapting to CC. A higher likeliness to undertake adaptation practices was found among male farmers in Ethiopia [65], and a higher awareness of CC was found among male farmers in Kenya [84]. However, in a Pakistan study, female farmers were found to be better at using adaptation practices and had a higher level of food security [39]. A cross-European analysis of individual perceptions of CC revealed that women in most European countries are more aware of CC, but the degree of awareness varied across countries [44]. Also, female representation in the national parliaments is related to the creation of stringent climate change policies and lower CO2 emissions [85]. Furthermore, the theory of socialization stated that female possesses stronger cooperation and carefulness, personal traits that are relevant to the success of CC action, than male [86].

This finding suggests that the gender e ffect is context- and location-specific. In this study, female farmers perceived a lower degree of impact. Based on the data, female farmers participate more in crop insurance than male farmers. The participation rate of female farmers in crop insurance is 0.255%, whereas that of the male farmer is 0.192%. Women farmers have a more positive perception regarding future farming conditions. Figure 4 shows the distribution of farmer perception of future farming conditions.

**Figure 4.** Perception of female and male farmers on future farming conditions.

The regression result revealed that age has no statistically significant e ffect on the degree of P-I. Age has often been associated with farming experience and increases the adoption of adaptation practices [65] and awareness of CC [67,84]. However, another study suggested that younger farmers are more likely to adopt adaptation practices [39]. A cross-European study obtained mixed results in terms of the age e ffects on individual perception of CC [44]. Thus, the results in this study and of the previous study indicate that, similar to gender, the age e ffect is location- and context-specific.

#### 4.2.2. Economics Factors

The results of this study revealed that land tenure decreases the degree of perceived climate change impact. Landholding and farm capital source did not have statistically significant effects. The effect of land tenure security on the rate of adaptation practices varies. A study in Ghana reported that land ownership increases a farmer's likeliness to undertake adaptation practices [87]. Conversely, another study in Pakistan found that land ownership reduces it [88]. This study suggests that other characteristics, such as education, household size, and resource access, might influence the identified adverse effect of land ownership. Similarly, a study in Pakistan found an adverse effect of land ownership, even after controlling for land size and wealth-related characteristics, but not for education, which was found to have a strong positive effect [39]. These findings sugges<sup>t</sup> that a stronger variable, such as education, might offset the effect of land ownership. The results of this study indicate that land ownership decreases the degree of P-I of CC.

Since we analyzed the P-I of CC on crop yield, the result indicates that farm owners are more willing to adopt yield-enhancing adaptation strategies than farm tenants and sharecroppers. Farm owners have a higher incentive to invest in adaptation practices [89]. A study in Pakistan and Indonesia found that insecure land tenure decreases adaptation practices [90] and makes farmers less cautious in terms of their farming, manifesting, for example, through an absence of soil conservation [91]. The positive effect of secure land tenure on adaptation practices, however, has potential disadvantages when the adaptation yielded unintended maladaptive outcomes. For example, a study on tomato growers in Ghana revealed that the perception of climate variability drives farmer to use more agrochemicals to retain crop production, and these chemicals cause the pollution of nearby water bodies and increase soil acidity above the optimum crop requirements [18]. Similar practices have been identified in rural areas of Indonesia, where watermelon farmers use excessive chemical pesticides to retain production, causing groundwater pollution that damages the quality of water for consumption [92].<sup>1</sup> Ultimately, adaptation becomes maladaptation, which erodes sustainable development and shifts vulnerability to other actors. Thus, it is essential to guide farmer's adaptation practices to avoid unintended maladaptive outcomes.

#### 4.2.3. Technical Factors

The probit estimation revealed that each variable in the technical category has a statistically significant effect. The regression coefficient of the irrigation variable indicated that farmers with access to technical irrigation perceive a lower degree of impact than farmers with non-technical and rain-fed irrigation. Irrigation infrastructure is crucial for reducing the adverse effects of CC. The availability of irrigation infrastructure (II) increases the efficiency of the distribution of limited water and decreases crop yield loss due to drought. In some cases, II becomes a drainage facility that reduces crop damage caused by flood. The establishment and improvement of II is a key framing device for CC adaptation in a rice-based country such as Vietnam [94]. In Vietnam, II reduces water availability during the rainy season and increases it during the dry season [95].

Technical irrigation in Indonesia has been developed since the 1970s with the support of the Green Revolution program. The establishment of technical irrigation infrastructure has increased the yields, cropping season, and cropping intensity of rice farming in Indonesia [73]. Conversely, both non-technical and rain fed agriculture rely mainly on rainfall to supply irrigation. Hence, those in such areas are vulnerable to drought occurrence and pay a higher cost of irrigation due to water scarcity. Rice and agricultural productivity in this area are lower compared to areas with technical irrigation, which means that farmers such areas are economically more vulnerable to CC.

The fertilizer variable demonstrated that the application of chemical fertilizer decreases the degree of P-I. In line with the previous discussion, farmers often apply more agrochemicals to retain their crop

<sup>1</sup> The example of data on extensive use of agrochemical in Indonesian farming can be found here [93].

production. A study in Nepal identified that farmers adapt to CC by applying more chemical fertilizer and this increases rice productivity [9]. This finding suggests that autonomous adaptation will likely become maladaptive if a broader perspective is included. This is not surprising as the primary goal of farmers is to increase crop yield or limit crop yield loss, and, in most cases, this is also the priority of agricultural policy. Thus, a further adaptation policy should guide farmers to implement appropriate adaptation practices to reduce unintended maladaptive outcomes.

The results also indicate that monoculture farming decreases the degree of perceived CC impact. Monoculture is often regarded as a threat to biodiversity. Thus, multiculture farming is commonly employed to reduce biodiversity loss in trees farming. However, monoculture in rice farming is less of a threat to biodiversity since its temporal dimension is short [96], and the majority of farmers practice crop rotation. The last variable, cultivation frequency, increases the degree of P-I. A higher cultivation frequency prolongs exposure to CC, which increases the probability of farmers being affected by CC and increases the degree of P-I.

#### 4.2.4. Institutional Factors

The estimation results show that access to extension services and participation in a farmer group significantly decrease the degree of P-I, whereas participation in a farmer's field school does not. Extension services are effective information channels used to raise farmer awareness of CC and drive adaptation practices. The importance of extension services as a means of delivering accurate information about CC has been addressed in many studies. A study of farm households in four provinces in Pakistan demonstrated that access to extension services increases how likely a farmer is to adapt to CC [39]. Similarly, a study on rice farmers in the Terai and Hill area of Pakistan indicated that farmers who receive information from extension agents are more likely to adapt [9]. Extension services have been the focus of studies in East Africa [97], Ethiopia [65], and Punjab, Pakistan [67], to enhance a farmer's resilience against CC. These finding indicate that information is crucial in shaping and directing a farmer's behavior with respect to CC. Access to timely and accurate information not only increases a farmer's awareness and adaptation to CC, but might also be used to inform farmers about adaptation practices that yield maladaptive outcomes.

Indonesian farmers receive extension services from several sources, such as state agricultural extension officers (PPL), a state pest-control officer (POPT), an officer from the office of agriculture at the district level (Diperta), and private extension services. Figure 5 shows the number of farmers who receive extension services based on the type of extension service officer. Only 25% of rice farmers in Indonesia have access to these services. The primary extension agen<sup>t</sup> is from the government. The data shows that non-governmen<sup>t</sup> extension agents vary in type, and these may be from a private corporation (such as a pesticide and fertilizer factory), a non-governmental organization (NGO), or may be associated with peer-farmer extension. Increasing the number and coverage of extension agents is essential for CC adaptation policy in Indonesia, since the majority of farmers (41%) receiving extension services are concentrated in Java, with 21.4% in Sumatera, 8.3% in Bali and Nusa Tenggara, 10.6% in Kalimantan, 14.4% in Sulawesi, and 1.2% and 1.9% in Maluku and Papua, respectively.

**Figure 5.** The number of farmers who receive extension services based on the type of extension agents.

The governmen<sup>t</sup> should encourage the establishment of farmer-to-farmer extension services given the potential human resource in Indonesian rice farming. Currently, 82% of the rice farmer population is 20–60 years of age, and 2% of rice farmers have a bachelor degree or higher. Of this educated group (bachelor degree or higher), 87% are aged between 20 and 60 years, with 25% in 20–40 years, and 61% in the 41–60 years old group. Facilitating the establishment of farmer-to-farmer extension provides a strong foundation for sustainable agricultural extension. Farmer-to-farmer extension provides a cost-effective training method for farmers, since trained farmers disseminate their useful knowledge to the non-trained farmer, and the non-trained farmer is eager to adopt the technology used by the trained farmer to increase yield and profit [98]. A strong positive effect of formal education on a farmer's resilience with respect to CC suggests that farmer-to-farmer extension provided by an educated farmer potentially increases the resilience of the recipient.

The estimation results indicate that participation in a farmer group decreases the degree of P-I. Participation in a farmer group is the primary requisite for obtaining farm inputs, extensions, and other programs provided by the governmen<sup>t</sup> [79]. Several studies have reported the importance of farmer groups in increasing farmer adaptation to CC [65,97], awareness to CC [67], and farm productivity and food security [39].

#### 4.2.5. Climatic and Regional Factors

This category focuses on identifying the type of impact that a farmer perceives as most severely damaging their farming. The category contains four variables: Flood, drought, heavy rain, and other hazards (such as landslides). The estimation results show that farmers perceive floods as causing the most severe impact, followed by drought, heavy rain, and other hazards. Flood and drought are the primary consequences of changes in rainfall frequency and intensity. The occurrences of drought and flood are the cause of agricultural yield loss in developing countries and generally affect large areas [99,100]. Figure 6 shows the distribution of farmers who have experienced flood and drought events in Indonesia. The distribution shows that more farmers have experienced droughts than they have floods. Even in provinces with high rainfall intensity, droughts are more frequent than floods. The spatial distribution of affected farmers shows that farmers outside Java are more vulnerable to CC. This information is vital in the implementation of adaptation policies. The governmen<sup>t</sup> should prioritize the increase in the climate resiliency of farmers outside of Java Island. The increase in resiliency can be achieved by expanding the coverage of agricultural extension services to farmers.

**Figure 6.** The distribution of farmers experiencing flood and drought in each province of Indonesia (the rainfall intensity unit is mm/year).

The regional variables demonstrate regions whose farmers perceive a higher degree of CC. The estimation results indicate that farmers in Sumatera are likely to perceive the impact of CC. Twenty-seven percent of Indonesian rice farmers are in Sumatera, so a large impact of CC on rice farmers in Sumatera is perceived, since the national rice supply significantly decreased. Thirty-eight percent of Indonesian rice farmers in Java perceive a low impact of CC. The agricultural infrastructure in Java is more advanced than other regions in Indonesia. Physical and institutional infrastructure have been established in Java since the 1970s. Thus, in the future, it is critical to emphasize agricultural regions outside Java. The order of regions from the most vulnerable to the most resilient is as follows: Sumatera, Bali and Nusa Tenggara, Maluku and Papua, Sulawesi, and Kalimantan.

#### 4.2.6. Policy Implications

The primary purpose of an agricultural CC adaptation policy is to reverse its adverse effects. Adaptation to CC limits agricultural yield loss and improves food security. In some cases, CC is beneficial to agricultural production if proper adaptation practices are implemented [101,102]. Whereas the majority of countries have formulated a specific adaptation strategy to be implemented at the farm level, the remaining challenge is delivering the strategy so that it is adopted by many farmers. The theory of risk perception suggests that individuals who perceive a higher degree of risk would be willing to take any action required to remove the risk [27–29]. In the case of the CC impact on agriculture, farmers will be more willing to adopt adaptation strategy if they individually perceive CC damage. However, this high likeliness to adapt becomes a problem if a farmer's autonomous adaptation creates maladaptive outcomes. Thus, identifying the determinants of a farmer's P-I is crucial in the implementation of adaptation policy. This information can be used to increase the rate of adaptation practices and limit maladaptation.

The results of this study show that the P-I of CC is affected by various factors. Female farmers are shown to be more resilience in managing the impact of CC than male farmers. The theory of socialization suggests that female possesses a stronger personal traits that are relevant to CC action (mitigation and adaptation) than male. Hence, increasing female farmer participation in adaptation activity and in the CC-related decision-making process is crucial to increase farmer resilience toward CC. Formal education and access to extension services have a strong e ffect in terms of reducing the degree of P-I, and this strongly agrees with previous studies in many developing countries. Therefore, information is crucial to increasing a farmer's resilience with respect to CC. Thus, we sugges<sup>t</sup> strengthening information about CC as a primary policy. Information about CC can be strengthened in two aspects: First, by increasing the coverage of current extension services and, second, by establishing farmer-to-farmer extension, where educated farmers (those with a bachelor degree or higher) provide education to their peers. Among other factors, land tenure is crucial in adaptation policy. Findings in this study sugges<sup>t</sup> that the security of tenure drives farmers to provide the investment and pay for the running costs of adaptation. Economically, the security of tenure ensures that farmers are incentivized to adopt of adaptation practices.

