Validating a Tailored Disaster Risk Assessment Methodology: Drought Risk Assessment in Local PNG Regions

10 Climate change is increasing the frequency and intensity of natural hazards, causing adverse impacts on vulnerable communities. Pacific Small Island Developing States (SIDS) are of particular concern, requiring resilient disaster risk management consisting of two key elements: proactivity and suitability. User-centred Integrated Early Warning Systems (IEWSs) can inform resilient risk management. However, an EWS is only effectively integrated when all components are functioning adequately. In Pacific SIDS, the risk knowledge component of an I-EWS is underexplored. Risk knowledge is 15 improved through efficient risk assessment. A case study assessing drought risk in PNG provinces was conducted to demonstrate the development and validate the application of a tailored risk assessment methodology. Hazard, vulnerability, and exposure indicators appropriate for monitoring drought in PNG provinces were selected. Risk indices for past years (2014-2020) were calculated and mapped in Geographic Information Systems (GIS). Risk assessment results were validated with a literature investigation of sources presenting information on previous droughts in PNG. The risk assessment indicated 20 a strong drought event in 2015-2016, and a moderate event in 2019-2020. The literature corroborated this, confirming the validity of the risk assessment methodology. The methodology and results can be used to inform improved disaster risk management in PNG, by advising decision-makers of their risk and policymakers on which provinces are of priority for resource allocation. The methodology can also be used to enhance the risk knowledge component of a user-centred I-EWS and guide the implementation of such a system for drought in PNG and other Pacific SIDS. 25

be improved. The drought risk index developed as part of the drought risk assessment by Rahmati et al. (2020) also had implications for utilising integrated Geographic Information System (GIS)-based mapping techniques to accurately map and 95 visualise drought risk levels of particular places to better inform drought relief preparedness strategies in those areas.
Integrated GIS-based mapping techniques for risk assessment include three key components: data integration into GIS, risk assessment tasks, and consideration of risk decision-making (Chen et al., 2003). The first component, data integration into GIS, consists of data collection and assimilation onto a GIS platform and data transformation and standardisation. Risk 100 assessment tasks are then performed on the GIS platform, including individual hazard, vulnerability, and exposure assessments with accompanying mathematic calculations (Hagenlocher et al., 2019). The consideration of risk decisionmaking is incorporated through efficient data visualization on GIS risk maps and appropriate dissemination of such products to decision-makers.

105
Although disaster risk assessments have been conducted for a variety of natural hazards in numerous countries throughout the world, there has been minimal risk assessment conducted for natural hazards in SIDSs. Out of the disaster risk assessments that have been conducted in SIDS, they have been conducted on a broader level rather than local area or community level (Hagenlocher et al., 2019). For risk assessments to effectively inform proactive and suitable disaster risk management in local areas and vulnerable communities, they must be tailored to the area of study (e.g. the specific country, 110 states or provinces, and/or local communities being investigated) (Wilhelmi and Wilhite, 2002). Tailored risk assessments would use specific hazard, vulnerability, and exposure indicators appropriate for monitoring hazard risk of the hazard under investigation, in the study area. Additionally, those risk assessments that have been conducted in SIDS have not utilised the integrated GIS methodology recommended by Rahmati et al. (2020), Hagenlocher et al. (2019), and Chen et al. (2003). Therefore, there is room for future investigation of risk knowledge in SIDSs to implement a tailored risk assessment with 115 specific hazard, vulnerability and exposure indicators, and map indices produced by such assessment using integrated GIS methodology.

Validating disaster risk assessments to ensure accuracy and usability of results
In addition to past disaster risk assessments not utilising the most efficient methodology, they also commonly lack adequate validation (Asare-Kyei et al., 2017). In a review of past disaster risk assessment methodology, Hagenlocher et al. (2019) 120 state that comprehensive validation "has proven to provide relevant information on the reliability, validity, and methodological robustness of risk assessments and their outcomes. However, its application in the field of risk assessment remains largely underdeveloped." Molinari et al. (2019) explain further that risk assessment validation is crucial; results can be used to inform large investments and allocation of resources, as well as other important risk management decisions, so results need to be credible. Among the few studies seeking to validate a risk assessment methodology, various validation 125 techniques have emerged. Validation through result comparison with historical data has been used in several studies, however the preciseness of this method has been criticised. To validate the agricultural drought risk assessment methodology which they developed for use in Nebraska (U.S), Wu and Wilhite (2004) estimated the probability of correct risk classification with independent, historical 130 crop data. This historical data was then compared to the risk assessment results to verify accuracy. Similarly, Fekete (2019) validated the results of a flood vulnerability assessment through comparison with social data from the time period assessed.
However, Fekete (2019) explains that the absence of globally accepted benchmarks for social, exposure and hazard data explicitly focused on revealing disaster risk, leaves too much to author interpretation when using this validation method. Molinari et al. (2019) also critics the validation through comparison with historical data technique, stating that there is "the 135 need of higher quality data to perform validation and of benchmark solutions to be followed in different contexts, along with a greater involvement of end-users".
An alternative technique, incorporating the views of end-users as a 'ground-truth' source, called participatory research is becoming increasingly utilised to validate drought monitoring outcomes, including risk assessment results. This technique 140 includes collaboration with stakeholders in a capacity building process as well as consideration of local peoples and expert observations into knowledge systems (Mckenna and Yakam, 2021;Fragaszy et al., 2020). For example, Fragaszy et al. (2020) used participatory validation by conducting interviews, focus groups and workshops to assess the extent of drought impacts experienced during the study period, to verify the results of a drought assessment conducted in the Middle East and North Africa. 145 Although participatory research is a promising validation methodology, past investigations using this method have used an additional 'ground-truth' source to strengthen validation adequacy. To verify results of remotely sensed drought risk monitoring in Morocco, Bijaber (2018) compared results to historical on the ground precipitation and crop production data at national scale as well as the views of experts regarding what was experienced on the ground during the investigated period. 150 Asare-Kyei et al. (2017) employed an analogous technique to validate flood risk assessment results for the urban area of Shanwei City in People's Republic of China. Records of impacts and results of household interviews were intended to be used as ground-truth sources for impact data which the risk assessment results could be compared to for verification. However, Asare-Kyei et al. (2017) found no systematically documented records of the impacts, and thus had to rely on local's recounts which were focused on the high intensity impacts, and often forgetful of small impacts. 155 In Pacific SIDS, data availability is scare, thus validation through comparison with historical independent data is unlikely to be credible. Overall, a strengthened validation methodology using multiple ground-truth sources seems most promising for future study regarding the verification of disaster risk assessments.

Disaster risk assessment for PNG 160
To continue upon past research regarding integrated GIS-based risk mapping (Rahmati et al., 2020) and I-EWS development (Bhardwaj et al., 2021a), PNG is deemed an appropriate country in which to investigate the risk knowledge component of an I-EWS through disaster risk assessment and mapping. PNG is one of Pacific SIDS, it is vulnerable to climate extremes and disaster events and is predicted to be increasingly affected by impacts from tropical cyclones, floods, and drought in the future. The El Niño Southern Oscillation (ENSO) and the Indian Ocean Dipole (IOD) are key drivers of climate variability 165 and resultant hazard events in PNG, and other Pacific SIDS.
In Pacific SIDS, ENSO alters the distribution of precipitation, often causing natural hazard events (Horton et al., 2021).
ENSO has two key phases: El Niño (warm phase of ENSO) and La Niña (cold phase of ENSO). La Niña-associated prolonged rainfall has contributed to floods, whilst El Niño-associated prolonged aridity has contributed to droughts in PNG 170 (Smith et al., 2013). Historically, the 1997-1998 El Niño contributed to severe drought in PNG causing immense loss of life, destruction of crops, and forest fires subsequently causing regional pollution problems (Nicholls, 2001). However, different regions of PNG experience varying climactic affects from El Niño and La Niña (Figure 1). For example, a moderate La Niña event which occurred in PNG during 2011-2012 resulted in drought conditions in several PNG provinces, particularly Milne Bay Province. 175 The effects of ENSO can be influenced by the IOD to further weaken or strengthen these trends in rainfall variability (Bhardwaj et al., 2021b). Defined as consistent changes in sea surface temperature variability across the tropical western and eastern Indian Ocean, the IOD can be negative, positive, or neutral, with each phase interacting with ENSO impacts differently (Bhardwaj et al., 2021b). The impacts of interactive IOD and ENSO phases experienced in PNG are shown in 180 PNG has a lack of coping capacity for managing the risks posed by the natural hazard events which occur across the country (Kuleshov et al., 2020). Particularly, drought poses an immense concern as it historically has disastrous impacts on PNG communities but has not been extensively investigated compared to other hazards like tropical cyclones and floods. 185 Considering the restricted knowledge of drought risk in the context of PNG, and the critical threat which it poses to communities, drought is an appropriate hazard to investigate in terms of assessing disaster risk to local areas in PNG.
Generally, drought can be described as an extended dry period resulting from rainfall deficiency. However, drought has many definitions for its various types: meteorological (when climactic factors result in dry conditions within an area), 190 hydrological (when water shortages occur after a period of meteorological drought), agricultural (when agricultural productivity is inhibited and crops are affected by meteorological and hydrological drought), and socioeconomic (when dry conditions restrict the supply and demand of commodities) (Wilhite et al., 2014). As drought impacts all major sectors (agriculture, economy, social, health, etc.), a drought risk assessment must not only use indicators tailored for monitoring drought in PNG, but also use a variety of sectoral indicators to encompass the overall drought risk to a local area. Remote 195 communities in PNG continue to have limited resources and capacity to effectively manage such a variety of sectoral impacts. Local areas, for example individual provinces, in PNG must be able to self-initiate strategies that are effective and appropriate to them. In this context, I-EWS and risk assessment informed community/provincial-scale DRR is an increasingly important focus for PNG (Webb, 2020).

200
This study will expand on previous research with an aim to address the risk knowledge components of a user-centred I-EWS for informing bottom-up resilient management on the local area scale in PNG. This research seeks to build capacity for the natural hazard monitoring through an I-EWS in PNG, as well as demonstrate the potential for tailored risk assessments to accurately inform on drought risk levels before, during and after a hazard event and thus contribute to more resilient disaster risk management in local SIDS areas. The study intends to use the most effective risk assessment methodology 205 recommended by Rahmati et al. (2020)  -Southern Region: Central, Gulf, Milne Bay, Oro (Northern), and Western (Fly River). 220 PNG is largely mountainous, and much of it is covered with tropical rainforest. The climate of PNG can be described as tropical throughout, however each region of PNG experiences differences in seasonal climactic factors ( Figure 3) (Bhardwaj et al., 2021b). PNG climate also varies between years, with a dominant driver being ENSO (Figure 1). PNG society consists of traditional village-based life, dependent on subsistence and small cash-crop agriculture, as well as modern urban life in the main cities. 225 Economic performance in PNG has historically been based on international prices for exports (including for agriculture), fiscal policies and construction activity. As of 2015, over 2 million Papua New Guineans are poor and/or face hardship, particularly those based in rural areas (Pacific Islands Forum Secretariat, 2015). Agricultural occupation is consistently important for local livelihoods, with approximately 80-85% of the rural population directly deriving their livelihood from farming (Pacific Islands Forum Secretariat, 2015). 230

Study Design
The methodology for this study was three-part: 1. Selection of tailored hazard, vulnerability and exposure indicators appropriate for monitoring drought risk in PNG provinces.
2. Calculation and GIS mapping of hazard, vulnerability, exposure, and risk indices for historical years (2014-2020) to 235 determine the occurrence of drought events in PNG in the past.
3. Validation of drought risk assessment accuracy through a comparison of the drought risk index results with literature detailing severity of drought conditions and impacts experienced on the ground at the time of each drought event indicated by the risk assessment.

Methodology: Part 1 240
Tailored risk indicators were selected for monitoring drought in PNG as the development of a region-specific drought risk index is the key to accurate drought risk calculation and mapping (Santos et al., 2014). All types of droughts were considered when selecting indicators, as well as all major sectors across PNG provinces. This was done to provide a holistic risk index for PNG provinces, as each type of drought is known to impact PNG communities (Kuleshov et al., 2020), with each major sector experiencing the effects (Bhardwaj et al., 2021b). 245 Hazard, vulnerability, and exposure indicators most applicable to drought risk assessment in the 22 provinces of PNG were determined by integrating information regarding the socio-economic, geographic, and climactic characteristics of PNG provinces and analysis of indicator selection used in earlier studies of characteristically similar areas (Refer to Appendix A for a detailed table describing the reasons for selection of each indicator). PNG National Weather Service advice was also 250 sought to approve indicator selection. Additionally, hazard indicators were assessed against recommendations made by WMO in their Handbook of Drought Indicators and Indices (Svoboda and Fuchs, 2016). Note, data was only available for certain indicators as data availability is poor in PNG, thus indicators which could have been more appropriate for use in hindsight had to be omitted. The most applicable and representative indicators were selected from what was available.

255
The indicators that were selected for use in assessing PNG provincial drought risk are shown in Table 1. Refer to Appendix B for a list of indicator data sources. It is key to note that space-based monitoring products were used when gathering data for hazard index calculations to ensure accuracy 1 .

Methodology: Part 2
Historical and current data detailing hazard, vulnerability, and exposure conditions in each of the 22 PNG provinces for each 260 year within the 2014-2020 period in PNG, was used to develop a risk index for each year in this historic period to see if it would have indicated high disaster risk and whether it is suspected that a drought event(s) actually occurred during this period. The historical assessment was conducted from 2014 onwards due to no data availability for space-based Vegetation Health Index (VHI) before 2014. Integrated-GIS methodology for mapping risk in each study region was used to display risk levels for the overall years 2014-2020. It was then determined whether a drought event was suspected as occurring across 265 PNG in each of the years assessed.
A nationwide drought event was suspected when the majority of provinces were in severe to extreme drought risk conditions and was not suspected when the majority of provinces were in mild to moderate drought risk conditions. This is deemed a fair assumption since in past drought events, when only certain provinces in PNG experienced drought conditions and direct impacts, other provinces encountered indirect impacts and PNG as a nation was adversely affected. For example, during the 270 1997-1998 nationwide drought event in PNG, dire social, health and economic effects were felt across the entire country (Kanua et al., 2016). Resources of provinces in non-dry conditions were pressured with PNG villagers from drought-affected provinces travelling to areas in non-drought conditions or to relatives living in urban areas seeking familial help and support (Allen and Bourke, 2009). Additionally, a major mine was closed in response to the dry conditions in Western Province, impacting the national economy (Kanua et al., 2016). 275 The years suspected of experiencing a nationwide drought event were recorded; this record was used in the validation of risk assessment results against literature review results. Risk levels were also determined for the months of November, and December in 2014, January to December of 2015 and November and December in 2016 to demonstrate the transition into and out of drought during any strong drought event indicated by the risk assessment. 280 Thresholds were applied prior to index calculations and mapping to determine the variance of indicator data between each of the PNG provinces. Thresholds suitable for PNG drought risk indices were adapted from earlier studies in similar areas to ensure accuracy (Dayal et al., 2018;Frischen et al., 2020b). Once indicator variance was confirmed, raw data was uploaded to ArcGIS Pro. 285 To calculate the hazard index, data was first reclassified by a linear function on a 1-10 scale and then standardised using fuzzy logic in ArcGIS Pro (Environmental Systems Research Institute (Esri) Inc., 2019). Data for the vulnerability and exposure indices was also standardised using fuzzy logic. Prior to the performance of the fuzzy function, fuzzy membership classes were assigned to each indicator, describing the relationship between it and drought risk as recommended in Rahmati 290 et al. (2020) and Aitkenhead et al. (2021). Two classes of fuzzy membership were assigned in this study: fuzzy small 2 and fuzzy large 3 . Fuzzy values scaled between 0-1 based on the possibility of the indicator data contributing to drought risk, where 0 was assigned to values unlikely to contribute to drought risk, and 1 was assigned to values most likely to contribute.
The default midpoint was not used when performing the fuzzy function; the midpoint used for each indicator was based on the mean value in the historical records for indicator data. This ensured that the data was standardised on both a spatial and 295 temporal scale.
The indicator fuzzy values for each province were mapped in ArcGIS Pro 4 . Fuzzy values of each indicator were used to calculate hazard, vulnerability, and exposure indices. Numerical weights were assigned to each indicator contributing to the hazard, vulnerability and exposure indices based on an expert weighting scheme informed by past studies and advice from 300 the PNG National Weather Service (Appendix C). The hazard, vulnerability and exposure indices were calculated using equations (1), (2) and (3), respectively for each province in the years and months under investigation. (1), (2), where HI is the Hazard Index, VI is the Vulnerability Index, EI is the Exposure Index, n is the number of Hazard, Vulnerability or Exposure Indicators, x i ′ refers to the standardised indicators and w i refers to the respective indicator weight.
Once the vulnerability, hazard and exposure indices were calculated, spatial maps of the area covering the 22 provinces of PNG, representing vulnerability, exposure, and hazard per unit area, were produced. The final drought risk index value for 310 each PNG province was determined through the integration of the drought vulnerability, hazard and exposure index maps using the Fuzzy Gamma Overlay function (using a gamma of 0.75) in ArcGIS Pro. A final drought risk map was then generated. The extent of drought vulnerability, hazard, exposure, and risk displayed on the respective maps was classified into four levels: mild, moderate, severe, and extreme. These classifications are commonly used in drought risk assessments (Dayal et al., 2018;Frischen et al., 2020a). This process was repeated to calculate a drought risk index for each year and 315 month under investigation.

Methodology: Part 3
Risk level accuracy was validated through comparison with documented records of observed impacts during the study period as a ground-truth source. Eight reputable literature sources detailing drought conditions around the time of each event indicated by the risk assessment (2015-2016 and 2019-2020) were analysed to determine the ground-truth of the drought 320 event severity and impact. As two drought events were investigated, and eight sources were assessed for each event, a total of 16 sources were assessed overall ( (Chua et al., 2020Gwatirisa et al., 2017;Burivalova et al., 2018;Jacka, 2020;Varotsos et al., 2018;Kuleshov et al., 2020;Schmidt et al., 2021;Rimes and  2020; Bang and Crimp, 2019)). Three severity levels were identified as being commonly implied in sources: mild, moderate, and severe to extreme. The level most clearly aligned with the details provided by each source was recorded. Additionally, any mention of specific provinces experiencing impacts was recorded.
The records in the literature were not extensive for the 2019-2020 drought event in PNG. An array of records was available 330 for the 2016-2020 drought event, but only a few were available for the 2019-2020 event. This may have been due to the 2019-2020 event being so recent, meaning that investigations of the event may still be ongoing and/or peer reviewed literature not being published as of when this research was conducted. To account for the limited availability of literature records for the 2019-2020 drought and to make the comparison with literature equal for both drought events assessed, an equal number of sources were selected for the analysis for each event (eight each). The small number of sources investigated 335 for each drought event was statistically analyzed; a two-tailed p-value was used to determine significance in the statistical tests as a two-tailed p-value accounts for smaller sample sizes and tests for the possibility of positive or negative differences in the samples.
Statistical analyses were conducted to determine if there were significant differences between the drought risk level indicated 340 by the risk assessment and the risk level indicated by the literature for each PNG province for each of the drought years under investigation (2015-16 and 2019-20) 5 . An F-test was firstly conducted to determine whether there were equal variances between the levels displayed in the risk assessment and the levels expressed in the literature for the 2015-2016 drought event. The F-value (test statistic), degrees of freedom and the two-tailed p-value indicating the level of marginal significance within the test, were recorded. A Student's t-test (assuming equal or unequal variances depending on F-test 345 results) was then conducted to determine the significance of difference of the drought risk levels indicated by the assessment and the levels indicated in literature for each province. The t-value (test statistic), degrees of freedom and the two-tailed pvalue were recorded. This process was repeated for the 2019-2020 drought event results. T-test assumptions were checked by plotting the data distribution on boxplots. All assumptions were met, thus the aforementioned tests proceeded. All statistical tests used α = 0.05. 350

Results
The 2014, 2015 and 2016 drought risk assessments determined that the majority of provinces had severe or extreme drought risk levels (Table 2), thus a drought event is suspected as occurring or commencing across the country during these years.
The 2017 and 2018 drought risk assessments indicated most provinces as having mild or moderate drought risk levels (Table   2), thus a drought event is not suspected, and these were likely non-drought years. In the 2019 and 2020 drought risk 355 assessments, slightly more provinces displayed a severe or extreme level than a mild or moderate drought risk levels (Table   2), therefore a drought event is suspected as occurring or commencing in this period.  (Table 3), whilst 2017 and 2018 were reported as non-drought years (Kuleshov et al., 2020).
In all but one source, 2014 was reported as a non-drought year. This is consistent with the drought risk assessment results, with 2014 being the exception as it was suspected as a drought year from the risk assessment results and was only mentioned 365 as a drought year in one of the literature sources investigated (Burivalova et al., 2018). Refer to Figure 4 for the mapped hazard, vulnerability, exposure, and risk results for 2014.
The 2014 anomaly was further investigated by the production of monthly drought risk maps throughout the year which were used to determine how the risk assessment was performing throughout the year. Results show drought conditions 370 commencing or occurring in March-July and again in November-December, with the risk levels in November and December being slightly more intense than those expressed in March-July (Table 4).
No statistically significant variation was displayed between the severity levels described in the risk assessment versus the literature for the 2015-2016 event (F 18 =0.86, p=0.37) and the 2019-2020 event (F 17 =0.71, p=0.25). There was no significant 375 difference between the severity levels recorded for the 22 PNG provinces given by the risk assessment compared to the literature for both the 2015-2016 drought event (t 36 =-1.70, p=0.10) and the 2019-2020 drought event (t 34 =1.51, p=0.14).
Refer to Table 5 (Table 5) Table 6 shows the heightening of drought risk from November 2014 to July 2015 for most provinces, with drought risk levels peaking in October-December 2015 and then slightly reducing at the commencement of 2016.

PNG drought events indicated by risk assessment 400
The drought risk assessment methodology used in this study was validated through a historical risk assessment paired with a literature review. As expected, the drought risk assessment identified a suspected drought event occurring or commencing in 2015-2016 as well as in 2019-2020; literature confirmed the occurrence of these suspected drought events in PNG.
It is widely reported that a strong drought event commenced in PNG at the beginning of 2015 and reached its peak during 405 2016 (Kuleshov et al., 2020;Chua et al., 2020;Gwatirisa et al., 2017;Jacka, 2020;Varotsos et al., 2018;Rimes and Papua New Guinea National Weather Service, 2017). Kuleshov et al. (2020) attributed the drought of 2015-2016 to a strong El Niño which occurred during these years. This strong El Niño phase was paired with a positive IOD phase; the interacting impacts of both climate drivers resulted in devastating negative rainfall anomalies across the entirety of PNG (Bhardwaj et al., 2021b). It is explained in the literature that the 2015-2016 drought event affected approximately 40% of PNG's 410 population, with drought-caused food shortages impacting half a million people throughout PNG's provinces (Kuleshov et al., 2020). The results also provided evidence as to which specific provinces were most at risk during each drought period. Central, West Sepik, Northern and Gulf Province were indicated by the risk assessment to be among the five most at-risk provinces for both the 2015-2016 and the 2019-2020 drought periods. This suggests that these four provinces are consistently at highrisk to drought compared to other PNG provinces, likely to persist in the future, and therefore should be of focus for 435 improved management resilience in the future. However, slight discrepancies were observed when the 2015-2016 period results were compared with literature findings, which challenges the validity of this conclusion.

Comparison to Literature Findings
The 2015-2016 drought event is consistently described in the literature as having extreme impact on local communities in 440 each PNG province. A poverty analysis in the lowlands of PNG conducted by Schmidt et al. (2021) stated that the severe El Niño event of 2015-2016 decimated a critical amount of PNG's local crop production which left PNG communities in a food crisis. A detailed survey found that such a climate shock had critical consequences for household welfare, contributing to a rise in households below the poverty line, particularly in rural and lowland areas (Schmidt et al., 2021) In comparison, the impacts of the 2019-2020 drought event are primarily discussed as moderate rather than severe or 455 extreme. However, the effects of the 2019-2020 drought event have not been widely discussed in peer-reviewed literature as it is such a recent event, but there are some sources that have similarly investigated drought conditions in PNG and the resulting impacts during 2019-2020. These sources have described the negative affect of dry conditions on agricultural production and food security (Food and Agriculture Organisation of the United Nations, 2021; Food Security Cluster et al., 2021). Areas mentioned as being of concern include the Gulf and Western Area, along with northern provinces and southern 460 coastal provinces; this is consistent with the risk assessment results. The moderate rather than extreme drought impacts on the agriculture sector, as a result of the 2019-2020 drought event, may be due to soil moisture levels being relatively well maintained across PNG during this time (2019).
There were no irregularities with what was reported by the risk assessment and the literature regarding the most at-risk 465 provinces for the 2019-2020 event, which suggests a high level of accuracy within the risk assessment results for 2019-2020.
Whereas, when comparing risk levels indicated for specific provinces, slight discrepancies were detected for the 2015-2016 drought event results. Central and Gulf Province were indicated among the five most at-risk provinces by the risk assessment but were included in the most at-risk provinces described by the literature. This might have been because the majority (five out of eight) of the 'ground-truth' sources used to investigate the impacts of the 2015-2016 drought event focused on only 470 one aspect of drought (meteorological, agricultural, hydrological, or socioeconomic), and thus did not consider the holistic impacts suffered by specific provinces like Central and Gulf Province (Chua et al., 2020;Burivalova et al., 2018;Varotsos et al., 2018;Schmidt et al., 2021;Gwatirisa et al., 2017). Comparatively, the risk assessment methodology of this study incorporated indicators for all types of drought's impacts to provide a comprehensive risk level for each province. It is not likely that discrepancy negates the overall validity of the risk assessment methodology as it is only slight, with all other 475 results proving the methodology to be accurate; further research should be conducted with a stronger 'ground-truth' comparison using first-hand local and expert perspectives (gathered through interviews) rather than what was recorded in the literature to verify.

The anomalous year of 2014
There was one discrepancy in the risk assessment results for 2014. The drought risk assessment indicated that it was a 480 moderate drought year, whereas most literature describe it as a non-drought year, with only one source including it as a year in the 2015-2016 drought event (Burivalova et al., 2018). Upon further consideration, it is not illogical that 2014 was indicated as a year in which a drought was commencing or occurring by the risk assessment. The risk assessment may have indicated 2014 to be a drought year as it was leading up to the extreme drought risk levels during the 2015-2016 drought event, and therefore may have reflected the strong risk which the following drought years posed. As the risk index provides 485 information on not only the hazard conditions at the time investigated, but also the vulnerability and exposure conditions of the area investigated, it may be able to give some indication on the chance of drought occurring within the investigated area in the future.
The monthly risk assessment conducted for all months during 2014 indicated two periods in which drought was suspected as 490 commencing or occurring, in March-July and November-December. In most PNG provinces, seasonal rainfall usually peaks between December-April with drier conditions commonly following in July-August (Regional Bureau for Asia & the Pacfic and Food Security Markets and Vulnerability Analysis Unit, 2015). Thus, the dry conditions indicated during March-July may have been due to normal seasonal rainfall patterns which usually cause drier conditions around July across PNG provinces. The November-December period is more of an anomaly as it is not consistent with the normal seasonal patterns of 495 PNG, which has rainfall peaking around December. However, this may be explained by the commencement of the strong El Niño event which then heightened into a widely reported drought event during 2015-2016. Reports of below-average rainfall were recorded as early as October 2014, for the 2015-2016 El Niño event (Regional Bureau for Asia & the Pacfic and Food Security Markets and Vulnerability Analysis Unit, 2015). For this study, this discrepancy does not reduce the accuracy or invalidate the risk assessment methodology as there is a logical reason for its occurrence. In the future research, the results 500 should be validated with further 'ground truth' investigation of what drought risk conditions were like in PNG throughout 2014 through surveys or interviews with local PNG people.

Increasing resilience through risk assessment and Integrated-Early Warning Systems
The combined results of this study demonstrate that the risk assessment methodology is valid; this novel methodology can be 505 recommended for use in the future to increase the disaster risk resilience of PNG communities and inform the risk knowledge component of an I-EWS for drought. The adverse impacts caused by drought events seriously threaten PNG provinces, and if resilience to such disasters is not increased in the future, heightened drought events under climate change are likely to decimate local communities (Kuleshov et al., 2020). An I-EWS like the one conceptualised by Bhardwaj et al. (2021a,b) would have the potential to efficiently inform community preparedness to drought events if implemented in PNG. 510 However, such a system would not be efficient without accurate risk knowledge. Thus, an accurate risk assessment methodology, such as the one developed in this study, could be vital for the development of an I-EWS for drought in PNG, as well as critical to informing proactive and suitable disaster risk management strategies in local PNG communities.

Study limitations
Although the risk assessment methodology was overall deemed accurate, this study was limited by several factors. The validation used literature sources discussing each drought period as the ground truth for what occurred during that time. A 525 more reliable ground-truth would have been the perspectives of local PNG people who personally experienced the drought conditions and ensuing impacts. Interviews could have been conducted like those executed by Mckenna and Yakam (2021) and Fragaszy et al. (2020). However, due to the COVID-19 situation in both PNG and Australia at the time of this study, interviews were not viable. Future research should consider interviewing local communities in each PNG province to determine a more robust ground truth of the conditions and effects of each drought event investigated. The validation method 530 was also constrained by the fact that there were limited numbers of scientifically robust literature sources reporting on the 2019-2020 drought event, as it was a recent event. The PNG National Weather Service was consulted to ensure that the results from the 2019-2020 literature sources were true and accurate.
Data was limited for the hazard indicator of VHI. Space-based VHI data is only available from 2014 onwards. Whereas the 535 SPI data record dates to 2001. To have a complete hazard index in the historical risk assessment, the historical period investigated had to begin from 2014. 2014-2020 is a shorter period of analysis, which limits the number of drought events and non-drought periods occurring within, resulting in lower confidence in results. A longer analysis would provide greater confidence in the risk assessment methodology.

540
Data availability was also limited for the exposure and vulnerability indicators, thus, the data available closest to the time investigated was used. This meant that the vulnerability and exposure indices were the same for both 2014 and 2015 as the data was not updated throughout those two years. However, as half the indicators in both the vulnerability and exposure are more static rather than dynamic (excluding agricultural occupation, key crop replacement cost, population density and access to safe drinking water), it is not expected that values would largely change on a yearly basis regardless, rather it would be 545 more likely for values to change every two or three years (Aitkenhead et al., 2021). Therefore, the limited data availability for vulnerability and exposure indicators in 2014-2015 will not likely have a large effect on the credibility of the results.
Data availability is constrained throughout many SIDS like PNG; investment in open-sourced and cloud-based data platforms would allow for collaboration between separate entities that have collected data so that all relevant data can be 550 combined, stored, and accessed from the same place (Sun et al., 2020).

Further research
The risk assessment methodology developed in this research is novel; it combined the most efficient approaches of past risk assessment investigations to formulate a holistic, accurate and tailored risk assessment methodology to effectively improve risk knowledge in Pacific SIDS. This methodology provides the foundation for further research regarding disaster risk 555 management and the implementation of an I-EWS for drought in SIDS like PNG. Future research on the communication of risk assessment results to local communities is required to ensure that the risk assessment results are user centered.
Additionally, further work is needed to integrate the risk assessment with the I-EWS being developed as part of CREWS activities.

560
At a policy level, it would be intended that the risk assessment would come in at a higher level than the I-EWS, so that local decision makers are informed of their disaster risk to know what to look out for in the warnings given by the I-EWS and how to act in response to such warnings (e.g. prioritizing resources in the most at-risk provinces, planning water restrictions in certain areas to avoid critical water shortages, formation and implementation of disease prevention and management plans in the most at-risk regions, etc.). Ideally, a risk assessment platform communicating risk information to local decision-makers 565 and a user-centered I-EWS would be developed and used as 'side-by-side' products aimed at informing proactive and suitable management of natural hazards in local communities.

Conclusion
The occurrence of natural hazards is expected to be exacerbated under anthropogenic climate change, with the impacts of hazards predicted to critically affect agricultural productivity, food security, and general economic productivity, severely 570 reducing the financial and social health of local communities in Pacific SIDS. The novel drought risk assessment methodology demonstrated in this study was overall deemed valid, and thus can be recommended for use in future disaster risk management practices in vulnerable Pacific SIDS. A strong foundation for tailored and accurate disaster risk assessments has been developed, with future research required to further verify the accuracy of the methodology by comparing the results to local and expert perspectives. The development of this tailored and accurate disaster risk assessment 575 methodology is vital to improving risk knowledge for the development and implementation of an I-EWS and resilient disaster risk management strategies in vulnerable communities.

Appendix A
Indicator selection for the Hazard, Vulnerability and Exposure Indices which contributed to the drought risk assessment of 580 each drought event investigated is shown below.

Index
Indicator -Indicator for Agricultural Sector.
-Use in reliable past studies investigating and assessing the effects of drought within study areas with similar socioeconomic characteristics as PNG (Nasrollahi et al., 2018;Mainali and Pricope, 2019 (14-30)).
-Indicator for Environment and Agricultural Sector (considers agricultural drought). -Use in reliable past studies investigating and assessing the effects of drought within study areas with similar socioeconomic characteristics as PNG (Antwi et al., 2015;Ayantunde et al., 2015). -Data is available for recent years from PNG National Weather Service (NWS) and United Nations Development Programme (UNDP).

Exposure
Land use (type) -Indicator for Environment and Agricultural Sector.
-Use in reliable past studies investigating and assessing the effects of drought within study areas with similar socio-geographic characteristics as PNG (Rahmati et al., 2020;Shahid and Behrawan, 2008 -Use in reliable past studies investigating and assessing the effects of drought within study areas with similar socio-geographic characteristics as PNG (Han et al., 2015;Sun et al., 2020). -Data is available from open-sourced GIS platforms. Population density -Indicator for Social Sector as it is an indirect indicator for infrastructure and health service accessibility. -Use in reliable past studies investigating and assessing the effects of drought within study areas with similar socio-geographic characteristics as PNG (Nasrollahi et al., 2018;Pei et al., 2018). -Data is available for recent years from PNG National Statistical Office. Access to safe drinking water (% of population with access to safe drinking water) -Indicator for Social Sector and Households (also considers hydrological drought). -Use in reliable past studies investigating and assessing the effects of drought within study areas with similar socio-geographic characteristics as PNG (Limones et al., 2020;Frischen et al., 2020b). -Data is available for recent years from PNG National Statistical Office.

Appendix C
An expert weighting scheme for the relative hazard, vulnerability and exposure indicators was developed, based on the relative importance and contribution of each factor for the specific index which it informs. This weighting scheme was developed on a 0-1 scale, with 0 indicating no probable contribution to the relative index and 1 being total probable contribution to the relative index (Frischen et al., 2020a;Dayal et al., 2018). The numerical weightings assigned to each 600 indicator were determined by investigating expert weights provided in earlier studies as well as seeking advice from PNG NWS. The weights assigned to each Hazard, Vulnerability and Exposure indicator are shown below.

Competing Interests 610
The authors declare no conflict of interest.

Author contribution
I.A. was lead for conceptualisation, methodology, software, validation, formal analysis, writing-original draft preparation and review and editing, and visualisation. Y.K. contributed to conceptualisation, methodology, writing-review and editing, research supervision, and funding acquisition. J.B. and Z-W.C. aided in formal analysis and writing-review and editing. C.S. 615 and S.C. contributed to writing-review and editing and supervision. All authors have read and agreed to the published version of the manuscript.

Acknowledgements
This research was supported by the Climate Risk and Early Warning Systems (CREWS) international initiative and the World Meteorological Organization (WMO) through "Weather and Climate Early Warning System for Papua New Guinea" 620 (CREWS-PNG) project.

825
6 As there is limited data for direct indicators of accessibility in terms of road accessibility and health service accessibility, population density has been used as an indirect indicator for accessibility as it is associated with the accessibility level for each province; provinces with low population densities have more rural communities which are expected to have reduced accessibility to infrastructure (e.g. roads) and health services compared to urban communities.

Exposure
Land use (type) Elevation (